Predictive Control of Thermal Power Plants: C. Aurora, L. Magni, R. Scattolini, P. Colombo, F. Pretolani and G. Villa

Download as pdf or txt
Download as pdf or txt
You are on page 1of 19

INTERNATIONAL JOURNAL OF ROBUST AND NONLINEAR CONTROL Int. J. Robust Nonlinear Control 2004; 14:415433 (DOI: 10.1002/rnc.

890)

Predictive control of thermal Power Plants


C. Aurora1, L. Magni1,n,y, R. Scattolini1, P. Colombo2, F. Pretolani2 and G. Villa2
1

" Dipartimento di Informatica e Sistemistica, Universita di Pavia, via Ferrata 1, 27100 Pavia, Italy 2 CESI, via Rubattino, 54, 20134 Milano, Italy

SUMMARY This work presents the results of a project aimed at verifying the applicability of industrial model predictive control (MPC) to thermal Power Plants. The research is motivated by the need to improve the eciency of power plants so as to cope with the high levels of competition induced by the deregulation of the energy market. A detailed plant simulator, already used for operators training and controllers tuning, is coupled to an industrial software package implementing the dynamic matrix control algorithm. The achieved results witness the great potentialities of MPC, with respect to classical decentralized schemes, in terms of economical savings, reduction of pollutants, improved exibility, easier tuning and better documentation. Copyright # 2004 John Wiley & Sons, Ltd.
KEY WORDS:

model predictive control; thermal Power Plants

1. INTRODUCTION The deregulation of the energy market in western Europe has dramatically increased the competition among the energy producers. This calls for higher and higher levels of eciency in the management of the operating units in order to fulll a number of requirements. Among them, the most important are: (i) the optimization of the steady state operating conditions according to precisely quantied economic criteria, (ii) the possibility to operate with exibility over the whole operating range with high rates of load variations, (iii) ecient dynamic control strategies explicitly coping with the constraints imposed by technological limits and by environmental restrictions, (iv) systematic procedures for the design and documentation of the control system. Due to its almost unique features, which fully comply with all these requirements, model predictive control (MPC), see e.g. the survey papers [1,2] and the books [3,4], has extensively been used during the past twenty years mostly in the chemical and the petrochemical industry, where nowadays it is unanimously considered as the proper approach to the control design, see Reference [5]. In spite of the many similarities between the problems of managing and controlling (petro)-chemical units and power generation plants, the application of MPC in the

Correspondence to: Professor Lalo Magni, Dipartimento di Informatica e Sistemistica, Universit" di Pavia, via a Ferrata 1, 27100 Pavia, Italy y E-mail: lalo.magni@unipv.it

Copyright # 2004 John Wiley & Sons, Ltd.

416

C. AURORA ET AL.

power industry has not yet been exploited in depth. Notable exceptions are the works reported in [69], where dierent control structures based on ad hoc MPC implementations have been considered to control a reduced order nonlinear plant model and in Reference [10], where arealtime laboratory simulator has been used in the testing phase. Experimental results concerning the application of MPC for control of the temperature and the pressure of the steam at the superheater outlet in a large coal-red power plant have been reported in Reference [11]. This paper presents the results of a detailed feasibility study on the practical applicability of MPC to thermal power plants. The plant is simulated with ALTERLEGO [12,13], a simulation environment developed by CESI, Centro Elettrotecnico Sperimentale Italiano, and validated over the years with real plant data. In ALTERLEGO also the classical co-ordinated scheme used in real plants is duplicated; it is composed by a number of PI regulators, by nonlinear static elements and by a number of logic blocks used during critical operating conditions, such as the plant start-up, and for safety reasons. The simulator is coupled with DMCplus by Aspentech, one of the most widely used commercial software environments for MPC implementing the popular dynamic matrix control (DMC) algorithm [14]. The MPC algorithm is applied to substitute in the co-ordinated scheme a number of single-input, single-output PI regulators controlling the steam pressure, the superheater and reheater steam temperatures and the opening of the reheat attemperator valve. Conversely, the generated power is controlled as in the classical decentralized approach in view of its fast dynamics and for safety reasons. Then, after the introduction of MPC, the resulting control scheme can be viewed as a boiler following. A number of results is reported to witness the many potential benets of the proposed solution. The paper is organized as follows. Section 2 is devoted to present the main characteristics of the plant under control, of the traditional co-ordinated scheme and the control structure considered in the MPC implementation. In Section 3 the DMC algorithm is briey summarized together with the main characteristics of the DMCplus package. Section 4 deals with the identication of the linear model used in the synthesis of the MPC controller. The main tuning guidelines of the MPC regulator and the simulation results are presented and discussed in Section 5, while the conclusions of this research activity are drawn in Section 6.

