An expert system for startup optimization of combined cycle power plants under NO/sub x/ emission...

414 IEEE Transactions on Energy Conversion, Vol. 11, No. 2, June 1996

AN EXPERT SYSTEM FOR STARTUP OPTIMIZATION OF COMBINED CYCLE POWER PLANTS

UNDER NOx EMISSION REGULATION AND MACHINE LIFE MANAGEMENT

H. Matsumoto, Member, IEEE Y. Ohsawa

S. Takahasi

Hitachi Research Laboratory, Omika Works, Hitachi, Ltd. Hitachi, Ltd.

832-2 Horiguchi, Hitachinaka-shi, Ibaraki-ken, 312 Japan

5-2-1 Omika-cho, Hitachi-shi Ibaraki-ken, 319-12 Japan

Abstract - This paper proposes an expert system which optimizes the startup schedule for gas and steam turbine combined cycle power plants. The speed-up and load-up pattem of the plant is automatically optimized through an iterative process. Plant dynamics models representing quantitative knowledge and fuzzy rules representing qualitative knowledge are alternately used in the optimization process to modify the schedule parameters. The rules represent expertise on causal relations between modification rates of the schedule parameters and operational margins for constraints, i.e. NOx emission and machine thermal stresses. Simulation analysis with a three pressure staged reheat type 235.7MW rated capacity plant shows that the system provides quick and economical plant startup under NOx emission regulation and reliable machine life management. Startup energy loss is reduced due to the reduction in startup time. Furthermore, optimum operating conditions are quickly reached with the expert system.

INTRODUCTION

Harmonizing machines operations is required not only for economical reason, but also for its environmental advantages, especially with the growth in capacity and complexity of gas and steam turbine combined cycle power plants. In accordance with the expanding power generation share of nuclear plants, fossil plants undergo heavy duty operation, involving frequent startup and shutdown, and quick load following. Furthermore, the fossil plants are required to meet stringent environmental regulations regarding air pollution [ 11. Consequently, time reduction for startup, and safe and economical operation under environmental regulations and reliable machine life management have become important needs [2]. Combined cycle power plants are receiving attention because of their potential advantages of higher efficiency, quicker startup, lower environmental pollution, shorter construction time, smaller site space, etc. An important topic in plant control systems, then, is finding a way to reduce the startup time without shortening the machine lifetimes and exceeding the environmental restrictions. Decisive factors for quick startup of the plant are thermal stresses which develop in the steam turbine and heat recovery steam generator (HRSG), and NOx emission from the HRSG.

To reduce the startup time, off-line scheduling and on-line control methods are used as state-of-the-art techniques. As for the off-line scheduling method developed for simple-steam-cycle plants, a startup schedule is generated through a metal matching chart using steam and

96 WM 028-1 EC A paper recommended and approved by the IEEE Energy Development and Power Generation Committee of the IEEE Power Engineering Society for presentation at the 1996 IEEUPES Winter Meeting, January 21-25, 1996, Baltimore, MD. Manuscript submitted July 19, 1994; made available for printing January 4, 1996.

T. Akiyama 0. Ishiguro

Hitachi Works, Kure Research Laboratory, Hitachi, Ltd. Babcock-Hitachi K.K.

3-36 Takara-maci, Kure-shi, Hirosima-ken, 737 Japan

3- 1 - 1 Saiwai-cho, Hitachi-shi, Ibaraki-ken, 3 17 Japan

metal temperatures just before startup [3j. Conventional control systems of combined cycle plants have basically applied the same method. In the on-line control method developed for startup of steam turbines, speed-up and load-up rates are periodically optimized using the estimated rotor stresses from measured temperatures of turbine casings [4]. To improve the accuracy and the stability of stress control, a turbine automatic startup system (HITASS) was developed 151. This system optimizes the speed-up and load-up rates periodically based on predicted turbine stresses using a plant dynamics model which consists of an autoregressive boiler model and a turbine heat transfer model.

However, through off-line scheduling methods, time required for startup tends to be prolonged, although an accurate startup completion time can be obtained. On the other hand, through on-line control methods, the startup completion time is less reliable, but the startup time is effectively reduced. It is important, but difficult to solve these two problems at the same time within the scope of conventional computer control technology, because of complicated physical relationships among process variables in power plants.

To satisfy these requirements, we focused on a synergetic approach with dynamics simulation and knowledge engineering. We have already proposed an expert system for operation support in simple- steam-cycle plants to reduce the work burden of plant operators in making and executing startup schedules [6j. Results obtained by this approach showed that knowledge engineering is a promising technology for utilizing knowledge and the way experts, such as operators, plant engineers, and control engineers, think.

This paper describes an expert system which is applied to combined cycle power plants in which there are complicated physical relationships between stresses in the steam turbine and HRSG, and NOx emissions appear during the startup process. The system optimizes the startup schedule automatically. Consequently, it provides quick and economical plant startup under NOx emission regulation and reliable machine life management. The system effectiveness is verified through simulation analysis with a three pressure staged reheat type 235.7 MW rated capacity plant.

