Electric Power Systems Research 60 (2002) 123 135 www.elsevier.

com/locate/epsr

Neuro-fuzzy controller for gas turbine in biomass-based electric power plant

Francisco Jurado a,*, Manuel Ortega a, Antonio Cano a, Jose Carpio b
b

Department of Electrical Engineering, Uni6ersity of Jae n, 23700 EUP Linares, Jae n, Spain Department of Electrical and Computer Engineering, Spanish National Uni6ersity for Distance Education, 28040 UNED Madrid, Spain
a

Received 20 February 2001; received in revised form 13 June 2001; accepted 13 June 2001

Abstract Biomass gasication is a technology that transforms solid biomass into syngas. The gas turbine controller regulates both the gas turbine and the gas turbine generator. Two fuzzy logic controllers have been developed using speed and mechanical power deviations, and a neural network has been designed to tune the gains of the fuzzy logic controllers based on the operating conditions of the biomass-based electric power plant. It is shown that by tuning the fuzzy logic controllers, optimal time domain performance of the system can be achieved in a wide range of operating conditions compared to xed parameter fuzzy logic controllers and PID controllers. 2002 Elsevier Science B.V. All rights reserved.
Keywords: Neural network; Fuzzy controller; Turbine generator; Modeling

1. Introduction A thorough introduction to gas turbine theory is provided in the book of Cohen et al. [1]. There also exist a large literature on the modeling of gas turbines. Model complexity varies according to the intended application. Detailed rst principles modeling based upon fundamental mass, momentum and energy balances is reported by Fawke et al. [2] and Shobeiri [3]. These models describe the spatially distributed nature of the gas ow dynamics by dividing the gas turbine into a number of sections. Throughout each section, the thermodynamic state is assumed to be constant with respect to location, but varying with respect to time. Mathematically, the full partial differential equations model description is reduced to a set of ordinary differential equations which facilitate easier application within a computer simulation program. For a detailed model, a section might consist of a single compressor or turbine stage. Much simpler models result if the gas
* Corresponding author. E -mail addresses: fjurado@ujaen.es (F. Jurado), mortega@ujaen.es (M. Ortega), acano@ujaen.es (A. Cano), jcarpio@ieec.uned.es (J. Carpio).

turbine is decomposed into just three sections corresponding to the main turbine components, i.e. compressor, combustor and turbine [4]. Instead of applying the fundamental conservation equations, as describes above, another modeling approach is to characterize gas turbine performance by utilizing real steady state engine performance data [5]. It is assumed that transient thermodynamic and ow processes are characterized by a continuous progression along the steady state performance curves, this is known as the quasi-static assumption. The dynamics of the gas turbine, e.g. combustion delay, motor inertia, fuel pump lag, etc. are then represented as lumped quantities separate from the steady-state performance curves. Very simple models result if it is further assumed that the gas turbine is operated at all times close to rated speed [6]. In this case, the modeling of gas turbine performance over a wide range of operating conditions, as is experienced during start up or shutdown procedures, is not attempted. For the purposes of developing a gas turbine model suitable for the evaluation of supervisory Combined Heat and Power (CHP) control strategies, it is necessary to capture the main dynamics of the gas turbine over a wide range of operating conditions. Conse-

0378-7796/02/$ - see front matter 2002 Elsevier Science B.V. All rights reserved. PII: S 0 3 7 8 - 7 7 9 6 ( 0 1 ) 0 0 1 8 7 - 0