2. PLANT MODEL AND REGULATION SCHEME The modelled plant is a conventional once-through 140320 MW multi-fuel power plant, reproducing one of the units of Rossano Calabro in Italy. A schematic diagram of the plant is shown in Figure 1. The main pressure must be maintained at 170 bar by acting on the fuel ow valve (FFV), while the air ow valve (AFV) regulates the percentage of oxygen in the exhausts. The feed water ow is driven by the pump (FW) into multiple heat-exchangers stages: the economizer (ECO), located in the last part of the boiler, is intended to improve the global plant eciency by pre-heating the water ow coming from the condenser (COND); the evaporator (EVAP), situated in the ame zone, causes water to turn into steam by radiation; the superheater, composed by a radiationconvection (SH2) plus a convection (SH1) stage, feeds the high-pressure stage of the turbine (HP), superheating the saturated steam to 5408C; the turbine governor valve (TGV), which plays the role of the main actuator in the closed-loop speed regulation of the generator (GEN), controls the steam ow to the HP turbine; the reheater
Copyright # 2004 John Wiley & Sons, Ltd. Int. J. Robust Nonlinear Control 2004; 14:415433

PREDICTIVE CONTROL OF THERMAL POWER PLANTS

417

Figure 1. A schematic power plant diagram.

(RH) takes the exhaust steam from the HP stage back to 5408C; and feeds the mediumlow pressure turbine stage (MLP). Two attemperators are provided to regulate the steam temperature: the rst one (ATSH), intended to compensate the slow fuel dynamics by means of a faster regulation during transients, is maintained at a 50% opening valve value in steady state operations; the reheater attemperator (ATRH), instead, should be sparingly used, due to its negative inuence on the boiler eciency. The reheater regulation is also performed by means of the gas re-circulation (GR) damper, which modies the combustion properties, thus allowing a balance action between radiation and convection. The plant simulator is built with ALTERLEGO, a modular simulation tool based on module libraries of the main plant components and on an ecient numeric solver. ALTERLEGO has been widely used in many applications for the design and validation of the control schemes subsequently applied to the real plants as well as for the operators training. In particular, the model adopted in this study has been tuned with data collected at the oil-red unit of Rossano Calabro, in the south of Italy. The traditional regulation implemented in the simulator and used as a standard in many applications is the classical co-ordinated control scheme schematically depicted in Figure 2, where at the right-hand side of the dotted vertical line there is the power regulation scheme, not detailed here for simplicity. The signals involved are listed in Table I, while the meaning of the blocks in Figure 2 is explained in Table II. The feedback regulators used in the scheme are mostly PI. Static nonlinear elements and logic switches are also used for performance enhancement and safety requirements. Finally, many loops are implemented according to a cascade structure. In particular, the PI regulator used to control the reheater steam temperature TRH produces the reference signal for the inner loop closed on the reheater attemperator.
Copyright # 2004 John Wiley & Sons, Ltd. Int. J. Robust Nonlinear Control 2004; 14:415433

418

C. AURORA ET AL.

Figure 2. The traditional co-ordinated regulation.