- SYSTEM DESIGN CONCEPT

Operational Constraints

There are many operational constraints in combined cycle power plants. Stresses which develop in the steam turbine and HRSG, and NOx from the HRSG are especially decisive factors for reducing startup time.

The turbine stresses occur on the rotor surfaces and rotor bores at the labyrinth packings behind the first stage of the high pressure turbine (HPT) and intermediate pressure turbine (IPT). In these areas, a heavy heat flux appears from the surface toward the bore during the startup, because of leaked steam having a high temperature and high velocity. The stresses of the steam turbine involve thermal and mechanical stresses. Thermal stress is caused by temperature difference between the surface and bore of the rotor. Mechanical stress is caused by centrifugal force in the rotor. Then, the more rapid the turbine

0885-8969/96/$05.00 0 1996 IEEE

415

it is difficult to get optimum solutions by ordinary computers, because plant procedures are dynamic, large scale, nonlinear and quantitative. Also, mathematical methods cannot easily handle object knowledge efficiently. Therefore, dynamics simulation and modeling of human expertise such as heuristic, fuzzy and qualitative thinking process are integrated by an AI technique to obtain quick solutions as shown in Fig. 1. In this approach, fuzzy reasoning is introduced into a qualitative evaluation of stress and NOx patterns, and schedule modifications.

Startup is, the greater the stress developed, and the more serious is its effect on shortening the service life of the turbine.

The HRSG stresses occur on the inner and outer surfaces at the headers of the high pressure super heater outlet and the reheater outlet due to quick and wide changes of the steam temperature and pressure during the startup. The stresses of the HRSG involve thermal and mechanical stresses. Thermal stress is caused by temperature difference between the inner and outer surfaces of the header. Mechanical stress is caused by the steam pressure. Then, the more rapid the gas turbine and HRSG startups are, the greater the stress developed, and the more serious is its effect on shortening the service life of the HRSG.

The NOx emission rate of the plant depends on dynamics characteristics of the gas turbine and the denitrifier which is installed in the HRSG. The NOx emission from the gas turbine is affected by its speed and load which generally depend on fuel input rate. Furthermore, NOx reduction rate of the denitrifier is affected by its catalyst temperature which strongly depends on the flue gas temperature. Then, the faster the gas turbine startup is, the less the NOx reduction is, due to the larger time delay of the catalyst temperature rise.

It is very important to know and manage these stresses and NOx emission in connection with startup schedules.

Expertise on Plant Operation

We consider how to optimize the startup schedule in relation to the stresses and NOx dynamics. Here, optimization means making the shortest schedule without exceeding any stress or NOx limits. After the gas turbine is ignited, the gas turbine, the steam turbine and the generator, which are connected by one shaft, increase their speed toward the rated speed. Then the generator is paralleled into the power system, and loaded up toward a target level. Then the startup session is completed.

In general, experienced operators and plant engineers can appropriately modify the schedule if a stress pattem and a NOx pattem are provided. They may reduce time periods where stress or NOx margins exist, or conversely extend them where excess stress or NOx appears. Therefore, we want to know how this system can solve the problem automatically.

System Framework

We think that well known mathematical methods such as nonlinear programming using dynamics simulation, are too time consuming and

Time Consuming 0

Quick Solution 0

Plant Properties Human Expertise Large Scale Heuristic

Knowledge Nonlinear 0 Qualitative 0 Quantitative n

Without Object 0 Dynamic Fuzzy

Fig. 1 System Framework

Strategies of Schedule Optimization

This system introduces strategies as shown in Fig. 2 to modify the schedule parameters which designate plant startup pattern. Startup schedules for the gas turbine and steam turbine are modified by predicting stresses of the steam turbine and HRSG, and NOx through the dynamics models.

The schedule for the gas turbine is modified by Gas Turbine Primal Scheduling (GTPS) which considers the NOx dynamics, and Gas Turbine Global Tuning (GTGT) which considers the stress dynamics. The schedule for the steam turbine is modified by Steam Turbine Primal Scheduling (STPS) which considers the stress dynamics, and Steam Turbine Local Tuning (STLT) which considers the NOx dynamics. The GTPS primary function in modification of the gas turbine startup schedule, because of direct effects on the NOx emission. The STPS primary function is modification of the steam turbine startup schedule, because of direct effects on the stresses. The GTGT tunes the gas turbine startup schedule globally, because of indirect effects on the stresses through heat exchangers in the HRSG. The STLT tunes the steam turbine startup schedule locally, because of indirect effects on the NOx emission through the heat exchangers.

Aforementioned expertise on plant operations is introduced to these strategies using fuzzy reasoning to represent the knowledge and the way experts think, and to utilize the strategies to reach the optimum schedule quickly.