124

F. Jurado et al. / Electric Power Systems Research 60 (2002) 123 135

quently, the model described in this paper retains the structure of Rowen [6] but extends the models range of validity. The concept of the gas turbine control system, which is applied in this paper, is based on the SPEEDTRONIC Mark 4 description as presented in the paper by Rowen [7]. Some considerations concerning the subject may be also found in Yacobucci [8] and Polimeros [9]. Despite huge advances in the eld of control systems engineering, PID still remains the most common control algorithm in industrial use today. Unlike the classical control design, which requires a plant model for designing the controller, fuzzy logic incorporates an alternative way, which allows one to design a controller using a higher level of abstraction without knowing the plant model. The recent growth in attention to neural networks (NNs) has led to many suggestions for combined use of fuzzy logic and neural networks in intelligent control. It is shown that by tuning the fuzzy logic controllers (FCs) with the aid of a NN, optimum response of the system can be achieved in a wide range of operating conditions compared to xed parameters FCs and PID controllers. The paper is structured as follows. Section 2 presents the modeling of the gas turbine. Some basic concepts of biomass gasication are presented in Section 3. Section 4 discusses the gas turbine control conguration. Section 5 depicts some simulation results. Finally, conclusions are presented in Section 6.

power developed by the turbine would be exactly matched by the power required to drive the compressor. The main modeling assumptions are as follows: 1. air and combustion products are treated as perfect gases; 2. specic heats are assumed constant for combustion products, air and injected steam; 3. ow-through nozzles are described by a one-dimensional adiabatic uniform polytropic process; 4. energy storage and transport delay in the compressor, turbine and combustion chamber are relatively small, thus steady state equations are applied; 5. inlet kinetic energy of gas ows into the compressor and turbine are treated as negligible; 6. air mass ow through the compressor is controllable via inlet guide vanes. The compressor is described by a one-dimensional steady ow nozzle equation for a uniform polytropic compression. This can be expressed as: w a = Ao



ma + 1 2ma /ma zipcin r 2 r c ma c pc(ma 1)

 

n

1/2

(1)

Compressor power consumption equation: Pc = waDhIC pcptrans (2)

Combustion energy equation: wgcpg(TTin 298) + wfDh25 + wacpa(298 Tcout) + wiscps(298 Tis) = 0 Power delivery equation: PT = pTwgDhIT (4) (5) (3)

2. Gas turbine The main components of an industrial power plant gas turbine are the compressor, the combustion chamber and the turbine. The gas turbine operates as follows. Air at atmospheric pressure enters the gas turbine at the compressor inlet. After compression of the air to achieve the most favorable conditions for combustion, fuel gas is mixed with the air in the combustion chamber, combustion takes place and the hot exhaust gases are expanded through the turbine to produce mechanical power. In terms of energy conversion, chemical energy present in the combustion reactants is transferred to the gas stream during combustion. This energy measured in terms of gas enthalpy or higher heating value (HHV) is then converted into mechanical work, by expanding the gas through the turbine. Thus the excess mechanical power available for application elsewhere, after accounting for the power required to drive the compressor, is derived ultimately from the combustion process. Without combustion, assuming 100% efcient compressor and turbine operation, the

Pmec = PT Pc Fig. 1 shows the block diagram of the gas turbine.

3. Biomass gasication Gumz is the earliest reference found describing the concept of combining a pressurized gasier with a gas turbine engine, although Gumz himself references an earlier work proposing this concept [10]. He also states that the combination could certainly benet from future development of pressurized hot gas cleaning to avoid excessive turbine blade wear. Gumz was speaking of coal-fuelled plants but the concept is similar when using biomass as fuel. Biomass gasication is a technology that transforms solid biomass into syngas (hydrogen and carbon monoxide mixtures produced from carbonaceous fuel). Current use of biomass, which stands at about 12% of the total energy supply to the world, is primarily used in combustion for immediate use. Small-scale gasication for CHP in distributed generation (in Europe

