Advanced ReviewSteam power plant configuration, design, and control Xiao Wu,1 Jiong Shen,1 Yiguo Li1 and Kwang Y. Lee2∗ This article provides an overview of fossil-fuel power plant (FFPP) configura- tion, design and especially, the control technology, both the conventional and the advanced technologies. First, a brief introduction of FFPP fundamentals and con- figurations are presented, followed by the description of conventional PID-based control system in the FFPPs and its short-comings. As the major part of this writing, different advanced control strategies and applications are reported, with their sig- nificant features outlined and discussed. These new technologies are collected from both the academic studies and industrial practices, which can improve the perfor- mance of the FFPP control system for more economic and safe plant operation. The final section presents a view of the next generation FFPP control technologies, emphasizing potential business and research opportunities. © 2015 John Wiley & Sons, Ltd. How to cite this article: WIREs Energy Environ 2015. doi: 10.1002/wene.161 INTRODUCTION will not change in a foreseeable future. For this reason, developing and operating the FFPPs according to the F ossil fuelled power plant (FFPP) refers to a group of power generation devices that convert the chem- ical energy stored in the fossil fuel such as coal, gas, most suitable available technologies are very impor- tant and should be the most effective and direct way to save energy and reduce the pollution. oil into thermal energy, mechanical energy and finally The history of FFPP can be traced back to the late electrical energy. In the past hundred years, FFPPs are 19th century, the simple D.C. generators were coupled the most widely used facilities in the power industry to coal-fired, reciprocating piston steam engines, pro- and play a fundamental role in social production and ducing electricity primarily for district lighting. These life. According to the 2013 Key World Energy Statis- initial plants typically operated at low temperature tics published by the International Energy Agency and pressure conditions (150∘ C, 0.9 Mpa) and could (IEA), in 2011, the annual generation of electric- only generate 30 kw electricity. Through a century’s ity from all types of sources was 22,126 TWh and technological developments, power plants have now FFPPs provided 15,054 TWh, accounted for 68% of been evolved into a highly complex system that can the total electricity generation. Although the rapid operate at supercritical conditions of 28.5 Mpa and increase of global energy crisis, combined with the concerns about environment issues has led to an exten- 600∘ C, generating 1300 MW of electricity with much sive promotion of nuclear and renewable energy, for higher efficiency. Although there are many variations the most parts of the world, the trend of conven- in power plant configuration and design, the basic tional fossil-fuel-dominated electric power generation working principle of the FFPPs keeps the same: fossil fuel is combusted, generating high pressure and tem- perature steam, which is then expanded to rotate a tur- ∗ Correspondence to:
[email protected] bine, and drives the generator to produce electricity.1 1 Key Laboratory of Energy Thermal Conversion and Control For the FFPP, the main task of the control of Ministry of Education, School of Energy and Environment, Southeast University, Nanjing, China system is to regulate the electrical power output 2 Department of Electrical and Computer Engineering, Baylor Uni- to meet the demand of the grid while maintain- versity, Waco, TX, USA ing the main thermal dynamical variables such as Conflict of interest: The authors have declared no conflicts of interest superheater/reheater steam temperature, throttle pres- for this article. sure, furnace pressure, drum water level, within © 2015 John Wiley & Sons, Ltd. Advanced Review wires.wiley.com/wene given tolerances to keep the power plant operating information reported here are collected from both aca- safely. Generally, such an objective is achieved via demic researches and engineering practices. the multi-loop proportional-integral-derivative (PID) A brief introduction of FFPP fundamentals and based controllers. The approach has been proven to configurations are presented first, followed by the be highly reliable and can attain a satisfactory per- description of conventional PID-based control system formance under normal operation maintained at base in the FFPPs and the associated problems. As a load, where plant characteristics become almost con- major part of this writing, different advanced control stant. strategies and applications are reported, with their However, during the past few decades, power significant features outlined and discussed. The final industry has undergone some significant changes, section presents a view of the next generation FFPP and as the primary devices of power production, control technologies, emphasizing potential business FFPPs have been endowed with higher operational and research opportunities. requirements: 1. The growth of electric power demand increases PLANT CONFIGURATION AND DESIGN the magnitude of the cyclic variation of the grid The essence of power production process in all types of load, and the renewable sources, such as wind, the FFPPs is energy conversion. In the vast majority of solar and hydro power, are severely influenced the FFPPs worldwide, water/steam is commonly used by the season and the weather condition; thus, as the working fluid, which is alternately vaporized the FFPPs have to participate in the grid power and condensed in a closed circuit following a thermal regulation frequently and respond to the load dynamic cycle. Within this cycle, the chemical energy demand variation quickly within a wide oper- of the fossil fuel is transformed into steam thermal ation range. energy by the boiler, then it is transformed into 2. The privatization and deregulation of the rotational mechanical energy by the turbine, and electricity industry has changed the power gen- finally it is transformed into electric energy by the eration business from a cost-plus, monopoly generator. This kind of FFPPs is also called steam environment with an obligation to serve, to power plant, and depending on the operating steam a competitive environment for the sale of its pressure, it can be classified into subcritical plants and product. For this reason, power plants are supercritical plants. increasing in size and becoming more complex in order to achieve high efficiency and the scale Subcritical Steam Plant of economy. Furthermore, the aforementioned In subcritical power plants, the steam parame- thermal parameters should be more stringently ters never exceed the critical point: 22.115 Mpa, controlled so that the plant can operate in an 374.12∘ C. Because under this critical point, liquid optimal mode at all times. water must go through a vaporization stage to become steam; in most of the large-scale subcritical plants, Therefore, control problems to deal with issues, drums are typically utilized to separate the steam out such as nonlinearity over a wide operation range, large of the boilers. inertial and time varying behavior, and strong cou- Figure 1 provides a simplified illustration of a pling among the multitude of variables, become severe coal-fired subcritical power plant, which is comprised in the FFPPs. Consequently, the conventional PI/PID of two basic systems: the fuel/air-flue gas system and based controllers are no longer sufficient in meeting the water-steam system.2–4 performance specifications, even if they are well tuned The fuel/air-flue gas system is also called the fire- at a given load level. On the other hand, with the help side of the plant. In this system, the raw coal is trans- of modern computer and instrumentation techniques, ported to the coal hopper by the conveyor and enters utilizing Distributed Control System (DCS) is now the the pulverizing mill; where grinding and crushing take routine rather than the exception, which makes the place. The qualified (smaller and lighter) coal parti- implementation of advanced controllers possible in cles are then separated and entrained in the air flow, the FFPPs. The primary purpose of this writing is to and carried into the burner. Finally, the combustion present an updated, representative snapshot of vari- occurs in the furnace, generating high temperature ous control strategies that are being applied to the (above 1000∘ C) flue gases. The air needed for com- FFPPs and describe how they can help in improving bustion is delivered to the furnace and mill by the the quality and performance of plant operation. The forced draught fans and an air preheater is installed in © 2015 John Wiley & Sons, Ltd. After leaving the superheater. are suspended on the horizontal passage at the top To improve the plant efficiency. Feed heater 20. Condenser 15. Cooling water pump 9. others are convection type.wikipedia. The induced draught fans work in conjunc. Finally. Boiler drum 24. Depending on the installed positions.org/wiki/Thermal_power_station#Boiler_furnace_ and_steam_drum). and finally out of the boiler to a given high speed (3000/3600 rpm). Economizer 3 3. which denser and delivered to the boiler by the feed pumps. The objective of this process is converting Water-steam system is also referred to as the the chemical energy in the fuel into the thermal energy waterside of the plant. The drum supplies the feedwater the steam within. the path to warm the air being fed utilizing the heat through the chimney. to reach higher temperature and pressure. The steam is then to reheat the steam or prewarm the feed-water and further heated through multiple stages of superheaters feed-air. which operates following the in the flue gases. Superheater 26. collected in the ash hopper and delivered to the ash through which the efficiency of the combustion can be system of the plant. Forced draught fan 27.2–4 improved. heat from fluid. Rankine cycle. Precipitator 5. Reheater FIGURE 1 | Simplified coal-fired subcritical power plant (Picture from http://en. Air intake 2. Pulverized fuel mill 23. high pressure feed heaters and an economizer are uti- some superheaters are radiant type. a series of low and of the furnace. Pylon (termination tower) 10. Steam governor 17. Deaerator 19. Coal hopper 22. Cooling tower 8. Boiler feed pump 14. The flue gases in the furnace trans. Air preheater 4 4. economizer and air preheater. design. Intermediate pressure turbine 16. . which absorb heat lized to heat the feedwater with the steam bled from by radiation. The falling slags and ashes are remaining in the exhaust gases leaving the furnace. Induced draught fan 6. Low pressure turbine 13. the extreme heat in the the dissolved gases in the feedwater by vigorously boil- flue gases is transfered to the superheater piping and ing and agitating it. the flue to the waterwall of the furnace to absorb the radiation gases pass over reheater. Through either type. then pull the flue steam expands along the turbines and rotates them gases into the precipitator. Coal conveyor 21. Chimney stack 7. Ltd. High pressure turbine 18. which then © 2015 John Wiley & Sons. WIREs Energy and Environment Steam power plant configuration. Ash hopper 25. Generator 12. some are a combination of the two The deaerator is also installed in this path to remove types. The procedure within this system fer the heat to the water wall by radiation and then begins with the feedwater being drawn from the con- flow through multiple stages of superheaters. and control 17 19 21 22 14 1 23 5 6 9 10 25 15 27 26 8 11 24 12 13 16 18 20 2 7 Key 1. absorbing the turbine and the remaining heat of the flue gases. the tion with the forced draught fans. heat and separate the resulting saturated steam from where almost all of their remaining heat is extracted the incoming saturated feedwater. Unit transformer 11. en_US). the steam generator operates at pressure greater than the critical in the ultra-supercritical power plants is expected to point. where the steam ation initiative: Vision 21. which will enhance the physical turbulence that characterizes boiling ceases plant efficiency to more than 50%.115 Mpa. by the year 2020. and then delivered to the medium and low pres. drive a generator to produce electricity. According to the USA DOE power gener- is another type of steam plant. The saturated steam leaving the low sumed and the emissions generated.2 For this reason. and for a higher plant supercritical/ultra-supercritical plant is now greatly efficiency. Figure 3).5 Mpa. evaluating. Advanced Review wires. compared with subcritical plants’ In contrast to the subcritical plant. there are still barriers for building this temperature (374. the reach 760∘ C and 38. 