In order to get a benet from the application of MPC to the power plant control, reducing at the same time the impact of its introduction on the plant operators, in this study it has been decided to substitute only a limited number of traditional regulation blocks, taking into account the following considerations. 1. In thermal power plants, the main problems are related to the control of the boiler, due to the strong couplings between the manipulated (U2; U3; U4 in Figure 2) and the controlled (TSH ; TRH ; aRH ) variables. For this reason, a multivariable centralized control scheme is advisable. Conversely, the steam turbine, and the related power regulation, is a simpler SISO system, characterized by a faster dynamics, where the reference is the load request, and the control variable is the steam ow rate; in this case it is not necessary to replace the traditional local PID regulator;
Copyright # 2004 John Wiley & Sons, Ltd. Int. J. Robust Nonlinear Control 2004; 14:415433

PREDICTIVE CONTROL OF THERMAL POWER PLANTS

419

Table I. Variables list. Variable P TSH TRH aRH %O2 Description Main steam pressure Superheater outlet steam temperature Reheater outlet steam temperature Opening of the spray valve at the reheater Oxygen percentage in the exhausts

Table II. Co-ordinated regulation blocks. Block SP X Reg X req X Description Set-point generation for variable X Feedback regulator of variable X Request generation for variable X

2. A decentralized control scheme is not the best choice to include the overall eciency improvement in the control objectives. In particular, the use of the gas recirculation as control variable in a decentralized feedback structure is usually prevented by the typical non-minimal phase behaviour (inverse response) in the transfer function between the gas recirculation and the pressure P: Then, only a feed-forward action is usually performed. This strategy is non-ecient and often causes an excessive use of the reheater attemperator, which negatively aects the boiler eciency. For all the above reasons, it has been a priori decided to substitute the PIs regulating the pressure P and the temperature TSH (grey boxes in Figure 2) with the MPC control algorithm, which must also compute the two additional signals U3 and U4 acting on the gas recirculation request and on the reference signal for the reheater attemperator respectively. The goal is to achieve a tighter control action of P; TSH and TRH during dynamic transients caused by load variations and to compute the steady-state values, in particular the value of aRH ; in order to enhance the plant eciency and the life duration of the plant components. Note that in this new n n multivariable scheme, schematically represented in Figure 3 (where P n ; TSH and TRH are the setpoint signals) the gas recirculation request acts in closed-loop. The resulting overall control scheme can be viewed as a boiler following one, which can cause as a drawback fast and large pressure variations due to load changes, see Reference [10]. Then, one of the goals of this study is also to check whether the MPC approach can eciently cope with this potential drawback.

3. MPC ALGORITHM AND TUNING GUIDELINES The adopted algorithm is DMC (dynamic matrix control) [14], one of the most popular and widely used MPC methods, with hundreds of applications in the process industry, (see e.g. Reference [5]). For readers convenience, the main characteristics of algorithm are briey
Copyright # 2004 John Wiley & Sons, Ltd. Int. J. Robust Nonlinear Control 2004; 14:415433

420

C. AURORA ET AL.

Figure 3. The multivariable control scheme.

reported in the sequel, with reference to SISO (single input, single output) systems. For a more detailed presentation see Reference [14]. In DMC, the plant under control, with input u and output y; is asymptotically stable and described by the truncated nite impulse response (FIR) model yt
n X i1

hi ut i dt

where t is the discrete-time index, hi ; i 1; . . . ; n is the ith impulse response coecient, and dt represents a disturbance signal. Letting si be the ith step response coecient hi si si1 with s0 0 and dut ut ut 1 be the control increment, the model can also be written as yt k
k X i1

si dut k i dt k

n1 X ik1

si dut k i sn ut k n

At time t; the future value of the disturbance is assumed to be given by dt k dt yt


n1 X i1

si dut i sn ut n

Then, the k step ahead prediction of the output y is yt k


k X i1

si dut k i fk t

where the sum is made over the sequence of future control moves DUt dut0 dut 10 . . . dut M 10
Copyright # 2004 John Wiley & Sons, Ltd. Int. J. Robust Nonlinear Control 2004; 14:415433