F. Jurado et al. / Electric Power Systems Research 60 (2002) 123 135

125

sometimes called embedded generation), and village power applications is a eld that has expanded very rapidly. Many villages and mini-grids can be served by biomass power generation in the size range of 1 kWe to 5 MWe. Biomass fuels are characterized by high and variable moisture content, low ash content, low density, and brous structure. In comparison with other fuels, they are regarded as of low quality despite low ash content and very low sulfur content. The residual biomass of the olive-grove in Spain with a potential energy use is classied into two groups. The rst group is constituted by residual biomass of olive in the extraction process of the olive oil. Depending on the extraction system, traditional, decanter in three phases or decanter in two phases, the available energy from the by-products is different. In case of a traditional or a decanter in a three-phase system, the byproduct is the foot cake (4.600 kcal/kg heating value), and the olive paste of second centrifugation (3.500 kcal/kg heating value) for the last extraction system. The second group of biomass is constituted by residual biomass from the olive tree, wood, small parts of the olive tree and the forest resources due to forestry works (bushes cleanliness, etc). The products of both above groups present, from an energy point of view, favorable aspects in their use, e.g. the ensured annual production, its relative concentration in a place, proper humidity conditions, low sulfur content and other harmful emissions, and nally, its high thermal value. Not using those resources yields environmental problems due to foot cake and olive paste storage, plague propagation and forest res. A variety of relatively large-scale biomass gasication technologies are at various advanced stages of development. Three gasier/gas clean-up designs are considered here: (i) atmospheric-pressure air-blown uidized-bed gasication with wet scrubbing, e.g. the technology under development by Waldheim et al. [11] its HHV is

1500 kcal/kg; (ii) pressurized air-blown uidized-bed gasication with hot-gas clean-up, e.g. the technology under development by Salo et al. [12] the HHV is 1300 kcal/kg; and (iii) atmospheric-pressure indirectly-heated gasication with wet scrubbing by Paisley et al. [13] the HHV is 4300 kcal/kg. Several scenarios point to the potential market for gasier power systems at about 10 000 MW by 2010. This power plant generates electric power using biomass from the olive tree. The gasier is capable of converting tons of wood chips per day into a gaseous fuel that is fed into a gas turbine. The gasier signicantly improves electrical generating efciency in a variety of applications. The biomass gasier enables the use of advanced power systems that will nearly double the efciency of todays biopower industry. The gasier heats the wood in a chamber lled with hot sand until the wood breaks into basic chemical components. The solids sand and char are separated from the gases, which then ow through a scrubber. The nal result is a very clean-burning gas fuel suitable for direct use in modern power systems such as a gas turbine.

4. Gas turbine control conguration The simplied gas turbine model is divided into two interconnected subsystems in this paper. The subsystems are: fuel system (fuel valve with actuator); turbine. The fuel system consists of the fuel valve and the actuator. The fuel ow out from the fuel systems results from the inertia of the fuel system actuator and of the valve positioner. Fuel system actuator equation: wf = kf e ~fs + 1 1 (6)

Fig. 1. Gas turbine.

126

F. Jurado et al. / Electric Power Systems Research 60 (2002) 123 135

Fig. 2. Gas turbine control system.

Valve positioner equation: e1 = a F bs + c d (7)

4.1. PID controller

This controller is widely used because of its versatility, high reliability and ease of operation [14]. A standard form of the controller is given in Eq. (10). u (t ) = Kpe (t ) + Ki

The input variable to the fuel system is Fd. Output variable from the fuel system model is wf. In this paper, a simplied gas turbine model is described. A single gas turbine does not require the digital set point feature. The kHHV factor depends on HHV. The kHHV and 0.23 factors cater for the typical turbine power/fuel rate characteristic which rises linearly from zero power at 23% fuel rate to rated output at 100% fuel rate. The turbine torque function is given by: T = kHHV(wf 0.23) + 0.5(D ) (8) Eq. (8) allows the turbine torque to be calculated algebraically. This torque is used in the differential equations which model the mechanical system Pmec = T (9) Input variables to the turbine are wf, D and . The output variable from the turbine is Pmec. The gas turbine controller regulates both the gas turbine and the gas turbine generator. One of the most popular designs at present is the SPEEDTRONIC system developed for General Electric gas turbines. For the purpose of this paper only modulating control of the mechanical side of the gas turbine is of interest. Thus the model, based on SPEEDTRONIC Mark 4, is signicantly simplied. The simplied model of the gas turbine controller in this paper consists of two inputs and one output. Inputs to the controller are Pmec and . Output from the controllers is Fd. The block diagram of the gas turbine control system is presented in Fig. 2. The diagram consist of two feedback controllers. LVG stands for Least Value Gate which transmits the minimum of two incoming signals.