22. Because above such a pressure. especially the steam the main choice in some countries due to its sim. supercritical plant 33%–39%. Thus an identifying. potentially lowering the amount of fossil fuel con- sure turbines. thus a well-designed control system is © 2015 John Wiley & Sons. power plants. are complex. there are more than 600 supercritical condenser. multivariable. because operating the plant at higher turbine. where the evaporation separation cracking in the critical sections of the plants as well as process occurs can be completely eliminated. the FFPPs. However.com/Industries/Power-Generation/Conventional-Steam/Flowserve-Products-Used-in-Supercritical-Units. boiler.4 Current Status of the Steam Power Plant CLASSICAL CONTROL OF THE FFPP The subcritical plant is still expected to remain As stated previously. and the the resulting lower reliability and higher maintenance feedwater circulates only once in the furnace in each costs. Therefore.2–4 and ultra-supercritical power plants with total capac- ity above 400 GW in the world (status 2010. and interac- plicity in operation and control. . after the expansion in the high pressure promoted. Ltd. supercritical plant.flowserve. the ‘once through potential alloy material is the major challenge for the steam generator’ is designed and employed in all successful implementation of supercritical technology. the liquid water immediately In spite of the great advantages of the super- becomes steam once is heated above the critical critical plants. to occur. Usually.com/wene Valve Steam turbine Steam turbine Generator Attemperator Main steam valve Cooling water Flue gas reture desulpherization Reheater Condenser (FGD) unit Reheat stop Boiler valve Condensate Condenser Superheater pump cooling water Cooling tower pump Carbon Feed Feed water capture water heater pump Feed water Feed water Deaerator heater booster pump FIGURE 2 | Simplified supercritical power plant (Picture from http://www. belief in higher tive processes. the drum used in type of the plant: the high thermal stresses and fatigue the subcritical plant. the are multiple stages of turbines.wiley. there reliability and low technical risk.12∘ C). Supercritical Steam Plant These supercritical plants can achieve efficiencies of more than 42%. pressure turbine is condensed back into liquid in the Currently. and instead. and qualifying cycle (Figure 2). the steam is extracted and reheated in the temperature and pressure can increase its efficiency. . so that high efficiency. WIREs Energy and Environment Steam power plant configuration. it should be most layer of the control system. i. the throttle pressure naturally represents a (subcritical plant). nevertheless. The advantage of this approach is a fast dinated control system (CCS) constitutes the upper.1.5. typ. and feedwater control system. thus gle entity. the Boiler-Turbine Coordinated Control System boiler awaits the actions of the turbine to match Current plant or unit control strategies allow gener. Exter- load demand and controlling relevant process vari. are two possible modes for coordinated control: coor- tion control system. Picture from http://www. design. to the changes in steam flow and pressure.e. Therefore.7.iea. Ltd. balance between the boiler’s energy supply and tur- ity and safety can be attained in the plant. currently. boiler following schemes were the first to be used. Then. the requested generation. nally. bine’s energy need. given tolerance. and it is responsible noted that such rapid response is basically achieved for driving the boiler-turbine-generator set as a sin. The basic put to the energy required by the turbine-generator to principle of the coordinated BF mode is illustrated match the unit load demand at all times. they match the boiler steam flow energy out. Evolved from multiple single-input der of this section will focus mainly on the con. pressure control since the boiler has a tendency to © 2015 John Wiley & Sons. single-output control loop (decentralized) configura- ventional boiler-turbine coordinated control system tions based on PID control algorithms. harmonizing the slow response of the boiler it is effective only for a small demand change. in dinated boiler-following (BF) mode and coordinated which the respective thermal dynamic variables are turbine-following (TF) mode. response to load changes. a cascade of PI/PID controllers based on ulate the power output to meet the demand of the single-input single-output control loops is designed grid while maintaining the throttle pressure within a in the plant to fulfill such tasks. drum water level internally. and control 800 700 600 Capacity (GW) 500 400 300 200 100 0 Australia Germany Japan Korea Russia South Africa United Kingdom 2010 2012 2014 2010 2012 2014 2010 2012 2014 China India United States Ultra-supercritical Supercritical Subcritical FIGURE 3 | Capacity of supercritical and ultra-supercritical plant in major countries (refers to capacity in 2010 unless specified otherwise.org).6 The coor. superheater/reheater plant’s power generation and grid’s power demand. The with the faster response of the turbine. the central task of the CCS is to reg- ically. required in the plants to ensure the correct operation For the FFPP. .7 The remain. the power demand..8 controlled separately. the boiler controls respond Mainly. durabil. The dominant behavior of the unit The boiler-turbine unit control schemes have is governed through the power and pressure control gone through several decades of evolution and. power output and throttle pres- of the entire process. steam temperature control system. in Figure 4. the power output reflects a balance between the ables such as: throttle pressure. to achieve disadvantage of the coordinated BF mode is that. furnace pressure. in its fast and stable unit response during load tracking pure form. rapidly following the grid sure are the two most important variables. The turbine control valves ation of the grid load demand while maintaining the regulate the steam flow into the turbine in terms of balance among the process variables within the unit.6.9 In boiler following mode. steam temperature. Historically. there (CCS). loops. etc. this approach shows a less stable throttle maneuvers and load disturbances. by using the stored thermal energy in the plant. combus. thus ing the efficient and safe operation of the boiler. Then the turbine controls respond This will be introduced in the next section. This coordinated control scheme is now widely boiler. P 0 : main steam pressure set-point. The power demand For this reason. used in practice. the PI/ PID control systems. so that the large inertial behavior rate. E : power output. various advanced modeling and is used by the combustion control at the boiler to multivariable control technologies are studied. However. which has a relatively quick response compared to the coal plant. which The regulation of the above three variable can are based on a cascade of separate SISO loops. it requirements are fulfilled by controlling relevant vari- is mainly used for a large base-load plant or a gas-fired ables in the plant. the turbine follows the actions of the boiler to output and throttle pressure. The main Under the CCS. the power demand is fed both to the boiler system (BF) (b) Excess air coefficient or optimal oxygen content and turbine system (TF)9 (this is shown as dotted-line in the flue gases to ensure appropriate air flow in Figures 4 and 5). Ltd.10 multi-output (MIMO) power plant. BD: boiler demand.com/wene Boiler main Turbine main control control BD + TD E0 – E Turbine sub- Boiler sub- control control system system + P 0 – PT ~ Boiler Governor valve Generator μT Turbine Fire rate μB FIGURE 4 | Working Principle of the Coordinated BF mode (E 0 : load set-point.wiley. The advantage of this approach is its very stable response to load changes Combustion Control System with minimal steam pressure fluctuations. TD: turbine demand). PT : main steam pressure output. mode). forced draught (FD) dampers and induced draught © 2015 John Wiley & Sons. match the requested generation. by adjusting the throttle valve openings to keep the pressure at the setpoint value. overshoot because it requires some time to match the different process variables in the nonlinear multi-input turbine. steam production. namely: plant. be achieved through the manipulation of fuel feed- cannot fully account for the interactions among the ers. Therefore. the mission of the combustion system disadvantage is that this approach does not make use is to provide enough thermal energy while guarantee- of the energy storage capability of the boiler. . Advanced Review wires. it is Turbine following began around the late 1960s still very difficult for the classical CCS to achieve and early 1970s. to improve the BF mode) or electrical power output (in TF performance of the CCS in both BF and TF modes. (a) Fuel flow rate to maintain throttle pressure (in It is worth mentioning that. of boiler can be partly compensated (in BF) and the turbine’s ability to respond quickly can be utilized (in (c) Furnace pressure to guarantee the safety of the TF). Such producing a rather slow response.9 In the coordinated TF mode of a satisfactory control performance in both power control.10 For this reason. aiming adjust the fuel and air into the furnace to modify the to realize a real coordinated control of boiler-turbine. being supplied to the furnace. in con- (ID) dampers. Because the fuel flow rate. BD: boiler demand.3. (1) where. it is direct to design a ratio energy can exactly match the load demand. Ltd. P 0 : main steam pressure set-point. and Pb is the drum Fuel Air Pressure regulator regulator regulator pressure representing the heat stored in the boiler. resulting in three indivisi.5 keeping the fuel/air ratio at the stoichiometric value.e. regardless of value of the fuel. and pressure regulator as shown in Figure 6. Cb is the heat storing coefficient of the boiler. E : power output.. respectively. An undersupply of the air will prevent the fuel from complete burning. in most of the FFPPs. and the feeder oxygen content in the flue gas is measured which can speed signal cannot reflect the variation of the heating reflect the combustion condition directly. © 2015 John Wiley & Sons. However. guarantee a suitable ratio between the amounts of fuel and air FIGURE 6 | Combustion control system (BD: boiler demand). is difficult to measure. the heat the fuel variation. especially for to set the ratio co-efficient in operation. flow rate. quantity signal M is used in the inner loop to rapidly Boiler main control deal with any disturbances due to the variation of the Coordination level fuel in heating values. therefore. the fuel regulator. In practice. WIREs Energy and Environment Steam power plant configuration. design. i. The air regulator is used to guarantee the effi- ciency of the combustion. PT : main steam pressure output. A cascade controller to keep the air flow rate matching the fuel control strategy is usually employed in this system. TD: turbine demand). – control control system system PT ~ Boiler Governor valve Generator μT Turbine Fire rate μB FIGURE 5 | Working Principle of the Coordinated TF mode (E 0 : load set-point. . the the coal flow. The task of the fuel regulator is to provide Since the fuel flow rate has already been deter- enough fuel to the boiler so that the generated thermal mined by the fuel regulator. it is a challenge mode plant. D is the steam flow rate representing the heat absorbed by the working substance. trast an oversupply of the air will absorb heat and thus ble sub-regulators. and control BD Boiler main control TD Turbine main control + E0 + P0 – E Turbine sub- Boiler sub. considering that the heating value which is shown in the left part of Figure 7 for a BF of the fuel can vary from time to time. which a certain amount of excess air is needed rather than form the combustion control system. air regulator increase the heat waste in the exhaust gas. to be specific. The heat quantity signal can be BD estimated through the equation M = D + Cb dpb ∕dt. Both can steam temperature/enthalpy out of the separator can result in catastrophic failures. tem. The steam flow rate signal D is used as furnace pressure is required to be maintained at the feedforward signal. heat quantity generated from and compare it with the optimal oxygen content value combustion and many other variables. feedwater circulates maintaining the drum level naturally represents the only once in the furnace in each cycle and there is no balance between the feedwater supply and steam clear disengagement surface between steam and water. plant: a high level will increase the risk of water being the steam temperature in the superheater would also carried over into the steam circle. Such a task feedwater flow rate respond quickly to the variation of is achieved by the use of feedforward-feedback the steam flow rate. a three-element cascade con- pressure. be controlled within the given range. if tolerance is important for the safe operation of the such a surface deviates far away from a designed level. the fuel signal and air signal to attain a basic fuel/air ratio drum water level can be influenced by feedwater flow and the outer loop receives the oxygen content signal rate. For the subcritical plant. Ltd. For the supercritical plant. There- only lead to a fluctuation of steam temperature. Pss : furnace pressure set-point.3.wiley. Thus. V: air flow rate. the inner-loop of the control system and a secondary controller is designed for a quick rejection of the dis- turbance inside the feedwater system. Fuel regulator Air regulator Pressure regulator A classic air regulator using the oxygen content The drum water level is determined by both the signal is illustrated in the middle part of Figure 7. a ratio controller is designed for the also cause fouling and damage of the superheaters. which may not have a deviation far away from the set-point. behavior of these disturbances and a ‘swell and shrink’ The task of the pressure regulator is to main.3 tain the furnace pressure close to the atmospheric For these reasons. so that the to be damaged from insufficient cooling. Ps : furnace pressure output. supercritical plant to regulate the feedwater flow rate. conversely. Advanced Review wires. generation. but fore. O2 : oxygen content in the flue gas. such a design can make the 20–50 Pa below the atmospheric pressure. so that the hazardous gases escaping and troller is typically used in the plant. The inner loop receives the the steam bubbles under the water level. . generally.5 enough water to the boiler to match the evaporation The main difference between a supercritical rate. steam flow rate. © 2015 John Wiley & Sons. BD: boiler demand). a low level will cause the waterwall piping keeping it matching the fuel flow rate. D: main steam flow rate. However. The feedwater flow rate signal is used to form Figure 7. which is illustrated cool air entering the boiler can be prevented. PT : main steam pressure output. thus avoid the effect of ‘swell and control system as shown in the right part of shrink’. both the fuel flow rate and feedwater flow Controlling the drum water level within a given rate can greatly influence the position of the surface. in Figure 8. because the separation plant and subcritical plant is at the water-steam sys- of steam from water always happens at the drum. and its control (which is calculated offline for a given load) to achieve presents a complex problem due to the large inertia a tight control of the air input. It is volume of the water in the drum and the volume of a cascade control system. effect. Usually. The drum water Feedwater Control System level H is finally fed back to the primary controller for The objective of the feedwater control is to supply an accurate regulation.com/wene O2 D pT p0 f1(x) – + – + PI1 PI5 BD ps pss f2(x) × V M + – + – + – + FIGURE 7 | Three sub regulators of boiler f3(x) combustion control system (P 0 : main steam pressure PI2 PI3 PI4 set-point. 