PREDICTIVE CONTROL OF THERMAL POWER PLANTS

421

to be properly selected by the MPC algorithm along the control horizon M > 0; while the term fk t yt
n1 X ik1

si dut k i sn ut k n

n1 X i1

si dut i sn ut n

is known at time t: The DMC control law is derived by minimizing, at any time t and with respect to DUt; the cost function tN X J wi yi0 Qwi yi dui 10 Rdui 1 1
it1

where N > 0 represents the prediction horizon, while Q and R are positive denite matrices and w is the reference signal. The minimization is performed under the following constraints jut ij4um i 0; . . . ; M 1 hard constraints i50 hard constraints i 1; . . . ; N soft constraints 2 3 4

dut M i 0 jwt i yt ij4em

Once the optimization problem (1)(4) has been solved, and according to the Receding Horizon principle, only the rst term du0 t of the computed optimal sequence DU 0 t is applied, while the whole procedure is repeated at the next sampling time t 1: Besides implementing the above dynamic control algorithm also for multi-input, multi-output plants, the software package DMCplus used in this study allows to estimate the step response coecients si : The identied models are preliminarily used for the tuning and the testing of the MPC regulator. Once a good controller tuning has been achieved, the resulting regulator is subsequently applied in real-time operations. Among the main features of the method which have been proven to be crucial in this study, the following ones have to be recalled. 1. The denition of priorities in the constraints fulllment, by ranking the manipulated and controlled variables and, in case of unfeasibility, by violating the lower priority constraints. This feature has been used to specify the following priorities: aRH ; P; TSH ; TRH for the lower limits, P; TSH ; TRH ; aRH for the upper limits (the higher priority has been assigned to the lower limit of aRH ; which constitutes a true hard constraint; the pressure P follows, by virtue of its direct link with the generated power, with a working range of 5 bar; nally, the operating limits 158C have been set for the two temperatures TSH and TRH ; to prevent excessive stress of the steam turbine, without penalizing the performances during transients). 2. In the computation of the dynamic control action, the penalties (matrices Q and R in the performance index (1)) on the future error variables and control increments can be made to explicitly depend on the sign and amplitude of the error components. In the tuning procedure [4], the diagonal values of matrix Q have been preliminary set to the inverse of the squared nominal values of the corresponding controlled variables. Then an asymmetric weaker weight to the reheater attemperator aRH has been used to allow for the full usability
Copyright # 2004 John Wiley & Sons, Ltd. Int. J. Robust Nonlinear Control 2004; 14:415433

422

C. AURORA ET AL.

3.

4. 5.

6.

of the spray valve opening during transients, while it has been strongly penalized in steadystate optimization as specied in the following. As for the elements of matrix R; the main temperature control variable, i.e. the fuel pilot, has been slightly more penalized to prevent too high temperature gradients. The possibility to specify dierent modes of operation, i.e. the tracking of a xed constant set-point, or the fulllment of oscillation bands. Specically, the variables P; TSH and TRH 0 have been controlled in tracking with set-point values P0 170 bar; TSH 5408C and 0 TRH 5408C; while aRH has been forced to remain in a prescribed band with a minimum value of 38C: The use of feed-forward and disturbance signals, which has been useful to force an instantaneous control movement at any new load variation. The possibility to include explicit constraints on the future variations of the control variables. This has been crucial to limit the variations of U3; so as to reduce the oscillations induced by an excessive use of the gas recirculation. The identied model can be also used to compute the optimal steady state working point by solving a linear programming problem dened according to economic and safety criteria specied by the user, and under constraints on the values of the manipulated and controlled variables. Specically, this feature has been considered to include in the optimization problem the goal to maintain, in steady-state conditions, the reheater spray valve as close as possible.