&

e (~ ) d~ + Kd

de (t ) dt

(10)

The measurable output y (t ) is subject to sensor noise n (t ) and the system disturbances d(t ), both of which can be assumed unknown. The control u (t ) is a summation of three dynamic functions of the error e (t ) from a specied reference (demand) output yref(t ). Proportional control has the effect of increasing the loop gain to make the system less sensitive to load disturbances, the integral of error is used principally to eliminate steady-state errors, and the derivative action helps to improve closed loop stability. The parameters Kp, Ki, Kd are thus chosen to meet prescribed performance criteria, classically specied in terms of rise and settling times, overshoot and steady-state error, following a step change in the demand signal. A standard method of setting the parameters is through the use of Ziegler Nichols tuning rules [15]. These techniques were developed empirically through the simulation of a large number of process systems to provide a simple rule. The methods operate particularly well for simple systems and those which exhibit a clearly dominant pole-pair, but for more complex systems the PID gains may be strongly coupled in a less predictable way. For these systems, adequate performance is often only achieved through manual and heuristic parameter variation. In this paper, PID parameters are tuned with Nonlinear Control Design Blockset of MATLAB [16]. Monte Carlo simulations generate uniformly distributed random plants between the upper and lower bounds. Then

F. Jurado et al. / Electric Power Systems Research 60 (2002) 123 135

127

after optimizing, simply using the initial response, we inspect the simulations.

4.2. Neuro -fuzzy logic controller

The fuzzy logic controller is very attractive for illdened systems or systems with uncertain parameters. With the help of fuzzy logic concepts, expert knowledge can be used directly to design a controller. Fuzzy logic allows one to express the knowledge with subjective concepts such as very large, too small, which are mapped to numeric ranges [17]. Similarities exist between the NNs and the FCs. Both techniques allow interpolate reasoning which makes them free from the true/false restriction of conventional logic systems [18]. In a multi-layer neural network of feed-forward type, input nodes record the features and pass activation values to the output layer through a hidden layer. The addition of the hidden layers to the two-layer perceptron networks allows these networks to represent any continuous mapping from input to output [19]. An appropriate training technique, adjusts the connection weights of the network to improve the match between the output of the network and the correct results. To design the FC some variables which can represent the dynamic performance of the system should be chosen

to be fed as the inputs. In addition to the proper input signals, signal gains and fuzzy subsets should be dened. It is common to use the output error and the rate o derivative of the output as controller inputs [20,21]. In this paper, the turbine speed deviation (D ) and its derivative (D %), the acceleration, are considered as the inputs of the rst FC and the mechanical power deviation delivered by turbine (DPm) and its derivative (DP % m) as the inputs of the second FC. After, D, D %, DPm and DP % m signals pass through four appropriate gains or scaling factors, and then are fed to the FCs. The outputs of the controllers are also scaled by passing through the output gains. To convert the measured input variables of the FCs into suitable linguistic variables, seven fuzzy subsets NB (Negative Big), NM (Negative Medium), NS (Negative Small), Z (Zero), PS (Positive Small), PM (Positive Medium) and PB (Positive Big) are chosen. Membership functions of these subsets are bell-shaped. Fig. 3 shows the membership functions. These membership functions are symmetrical and each one overlaps with the adjacent functions by 50%. In this paper, both inputs of the FCs have seven subsets. Thus, two fuzzy rule tables with 49 rules are constructed. A rule table, which is formulated based on the past experience of manual tuning of a controller, is shown in Table 1 and Fig. 4 illustrates the control surface. The center of gravity method is employed.