5 1. the steam flow rate has to match the load demand. A classical SST control system is illustrated in Steam Temperature Control System Figure 10. design. moreover. For the supercritical plant. © 2015 John Wiley & Sons. during the operation.3. thus threatening the the flow of the flue gases across the reheater tube safety of the plant. as analyzed in section rial damage on the superheater/reheater steam 3. bance in the spray water. Therefore. heat transfer from flue gas and the rate of spray water flow. Δp the regulation of heat transfers (for example. which receives the economy reasons: final steam temperature signal T1 to achieve an accu- rate control performance. there are many factors expands in the high pressure turbine and will reduce that will influence the SST: mostly the rate of the the efficiency of the whole plant. The first two have a relatively KZ quick influence on the SST while the spray water’s influence is quite slow. the FIGURE 8 | Three elements cascade feedwater control system for a burner tilt or the gas recirculation) will influence the √ efficiency and security of combustion. thermal stress of the piping material and mag. the drum type boiler (Δp is the differential pressure transmitter. employed. Kz is the actuator of the the SST and the cascade control system is generally feedwater flow control valve). Excessively high temperature will lead to mate. Lower temperature will reduce the efficiency of is mainly controlled by regulating the fuel/feedwater the plant. Large temperature variation will increase the temperature. banks. aW steam flow. for the following safety and much faster than the outer loop. spray water is then used for a tighter control of the superheater steam 3. very limited. bances originating upstream as well as the self distur- They must be tightly controlled within a small range. the regulation capability for the spray water is pipes at the inlet of the turbine. The inner loop receives the steam temper- Superheater steam temperature (SST) and Reheater ature signal T2 immediately after the attemperation. as shown in Figure 9. because it will reduce the amount of steam which For the subcritical plant. . and control Super heater Steam flow D Dangerous operation region T_sp+10 Drum Drum water level H Δp Temporary operation region T_sp+5 Long term operation region T_sp Long term operation region Δp aD Economizer T_sp-5 Temporary operation region T_sp-10 PID1 Low efficiency operation region FIGURE 9 | Operation regions of superheater steam temperature PID2 (T_sp is the temperature set-point and the numbers represent temperature deviations in Celsius degree). of course. Ltd.2. is the square root extractor. and the superheater steam temperature 2. The spray water is only used in emergency case. it will build up the steam ratio. WIREs Energy and Environment Steam power plant configuration. However. The inner loop is. steam temperature (RST) are two of the most critical which is required to reject the temperature distur- variables to be controlled in a steam power plant. aD are the sensitive coefficients of feedwater spray water flow becomes the only variable to control flow rate and steam flow rate signals. The steam temperature/enthalpy out of the humidity in the rear of the low pressure turbine separator is first controlled by keeping the feedwater that would erode the turbine blades. aw . The control of the reheater steam temperature is nify the variations of the air gap between rotor mainly attained by regulating the dampers that control and stator of the turbine. and flow rate matching the fuel flow rate. which uses an inner loop to handle the large inertia property. com/wene Attemperator when the controller parameters are well tuned. PI2 PI1 Therefore. auto-tuning inertia behavior. The most common method of tuning these controllers in the FFPPs is the so (a) The main drawback of the PI/PID control called ‘trial and error’ method. Ltd. fuel variation. the PI/PID controller parameters are attain a satisfactory overall performance. envi. Such a which implement the state-of-the-art design or tun- design has also been proved for its value for regulation ing technologies on the basis of conventional PI/PID and disturbance rejection during normal operation control loops. . all controller © 2015 John Wiley & Sons. T2 is the steam temperature signal immediately after the attemperation. loops. the conventional PI/PID strategies Auto-Tuning PI/PID Control are no longer sufficient in meeting performance As stated before. where an objective function reflecting an overall dle the constraints of manipulated variables dynamic control performance is utilized and through in the controller calculation stage. when wide-range load follow.wiley. expertise and experience and may still be difficult to (b) In general. A migration from classical Attemperator PI/PID based control system to new concepts based on valve advanced control techniques in FFPPs will take place in a foreseeable future. An alter- optimized at a given operating point and then native method is to design an analytical compensa- fixed. etc. as the tem has already been widely accepted and employed FFPPs increase in size and participate in grid load throughout the FFPPs. the control loops sequentially. will result in sively studied methods in power plants. This may also cause Steam γT2 γT1 the integral windup when a sharp change of the power demand occurs. 𝛾 T1 and 𝛾 T2 are the temperature transmitters. dle the interactions among process variables and (c) The PI/PID controllers are not possible to han. is that they do one with least interaction. due to severe non. tion of control-loop interactions for wide-range plant ing is required for the FFPPs. significant disturbances and plant behavior variations. tight operating constraints and large plant engineers and operators.. For complex MIMO sys- not account for the interactions of the different tem like FFPP. steam temperature. the power plant control system con- specifications because of the following drawbacks: sists of many SISO PI/PID loops which are strongly interacting to each other. control of FFPP unit has effectively improve the operation performance with- been shown to be a challenge. various advanced control strategies have been proposed in both academic studies and industrial practices. thus even minimization of this objective function.11 It generally tunes systems. Recently. Kz is the actuator of the attemperator valve). the performance operation. aiming at improving the per- KZ formance of the FFPP control system for economic and safe plant operation. such a method can directly and regulation more frequently. the equipment wear. However. The next section introduces four types of different control strategies with diverse FIGURE 10 | Classical cascade SST control in the FFPP (T1 is final applications in the FFPPs. beginning from the single-output (SISO) loops. Advanced Review wires. Moreover.12 of the plant operation is decreased because the Auto-tuning PI/PID control can potentially han- nonlinearity becomes significant. operation linearity in the multitude of variables over a wide procedures and concepts which are well understood by operation range. such a method requires considerable thermal properties in the plant. Superheater I T2 Superheater II T1 the performance is still decreased when phys- ical limitations (both magnitude and rate) of the valves are involved. and gain scheduling PI/PID controls are two exten- ronmental changes. Consequently. Therefore. out altering the original simple structure. Because the PI/PID based control sys- maintained at/around a base load. based on separate single-input. ADVANCED CONTROL OF THE FFPP The Drawbacks of the Conventional Control Advanced PI/PID Control The conventional PI/PID based control system has The advanced PI/PID control refers a class of methods been successful in FFPPs and are very reliable. which could include time increment. nonlinear analytical eter candidates randomly in the given solution or identified model must be employed due to the space. and control PID controller FFPP Reference Plant output Update Objective Model function Candidate controller parameters Run Search mode mode Heuristic search FIGURE 11 | Block diagram of a typical tuning method for PI/PID parameters. develop. Evaluation: simulate the power plant model sidered essential to this method is the objective func. Consequently. Ltd. accuracy and complexity of the model. Model plays a fundamental Algorithm (GA). However. thus the optimiza- optimally. it is worth mentioning here that mulated into a quadratic programming (QP) prob. otherwise. design. 3. However. the plants over wide operation range is difficult to minimization of the objective function can be for. Therefore.14 Evolutionary Programming (EP). which brings difficulties in mak- where k is the current instant. which is computationally efficient. recently. nonlinear behavior of the plant. model which can capture the dynamics of the power If linear model is employed in the method. Modification: modify the parameters and con- tion. model-free direct tuning methods using the lem. nonlinear models are utilized to power plant to achieve an optimal tuning of the © 2015 John Wiley & Sons. r represents the corresponding Another drawback for this auto-tuning method references to these output variables. another feature that is con. tion procedure becomes a non-convex problem with Figure 11 shows the block diagram of a typical heavy computational load. and Q is the is the model-dependence character. such as Genetic of controller parameters. function over a finite horizon N is commonly used: The results in works13–15 show that. An accurate weighting matrix. the PI/PID model output vector at instant k + t based on the controller gains do not have to be updated for each candidate controller parameters. 2. Initialization: generate initial controller param- a good performance. A basic work- structure. tinue to evaluate the performance until finding The following quadratic output performance objective the satisfactory parameters. with these parameters and evaluate the corre- tion. and may get stuck in a local tuning method in Refs 13–15. in which a dynamic minimum far from the optimal value. WIREs Energy and Environment Steam power plant configuration. Any output or input variables of interest in the sponding objective function. If ing principle for these heuristic search techniques is: an operation around a given loading condition is considered.15 pressure. . ŷk + t is the expected ing frequent online updates. the nonlinear optimization is still time consuming. Fortunately. and various types of functions can be designed.15 are proposed to computational requirement depend greatly on the find the optimal controller parameters. such as power output. parameters can be determined simultaneously and better model the plant behavior.13 role in this method. Besides the model. model of the plant is used to test different groups various heuristic search algorithms. throttle sen to tune the gains. where the tuning result and Particle Swarm Optimization (PSO). thus a large window size can be cho- the interested variables. linear model may be enough to achieve 1. controller tuning can be included in the objective func. in closed-loop experimental data are employed in the most of the cases. quality solu- ∑( t=N )T ( ) Jk = ̂yk+t − r Q ̂yk+t − r (2) tions and fast convergences can be provided in many t=1 applications. and so on. For uncertainty. and secondly. the PID con- search based minimization of these objective functions troller parameters are tuned offline. but instead. basic idea of the H∞ control comes from the theory in The expert knowledge rule-table could be pro. according to the value of the scheduling variable. an H∞ approach with mixed is to change the controller parameters according to sensitivity has been utilized in the controller design.com/wene PI/PID controller parameters. In these works. Ltd. extremum seeking (ES) to adjust the controller parameters. One attractive tion. a small integral gain is desired changes. In Ref 11. to make the controlled the plant. environmental have large time-delay. feature for these model-free based approaches is that the PID parameters can be determined through some they potentially eliminate or reduce the modeling interpolations between the parameters predesigned effort. and a fuzzy FFPPs over the past few decades. When the output response is In contrast to the adaptive approach which near the set-point. technique is utilized to find the optimal PID gains of For the FFPPs. © 2015 John Wiley & Sons. the power load is usually selected the superheater steam temperature control system in as the scheduling variable since its variation nat- the FFPP. in Refs 22–25 is more practical to deal with the limitation of the H∞ control approaches are proposed to the fixed parameter PI/PID controller and attain