GTGT I I

GTPS: GT Primal Scheduling GTGT: GT Global Tuning STPS : ST Primal Scheduling STLT : ST Local Tuning

-1 - - - - - - - - - - - - - - -

Plant Dynamics Models I & I *

I GT : Gas Turbine I ST : Steam Turbine

I

I

I

HRSG : Heat Recovery Steam Generator

Fig. 2 Strategies of Schedule Optimization

ALGORITHM OF SCHEDULE OPTWITZATION

Functional Structure of the System

416

The functional structure of the system and configuration of the plant are shown in Fig. 3. The system consists of functions for startup schedule optimization and machine control. The latter executes schedules which are generated by the function for schedule optimization. The main process of schedule optimization consists of functions for schedule assumption, dynamics models, dynamics evaluation, convergence judgment, schedule modification, schedule presentation, and optimum schedule setting.

The object plant with which the system is studied has a three pressure staged reheat type 235.7 MW rated capacity. The denitrifier is installed between two high pressure evaporators. Speed and load of the gas turbine are regulated by the fuel control valve (FCV). The steam turbine consists of a high pressure turbine (HPT), an intermediate pressure turbine (IPT) and low pressure turbine (LPT). The steam inlet flows to the turbines are regulated by a high pressure control valve (HPCV), an intermediate pressure control valve (IPCV), and a low pressure control valve (LPCV). The turbine bypass steam flows are regulated by a high pressure bypass valve (HPBV), an intermediate pressure control valve (IPBV), and a low pressure control valve (LPBV). Temperature of the steam from the intermediate pressure drum is adjusted to the HPT outlet steam temperature by an intermediate

Center

Plant Control Svstem

! ! Schedule Optimization ! I Dynamics Models ! I

1 i-..

['Machine Control System

I i ST Control

FCV: HPCV: IPCV: LPCV:

Fuel Control Valve High Press. Control Valve Intermediate Press. Control Valve Low Press. Control Valve

HPBV: High Press. Bypass Valve IPBV: Intermediate Press. Bypass Valve LPBV: Low Press. Bypass Valve IPSV: Intermediate Press. Stop Valve

Fig. 3 Functional Structure of the System

pressure stop valve (IPSV). The control center of the power system commands a startup

completion time and a target load level as a startup demand. The function for schedule assumption makes an initial schedule along with the command, and sends it to the dynamics models. In the dynamics models, the plant startup process is simulated as if the gas turbine, the HRSG and the steam turbine were started in accordance with the initial schedule. The stresses of the steam turbine and the HRSG, and NOx which are also simulated with the model, are forwarded to the function for dynamics evaluation. This function for dynamics evaluation calculates margins for the operational constraints and the startup time. Then, the function for convergence judgment decides whether the schedule has reached to the optimum point or not. The optimum point means the shortest schedule without exceeding any stress or NOx limits. Therefore, the initial schedule may not be the optimum, so the evaluated margins must be forwarded to the next function for schedule modification. The schedule is modified by fuzzy reasoning in this function using fuzzy rules prepared through the strategies of the schedule optimization described before. Then, the modified startup schedule is sent to the dynamics model again. Through the iterative manner described before, convergence to the optimum schedule can be expected. Convergence of the optimization is judged when the

Steam 2onditior

GT

7 Main Stesm Press. x27 I /

4 1x28 HPT Outlet Press. !

I -

<Object Parameters of Schedule Optimizatlon > xi: Changing Rate of Speed x2: Hold Time at Rated Speed x3: Initial Load of GT x4: Hold Time at Initial Load of GT x5: First Stage Load - up Rate of GT x6: Hold Time at Second Stage Load of GT xi': Second Stage Load - up Rate of GT x8: Third Stage Load - up Rate of GT x9: Closing Speed of HPBV XlO: Closing Speed of IPBV xi 1: First Stage Opening Speed of HPCV x12: Second Stage Opening Speed of HPCV

Fig. 4 Plant Startup Schedule Parameters

417

- - - __ - - - - - -

reduction rate of startup time by the function for schedule modification becomes sufficiently small.

After finding the optimum schedule, the schedule parameters are sent to the function for machine control by the function for optimum schedule setting. The iterative search process for the optimization is presented on a CRT display through the function for schedule presentation as a man-machine interface. The schedule optimization is executed by a unit computer for each generation unit, or a dedicated computer for the startup scheduling. Therefore, nothing needs to be modified in the major functions of the conventional machine control systems.

__ - - NOX Outlet

Instantaneous

Schedule Parameters

Gas Temp, 0 I

Schedule parameters (Xi - x28) which designate startup pattems of the plant are shown in Fig. 4. Among these, 12 parameters (Xi - X22) are selected as optimizing objects considering their effective on the operational constraints, consequently, the time reduction for startup. The other parameters for control and operating timing are defined in a basic engineering phase.

Textbookish optimization algorithms which need hundreds of iterative calculations for optimization of 12 parameters are time consuming and not practical. A heuristic optimization applying fuzzy reasoning is attempted to obtain rapid convergence. By knowing the stress and NOx pattem, experienced operators can modify the schedule parameters easily. Such heuristic knowledge is used for the schedule optimization process.