4. IDENTIFICATION PROCEDURE It is well known that thermal power units are highly nonlinear plants, with gains and time constants which monotonically depend on the load, see Prasad et al. [8]. In order to assess the extent of the plant nonlinearities, to obtain an initial rough estimation of the dominant time constants involved, and to examine the plant cross couplings, a preliminary set of open-loop experiments has been performed. Specically, the PI regulators Reg P; Reg TSH and Reg TRH , and the gas recirculation open-loop request in Figure 2 have been disconnected and step changes have been imposed to the manipulated variables U1U4 starting from a steady state condition corresponding to the operating point of 240 MW: Some of the results achieved are reported in Figure 4. This gure shows that the single transfer functions exhibit a quite complex behavior, with time constants of dierent orders of magnitude, and with some inverse responses which make the control problem dicult. From these results, it has been computed the following plant relative gain array (RGA) matrix [15]. The RGA matrix shows that there are strong coupling between the control variables U2; U3; U4 and the controlled variables TSH ; TRH and aRH : Moreover, the use of U2 for control of TSH as in the decentralized scheme of Figure 2 causes signicant problems (the corresponding element in RGA is negative). On the contrary, the pressure P in steady state conditions mainly depends on the control variable U1; however it has been decided to include P between the variables controlled by MPC, so as to regulate the main plant variables according to a boiler following strategy. The extent of the plant nonlinearity can be appreciated with Figures 5 and 6, which show the transients of the controlled variables P; TSH and TRH in front of a step variation of amplitude 1% of the control variable U2 starting from dierent operating points. From this gure it is
Copyright # 2004 John Wiley & Sons, Ltd. Int. J. Robust Nonlinear Control 2004; 14:415433

PREDICTIVE CONTROL OF THERMAL POWER PLANTS

423

Figure 4. Plant step responses.

Figure 5. Response of P to a step variation in U2: Operating points: thin line 300 MW; bold line 220 MW; dashed line 140 MW:

apparent that the plant is much slower at low loads than at full load. Moreover, the plant gains vary signicantly in the operating range. The same conclusion could be drawn by examining the other links between the control and controlled variables, and are in perfect agreement with the results reported in Reference [8]. From Figure 4, it can be seen that at 240 MW; the dominant time constants range from a minimum of about 90 s for the pressure to a maximum of about 1250 s for the superheat steam temperature. In view of the fast pressure dynamics, the sampling time used in identication and control has been set equal to 15 s and the identied step response, single-input, single-output,
Copyright # 2004 John Wiley & Sons, Ltd. Int. J. Robust Nonlinear Control 2004; 14:415433

424

C. AURORA ET AL.

Figure 6. Response of superheater and reheater temperatures TSH and TRH to a step variation in U2: Operating points: thin line 300 MW; bold line 220 MW; dashed line 140 MW:

FIR models have been composed by 480 coecients, which is sucient to consider the settling time of all the controlled variables. These models have been identied by using as input variables the signals U1U4 obtained combining step variations, for a good identication of the static gains, and PRBS (pseudo random binary signals), necessary to excite the system in the frequency range of interest. The amplitude of the PRBS has been limited to 510% of the nominal values of the corresponding signals, which represents a feasible perturbation also for real plants. Other characteristics have been selected according to the guidelines given in Reference [16].