a better boiler-turbine coordinated system. A measurable feedback tuning (IFT) technique is employed for a process variable. which is descriptive of the operating simultaneously tuning of six main control loops in condition. The last technique introduced in this part is the expert knowledge-based PI/PID auto-tuning Robust Control approach. iterative the variations of process dynamics. modeling mismatches are difficult example. but the gradient points are selected and at each point. Also. plant. fast rise time. developing a controller which could min- imize the H∞ norm as an objective function will natu- Gain Scheduling PI/PID Control rally reduce the closed-loop impact of a perturbation. which have to be a typical gain scheduling PID control is shown in designed carefully to make sure the operation of the Figure 12.20. plant is not under too much stress. no optimality can be achieved mum gain between bounded input energy and output since no optimization is performed in this approach.21 of the PI/PID parameters leads to the development To guarantee robust performance of the con- of gain-scheduling PI/PID control methodology. The computational complexity of the online-tuning and enhance the stabilization or performance. Advanced Review wires. and once finished. The H∞ norm of the transfer duced off-line. Then several typical loading be designed in these approaches. the PID parame. which requires a big proportional gain the unknown disturbances commonly exist all over and a small derivative gain. the proportional gain and integral attempts to learn the uncertainties of the plant and gain should be changed from large to small and from eventually adjusts the controller to be best suited for small to large. and so on. function (the maximum singular value of the func- ters can be determined directly and efficiently by the tion over the H∞ space) can be interpreted as a maxi- look-up table. which trollers under unexpected uncertainties. the robust control has a fixed structure output converge to the set-point quickly. The controller with fixed parameters. a number of experiments are at typical loading points. However. First. Because the robust controllers are rules are generated in Ref 17 to adjust the PID simpler to implement without online-tuning to the parameters according to the current output error plant variations. they have been extensively studied for e(k) and its first difference value Δe(k). which yields acceptable performance for a given plant On the above classical tuning knowledge. thus. auto-tuning PID controller is designed for the main The H∞ control is the subject of the largest share steam pressure loop in the FFPP. heating/cogeneration industrial boilers and gas tur- The essential idea of the gain scheduling control bine. urally represents various operating points of the Similar to the ordinary model based methods.wiley. a large to avoid due to the complex dynamics of the plant control signal is needed in the beginning to achieve a and the desire to use simplified model. where the expert knowledge is utilized to One major issue in the control of FFPPs is the determine and coordinate the controller gains. is known as a scheduling variable and used the power plant. fuzzy uncertainty set. the frequency domain. to reduce the overshoot. In Ref 16. energy. fuel variation. In online opera- is time consuming and lacks robustness. gas and oil-fired wide-range operation performance in the power plant. especially when the plant is operating under a different overall control objective functions can also ‘constant-pressure’ mode. respectively. to obtain a desired step response. which can enhance the robustness and control performance of the PID of the robust control researches in the FFPPs. because processes the plant due to the equipment wear.17–19 The block diagram of required for gradient computation. . R(s). d.. © 2015 John Wiley & Sons. e. ||T(s)||∞ reflects the allowed amplitude of Figure 13 shows the block diagram of a typical multiplicative uncertainties (I + Δ(s))G(s). where G(s) is the A generalized plant G (s) can be built as follows: transfer function of the plant. Ltd. disturbance input and system ⎢z2 ⎥ r ⎢ 0 W2 (s) ⎥ r output. H∞ mixed sensitivity problem. respectively. and y are the reference. the variables r. and T(s) for the control system design Here. z2 . design.. u. (6) W3 (s) G (s) ⎥ u weighting matrices. shaping the closed loop transfer functions S(s). W 1 (s). (3) constructed as: ⎡ W1 (s) S (s) ⎤ R (s) = K (s) (I + G (s) K (s))−1 = K (s) S (s) . (4) Φ (s) = ⎢W2 (s) K (s) S (s)⎥ . which can achieve a trade-off between the robustness reflects the allowed amplitude of additive uncertainties and performance. z2 . W 2 (s). and W 3 (s) are ⎢z ⎥ = G (s) u = ⎢ 0 ⎥ . z3 ]T can be S (s) = (I + G (s) K (s))−1 . (5) Then. Performance and robustness are characterized by various well-known closed-loop functions. in partic. K(s) is the controller. u = K (s) e. (7) ular the sensitivity function S(s). and a transfer matrix from the disturbance input d tion T(s): to the control system outputs z = [z1 . control input. and z1 . (8) ⎢ ⎥ ⎣ W3 (s) T (s) ⎦ T (s) = G (s) K (s) (I + G (s) K (s))−1 = I − S (s) . G(s) + Δ(s). z1 z2 z3 W1(s) W2(s) W3(s) d r e u y K(s) G(s) – FIGURE 13 | Block diagram of a typical H ∞ mixed sensitivity problem. . ||S(s)||∞ reflects the disturbance rejection is converted to find the controller K(s) which can and reference following ability of the system. the input sensitivity function R(s) and the complementary sensitivity func. PID_n Scheduling variable Controller parameters Driving Plant Reference error output PID FFPP Controller Control signal FIGURE 12 | Gain Scheduling PID control. control ⎡z1 ⎤ [ ] ⎡W1 (s) −W1 (s) G (s)⎤ [ ] error. WIREs Energy and Environment Steam power plant configuration. ||R(s)||∞ minimize ||Φ(s)||∞ . and control Gain scheduler PID_1 PID_2 . z3 are corresponding ⎢ 3⎥ ⎢ ⎣e⎦ ⎣ I −G (s) ⎦ weighted control system outputs. wiley. and employed in recent years. such as 𝜇-synthesis and Linear Matrix Inequality (LMI) technique are and H2 approaches through the simulation on a employed in these methods. there are some drawbacks which may input constraints during the controller design stage. inputs through the minimization of a dynamic objec- 3. and constraints and optimization. Lyapunov theory modern control design methods. the performance of the controllers will be degraded. the order. In Refs 30 and 31. The basic idea and principle of the MPC is A bumpless switchover mechanism is designed to shown in Figure 14. output disturbances and Although H∞ approach is the most popular sensor noise in the FFPPs. . the resulting controller has high tive function within the predictive horizon. and a robust tracking boiler-turbine coordinated system model. three steps: ating regions. thus order reduction has to be carried predictive control has progressed steadily and gradu- out without degradation in performance and ally made a significant impact on the FFPPs control. Relying on the linear model of the plant. In general. based on the available infor- controllers are designed for the boiler-turbine unit mation and predictive model. 1. Based on the response of the plant within the predictive © 2015 John Wiley & Sons. thus a wide range load following is attained. and can be briefly explained in achieve a smooth transition between the two oper. time-delay behavior of the plant. modeling uncertainty. which utilize an explicit process model to predict the ever. limit the application of H∞ design approach in Therefore in practice. H∞ controller is compared with other state-space type of local models. In Ref 29. two transfer function models 4. even if the robustness can be 2. The results show that. Other robust control attainable by other designs. The tuning of the parameters is easy and intu- are developed at different operating point of the itive. no special attention needs to be placed on boiler-turbine unit of a coal-fired power plant. Cannot deal with the input constraints effec. becoming insufficient.34 Origi- nally applied mainly in the petrochemical industry. Ltd. the H∞ loop shaping technique loop transfer recovery32 (LQG/LTR) and 𝜇-synthesis33 is utilized to design multi-variable robust PI/PID con. As the FFPPs are required to operate in a wide which frequently leads to more rapid response loading range. Advanced Review wires. none of them can effectively deal with the However. approaches such as linear quadratic Gaussian with In Refs 26–28. robust H∞ tracking 1. the setting and tuning of them is indirect future response of a plant and calculate the control and inconvenient. multi-model based H∞ 3. The weighting matrices W 1 (s). W 3 (s) are important to attain a balance among MPC refers to a class of control approaches various control performance objectives. W 2 (s). two H∞ controllers are built based on these models to guarantee the robustness in each local-region. the proposed controllers Although better control performance can be attained can achieve better performances than the conventional by the use of advanced PI/PID controller or robust PI/PIDs. the liner model based H∞ approach is and more profitable operation. It can effectively deal with the large inertial and approaches are proposed recently. It can effectively handle the actuator limitations and allow the operation closer to constraints. the H∞ controller can one in the area of FFPP robust control. when physical limitations of practice: valves in the FFPPs are involved. 1. are also applied to the FFPPs to enhance the system trollers for the FFPPs. predict the future via Takagi-Sugeno (T-S) fuzzy model.com/wene In Ref 22. In this context. Their test control of the boiler-turbine is achieved in a wide results show that under the presence of complicating operation range with the stability of the closed-loop factors such as coupling between multi-variables. It handles multivariable control problems natu- guaranteed within a certain range around the typical rally. time system being guaranteed. the model predictive control (MPC) has been widely 2. how. robustness. to the advantages of the H∞ design approach such as: yielding MIMO control laws. At current time k. delay. The main reasons for its success in FFPP studies and 4. Therefore. providing strong Model Predictive Control robustness of the system. controller. owning robustness and performance. applications are: tively (although tuning W 2 (s) can impose limi- tations on the inputs to some extent). operating point. it is by provide performance and robustness superior to that no means the only approach. the boiler-turbine. nonlinear plant model has been [ ( ) ( )] used instead of the linear model in the MPC design to ×Q ̂ y k + j|k − r k + j achieve a wide range plant operation. due to the fact that N the linear model works only for linear system. In Ref 35. uk+2 . … . compared with PID. drum pressure and water level.̂ y . (9) the nonlinear analytical model of the plant is employed as the predictive model. FIGURE 14 | Basic working principle of the MPC. smaller overshoot and shorter adjusting time. In Ref 43. design. . in which linear model for the plant is utilized and the optimal control inputs are determined Nonlinear MPC through minimizing a finite horizon quadratic perfor- Although the linear MPC can effectively improve the mance objective: control performance of FFPPs.{and express the future predic. and an MPC is problem with all input output constraints collected designed based on the model. the Linear MPC adaptive GPC is tested in a real power plant and The first generation of MPCs applied to the FFPPs is evident improvements were observed in the results. … . least square based adaptive algorithm is added to the GPC. Simulation results show © 2015 John Wiley & Sons. multi-variable { k+1 k+2 } GPCs are developed for a coordinated control of future inputs uk+1 . the GPCs can not only objective function. As Past Future QP is one of the simplest possible optimization prob- lems. … . and control into a matrix inequality involving the input vector. For this reason. uk+Nu . the results demonstrate that better control Predictive horizon Ny performance can be achieved as compared to the conventional PID control.34 Because the linear step response model of the rk + p plant can be easily obtained. modeling and computational requirement. its ( ) ∑y [ ( ) ( )]T application is limited to a small operating region of a J k = ̂ y k + j|k − r k + j j=1 plant. (GPC). In Refs 36 and k k+1 k+Ny 37. ∑ Nu To overcome the issues associated with nonlinear ( )T ( ) + Δu k + j − 1|k RΔu k + j − 1|k . where auto regressive moving-average model } tive outputs ̂ y . designed to regulate the superheater and reheater steam temperatures of a 200 MW power plant. u∗k+2 . the linear MPCs. neural network-auto regres- model and the objective function can be re-written in sive exogenous (NN-ARX) model is identified for a the form of a standard quadratic programming (QP) 200 MW oil-fired drum-boiler plant.38–41 In Refs 38 and 39. DMC ŷk+p is applied to a drum-type boiler-turbine system to yk–1 achieve a simultaneous control of electrical power. In Ref 42. recursive subsequent time. and GA is utilized to cal- The future projected output ŷ can be related culate the optimal control sequence under the input directly back to the input vector Δu through the linear constraints. In Ref 40. In Ref 41. the optimal inputs can be found efficiently. DMC techniques have been applied to control k+Nu the superheater/reheater steam temperature of the Control horizon Nu FFPPs. To 3. but u which can drive the predictive output ŷ to also has stronger robustness. u∗k+N . GPC is follow the reference r in an optimal way. Through minimizing a given dynamic control that. It shows that the uk–1 u*k+p step-response model based on the test data is better suited than the linearized model. their simulation results indicate 2. WIREs Energy and Environment Steam power plant configuration. dynamic matrix control (DMC) has been extensively studied. Send only the first input in the optimal sequence further enhance the robustness of the system and to the plant.