Denitrification Moving I Average Reaction I fi-*

I

Plant Dynamics Models

A dynamics model of the HRSG is produced by considering transient heat and mass balances in all heat exchangers, drums, pipings, and gas ducts. The gas turbine and steam turbine are simulated as static models, except for the steam turbine stresses, because of quicker responses compared with that of the HRSG to study startup characteristics.

Thermal stresses of the steam turbine are generated by temperature differences between the rotor surface and the bore. If the rotor temperature distributions are known, they can be calculated. However, it is difficult to measure these distributions directly, and the corresponding casing temperatures do not always coincide with the rotor temperatures. Furthermore, even measuring the steam temperature behind the turbine first stage is difficult because of its mechanical structure. Considering these difficulties in measuring the temperature, dynamics models of the steam turbine rotor stresses are developed. The main calculation steps of the models are estimation of steam conditions and heat transfer coefficients at the rotor surfaces using steam conditions, i.e. temperature and pressure, at the inlets of the control valves, and computing temperature distributions and stresses at the rotor surfaces and bores. The rotor stresses combine the thermal stresses with the mechanical stresses which are generated by the centrifugal force corresponded to the rotating speed. Details of the model can be found in reference [5].

Thermal stresses of the HRSG can be calculated in a similar manner to the turbine stresses. The main calculation steps are estimation of heat transfer coefficients at the inner surface of the headers using the steam flow rates and steam conditions, i.e. temperature and pressure, and calculation of the temperature distributions and stresses of inner and outer surfaces of the headers. The stresses of the HRSG combines the thermal stresses with mechanical stresses which are generated by the steam pressure.

The denitrifier model consists of an equipment model and its controller model as shown in Fig. 5. The NOx from the gas turbine is reduced by a complicated denitrification process under dynamics of temperature distributions of the flue gas and catalyst, and distribution of absorbed ammonia (NH3) in the catalyst. Therefore, the equipment model is divided into ten sections in the direction of gas flow. Each

section has calculation modules for heat balance and the denitrification reaction. Instantaneous values of the NOx emission and NH3 leakage, and moving average of NOx emission are obtained from this equipment model. The controller model is used to simulate control actions which determine the NH3 injection rate. All the input variables of this denitrification model come from the dynamics model of the HRSG.

GT Load 0 tl i

Hy6 8-4 Controller Model I

I I T Heat Balance I I Equipment Model A

Fig. 5 Denitrifier Model

Evaluation of Startup Dynamics

The whole startup period is divided into time sections to evaluate the startup dynamics; five sections for the stresses, and six for the NOx as shown in Fig. 6. The whole startup period t8 is defined as follows.

t8 = MaX ( t81, t82 ) (1 )

t81 = # + te t82 = t9 i te n: Time needed for gas turbine startup which is the time

t9: Time needed for plant startup which is the time from

te: Time extension for evaluation (1 800 seconds is assumed

from gas turbine ignition to reach the rated load

gas turbine ignition to reach 97% of the steam turbine load

in this system)

The margin (m) for each variable is generally defined as follows.

GTLoad

I x34

to t l t2 t3 t4 t5 t6 t7 t8

Fig. 6 Time Sections for Dynamics Evaluation

41%

m = S L - S (2)

S: Calculated variable through dynamics models SL: Operational limit

mT -

Minimum stress margins calculated for each time section (i=I-5) about four locations of critical stress are designated as follows.

mHS(i): Minimum stress margin at HPT rotor surface m ~ ~ ( i ) : Minimum stress margin at HPT rotor bore mrs(i): Minimum stress margin at IPT rotor surface mIB(i): Minimum stress margin at IPT rotor bore

Stress margins of the HRSG can be evaluated in a similar manner to the turbine stresses. Locations of critical HRSG stresses are the outer surfaces of headers of the high pressure super heater and reheater. Minimum stress margins calculated for each time section (i=I- 5 ) are designated as follows.

m w ( i ) : Minimum stress margin at outer surface of the high

mIHD(i): Minimum stress margin at outer surface of the reheater pressure super heater header

header

NOx margins for each time section (i=I-6) can be evaluated in a similar manner to the turbine stresses. In this case, margins for the instantaneous value and moving average are calculated. Minimum NOx margins calculated for each time section are designated as follows.

mPs(i): Minimum NOx margin for instantaneous value mPA(i): Minimum NOx margin for moving average

Critical values of the margins should be noted in startup the plant without exceeding any operational limits. Therefore, the respective smallest values are selected from the turbine stresses margins, the HRSG stresses margins and the NOx margins as follows.

mT = Min ( -mHS, mHB, -mIS, mIB ) mB = Min ( m m , mlHD ) mp =Min ( mps, mPA )

These mT, m~ and mp are used for the fuzzy reasoning.