5. EXPERIMENTAL RESULTS Many simulation experiments have been performed to assess the performance of the MPC algorithm. Among them, in the sequel only three sets are reported for conciseness. The rst one refers to ramp variations from 140 to 320 MW; and it is here discussed to illustrate the ability of the MPC algorithm to cover all the operating range in spite of the plant nonlinearities. The second set of experiments shows the performance of the predictive controller in front of a 40 MW step-like variation of the load, which represents a critical test for the control algorithm. Indeed, it is important to check the performance of MPC with respect to these perturbations in view of the higher and higher rates of load variations required by the market. Finally, the third set of simulations presents the results achieved by forcing a prole of load variations recorded on a real plant during a period of 4 days. This is useful to draw some conclusions on the average performance of MPC in standard
Copyright # 2004 John Wiley & Sons, Ltd. Int. J. Robust Nonlinear Control 2004; 14:415433

PREDICTIVE CONTROL OF THERMAL POWER PLANTS

425

operating conditions and to compare them to that provided by the classical co-ordinated control scheme. 5.1. Experiment 1 Starting from the operating regime corresponding to the minimum load 140 MW; the load request is brought to the maximum value of 320 MW at a rate of 3:2 MW=min: Correspondingly, the transients of the main control and controlled variables are shown in Figures 79. From these gures, it is apparent that MPC produces a much tighter control action than the co-ordinated scheme, as also witnessed by the value of the root mean square (RMS) error of P; TSH ; TRH computed for the two algorithms and reported in Table III. This Table also shows the mean value of aRH and of the gas recirculation request U3: It is apparent (see also Figure 9) that in steady state conditions MPC maintains more closed the spray attemperator valve by means of a better use of the gas recirculation (by virtue of the closed loop regulation of this variable, not implemented in the traditional control scheme), so increasing the plant thermal eciency. On the contrary, during the transient caused by a load variation, the valve is used to achieve a better and faster control action. 5.2. Experiment 2 A load step variation of 40 MW has been imposed starting again from the operating point of 240 MW: The obtained results are reported in Figures 1012, while the RMS and mean values of the main variables are summarized in Table IV.

Figure 7. Experiment 1: load request and steam main pressure P with the traditional co-ordinated control scheme (thin line) and the MPC regulator (bold line).
Copyright # 2004 John Wiley & Sons, Ltd. Int. J. Robust Nonlinear Control 2004; 14:415433

426

C. AURORA ET AL.

Figure 8. Experiment 1: superheater TSH and reheater TRH temperatures with the traditional co-ordinated control scheme (thin line) and the MPC regulator (bold line).

Figure 9. Experiment 1: reheater spray valve opening aRH and gas recirculation ow with the traditional co-ordinated control scheme (thin line) and the MPC regulator (bold line).

Copyright # 2004 John Wiley & Sons, Ltd.

Int. J. Robust Nonlinear Control 2004; 14:415433

PREDICTIVE CONTROL OF THERMAL POWER PLANTS

427

Table III. Performance of the controllers in Experiment 1. RMS MPC P TSH TRH aRH Flow gas recirculation 0:222 bar 1:1368C 0:5978C Co-ordinated 0:452 bar 1:4428C 1:5208C 10.1% 39:281 kg=s 16.2% 54:745 kg=s MPC Mean value Co-ordinated

Figure 10. Experiment 2: load and steam main pressure P with the traditional co-ordinated control scheme (thin line) and the MPC regulator (bold line).

In this case, it can be noted that the pressure P exhibits a larger undershoot with MPC then with the classical control scheme of Figure 2, as expected in view of the boiler following nature of the MPC implementation. However, its variation of about 3 bar is largely inside the prescribed band of 5 bar: Conversely, the regulation of the temperatures is signicantly improved as well as the use of the reheater spray valve and of the gas recirculation. 5.3. Experiment 3 The recorded transients of the load request registered along a period of 4 days in a real plant have been used to test the performance of MPC in real operating conditions. The results reported in Figures 1318 and in Table V conrm the conclusions drawn from Experiments 1
Copyright # 2004 John Wiley & Sons, Ltd. Int. J. Robust Nonlinear Control 2004; 14:415433

428

C. AURORA ET AL.

Figure 11. Experiment 2: superheater TSH and reheater TRH temperatures with the traditional co-ordinated control scheme (thin line) and the MPC regulator (bold line).