̂ yk+Ny in terms of is employed. which can tune the controller parameters online using the real-time input/output data. Ltd. and repeat the entire calculation at attain a wide range operation of the plant. Another linear MPC that has been widely used in the FFPPs is the generalized predictive control horizon Ny . artificial j=1 intelligence techniques have been applied. calculate { the optimal future } track the target value smoothly and rapidly with control sequence u∗k+1 . which forces the states at the Multi-Model MPC end of the finite prediction horizon to lie within a The essential idea of the multi-model technique is prescribed terminal region. . however. coordinated control of a 500 MW plant.wiley. To improve the the FFPPs. To improve the computational Based on the Radial Basis Function (RBF) neural performance of the nonlinear optimization. multistep predictive controller. the multi-model techniques bring an ensure the stability.59 GA61 and 160 MW boiler-turbine unit. interesting to note that. an infinite-horizon objective function behavior of the plant. network and Adaptive Neuro-Fuzzy Inference System ticle swarm optimization (PSO) and its modifications (ANFIS) approaches. as shown in Figure 15. a The earliest multi-model MPC used in power disturbance observer and steady-state target calculator plant control is presented in Ref 51. GPC is then designed on and temperature and reheat steam temperature can the basis of the global model to achieve a wide range be attained during load-cycling and other severe plant control of the power plant. where networks of (SSTC) are added to the MPC structure to achieve dynamic local linear models are created after dividing a tracking control of the boiler-turbine unit even in the whole operating region into a number of zones and the case of significant unknown disturbances or plant the global model is built by using the interpolation parameter variations. a state-space type of local lin- trol of the nonlinear FFPPs. Based on affine (PWA) models or Takagi-Sugeno (T-S) fuzzy the technique of input-output feedback linearization. a recurrent neural network (DRNN) models have been fuzzy interpolated DMC is proposed in Ref 54.51–66 linear matrix inequalities (LMIs). 66. Advanced Review wires. robustness and suffers from computational require. and its by minimizing this upper-bound while subjecting to integration with the MPC approach has been shown the stability and input constraints in the form of to be effective to control the power plants. © 2015 John Wiley & Sons. 56.60 LMIs. model using various kinds of objective functions and GPC is combined with the nonlinear state feedback computational tools.62 Besides Ref 62. it is The nonlinear MPC is naturally suited for con.63–66 controller in Ref 49 to solve the control problem of a multi-parameter programming (MPT). the basis of neuro-fuzzy networks (NFNs). Simulation studies on better control performance is shown compared to the a 600 MW FFPP indicate that owing to the ability of linear GPC. In Refs 57–66. In Refs 59 and 61. In Refs 49. simulation studies on the boiler-turbine coordinated and then used to establish a constrained multivariate system and superheater steam temperature system. the multi-model based control system is achieved and the overall control predictive controllers are developed recently. Through the the plant nonlinear properties is identified off-line. The control input can be solved alternative way to handle the nonlinearity.49. 55. Ltd. both the in Ref 48.58. In Refs 61.53 Although the advantages and effectiveness of these approaches the multi-model strategy seems more complicated than have been clearly shown through simulations of the the direct nonlinear approaches.57. which greatly limit its application: (1) because of the advances in multi-variable systems the satisfactory nonlinear dynamic model is difficult and the-state-of-the-art control techniques for linear to build. the stability of the closed-loop ment. which uses a combination of terminal state penalty cost in the objective function. there are two ear model is adopted in most of the multi-model MPCs main issues. a load-dependent exponential ARX local linear model weighted and controller weighted (Exp-ARX) model that can effectively describe approaches are given in these works. a plant-wide control. iterative learning control (ILC). the par. and PSO can be viewed as an extension of the linear DMC in based nonlinear MPCs are then designed to achieve Ref 35. the nonlinear model capturing the plant dynamics. piecewise without using the online-update method.50 Because the advances in linear is adopted and a Lyapunov function is constructed to modeling and control theory can be directly taken find the upper-bound of the objective function and into account. operating conditions. 65.52. which developed for 500 and 1000 MW FFPPs. and (2) the nonlinear optimization lacks in systems. in fact.com/wene that a satisfactory control of main steam pressure among these local models. multi-model MPCs are pro- much better control performance can be attained posed on the piecewise linear (PWL) models. such as QP. performance is improved by including a terminal inequality constraint. GPCs are developed on Besides the intelligence-based nonlinear MPC. local linear MPCs developed have been used in Refs 44–46 to search for the opti. it is easier superheater and reheater steam temperature control of and more efficient to implement. In Refs 15 and 47 online-update diagonal wide-range operation of a boiler-turbine system. at different loading points are combined for the mal control sequence in the NN based nonlinear MPC. and by adding a quadratic ‘divide and conquer’. several linear models to approximate the nonlinear In Refs 63–66. To overcome these issues. recurrent or feedback connections.j is the it provides an effective way to develop the model for weight that connects node i in the hidden layer to node 3 the nonlinear power plant using only the input-output j in the output layer.j ui k + Wi. particle swarm optimization. tant step in the advanced controller design. respectively. and accomplish the tasks. with the ability to The basic structure of an ENN with M inputs learn and discover knowledge. The ci k = xi k − 1 . input layer to node j in the hidden layer. ( ) ( ) eral depending on the purpose of application.67 (M ) As was already introduced in the advanced ( ) ∑ ( ) ∑R ( ) PI/PID control and MPC sections. computa. design. yj k = g Wi. The outputs can autonomously achieve a high level goal even in the in each layer of an ENN are given by: complicated or unexpected case..68 Beside this..j is the weight that connects data. Wi. and their structures are modified in gen. which makes the techniques to develop the computer-based control sys. © 2015 John Wiley & Sons. where for controller design. genetic (R ) algorithm. network sensitive to the history of input and output tems. Wi. node i in the context layer to node j in the hidden layer. The ENN differs from the tional complexity and other problems associated with conventional feedforward networks in that it includes large-scale distributed complex power plants. Wi. hidden patterns in data sets. evolutionary programming. and control Multi-model MPC Optimal control Reference Moving horizon input Plant output optimizaiton FFPP Global model Power load Weight calculation Local model #1 Local model #2 . Ltd. Local model #i + FIGURE 15 | Control diagram of model-weighted multi-model MPC.j xi k . where the tise and decision making. or where. Elman neural plant control are increasing steadily in recent years networks (ENN) and their modifications are utilized owing to its advantages. fuzzy logic. in Refs 44–46. emulate human exper. suitable for dynamic system modeling. multi-agent ( ) ∑ ( ) systems. Since the NN 1 has outstanding abilities in knowledge discovery. The delay in these Intelligent control is a class of control tech. connections store values in the previous time-step and niques that use various computational intelligence use them as inputs in the current step. WIREs Energy and Environment Steam power plant configuration. it recurrent neurons are in the context layer. . many different xj k = f 1 Wi.j ci k . 2 (12) i=1 The first intelligent technique investigated in the section is the neural network (NN). (11) most popular intelligent techniques are neural net- work. Intelligent Control (such as BP or RBF neural networks) with tapped The development of intelligent techniques in power delays. to approximate the behavior of the steam temper- nificant nonlinear and uncertain dynamics. which is generally the first and foremost impor. ature systems of FFPPs. as humans or species in nature. the intelligent control system behaves data. 3 (10) i=1 i=1 intelligent techniques have been applied to power plant control. the usual NN modeling is f (·) mostly takes logsig or tansig function and g(·) often combining the conventional feedforward networks takes purelin function. and f (·) and g(·) are the transfer functions of the hidden To provide dynamic information about the plant layer and the output layer neurons. Because. as well as their combinations. and N outputs is shown in Figure 16. such as overcoming the sig.j is the weight that connects node i in the 2 discovering underlying. the aforementioned y(k + 1) in Equation (13) with the desired output. to find the control input u(k) to drive the system to an With the dynamic process model being suc.wiley. utilized in Refs 72 and 73 to regulate the superheater © 2015 John Wiley & Sons. various advanced controller a controller by replacing the future expected output can be designed. In Ref 71. (EP). (13) suited for real-time application. by introducing a natural selection mecha- … … nism. Figure 17 illustrates how the techniques.com/wene y1(k) yN(k) … Δ Δ yref NNIC Plant u(k) y(k+1) x1(k) … xR(k) Δ Δ FIGURE 17 | Control system of the NNIC. and the PID compensator is serving as the feedback However. thus shorten the stabilization basic GA and PSO algorithms are compared and the time and reduce the overshot of the control process. the control To generate solutions to the highly complex nonlin. The resulting FIGURE 16 | Basic structure of the ENN.13. vergence compared with the conventional nonlinear a PID compensator is usually augmented to the sys- programming.13. Besides the traditional model based controller.y k − 1 .15. the diagonal recur. Ltd.y k − 1 .42 or particle As the NNIC is a feedforward controller. input/output mapping: lapsed to the hidden layer. fewer ( ) ( ) ( )) training iterations are needed and the DRNN is better u k − 1 . hybrid PSO can have a higher performance. Then these DRNNs are connected by using the u k − 1 . and the resulting control structure is shown The basic principle for all these heuristic opti. … . design the model-free controllers directly. which provides the main contribution to tion.69 In order to capture the complex dynamic of the which can be viewed as a nonlinear inverse mapping whole power plant.y k − n . to the set-point changes. area of FFPP dynamic modeling. tem. which eliminates the cross ( ) ( ( ) ( ) ( ) ( ) talks and reduces the number of weights between the u k = g y k + 1 . providing quality solutions and fast con. and it is capable simulator. in order swarm optimization (PSO)15. for example. Therefore. Therefore.y k − n . and the global best value (gbest). the context layer of DRNN is col. because the search procedure by the PSO controller to eliminate the steady-state control error strongly depends on the agent best value so far (pbest) induced by the NNIC. (14) outputs of a DRNN as the inputs of another DRNN. thus can be used as the plant tified by using the ENN or DRNN. evaluation and modification as we introduced in the control signal to make the plant respond quickly the Auto-tuning PI/PID Control section.70 genetic algorithm (GA)14.69 Compared The essential idea of the NNIC is to identify an with the ENN. controller.y k . … . the search area can be The feedforward-feedback control system is limited by them. input produced by the network will drive the system ear optimization problem. results show that PSO can attain a better performance. … . … . Advanced Review wires.44–47 have been employed to deal with the unavoidable modeling mismatches extensively. the neural network technique can also be used to As a special case of the ENN. C1(k) Cp(k) u1(k) uM(k) the effect of pbest and gbest is gradually curtailed and Context layer a broader search area can be realized. . heuristic optimization output y(k + 1) to yref . which is rent neural network (DRNN) is also popular in the called the neural network inverse control (NNIC). and the unknown plant variations and disturbances. such as evolutionary programming NNIC is applied as a controller in the system. which is usually performed by the EP and GA. The NNIC is used as a feedforward mization algorithms can be concluded as: initializa. yref . If the network represents the exact inverse. u k − 2 .u k . in Figure 18.47. the DRNN based combined model of the conventional dynamic model: is proposed in Refs 15 and 69. which may lead to the local minima. context layer and the hidden layer. where each primary ( ) ( ( ) ( ) ( ) ( ) component of the plant is approximated by a DRNN y k + 1 = g y k . expected point y(k + 1). Auto-tuning PI/PID Control and nonlinear MPC. The composed hierarchical structure can achieve a The inverse mapping function g(·) can be iden- satisfactory accuracy. it can be used as cessfully developed. u k − m . Therefore.u k − m . ( ) ( )) model. 