Fig. 3. Membership functions. Table 1 Control rules Turbine speed D NB D % NB NM NS Z PS PM PB NM NB NB NB NM NM NS Z NS NB NB NM NM NS Z PS Z NB NM NS NS Z PS PM PS NM NM NS Z PS PM PM PM NM NS Z PS PS PM PB PB NS Z PS PM PM PB PB Z PS PM PM PB PB PB

128

F. Jurado et al. / Electric Power Systems Research 60 (2002) 123 135

Fig. 4. Control surface.

Sequential de-centralized control means design of each modulation controller one after the other, so that the dynamics of previously designed controllers are taken into account in designing the next controller [22]. Use of this technique eliminates the requirement of considering several input variables in designing a fuzzy logic modulation controller and allows us to realize the control objectives with the use of simple fuzzy logic controllers.

By time weighting the error signal, less emphasis is placed on the initial error, which is largely unavoidable, and greater emphasis on reducing long-duration oscillations. A technique called the Levenberg Marquardt method is used to train the NN.

5. Results A plant consisting of a load and induction motor is fed from a synchronous generator/gas turbine unit. The gas turbine, fuzzy controllers, neural network, synchronous generator and governor system are modeled by MATLAB (Fig. 6). Initially, the induction motor develops a mechanical power and the synchronous generator provides active power. The generator controls the voltage at 1 per unit and generates active power. Mechanical power from the synchronous generator increases from its initial value to the nal value required by the load and induction motor. The active power load P, reactive power load Q and mechanical torque in asynchronous motor Tm are selected for input signals to represent the operating condition of the system. A number of patterns related to the system conditions are assembled as training sets to ensure adequate training is presented to the NN during the training phase. The selection of the training features used to form the training pattern is of critical importance to the success of operating a NN. Techniques exist to determine the proper number of training examples from a data set [23]. N] 32W 32M ln m m

4.3. Tuning the fuzzy controllers

In order to tune the FCs, the D is scaled according to the relation D* = Ge D and D %, D % * = Gr D %. The DPm is scaled according to the relation DPm* = GePmDPm and DP % m, DP % m* = GrPmDPm. Also, the output of the rst FC is scaled by Gu and the output of the second FC is scaled by GuPm. In the aforementioned relations, Ge, Gr, Gu, GePm, GrPm and GuPm are the scaling factors or gains. The gains of the FCs are tuned with a neural network, making the FCs adaptable to changes in operating conditions. The FCs are tuned by computing optimum gains, exploiting a neural network. The neural network is composed of three layers (Fig. 5). The purpose with the controllers is to improve the performance of the system response. For various sets of input data to the NN, the optimum values of gains are searched sequentially using simulations. The evaluation of the optimality is checked by time domain performance specications such as peak time, percent overshoot, rise time and settling time. The optimization was conducted by minimization of integrated time and squared error J=

&

TS

~ (D (~ ) + DPm)2 d~

(11)

 

(12)

F. Jurado et al. / Electric Power Systems Research 60 (2002) 123 135

129

Fig. 5. Neural network.

N\

W m

(13)

5.2. Case 2
Operating point P = 0.9 MW, Q = 0.4 Mvar and Tm = 8000 N m. The system responses are shown in Figs. 9 and 10. As shown in the gures, the NFCs have the optimum responses and the responses are smooth and fast. They have less settling time, compared with the other controllers.

A data set has been generated by computer simulation by varying the input signals. This data set comprised 1500 patterns. From these, 750 were used for learning purposes, 375 for test and 375 for validation, using a random selection with uniform distribution. The training performance of the NN was assessed by examining the root mean square error. Assuming that if the error was less than 0.006, then the NN was considered to be satisfactorily trained. For the power plant with the three controllers, i.e. PIDs, FCs and Neuro-Fuzzy Controllers (NFCs), the system responses for four different conditions were obtained using simulations over a suitable simulation time (TS = 1.5 s) following an applied step in load at t = 0.

5.3. Case 3
Operating point P = 0.8 MW, Q = 0.3 Mvar and

5.1. Case 1
Operating point P = 1 MW, Q = 0.5 Mvar and Tm = 8000 N m. Since FCs and NFCs are designed for this operating condition, their responses are optimum and coincident with each other in this case, as is observed from Figs. 7 and 8.