Figure 12. Experiment 2: reheater spray valve opening aRH and recircle gas ow with the traditional co-ordinated control scheme (thin line) and the MPC regulator (bold line).

Copyright # 2004 John Wiley & Sons, Ltd.

Int. J. Robust Nonlinear Control 2004; 14:415433

PREDICTIVE CONTROL OF THERMAL POWER PLANTS

429

Table IV. Performance of the controllers in Experiment 2. RMS MPC P TSH TRH aRH Flow gas recirculation 0:477 bar 0:5898C 0:6188C Co-ordinated 0:319 bar 1:0668C 0:9248C 5.50% 23:031 kg=s 14.2% 40:174 kg=s MPC Mean value Co-ordinated

Figure 13. Experiment 3: load during the long-period test.

Figure 14. Experiment 3: main pressure P with the traditional co-ordinated control scheme (upper graph) and the MPC regulator (lower graph).
Copyright # 2004 John Wiley & Sons, Ltd. Int. J. Robust Nonlinear Control 2004; 14:415433

430

C. AURORA ET AL.

Figure 15. Experiment 3: superheater TSH temperature with the traditional co-ordinated control scheme (upper graph) and the MPC regulator (lower graph).

Table V. Performance of the controllers in Experiment 3. RMS MPC P TSH TRH aRH Flow gas recirculation 0:420 bar 0:7908C 0:8028C Co-ordinated 0:316 bar 1:4488C 1:2858C 8.79% 29:058 kg=s 17.20% 48:796 kg=s MPC Mean value Co-ordinated

and 2. Specically, the regulation of the pressure provided by MPC is worse than that of the traditional decentralized scheme, although the variations with respect to the set-point value are largely inside the prescribed band. Conversely, the steam temperatures TSH and TRH are controlled much better and the use of the reheater spray valve and of the gas recirculation is optimized. In this experiment, the specic fuel consumption has also been computed, it is 14:2864 kg=s for the co-ordinated scheme and of 14:2530 kg=s for the MPC regulator; this reduction is equivalent to a considerable saving of money. Moreover, on the basis of a qualitative analysis, it is possible to appreciate a pollutants decrease, directly consequent on the fuel consumption reduction (Figures 1618).
Copyright # 2004 John Wiley & Sons, Ltd. Int. J. Robust Nonlinear Control 2004; 14:415433

PREDICTIVE CONTROL OF THERMAL POWER PLANTS

431

Figure 16. Experiment 3: reheater TRH temperature with the traditional co-ordinated control scheme (upper graph) and the MPC regulator (lower graph).

Figure 17. Experiment 3: reheater spray valve opening aRH with the traditional co-ordinated control scheme (upper graph) and the MPC regulator (lower graph).

Copyright # 2004 John Wiley & Sons, Ltd.

Int. J. Robust Nonlinear Control 2004; 14:415433

432

C. AURORA ET AL.

Figure 18. Experiment 3: recircle gas ow with the traditional co-ordinated control scheme (upper graph) and the MPC regulator (lower graph).

6. BENEFITS FOR INDUSTRIAL APPLICATIONS AND CONCLUSIONS This study allows to draw the following conclusions on the use of MPC for control of thermal power plants. 1. MPC produces a signicant increase of the overall thermal eciency. This is achieved with a more ecient management of the reheater spray valve and of the gas recirculation. In turn, this improved eciency leads to a reduction in the fuel consumption and to signicant economic savings. Moreover, a reduced fuel consumption also means less pollutants in the atmosphere: this is of major importance in view of the more and more stringent legal requirements on the pollutants emissions; 2. The improved regulation of the steam temperatures TSH and TRH produces minor thermal stresses of the materials, which means reduced costs for the maintenance and supervision of the plant. Moreover, this tighter control action could allow for higher set-points for TSH ; TRH ; with an improvement of the plant thermal eciency; 3. Thanks to the better overall regulation, it is possible to work on a wider operating range and to considerably increase the exibility in the management of the production units; in turn, this allows to quickly change the operating conditions and, consequently, to have access to dynamic services; 4. The optimal regime can be computed by minimizing an economic criterion which can be frequently modied to cope with changing external conditions; 5. The synthesis procedure is easier and better documented than in the case of co-ordinated decentralized control schemes. This has a great importance in the management of power plant units.
Copyright # 2004 John Wiley & Sons, Ltd. Int. J. Robust Nonlinear Control 2004; 14:415433