Schedule Modification by Fuzzy Reasoning

The schedule modification algorithm using multi-fuzzy reasoning is structured as shown in Fig. 7 based on the aforementioned system concept. Fuzzy reasoning for the STPS and GTGT regulate the turbine and HRSG stresses, and STLT and GTPS regulate the NOx emission. The fuzzy reasoning provides candidates ( A X r ( i ) , AxB( i ) (i=1,4- 12), AXp(i) (i=I-11)) for modification rates ( A X ( i ) (i=l-12)) of the schedule parameters X(i) (i=I-12). Here, the subscripts T, B and P mean their candidates gotten from the margins mT, mB and mp,

respectively. Final modification rates ( AX( i ) (i=I-12)) are selected from these candidates through priority decision gates which consist of high value gates and low value gates. These gates select safer values to keep the process variables within the limits. The larger the parameters X(4) and X ( 6 ) become, the safer the startup is, because these two parameters correspond to load hold periods of the gas turbine. The other parameters hasten the startup if they are reduced.

The margins mT, m and mp are evaluated qualitatively using membership functions as shown Figs. 8(a-1) and (b-1) to extract qualitative features of the dynamics characteristics of the plant. The membership functions shown in Figs. 8(a-2) and (b-2) are used to determine schedule modification coefficients ks and kc. The meaning of ks and kc are given later. The k s and kc in Fig. 8(a-2) are

mF -

Turbine Stress I Regulation i

AXT(I) h X ~ ( 1 ) AXP(I) A Xp(2)

AXp(3)

A X B ( ~ ) AXp(4)

A XT(4)

A XT(5) A XB(5) A Xp(5)

A XT(6) A xB(6) A XP(6)

A XT(7) A X B ( ~ ) A Xp(7)

A XT(8) A XB(8) A xP(8)

HRSG Stress I Regulation .- j

Fuzzy Reasoning

AXT(IO)+ L AXB(IO)+ v AXP(10)+ D G

A XT( 12) h X ~ ( 1 2 )

I N Ox 1 Priority Regulation i 1 Decision Gate:

U

mT : Turbine Stress Margin mB : HRSG Stress Margin mP : NOx Margin

HVG : High Value Gate LVG : Low Value Gate

Fig. 7 Schedule Modification Algorithm

determined by the fuzzy reasonings for STPS and GTGT using mT and m ~ , respectively. The ks and k c in Fig. 8(b-2) are determined by the fuzzy reasonings for GTPS and STLT, respectively, using m p . PB, PS, ZO, NS, NB are symbols attached to the membership functions. They have the following qualitative meanings.

PB: Positive Big PS: Positive Small ZO Zero NS: Negative Small NB: Negative Big

The extracted qualitative features of the dynamics characteristics of the plant are compared with fuzzy rules. Example rules are shown in Tables l(a) - (d) which are used by the fuzzy reasoning for STPS, GTGT, GTPS and STLT, respectively. Rules in the tables give fragmentary knowledge for determining schedule modification rates in

419

relation to stress and NOx pattems. Considering effects of the schedule parameters on the startup dynamics is important for making the fuzzy rule tables of minimum size and good efficiency. These tables were generated on the basis of this concept. The rules in Table l(a) are used for determining the schedule modification coefficients k s ( l ) , ks (2) , ks(3) and ks(4) in qualitative relations with two steam turbine stress margins out of mT(2), mT(3), mT(4) and mT(5). The rules in Table 1 (b) are used for determining the schedule modification coefficients k c ( l ) , kc(4), kG(5), kc(6), kc(7) and kc(8 ) in qualitative relations with two steam turbine stress margins out of mT(I), mT(2), mT(3), mT(4) and mT(5). These two rule tables are similarly used for the STPS and GTGT in the functions of HRSG stress regulation.

The rules in Table I(c) are used for determining the schedule modification coefficients k c ( l ) , kc(2), kc(3), kc(4) , kc(5) , kc(6) , kc(7) and kc(8) in qualitative relations with two NOx margins out of mP(I) , m ~ ( 2 ) , mp(3), m ~ ( 4 ) , mp(5) and m ~ ( 6 ) . The rules in Table l(d) are used for determining the schedule modification coefficients k s ( l ) , ks (2) and ks(3) in qualitative relations with two NOx margins out of mP(l), mp(2) and m ~ ( 3 ) . Null parts in these tables mean that relations between the schedule parameters and the operational margins are negligibly small.