Fig. 6. System model.

130

F. Jurado et al. / Electric Power Systems Research 60 (2002) 123 135

Fig. 7. Turbine speed for Case 1.

Fig. 8. Mechanical power delivered by turbine for Case 1.

F. Jurado et al. / Electric Power Systems Research 60 (2002) 123 135

131

Tm = 8000 N m. The system responses are shown in Figs. 11 and 12. It is observed that the system response for the NFCs is optimum in this case.

5.4. Case 4
Operating point P = 1 MW, Q = 0.5 Mvar and Tm =

Fig. 9. Turbine speed for Case 2.

Fig. 10. Mechanical power delivered by turbine for Case 2.

132

F. Jurado et al. / Electric Power Systems Research 60 (2002) 123 135

Fig. 11. Turbine speed for Case 3.

Fig. 12. Mechanical power delivered by turbine for Case 3.

8000 N m. Pressurized air-blown uidized-bed gasication with hot-gas clean-up is the technology under development in Spain and the HHV is about 1200

kcal/kg. Hence, the HHV of biogas is considered to be 1100, 1200 and 1300 kcal/kg, respectively. Figs. 13 and 14 depict the system responses using NFCs.

F. Jurado et al. / Electric Power Systems Research 60 (2002) 123 135

133

6. Conclusions Despite huge advances in the eld of control systems

engineering, PID still remains the most common control algorithm in industrial use today. A new simplied model of gas turbine controller is

Fig. 13. Turbine speed for Case 4.

Fig. 14. Mechanical power delivered by turbine for Case 4.

134

F. Jurado et al. / Electric Power Systems Research 60 (2002) 123 135

presented in this paper. This controller consists of two inputs and one output. Inputs to the controller are the mechanical power delivered by the turbine and the speed of rotation, related to electrical frequency. Output from the controller which is input to the gas turbine model is the fuel ow. Two rule-based FCs are designed and six gains are introduced for tuning purposes. A neural network is designed to tune these gains based on the operating conditions of the system. It is shown that by tuning the FCs with the aid of a neural network, optimum response of the system can be achieved under a wide range of operating conditions compared to xed parameter FCs and PID controllers.

t T Tcout Tis Tm TS TTin u (t ) W wa wf wg wis y (t ) yref(t ) Dh25 DhIC

Appendix A. Nomenclature Ao a, b, c cpa cpg cps d (t ) e (t ) e1 Fd Ge GePm Gr GrPm HHV J Kd Ki kf kHHV Kp M ma N n (t ) P Pc pcin pcout Pmec PT pTin pTout Q rc compressor exit ow area valve parameters specic heat of air at constant pressure specic heat of combustion gases specic heat of steam system disturbances error signal valve position fuel demand signal turbine speed deviation scaling factor power deviation scaling factor turbine speed deviation derivative scaling factor power deviation derivative scaling factor higher heating value optimization function PID parameter PID parameter fuel system gain constant factor which depends on HHV PID parameter number of hidden neurons polytropic index number of random examples sensor noise active power load compressor power consumption air pressure at compressor inlet air pressure at compressor outlet mechanical power delivered by turbine total mechanical power delivered by turbine pressure of combustion gases at turbine inlet pressure of combustion gases at turbine outlet reactive power load pressure ratio (outlet/inlet) DhIT DPm DPm* DP % m DP % m* D D* D % D % * m pc pT ptrans pc zi ~f

time mechanical torque delivered by turbine outlet air temperature temperature of injected steam mechanical torque in asynchronous motor simulation time turbine inlet gas temperature control signal sum of weights air mass ow into the compressor fuel mass ow turbine gas mass ow injection steam mass ow measurable output reference output specic enthalpy of reaction at reference temperature of 25 C isentropic enthalpy change for a compression from pcin to pcout isentropic enthalpy change for a gas expansion from pTin to pTout power deviation delivered by turbine scaled power deviation power deviation derivative scaled power deviation derivative rotation speed deviation of the turbine scaled turbine speed deviation turbine speed deviation derivative scaled turbine speed deviation derivative error overall compressor efciency overall turbine efciency transmission efciency from turbine to compressor compressor polytropic efciency inlet air density fuel system time constant rotation speed of the turbine