PREDICTIVE CONTROL OF THERMAL POWER PLANTS

433

Finally, it has to be noted that the MPC algorithm has been used to substitute only a part of the classical co-ordinated control scheme, so that further advantages could be achieved by using it as the unique high level controller generating the reference signals for lower level traditional single-input single-output regulators.

ACKNOWLEDGEMENTS

The authors gratefully acknowledge the support of Dr. Aleotti, of Aspentech, for providing the software package DMCPlus used in this research.

REFERENCES 1. Garcia CE, Prett DM, Morari M. Model predictive control: theory and practice}a survey. Automatica 1989; 25:335348. 2. Mayne DQ, Rawlings JB, Rao CV, Scokaert POM. Constrained model predictive control: Stability and optimality. Automatica 2000; 36:789814. 3. Camacho EF, Bordons C. Model Predictive Control in The Process Industry. Springer: Berlin, 1999. 4. Maciejowski J. Predictive Control with Constraints. Prentice-Hall: Englewood Clis, NJ, 2001. 5. Qin SJ, Badgwell TA. An overview of industrial model predictive control technology. Fifth International Conference on Chemical Process Control, Kantor JC, Garcia CE, Carnahan B (eds). 1996; 232256. 6. Rovnak JA, Corlis R. Dynamic matrix control for fossil power plants. IEEE Transactions on Energy Conversion 1991; 6(2):320326. 7. Prasad G, Swidenbank E, Hogg BW. A neural net model-based multivariable long-range predictive control strategy applied in thermal power plant control. IEEE Transactions on Energy Conversion 1997; 13(2):176182. 8. Prasad G, Swidenbank E, Hogg BW. A local model networks based multivariable long-range predictive control strategy for thermal power plants. Automatica 1998; 34(10):11851204. 9. Prasad G, Irwin GW, Swidenbank E, Hogg BW. Plant-wide predictive control for a thermal power plant based on a physical plant model. IEEE Proceedings}Control Theory Applications 2000; 147(5):523537. 10. Lu S, Hogg BW. Predictive co-ordinated control for power-plant steam pressure and power output. Control Engineering Practice 1997; 5(1):7984. 11. Oluwande G, Boucher AR. Implementation of a multivariable model-based predictive controller for superheater steam temperature and pressure control on a large coal-red power plant. European Control Conference, Karlsruhe, 1999. 12. Castiglioni L, De Chirico M, Pretolani F, Spelta S. A real time simulator implementation in the unix environment. BIAS 25th edition, Milano, 1993. 13. Cori R, Spelta S, Guagliardi G, Pretolani F, Maltagliati P, Persico F, Sommani M. The legocad system: a computer aided power plant modelling environment (in italian). Proceedings of XC National Symposium AEI, Lecce, 1989. 14. Cutler CR, Ramaker BC. Dynamic matrix control}a computer control algorithm. Proceedings Joint Automatic Control Conference, San Francisco, CA, 1980. 15. Bristol EH. On a new measure of interactions for multivariable process control. IEEE Transactions on Automatic Control 1966; 11:161166. 16. Rivera DE, Jun KS. An integrated identication and control design methodology for multivariable process system application. IEEE Control Systems Magazine 2000; 20:2537.

Copyright # 2004 John Wiley & Sons, Ltd.

Int. J. Robust Nonlinear Control 2004; 14:415433

You might also like