2. which has advantage in capturing ing conditions.74 Its basic idea is using els and form the integrated T-S fuzzy model of the the fuzzy set to divide the whole operation region FFPP. Di ) are local system matrices. Unlike a conventional ber of fuzzy rules. then according to the expert knowl- the tacit knowledge of the system. It has been shown © 2015 John Wiley & Sons. Ci . zN k is MiN . yref By using fuzzy blending. N) the fuzzy sets. 64–66 an approximation or Another intelligent technique being widely used transformation of the nonlinear system has been used in the area of FFPP modeling and control is the fuzzy to obtain the linear state-space model at different load- logic control (FLC). In Refs 30. In Ref 75. The Takagi-Sugeno edge or the nonlinear investigation of the plant. The fuzzy technique also provides an effective els. In Refs 76 and 77. the T-S fuzzy model naturally provides a bumpless way to design the controller. Mij (j = 1. is proposed. FLC. L the num. for instance. a data-driven fuzzy modeling strategy of the system into several overlapping local regions. zN (k)) is the antecedent vector history at each sampling time. is Mi2 . the T-S fuzzy model can be responding to the current output error e(k) and its first described as follows: difference value Δe(k) . 2. and determined from the crisp values of the input–output Z(k) = (z1 (k). WIREs Energy and Environment Steam power plant configuration. … . … . Ltd. the the T-S fuzzy model can approximate the nonlinear subspace identification (SID) method is extended to behavior of the plant using the combination of several extract the local state-space model parameters. these linguistic values are matrices (Ai . the resulting be used to overcome the nonlinear control problem. The in the FARMA controller. Cz . PID control. THEN ∶ { ( ) ( ) ( ) respectively. and all other matrices Bz . Then by combining the input data with weighted summation of all the local models. which is composed by current and Δ Δ past measurable variables of the plant. Mij zj k . The results in these Mi Z k ∕ i=1 Mi Z k . Owing linear models. and control of the fuzzy model. z2 k membership functions to control the nonlinear ( ) boiler system and boiler-turbine system in the FFPP. … L ing Average (FARMA) controller is applied to the boiler–turbine system. Fuzzy logic controllers are then ( ) ( ) developed based on Gaussian and Triangular-shaped Ri ∶ IF z1 k is Mi1 . operation region and develop the structure of the T-S and then the global model is finally derived from the fuzzy model. . z2 (k). u(k) ∈ ℜm the control antecedent and consequent variables and makes rules. wi Z k = ( ( )) ∑L ( ( )) ( ( )) steam temperature in the FFPPs. and j=1 formance can be realized than the original cascade Dz are defined in the same way. the (T-S) fuzzy modeling technique is now the most pop. ∑L ( ( )) ( ( )) where AZ = i=1 wi Z k Ai . … . 60–62. fuzzy model can represent the boiler-turbine unit of Moreover. Because the corresponding fuzzy membership functions. a simple linear model is devel. control switchover. since the overlaps between different regions an FFPP very closely for advanced controller design. the controller is an inverse (15) mapping of the FARMA model. a Fuzzy Auto-Regressive Mov- ( ) ( ) ( ) y k = Ci x k + Di u k i = 1. better control per. Bi . design. historic input–output data. the advanced linear control theory can to the advantages of the both methods. which finds the control input according to the reference value and where Ri denotes the ith fuzzy rule. ing is used to provide an appropriate division of the oped to represent the local dynamics of the system. x k + 1 = Ai x k + Bi u k In Ref 78. fuzzy rules are developed to connect the local mod- ular FFPP modeling strategy. Mi Z k = works show that owing to the advantages of both ∏ N ( ( )) NNIC and the PID compensator. and y(k) ∈ ℜl the output vector. where an expert gives the linguistic values of the x(k) ∈ ℜn the state vector. can guarantee smooth transitions between local mod. rules are first given based on the expert knowledge to Take the discrete-time. the dynamic fuzzy NNIC + u(k) Plant model (15) can be expressed by the following global y(k+1) model: { ( ) ( ) ( ) PID x k + 1 = AZ x k + BZ u k compensator ( ) ( ) ( ) (16) y k = CZ x k + DZ u k FIGURE 18 | NNIC augmented by a PID compensator. where Gaustafson-Kessel (G-K) cluster- in each local region. state-space form of local determine the linguistic value of the control signal cor- model. input vector. and performed to adjust the power-pressure mapping their effectiveness in dealing with the nonlinearity to optimally accommodate different operating sce. Therefore. data-driven modeling methods such as After receiving and confirming the objective through ANN. combustion control system operate at a higher level of automation. proactive. to the energy shortage and environmental concern intelligent distributed parameter estimation coupled worldwide as well as competition among utilities and with the fiber-optic sensor system promises better other society driven forces. but also the A monitoring agent is also placed in this layer to monitoring of electric power plant components such select. there has used in large-scale FFPP simulators to achieve a plant been growing interest in multi-agent systems (MASs) wide optimization of the set-points. Advanced Review wires. and then the centralized control schemes or loosely decentral- PSO algorithm is employed to calculate the optimal ized control schemes because a single failure can bring set-points. In the highest level of power as inexpensively and as reliably as possible. and heat transfer compo- procedure to update the rule base. Since the has intelligent and autonomous properties because development of an accurate analytical model for a it is reactive. In Refs 13. recently. been applied extensively in the area of power plant In Refs 79–81. and it is dif- capture the steady-state nonlinear behavior between ficult and dangerous to manage the system using only multi-variables in the boiler-turbine system. of the system. in order to deal successfully with the complexity and To reduce the nitrogen oxides (NOx ) emission distributed problems in power plants and make the during coal combustion in the FFPPs.wiley. to different agents and investigates the performance ingly difficult over the past 30 years. Environmental and problems of the lower level. 71. However. intelligent agents that have temperature distributions of high-pressure liquids. environmental and through simulations. social. basic approach in developing the intelligent mon- tion. the similar methods are down the entire system. more stringent require. mal openings for the primary. Traditional sensors have not exhibited sufficient Meeting the power demand of the grid has stability and long-term accuracy without requiring been the primary FFPP operation task. where more economic. due expensive maintenance and recalibration. 15. and uncertainty of the plants has been demonstrated narios. not only the distributed parameter system (DPS) models to map conventional dynamic control performance. better controls can nents in extremely adverse power plant environments. optimization and monitoring. the MAS. support vector machine (SVM).85. This has required major improvements in information directly with the information requesters. both the FFPP and its safety concerns are taken into account. 3D boiler temperature distribution based on the 1D ous computational intelligence techniques are utilized fiber-optic sensors. asks a certain agent to share the and oil. combustion gases. It has become a challenge to measure high plant.86 The maximization of duty life and minimization of pollu. . Therefore. fuzzy c-means the communication with other agents. flexible and robust. etc. and economic concerns pressed the utility industry the mediate agent coordinates the cooperation of the to develop clean and efficient ways of burning coal lower agents such as.. The agents are arranged into a hierarchical The electric utility industry is charged to deliver structure and form the MAS. such as minimization of fuel consumption. multi-objective optimization is modeling. have to be fulfilled by the power plant to itoring system is in two folds: (1) development of achieve an optimal operation. where vari. optimization is studied in Refs 82–85 to find the opti. However. and (2) development of an intelligent tioned multiple objectives is important to realize an monitoring system for real-time monitoring of the integrated optimal control of the FFPP.com/wene in the simulation that by using the self-organizing steam. coal-fired boiler is difficult owing to the complexity and has multiple algorithms for solving problems. furnace and for improved combustion. multiple algorithm modules are cooperating together © 2015 John Wiley & Sons. for the furnace. Steady-state operation system are large-scale complex system con- ANN models are first developed in these works to sisting of many subsystems and functions. not only the optimization and control. In the middle level. However. store and analyze the sensing data of the power as boilers. be applied as time progresses. but also the three-dimensional (3D) temperature distribution the set-point optimization considering the aforemen. secondary and over-fire The basic unit in the MAS is the agent which air valves at different layers of the boiler. In the lowest level. control. the agent will clustering as well as their mixtures are proposed to choose a plan for the objective and select an algorithm build the steady-state relationships between these air to launch the plan. valves and NOx emission. the task delegation agent allocates tasks Meeting these dual obligations has become increas. estimate of the temperature distribution of a boiler ments. to overcome the nonlinear modeling and calculation The aforementioned intelligent techniques have issues of the FFPP. flexibility and robustness. Ltd. coupled with advanced diode PRODUCTS FOR THE FFPP lasers and optical fibers developed by the telecommu- nications industry. Therefore. and (3) accurately measuring the the technology development in the industry. fuel variations.). WIREs Energy and Environment Steam power plant configuration. increased FFPP control where the target gas is contained. Moreover. there is essen- bility. by using the data in the combustion zone of nies to demonstrate the effectiveness of the advance the furnace. the environment of growing beam through the gas mixture in the boiler furnace energy and pollution issues. by using the zone according to the experience. model. fuel variation through multiple computational intel- ble to measure the combustion condition (tempera. envi- tion optimizer on the basis of direct monitoring of ronmental temperature. different from these absorption lines. Zirconium oxide detector. 79. 67. the high temperature (beyond 1500∘ C) in the can be calculated for different loading condition and combustion zone of the furnace makes it impossi. radiation energy signal as a feedforward signal. Thus. more accurate combustion model can be technologies in the FFPP practice. etc. for example. a novel laser based measure- as well as its implementation in the FFPPs are well ment product is developed by the ZOLOBOSS based introduced in Refs 13. and the performance of the combustion (boiler optimization. tuning parameters and train. More- of commercial products provided by different compa. The new control software could be the computer aided calculation and tomography. ligence algorithms to realize an optimal operation of ture. Thus.). the flue gas temperature in the fur- thermocouple. developed or updated. so that through trol approaches. For these reasons. etc. CO. NOx emission. prevent the dust deposition and several companies have presented their own commer- slagging in the furnace. etc. . multiple light paths are assigned in the grid form have facilitated the employment of advanced con. In gen- plant and the development of computer technologies eral. the surement data around the rear flue gas channel. how. law. complex design. nace can be used to calculate the equivalent radiation tor. which can represent a balance can only estimate the parameters in the combustion between the boiler and turbine. The BCOS-ZOLO proposes the The first product introduced is the BCOS-ZOLO use of the ANN to capture the dynamics between the offered by ZOLOBOSS. reduce the deviation of the cial FFPP control solution products. CO detec. gas) using the traditional instrumentations such as the Furthermore. on the certain layer of the furnace. so far. the widespread tration of target gas molecules as well as the flue gas use of distributed control system (DCS) in power temperatures distributed over the beam’s path.87 which is an FFPP opera. Ltd. 80. distributions of fuel and the combustion zone in the furnace and combustion air. At wavelengths slightly assessed procedures due to the reasons such as: relia. The principles are straightforward: Although the power plant industry may still be reluc. the installed in the programmable logic controllers (PLCs) temperature and gas concentration distribution can be or industrial computers and communicates with the obtained and viewed. design. the combustion in efficiency. in the flue the boiler. tially no absorption. and (2) tuning the challenges as well as the significant improvements wavelength of the beam to one of the absorption lines reported by the academic studies has already led to of the target gas.. and so on. etc. which issues of large inertia behavior of the steam pressure is inaccurate and thus brings great challenges in the and temperature control system can be overcome to a © 2015 John Wiley & Sons. boiler operation condition (load. based on the mea. the plant personnel energy in the boiler. Design of single agents and the integrated MAS For this reason. aiming at improv- flue gas temperature at the exit of the furnace to pre- ing the operation performance of the FFPPs. DCS through the object linking and embedding for Knowing the combustion condition in