Plural membership functions are selected for each schedule modification coefficient after pattem matching of the qualitative features of the startup dynamics with these rule tables. After that, a single value for each schedule modification coefficient is defined by calculation of the balanced point of the selected membership functions. An example of the calculation for schedule modification coefficient ks is shown in Fig. 9. In this case, the value of the balanced point ksc is actually used to regulate the steam turbine stresses. The following equation gives the balanced point ksc.

o/m~(4) Olmp(5)

NS ZO PS

I I

- 3 0 3 5 10 20

NS zo PS PB

~~1-s I PE 1 NS [ZO 1 PS PB 1 NS I zo 1 PS 1 PB I N S /ZOPS/Pe

- 1 mT, me (kg/mmz)

(1) Stress Margin Evaluation

NB NS ZO PS PB

0.9 - 0.4 - 0.1 0.1 0.3 0.E -0.5 -0.2 0 0.15 0.35

ks, kc (2) Schedule Modification

NS ZO PS

- 2 - 0.5 0 0.5 2 4 5

(p"ph (1) NOx Margin Evaluation

NB N S Z O P S PB

0.6 -0.3 -0.1 0.1 0.3 O.( -0.35 -0.15 0 0.15 0.35

ks, kG

(2) Schedule Modification

(a) Membership Functions for STPS and GTGT

(b) Membership Functions for GTPS and STLT

Fig. 8 Membership Functions

Table 1 Example Rule Tables for Fuzzy Reasoning

(a) Example Rule Table for STPS

(b) Example Rule Table for GTGT

(d) Example Rule Table for STLT

4 4

kSG = 2 W ( i ) ks(i) / 2 W ( i ) = 0.0273 (6) i = I i = I

Here, W ( i ) ( i=I-4) are weights of the membership functions. Then, the candidate for schedule modification rate A X r is defined by the following equation.

(7) A X T = (Y kSG (XMAX - XMIN )

Here, (Y , XMAX and XMIN are a tuning coefficient, and maximum and minimum allowable limits of the schedule parameter, respectively. The tuning coefficient (Y for each schedule parameter is tuned through a simulation study described next. The XMAX and XMIN are uniquely designated for each schedule parameter in view of safe and practical plant operation.

NS ZO PS PB

-0.1 0.1 0.3 0.6 ks

__ ks

0.09

Fig. 9 Example Calculation of Schedule Modification Coefficient Ks

420

SIMULATION STUDY

................................................................................ , : I .--.; .... : . . . : , . , : a 4 : 4 : 4 4

Values of Operational Constraints

In comparison of startup characteristics between conventional methods and the proposed method, evaluations under the same stress and NOx level developed at the maximum point are necessary. Therefore, startup characteristics are evaluated along the following two steps.

(Step 1) The maximum stress and NOx levels are obtained through a simulation of the conventional method which introduces a metal matching chart to make startup schedules as described in reference [ I ] .

(Step 2) The optimization process is obtained through a simulation of the proposed method. In this step, 12 schedule parameters are modified under the operational constraints which are the same level as the maximum values obtained in step 1.

A simulation result for the conventional method is shown in Fig. 10. The required startup time is 108 minutes. In this case, the maximum values of the rotor surface stress, the rotor bore stress, the HRSG header stress, and NOx emission are -31.2 kg/mm2, 36.8 kg/mm2, 13.5 kg/mm2, and 8.3 ppm, respectively. These values are used as the operational constraints in the following simulation study.

..... ...... ....... 1 1~ .......!.......... i./ j - - r ~ ,

6 6 ' . , ......... L ......... ; .......... L ...... ~; j 1 ........

2 I.GT i 0 2 5 0 ) GTLbad ( 0 . Z S O ) STLoad

4 AHPBV ( -3 2 1 HPBVOpening S:AlPBV ( -1. 3 ] IPBVOpening 6 A H C V ( -4 1) HPCVOpening

(a) Startup Schedule

(b) Dynamics of Stresses and NOx

Fig. 10 Results by the Conventional Scheduling Method

Optimization Process

The tuning coefficients N (i) (i=l-12) are tuned considering stability of convergence of schedule optimization. Table 2 shows the tuning steps. Two steps are enough to get stable convergence. Three of the coefficients are changed considering the converging stabilities of the schedule parameters during the iterative optimization process. a (7) was reduced for the step 1, because the 7th schedule parameter showed oscillatory changes. Conversely, cy (9) and N (1 1) were enlarged for the step 2, because the 9th and 11th parameters showed slow modifications. Simulation results are shown in Fig. 11. Startup energy loss is remarkably reduced with startup time.

Converging characteristic of step 2 is shown in Fig. 12. The ninth iterative step gives the optimum startup schedule which can minimize the startup time and energy loss. Both can be remarkably reduced compared with the conventional method. This means that the proposed method provides quick and economical plant startup under NOx emission regulation and reliable machine life management. The operational constraints are satisfied through the optimization process, while the convergence characteristics are somewhat oscillate trajectory. The oscillation is small enough for this application. Even though, additional tuning of (Y will improve the stability. The simulation results along with the optimum startup schedule are shown in Fig. 13. The load up pattem of the gas turbine shown in Fig. 13(a) is drastically

Table 2 Tuning Coefficient LY

- Startup Time * ........... * Startur, Loss c

h

C .- E a, E t-

v

._

Q 3 Y

z G I terat ion Step

Fig. 11 Optimization Process

42 1

Computation time depends on the time range of the startup process and the iteration number of the schedule optimization. Using an engineering workstation (Hitachi Engineering Workstation 3050RW230) which has a 105 MIPS computation speed, takes about 2.5 minutes for the optimum schedule to be reached in step3. This is quick enough for actual use.