References
[1] H. Cohen, G.F.C. Rogers, H.I.H. Saravanamuttoo, Gas Turbine Theory, 3rd Edition, Longman, London, 1987. [2] A.J. Fawke, H.I.H. Saravanamuttoo, M. Holmes, Experimental verication of a digital computer simulation method for predicting gas turbine dynamic behaviour, Inst. Mech. Eng. Proc., 186 (27) (1972). [3] T. Shobeiri, Digital computer simulation of the dynamic operating behaviour of gas turbines, Brown Boveri Rev. 3 (1987). [4] A. Hussain, H. Sei, Dynamic modeling of a single shaft gas turbine. Proc. of the IFAC Symposium on Control of Power Plants and Power Systems, Munich, Germany, 1992, Pergamon Press, pp. 43 48. [5] W.W. Hung, Dynamic simulation of gas-turbine generating unit, IEE Proc. C 138 (4) (1991). [6] W.J. Rowen, Simplied mathematical representations of heavy duty gas turbines, ASME J. Eng. Power 83-GT-63 (1983) 865 869.

F. Jurado et al. / Electric Power Systems Research 60 (2002) 123 135 [7] W.J. Rowen, Speedtronic Mark IV control system, Alsthom Gas Turbine Reference Library, AGTR 880, 1988. [8] R.B. Yacobucci, A control system retrot for a GE frame 5 turbine/generator unit, IEEE Trans. Energy Conversion 6 (2) (1991) 225 230. [9] G. Polimeros, Energy Cogeneration Handbook, Industrial Press Inc, 1981. [10] W. Gumz, Gas Producers and Blast Furnaces, Wiley, New York, 1950, pp. 166 167. [11] L. Waldheim, E. Carpentieri, Update on the Progress of the Brazilian Wood BIG-GT Demonstration Project, ASME Paper GT-472, 1998. [12] K. Salo, A. Horvath, J. Patel, Pressurized Gasication of Biomass, ASME Paper GT-349, 1998. [13] M. Paisley, D. Anson, Gasication for Gas Turbine Based Power Generation, ASME Paper GT-5 1997. , stro [14] K.J. A m, T. Hagglund, PID controllers: Theory, design, and tuning, Research Triangle Park, Instrument Society of America, NC, USA, 1995. [15] J.G. Ziegler, N.B. Nichols, Optimum setting for automatic controllers, Trans. ASME 64 (1942) 759 768.

135

[16] MATLAB, Math Works, Inc., Natick, MA, USA, 2000. [17] L.A. Zadeh, Outline of a new approach to the analysis of complex systems and decision processes, IEEE Trans. Syst. Man Cybernet. 3 (1) (1973) 28 44. [18] J.-S.R. Jang, C.-T. Sun, Neuro-Fuzzy modeling and control, IEEE Proc. 83 (3) (1995) 378 406. [19] L.L. Lai, Intelligent System Applications in Power Engineering: Evolutionary Programming and Neural Networks, Wiley, Chichester, UK, 1998. [20] M. Lown, E. Swidenbank, B.W. Hogg, Adaptive Fuzzy logic control of a turbine generator system, IEEE Trans. Energy Conversion 12 (4) (1997) 394 399. [21] J. Talaq, F. Al-Basri, Adaptive fuzzy gain scheduling for load frequency control, IEEE Trans. Power Syst. 14 (1) (1999) 145 150. [22] D.J. Trudnowski, J.R. Smith, T.A. Short, D.A. Pierre, An application of Prony methods in PSS design for multimachine systems, IEEE Trans. Power Syst. 6 (1) (1991) 118 126. [23] S. Haykin, B. Kosko, Intelligent Signal Processing, IEEE Press, USA, 2001.