the fur- process control (OPC) or Modbus protocols. Gas molecules absorb energy at specific wavelengths in tant to accept changes that would revolutionize well the electromagnetic spectrum. the best distributions of the fuel and air ever. The TDLAS instruments rely on well-known spectroscopic principles and sensitive ADVANCED CONTROL TECHNOLOGY detection techniques. and control to identify the system. As is well known. one can deduce the concen- over the course of the last two decades. The purpose of this section is to introduce several examples vent the explosion of the heating surface. absorption of that beam. over. and the contents of O2 . which nace will provide a direct guidance for the FFPP oper- will then collect the required data and send the com- ators to achieve a safe and efficient operation of the mands to regulate the actuators. on the technology of tunable diode laser absorption spectroscopy (TDLAS). On the basis of the the furnace is the energy source of the FFPP. by (1) transmitting a light ing of plant personnel. boiler. perform the multi-objective combustion control and the resulting difficulties in optimization and determine the set-points and control steam temperature and pressure control. and optimizer. gies to optimize the combustion. the PROFI UCC offered by SIEMENS89 is the most widely used one. issues of FFPPs such as wide range load following. By using the MPC’s ability advanced FFPP control works. optimization and control. nonlinear or multi-model based MPCs to the process model to approximate the dynamics of handle the large-inertia behavior and the strict the plant. many companies such as SIEMENS. bustion zone of the boiler through the TDLAS. But.wiley. linear issues of the plants. fuel and plant behavior variations. The essential idea of the PROFI UCC is devel. intelligent and model-based On the basis of model prediction. in order to solve the control 1. Using the computational intelligence technolo. is the best and will be the future trends of the a pre-action command can be given to the fuel feeder FFPP control. and also offered by the SIEMENS SPPA-P3000 Process made a solid foundation for the people’s lives. control practices and we cannot answer which one mal energy’ produced by the boiler. gies such as multi-variable control methods. which leads to 1. Advanced Review wires. development in nonlinear. and (2) utilizing advanced control strate. The common features disturbances. and achieve a more 2. solving the non- performance of the plant. plant operation. the academic researches and industrial applications. Design of FFPP applied in more than 20 FFPPs in the world. Germany and Korea. Among these commercial products. mostly has evolved dramatically in the area of scale. working in the United States. Developing the combustion model using the safe and efficient operation.com/wene large extent. Over the past hundred years. The main parameters.92 have launched their own FFPP and plant behavior variations. Computational intelligence techniques in model- or intelligent control algorithms to improve the ing. it is safe to say that much more temperature can be achieved. FFPPs have been the A similar TDLAS based FFPP optimization is primary power generation plants in the world. steam pressure and However.89 EMERSON. monitor the pulverized coal concentration and where these advanced plant operation optimizers are primary air flow rate from pulverizing mills to implemented. the large-scale power plant. challenging the classical PI/PID based control system. . Advanced PI/PID controllers using auto tuning or gain scheduling techniques to improve the To improve the control performance of operation of the FFPPs in a wide load range. through the investigation of many temperature control. input–output constraints of the plants. Robust controllers in order to deal with the fuel HONEYWELL. Improving the control performance of the steam pressure and temperature. For example. Linear. of these commercial products are: (1) identifying 3. Measurement of key parameters during the quicker load following and more efficient and prof. In recent 20 years. novel mea- itable operation. the that may need to be studied further in the future: MIMOROC allows operation closer to the constraints compared with conventional control. considering the benefits brought by to compensate for the inertia and delay behavior of these advanced technologies and the facilities provided the boiler so that a quick load following performance by the wide use of DCS and fast development of and a smooth response of the main steam pressure and computer technologies. configurations and techniques to meet the features of them can be concluded as: growing power demand. a multi-input MIMO control of FFPPs is ahead of us. and a better load following operation can DISCUSSIONS AND CONCLUSIONS directly be achieved. The major FFPP control developments include: 4. Ltd. MPCs 4.88 Both the two products have now been social and industrial developments. various advanced FFPP ANN technique. These advanced control methods as well as their oping/updating a nonlinear model of the boiler and mixtures have all been implemented in the FFPP using the technology of MPC to estimate the ‘ther. uncertainties and coordinated control optimizers. there are several issues of dealing with the input–output constraints. the burners (Some products have already been © 2015 John Wiley & Sons. control techniques have progressed swiftly in both 3. Remarkable plant efficiency and surement technologies can be developed to profit improvement have been reported on the FFPPs. optimization of the combustion. Direct measuring of the parameters in the com. 1.90 ABB91 and 2. through which. and the multi-output rate-optimal controller (MIMOROC) advanced control techniques will replace the classical is proposed by the HONEYWELL’s UES92 for the PI/PID controllers in a foreseeable future. we have to admit that. because ing current fixed and non-optimal set-points the multi-variable controller would work in a given at specific loading condition. These are where both the dynamic control objective and also the main reasons that the plant person- other non-quadratic economic objectives are nel are reluctant to employ the modern control integrated together in the controller. but failed to consider the model PID loops present in the control structure. therefore. so that more optimal and robust control knowledge of analytical model and advanced can be attained through its estimation. it has optimal set-points for the control layer. In that case. how to design an them. for example. design. modern control theory. in the suited for the FFPP is a possible future research context of global energy and environmen. However. information which will guide the controller modynamic knowledge and plant operation design. a advanced controller design. advanced multi. will effectively deal with the control issues of ous and multiple objectives of plant economic the FFPP and greatly improve the operational operation. are all rently the variation of the fuel is one of the important issues in the FFPP practices. single-output PI/PID control objective optimization with Pareto optimality loops currently used in FFPPs. Ltd. topic. con. Reliability. how to better integrate the data. the researchers focused on the control of the system is keeping the well-assessed SISO algorithm. Closed-loop data-driven modeling. Therefore. Therefore. even and balanced combustion in the furnace. for example. Modeling Besides careful design and testing of the con- is the first and foremost important step in troller and training of the plant personnel. study of bumpless switchover mech- sidering the complexity of real FFPP. the PfMaster and design specifications of many components. compared mance. 4. param- dynamics and environmental changes to provide eter tuning and computation. Although the advance controllers of dynamic control optimality to simultane. nonlinear analytical model. key problems in the coal-fired power plants. In most of the possible way to directly ensure the reliability studies. soft measurement best one to accurately portray the dynamics technologies of the coal properties are expected of nonlinear FFPPs. For be the main factors causing the uneven distri. performance. the the new control system could be disconnected at results are relying on the availability of the anytime. and control developed recently. Cur. through the measurement of model in practice. In this case. control technology to develop a controller better 2. WIREs Energy and Environment Steam power plant configuration. induce a failure of the entire plant. more simple and efficient method such tions. Therefore. These parameters are which becomes the major limitation for the unknown in the FFPPs. but have been found to application of advanced control methods. examine the outlets of the pulverizing mills to guarantee and polish the data. replac. by ABB/Greenbank). 3. should be broadened from a simple objective 5. the control objective of the FFPP model based nonlinear predictive controller. In many of their works. a failure of the control algorithm would as novel economic MPC can be investigated. emission and dynamic control perfor. development. on the other centralized control unit in most of the applica- hand. it is anism between advanced controller and PID difficult to develop an accurate analytical model controller could be made. Thus. an analytical model is after all the plant operation. leading to an analytical tal issues. an efficient way to develop a control oriented heaters. action with the identification theory. it can also provide much to be developed using the combination of ther. Furthermore. . Optimal control. to the single-input. Better combination of the analytical model and the uncertainty greatly influences the safety. limitations in reliability. additional regulators can be installed in optimal input signal for identification. these advanced can be developed based on the plant models and controllers are complicated in structure. As mentioned before. without the knowledge of thermodynamics © 2015 John Wiley & Sons. this reason. Thus. techniques. without compromising the plant safety. and test the model sufficiently for healthy inter- Another example is the coal properties. select the model structure. on the one hand. system identification methods using bution of the temperature field in the furnace the closed-loop plant operation data provide and the explosions of the waterwalls and super. Although difficult to stability and efficiency of the combustion and develop. Int J Robust Nonlin Cont 1994. IEEE Trans Autom Control 10. Boiler-turbine control system design 25. 40th ed. Colorado Springs. concepts for a classical/modern synthesis. Greece. Souza GFM. In: Proceedings of Symposium operation. Robust control of speed and using a genetic algorithm. 10:752–759. Zames G. the Natural Science Foundation of Jiangsu Province. Taft C. CO. Bentsman J. Lee KY. superheated steam temperature control system using Research Triangle Park. Power Plant Engi. 16. ISA Trans 1995. boilers. H∞ controller design for 16–21. 5693–5700. Stein G. ISA Trans 2012. In: Proceedings of the IEEE World Congress on ence. USA. 40:952–956. In: Proceedings methodology. control strategy for boiler-turbine system of coal-fired 5. Van Sickel JH. In: Steam: Its Gener. multiplicative seminorms. Raja AK. Garduno-Ramirez R. Dimeo R. 17. Fuzzy gain-scheduling 6. Hussey A. Liu H. IEEE Trans 1992. S-H. Zhao H. New York: Babcock & Wilcox. Kohala Coast. paper #88-0417. Advanced Review wires. Lee KY. Power-Plant Control and Instrumentation: power plant.com/wene ACKNOWLEDGMENTS This work was supported in parts by the National Natural Science Foundation of China (NSFC) under Grant 51036002. Power plant fuzzy PID of America. Lee KY. Stevenage. The control philosophy for a unit ligent System Application to Power Systems (ISAP control system for co-ordinated operation of a boiler 2003).pdf. Baden. June 2004. Kaohsing. Landis R. Ltd. 2003. UK: IEE Press. Role of power station control in PID+decoupling control for power plant wide-range overall system operation. J Southeast Univ 2010. Lee KY. Xiang X. The evolution of coordinated control. Stevenage. Cai WJ. Pellegrinetti G. 1988. 18. 2007. extremum seeking algorithm. Petrus B. Canada. Dwivedi M. Compensation of Applications. 215–257. Lindsey D. Grant 11190015 and Grant 51306082.wiley. China under Grant BK20141119. February:19–23. Energy Conver 2010. 2000. Zhang S. Xie LH. Gery HC. 13: The Control of Boilers and HRSG Systems. Quazza G. Feedback and optimal sensitivity: model ment Society of America. 4. Ferrari E. 4:645–671. ed. 51:304–308. CD ISAP03-086. 7. 1988. and approximate inverses. IEEE Trans Contr Syst Technol 2005. the Cooperative Innovation Foundation of Jiangsu Province-Prospective Joint Research Project under Grant BY2013073-07 and the U. . 11. Soh YC. Babcock & Wilcox. 802–807. Chapter 41: Controls for fossil 1981. Electron 1988. Utility front end controls. In: Pro- ceedings of the International Conference on Intel- 8. Taft CW. 22. Heidelberg: Springer. 943–954.S. Russell T. 113–122. Instru- 20. Li S. 21. Li W. National Science Foundation under Grant ECCS 0801440. Vancouver. 31–September 3. Fei W. 109–112. 2012. 2005. 26:301–320. Taft CW. Kim Control. Zheng K. 1972. Lee KY. 233–238. fuel-fired steam generating plants. 2003. Robust controller Simultaneous gains tuning in boiler/turbine PID-based design for simultaneous control of throttle pressure and controller clusters using iterative feedback tuning megawatt output in a power plant unit. Control Eng Pract 2005. Controller design for a large-scale ultra super 2. Srivastava AP. 26:4–16. Advance boiler/turbine operation. Lee KY. Doyleand JC. Gilman GF. control-loop interaction for power plant wide-range 23. August and turbine. 25:1063–1070. November 5–8. 9. IEEE Trans Energy Conver temperature in a power plant gas turbine. 