Initial Schedule

-- 60: 70 80 90 100 110

186' I 62 ~ 64 66 68

j Startup Time (min) ........... Optimum Schedule .......

i Startup Time (min)

Fig. 12 Convergence Characteristics

6 6 1 6 6 6 s

. .....,.......... ....... . . ......... .......

............... .. ,~~~ . j .........,..... ~ ..,..........

.....; 4~~~ .; 1

...................................................................... : . :

a : * ' m + : a : k : 4 : 6 1 4 : 4 , ~ ~.

............................................... . . ....... .... ....... ....... . .

......... ......................

0 TIME (sec) lwoi

1 NGT ( 0 2owo) G T S p d kpml 1: LGT ( 0 - 2 5 0 ) G T b a d IS1

4:AHPBV ( -3 2 ) HPBVOpening 1P.U 1 S A I P B V ( -2 3 ) I P B V O p i n g 1P.U I

I ) HPCVOpning [ Q U I

1x1 3: LST ( 0 - 2 5 0 ) S T b a d

6 A H C V ( - 4 '

(a) Optimum Startup Schedule

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .,&Hff*Limit:;.3 i, ; ......... J ....

. I I

, ~~ ............ ~ ............ .~

1 . I TIME($=) IWW

mm

1 N O X O U T A ( 0 50 ) NOx Emiaion M A Bpml 2. SHS ( - lw l w ) HPTRotor SurfaccStrar ""4 3 SHB ( -100 ' 100 ) HPT Romt Bore Suers W"2l 4" ( ~ 100 ' 100 ) IPT Romr Surface Suers W " I 5 SIB ( -100 IW ) IPT Rotor BoreSUni W m m 4 6 SHHDOT ( -70 30 ) HP Header Oliter S e a r W " 1 7 SIHDOT ( -70 30 ) I Q Header Ourer SVBI W"21

(b) Dynamics of Stresses and NOx

Fig. 13 Results by the Proposed Scheduling Method

modified compared with that of the conventional method as shown in Fig. 10(a).

Computation Time

DISCUSSION

Flexibilites of the Expert System

This expert system deals with the stresses in the steam turbine and HRSG, and NOx emission as the operational constraints. In a similar framework, other constraints such as differential expansion of the turbine rotor and casing, drum stress, and SOX and CO emissions, can be considered depending on individual machines, plant structural arrangement, operational objectives, and environmental regulations. Stresses can be managed indirectly using the temperature varying speed of the steam and metal instead of using the entire dynamics models directly.

The startup schedule is optimized on the base of startup time in this system as described before. Additional factors such as startup energy loss, life consumption rate of machines, or NOx emission rate, can be introduced to evaluate optimality using a weighted function. Then, flexible startup schedules can be obtained by varying the weights with environmental, seasonal and daily requirements.

The optimum schedule is intended to be able to keep the stresses and NOx within their limits. Even then, an unscheduled speed hold or load hold might be caused by accidents after the gas turbine is ignited. The startup schedule can be reviewed to minimize the time deviation from the original schedule by the function for rescheduling in the same manner as for the optimization using the current states as the initial conditions for the dynamics models.

Potential Applicabilities of the Optimization Algorithm

The optimization algorithm can be applied to various purposes which require parameter optimization in relation to plant dynamics, such as engineering tools for plant design and operation, training simulators for plant operators, and design tools for control systems.

In the engineering tools for plant design, the algorithm can function to optimize machine structure, strength, capacity and materials considering cost benefit, reliability, environmental adaptability, controllability, etc. In the basic planning of plant operation, this algorithm takes the part of experts to improve reliability and economics of the plant operation. The plant operators work burden can be lessened in acquiring the knowledge and the skills needed for plant operation by using training simulators which apply this algorithm.

Future Improvements

The expert system proposed in this paper deals with startup scheduling problem for a single shaft designed unit. The fuzzy rules introduced in this expert system can not be directly used for the multi- shaft designed unit which consists of one or more gas turbine and generators, one HRSG, or one steam turbine and generator, however, the basic approach taken in this expert system is applicable to such a design unit.

This system selects 12 objective parameters to optimize the startup schedule. Further simulation study with additional parameters might show a possibility for improved startup characteristics.

The size of the rule tables for the fuzzy reasoning is directly related to the number of the time period divisions to evaluate the startup dynamics. Therefore, further study on the evalustion method, construction and tuning method of fuzzy rules is needed to reduce the table size without deteriorating convergency of the optimization algorithm.

422

CONCLUSIONS

We have developed an expert system for optimizing the startup schedule of combined cycle power plants. This system harmonizes machines operations to minimize startup time and energy loss to meet NOx emission regulations and reduce machine stresses. Plant dynamics models representing quantitative knowledge and fuzzy rules representing qualitative knowledge are used alternately to optimize schedule parameters which designate the startup pattem.

Simulation results with this system showed the following. (1) Synergetic organization of fuzzy reasoning and plant dynamics

models provided flexibility in minimizing the startup time automatically under various operating constraints.

(2) The schedule parameters converge more quickly to the optimum values compared with the conventional method.

(3) Startup energy loss was reduced due to the reduction in startup time.

This expert system can meet the requirements of middle-load operation with the frequent startup and shutdown as found in combined cycle power plants.

REFERENCES

T. Arakawa and Y. Hishinuma, "Thermal Power Generation Technology Giving Considerations for Global Environment," Hitachi Hyoron, Vol. 73, No. 11, pp. 4 - 8. T. Akiyama, T. Matsushima, N. Nagafuchi, A. Nakajima, H. Yoshizaki, J. Matsumura, "Dynamic Simulation of an Advanced Combined Cycle Plant with Three Pressure and Reheat Cycle," JSME - ASME Intemational Conference on Power Engineering - 93 (ICOPE - 93), Tokyo, Japan, September 1993. F. J. Hanzalek and P. G. Ipsen, "Thermal Stress Influence Starting, Loading of Bigger Boilers," Electrical World, Vol. 165, No. 6, pp. 58 - 62, February 1966. R. G. Livingston, "Computer Control of Turbine-gznerators Startup Based on Rotor Stress," Joint Power Conference ASME and IEEE, 1973. H. Matsumoto, Y Sato, F. Kato, Y. Eki, K. Hisano, K. Fukushima, "Turbine Control System Based on Prediction of Rotor Thermal Stress," IEEE Transaction on PAS, Vol. PAS-

H. Matsumoto, Y. Eki, A. Kaji, S. Nigawara, M. Tokuhira, Y. Suzuki, "An Operation Support Expert System Based on On-line Dynamics Simulation and Fuzzy Reasoning for Startup Schedule Optimization in Fossil Power Plants," IEEE Transaction on Energy Conversion, Vol. 8, No. 4, pp. 674 - 680, December 1993.

101, NO. 8, pp. 2504 - 2512, August 1982.

Hiroshi Matsumoto received his B.S. degree in electrical engineering from Doshisha University, Kyoto, in 1970.

He joined Hitachi, Ltd., in 1970 and has been engaged in research and development of computer control systems for fossil power plants at Hitachi Research Laboratory. He is a senior researcher of New Thermal Power Planr Systems Group in Second Dept. of Energy Systems Research of Hitachi Research Laboratory.

His current research interests concem enhancement of operation and control systems for power plants based on operations research, artificial intelligence, fuzzy theory and neural networks.

He is a member of the IEEE, the Institute of Electrical Engineers of Japan, the Japan Society of Mechanical Engineers, Society of Instrument and Control Engineers, and the Japan Neural Network Society.

Yo Ohsawa received his M.S. degree in mechanical engineering from Tohoku University, Sendai, in 1992.

He joined Hitachi, Ltd., in 1992 and has been engaged in research and development of computer control systems for fossil power plants at Hitachi Research Laboratory. He is a researcher of New Thermal Power Planr Systems Group in Second Dept. of Energy Systems Research of Hitachi Research Laboratory.

His current research interests concem control systems for gas and

He is a member of the Japan Society of Mechanical Engineers. steam combined cycle power plants.

Shoei Takahashi received his B.S. degree m electrical engineering from Iwate University in 1976.

He joined Hitachi, Ltd., in 1976 and has been engaged in planning and designing of instrumentation and control systems for gas turbine and combined cycle power plants at M a Works. He is a senior engineer of Thermal & Hydroelectric Power Control & Instrumentation Systems Engineering Dept. of the

Works. His current interests concem designing of control systems for

He is a member of the Institute of Electrical Engineers of Japan, Advanced combined cycle power plants.

and Thermal and Nuclear Power Engineering Society.

Takao Akiyama received his B.S. degree in instrument & control engineering from Kobe University in 1967, and his M.S. degree in control engineering from Osaka University in 1969.

He joined Hitachi, Ltd., in 1969 and has been engaged in research and development of safety analysis and dynamic control system for nuclear plants and stayed at Purdue University 1973 - 1974 as a visiting scholar of nuclear engineering department.

He shifted his major interest to fossil power plants since 1992 at Hitachi works and was involved in developing a control strategy decision system with aids of plant simulators.

He is a member of IEE, SICE, JSME of Japan and ANS.

Okikazu Ishiguro received his B.S. degree in mechanical engineering from Okayama University in 1965.

He joined Babcock-Hitachi K.K. in 1965 and has been engaged in research and development of control systems for fossil power plants at Kure Research Laboratory. He is a senior researcher and manager of the 1st Lab. in the 4th Dep. of Kure Research Laboratory.

His current research interests concem development of control systems for Denox Reactor of power plants based on fuzzy theory, and concern enhancement of boiler design based on dynamic characteristics analysis.

He is a member of the Japan Society of Mechanical Engineers, Society of Instrument and Control Engineers.