13:1475–1487. scheduling control over full operating space. Autom Control 1981. paper #88-0418. Garduno-Ramirez R. Li Y. Garduno-Ramirez R. In: Proceedings of the 14th Annual Joint ISA heuristic optimal control system for a large-scale power POWID/EPRI Controls and Instrumentation Confer- plant. reference transformations. IEEE Trans sis. 14. 51:609–621. control and its benchmarking in a coal-fired power 13. ISA. Heo JS. Multivariable feedback design: ation and Use. on Intelligent Systems Application to Power Systems. © 2015 John Wiley & Sons. critical once-through boiler power plant. Taiwan. 2006. neering. UK: IEE Press. Optimization of 3. Jung W-H. New Delhi: New Age International Publishers. Lemnos. In: Proceedings of International Conference on Real-Time Control of Electric Power Systems. 2012. Wulfsohn E. Thermal Power Plant Simulation and 15. Shen J. Flynn D. A multi-agent system-based intelligent plant. Switzerland. Bentsman J. Computational Intelligence. HI August 1999. Instrument Society 19. Thermal Power Plant Performance Analy. Ramezani MH. Najimi E. Bentsman J. Bentsman J. the Doctoral Fund of the Ministry of Education of China under Grant 20130092110061. July 24. Hoffman JA. A new coordinated 2006. REFERENCES 1. Boiler control system engineering. of the 1999 IEEE International Conference on Control 12. 42. controller with elastic search space for STC of a LSSPU. 18:1116–1125. 3950–3954. 1997. An improved predictive optimal based on fuzzy Lyapunov functions. Kohala 43. Ma L. Lee J. China. Kim B. steam-electric power plants. 7:1–10. In: 2007 33. Multiple Model temperature regulation of thermal power plants. unit. Nguang SK. 10:256–262. Maui. 196–201. Lee KY. Moelbak T. Electr Approaches to Modeling and Control. Automatica 1998. 11:733–764. In: 2010 Chinese Control and Decision Confer- 2011. In: 2009 Inter- MW Unit Plant Coordinated Control System. Swidenbank E. Weng C. In: 2010 IEEE strategy for thermal power plants. Multi-model predic- predictive control for a multivariable boiler-turbine tive function control based on neural network and its unit. Simulation research of 53. Control Eng Pract 2003. Liu J. Syst Technol 2002. 33:96–103. Liu J. 11:407–413. Johansen TA. 51. Tampa. Zheng K. 10:735–742. 811–814. TX. Robust wide-range control of IEEE PES General Meeting. Marquez HJ. predictive optimal control using neural network-based combined model for large-scale power plants. Kim S. Li Y. 2012. Hogg BW. Ltd. A neural net Coast. Moon U. Hou G. PES General Meeting. IEEE Trans Control Syst Technol 2002. Lee KY. power plants. Calgary. Krug M. 6:282–288. 45. 26:779–785. China. Han P. In: Proceedings of the 1999 IEEE International system. 31. Murray-Smith R. El-Rabaie NM. WIREs Energy and Environment Steam power plant configuration. . J Dyn Syst Meas Control 2008. 13:176–182. Modified plants. Tan W. Predictive con- the multi-variable generalized predictive control in 500 trol for boiler-turbine unit using ANFIS. Jung W. Ma L. London. 34:1185–1204. application to the coordinated control system of power Automation and Systems. Hogg BW. Lee KY. 49:369–375. © 2015 John Wiley & Sons. Liu G. Advanced control of superheater steam governing steam turbine power generators. Taylor & Francis. H∞ control for a boiler-turbine linear predictive switching control for a boiler-turbine unit. Liu X. Japan. Intelligent 30. IEEE Trans Control Syst Technol 1997. Robust H∞ modified predictive optimal control of reheater steam tracking control of boiler–turbine systems. Shen J. Proc CSEE 2013. 39. IEEE Trans Energy Conver 1995. Boo C. An auxiliary LQG/LTR robust controller design for cogeneration 47. Ramirez A. Arroyo-Figueroa G. Lee KY. Wang D. A survey of industrial model able generalized predictive control strategy for thermal predictive control technology. 44. Zhang J. 041011-14. 2010. In: 11th International Conference on Control. ence. Xuzhou. design. Swidenbank E. Park Y. Lee KY. Jung W. 10:100–106. Step-response model development for 49. 48. Shen J. Hogg BW. Nguang SK. Niu Y. Hoffman JA. Tahami F. Meng L. Sanchez-Lopez A. Guan P. Wu J. Prasad G. Sun Y. Hong International Conference on Advanced Mechatronic Kong. Taft CW. Multivariable generalized trol with fuzzy-interpolated step-response model for a predictive control of a boiler system. UK: Power Energy Syst 2004. In: 51th IEEE Conference on Decision and Control. 7024–7029. 38. Control Eng Pract 1999. model-based multivariable long-range predictive con- 28. Cao X. Tokyo. Huang B. Advanced control algorithms for steam 50. November 2009. 54. Liu S. Bentsman J. Peng H. App Mech Mat 2012. IEEE Trans Control IEEE Trans Energy Conver 1998. Moon U. trol strategy applied in thermal power plant control. Ray A. Wu J. IEEE Trans Energy Conver 2009. Korea. Nademiand H. Mohseni R. Badgwell TA. Lee KY. Power Engineering Society General Meeting. Kim S. Wu J. IEEE Trans Energy Conver 1991. Tan W. Minneapolis. Conver 2011. 130:041011-1– 128-129:1065–1069. 26:393–401. Chan CW. Robust controller design for 41. Zhengzhou. Nonlinear multivariable dynamic matrix control of a drum-type boiler-turbine power plant coordinate control by constrained pre- system. China. 2007. Chen T. plants. dictive scheme. Li Y. 910–914. Ozaki T. HI August 1999. Kim W. Canada. 2010. An adaptive dynamic matrix con- 40. Ge Y. 24:423–430. temperatures—an evaluation based on practical appli- tional Conference on Electrical Machines and Systems. In: Interna. 2009. Villavicencio. FL. ISA Trans temperature in a large-scale boiler unit. 35. Xi Y. J Contr Theor Applicat 2012. Hou G. GA-based non- 27. Conference on Control Applications. troller design for a boiler system. A nonlinear exponential ARX model-based multivari- 34. Liu X. 32. 36. In: 2011 national Conference on Test and Measurement. Design of tracking controller for coordinated boiler-turbine control system 46. MN. Ge Y. Multivariable robust con. 37. Qin SJ. Superheated steam temperature 29. Jung W. IEEE Trans Energ drum-type boiler-turbine system. 2011. Shen J. 5:74–88. Moon U. Haggan-Ozaki V. 2009. A local model net- Once-through boiler steam temperature control using works based multivariable long-range predictive control Dynamic Matrix Control technique. IEEE Trans Contr Sys Technol 2010. Zhang J. cations. Toyoda Y. Li Y. Generalized model 52. Choi M. Kim S. Karampoorian HR. and control 26. Prasad G. Heo JS. In: 2009 IEEE 2010. Full operating range control based on improved recurrent neural network robust hybrid control of a coal-fired boiler/turbine and simplified PSO algorithm. Systems. Liu H. Lee KY. Kintex. Ma L. Xiamen. Wu X. ISA Trans 2014. Minneapolis. 2002 Congress on Evolutionary Computation 2002. Li Y. 1264–1265. Gomes JR. June 44–6. 71. Finland. 72. In: Proceedings Canada. on Power Plants and Power Systems Control. Multiobjective tive control of a nonlinear power plant reheater with control of power plants using particle swarm opti- switched dynamics. Hierarchical optimization Power Symposium. Control Eng Pract 2014. Nonlinear fuzzy model predictive and predictive control for boiler-turbine unit using iterative learning control for drum-type boiler-turbine fuzzy clustering and subspace methods. Superheated steam temperature predictive 2011.wiley. to boiler-turbine coordinated system. MI. Montreal. 2:1308–1313. Canada. Lee KY. PPPSC 2012. Gomes JR. In: 2006 IEEE Power Engineering Society for power plants. July 5–8. IFAC Proc 2012. Albuquerque. Hubka L. Modeling and predic. Intelligent predictive control 57. control of boiler steam temperature. Ghezelayagh H. Multi-agent 67. Ge Y. Wu X. Piecewise affine modeling and control of a dynamic process models. Man Cybern 1985. July 13–17. Ltd. 284–289. Lv J. September 26–28. In: 16th WSEAS Interna. 61. Tampere. Chan CW. Ma L. Tuma L. Lee KY. Simul Model Pract Theory 2012. Xiang W. optimal control based on external time-delay BP neural 81. 8:201–206 8th Power Plant tation of a multi-agent system for optimized multiob- and Power System Control Symposium. Wu X. 2011. In: 16th International Conference mization techniques. 53:699–708. Lee KY. Jahed-Motlagh ture control for a 300 MW boiler unit with inverse MR. tive control. mizer and neuro-fuzzy identifier.com/wene 55. In: Proceedings of the 1997 American Control In: 2010 8th IEEE International Conference on Control Conference. Multiagent con- 64. . 30: neering Society General Meeting. IEEE Trans Energy Conver 2006. trol system for a fossil-fuel power unit. 21:552–561. Hlava J. 75. Zhang T. In: 2010 IEEE Power Engi- boiler-turbine unit. trol using a fuzzy auto-regressive moving average 3356–3361. 2006. Williams CS. In: Proceedings of the IFAC Symposium 1115–1120. and Automation (ICCA). NM. Liu J. 2010. ica Sinica 2006. 15:116–132. Greece. Li Y. Lee KY. In: 2003 IEEE tive control based on TS fuzzy model with application Power Engineering Society General Meeting. Fuzzy logic strategy on boiler control nated system using multiple-model predictive approach. system. Lee KY. Williams CS. Lee KY. Chang J. Kos Island. tracking control for boiler-turbine coordinated control 79. IEEE Trans Energ Conver 2003. In: Proceedings of the tional Conference on Circuits and Systems. Appl Therm Eng 2010. In: 2007 IEEE PES General Meeting. 2010. 2011. 80. Liu X. July 24–28. 26:77–95. Kong XB. boiler unit. Implemen- system. Lee KY. Heo JS. Inverse control for nonlinear boiler-turbine system based on dynamic neuro-controller for superheater steam temper- fuzzy gain scheduling. Fuzzy identification of systems and predictive control of a boiler-turbine system based on its application to modeling and control. (FARMA) model. In: Proceedings of the 2008 IEEE ature control of a large-scale ultra super critical (USC) International Conference on Automation and Logistics. Hoffman JA. Chalupa P. 2003. Shen J. Miedzyzdroje. Acta Automat. 76. Head JD. IEEE Trans Syst genetic algorithm. 70. Lee KY. China. 69. Li Y. Lee KY. Stable model predictive 142–148. 2008. Control of boiler-turbine coordi. on Methods and Models in Automation and Robotics. Lee KY. In: 2011 IEEE Power Engineer- General Meeting. Keshavarz M. Detroit. Lee KY. Li Y. 23:1023–1040. Barkhordari Yazdi M. Liu XJ. Boo CJ. Novak J. Li Y. Shen J. Constrained power plant coordinated pre. Intelligent techniques applied to power plant system based intelligent distributed control system control. Head JD. dictive control using neurofuzzy model. network-based modeling for a large-scale power plant. 781–791. 2012. TX. 30:112–123. Stable model predic. 2009. of the 50th IEEE Conference on Decision and Con- 78. 1997. Liu X. A boiler-turbine system con- trol and European Control Conference (CDC-ECC). Heo JS. Jung W. Shen J. Chai T. Wu X. Jung WH. IEEE Trans Energy July 25–27. June 9–11. Garduno-Ramirez R. Wu K. 2012. Tampa. Toronto. ing Society General Meeting. Lee KY. Sugeno M. 58. Arlington. 2010. Lee KY. Ma L. Neuro-fuzzy generalized predictive of the 31st Chinese Control Conference. Predictive control of a of a power plant with evolutionary programming opti- boiler-turbine system. J Process Control 2013. Hefei. jective power plant control. Dynamic mul- network and a simpler PSO algorithm. Heo JS. Moon U. Offset-free fuzzy model 74. Advanced Review wires. In: 2010 North American 66. Takagi T. Superheater steam tempera- 59. FL. Neural 56. 60. In: Proceedings tiobjective optimization of power plant using PSO © 2015 John Wiley & Sons. 32:785–790. of boiler-turbine unit using fuzzy stable model predic. Conver 2006. 63. Data-driven modeling 62. problem. Garduno-Ramirez R. Wu X. Lee KY. Shen J. Lin Y. Liu XJ. 68. 2007. Shen J. China. Kim S. Shen J. Li Y. Ma L. MN. 18: 65. Lee KY. Kim SH. Poland. Garduno-Ramirez R. 21:900–908. Model predictive 73. 77. Liu X. Lee KY. EMERSON Process Management. A novel least squares nated control system. Kim BH. (Accessed March 2014). Lv Y. In: © 2015 John Wiley & Sons. Zhou H. Liu J. Coal power plant combus- 82.com/en-us/brands/smartprocess/pages/ 85. boiler with artificial neural networks. 29:167–183. Energy 2004. Lee KY. 55:319–329. intelligent monitoring system with high temperature 91.com/ the NOx emission characteristics of a tangentially fired power-generation/coal-power-plant-efficiency/. (Accessed March 2014). Development of an index. Energy 2013. Available at: http://www2. 02852573a3004c9a5a/$file/plant_optimization_s_ CO. Ind Eng Chem Res 2012. plant optimization solutions for distributed fiber-optic sensor for fossil fuel power plant. Lv Y. ABB. 16092–16100. 90.abb. 2000. San Francisco. June 6–10. Denver. Honeywell UES. optimization solutions. SPPA-P3000 83. system type neural networks and semigroup theory. PROFI unit coordi- 84. Velas JP. and control techniques. San Francisco. emersonprocess. 87. Available at: http://www05. KWU LPSC Technology Center. WIREs Energy and Environment Steam power plant configuration. Available at: http:// with error-based LSSVM and its application to www. Yang T. Ltd. 88. 2005. ZOLO Technologies. 86. (Accessed December 2013). 89.energy. Available at: https:// monitoring system for fossil-fuel power plants using www. (Accessed February 2014). CA.com/hq/en/automation/power- NOx modeling. OPTIMAX. Smart process sion prediction of a coal-fired boiler. Lee KY. Nonlinear PLS integrated solutions for process optimization.nsf/veritydisplay/e15f7798384e6f General Meeting. Siemens AG. June 12–17. SIEMENS Power Plant Automation.com. power generation. Kim BH. Velas JP. Modeling and optimization of tion optimization. Fan J. Available at: http://zolotech. support vector machine ensemble model for NOx emis.PDF. Zeng D. Liu J. July 12–17. Honeywell. Yang T. CA.honeywell. Proceedings of the IEEE Power Engineering Society ety General Meeting. deabb_1291_06_e. design. In: 2005 IEEE Power Engineering Soci. .com/ In: Proceedings of the IEEE Power Engineering Society global/scot/scot267. 51: generation/sppa-p3000/. 2005. PESGM2004-001350. Development of intelligent 92. Cen K. (Accessed May 2014). General Meeting.aspx.siemens.pdf. 2004.