ANFIS Based Design of Controller for Superheated Steam Temperature Non Linear Control Process Subhash Gupta, L. Rajaji, Kalika S. Research Scholar SVU, UP; Professor P.B.College of Engineering, Chennai Abstract The objective of this paper is to develop Adaptive Network based Fuzzy Inference Systems (ANFIS) and effectively use for generation of membership functions to model the process variables of a non linear superheated steam temperature control process. The inputs to the system are obtained by making real time measurements of process variables. The performance of the various membership functions in tracking the input output data set is compared and the design of an ANFIS based controller is carried out. The rule base for various membership functions is automatically generated and the surface plot is shown for a combination of inputs and output. Index Terms Non Linear, Super Heated Steam Temperature. I. INTRODUCTION The integration of neural network models with fuzzy logic control is particularly appropriate since both techniques are best suited when detailed analytical understanding of a process is not available [1]. Adaptive Network based Fuzzy Inference System (ANFIS) can play a particularly important role in the induction of rules from observations. It is a powerful alternative strategy to fuzzy systems, since it is capable of learning and providing If-Then fuzzy rules in linguistic and explicit forms. Amongst such models ANFIS has been recognized as a reference framework mainly for its flexible and adaptive nature [2]. The salient features of ANFIS are: it approaches any linear and nonlinear functions, has quick converging speed, decreases precision errors and needs less data [3]. The design and stability aspects of ANFIS have been given by Wang [4]. Rizzi et al [5] have proposed a scheme for the automatic training of ANFIS networks. Adaptive neuro fuzzy controllers have recently found various applications. They have been used earlier for power system stabilizer.in this paper, the design of an ANFIS based control system is proposed for the control of the superheated steam temperature control process. The inputs to the process are temperature of steam at inlet of the final super-heater, burner tilt angle, steam flow rate and temperature of steam at the outlet of the final superheater. The output of the process is the spray water level. The data set obtained from real time measurements of the process variables was used for the design of the ANFIS system. The following membership functions are considered for modelling the process: Triangular, trapezoidal, gbell, guass1, guass2, pi, psigmoid and sigmoid. A comparison is made on the performance of these membership functions. Fuzzy inference rules are automatically generated by ANFIS system for all these membership functions. II. ADAPTIVE NETWORK BASED FUZZY INTERFERENCE SYSTEM (ANFIS) ANFIS uses a hybrid learning algorithm to identify parameters of Sugeno type fuzzy inference systems [5]. It applies a combination of the least squares method and the back propagation gradient descent method for training FIS membership function parameters to emulate a given training data set. The learning in an ANFIS is a twostage process. During the forward pass the consequence parameters are updated using the least square estimate method or recursive least square estimate method. In the backward pass the premise parameters are updated using back propagation method [6]. This learning process is continued until the change in output is zero. III. STEPS FOR CREATING ANFIS MODEL The various steps involved in creating a Fuzzy inference system (FIS) model from the input output data are as follows: Load data (training, testing, and checking). Generate an initial FIS model or load an initial FIS model. Choose the FIS model parameter optimization method: back propagation or a mixture of back propagation and least squares (hybrid method). Choose the number of training epochs and the training error tolerance. Train the FIS model by clicking. This training adjusts the membership function parameters and plots the training (and/or checking data) error plot(s) in the plot region. View the FIS model input versus the training, checking, or testing data output. Verify the test data against the FIS output in the plot region. IV. ANFIS CONTROLLED SUPERHEATED STEAM TEMPERATURE CONTROL PROCESS A detailed description of the typical control scheme for the realization of super heater outlet steam temperature control process is given in Figure 1 [7]. One factor which affects the heat input to the final super heater is the spray water level. Another factor is the angle of tilt of the burner block. This determines the elevation of the fireball in the boiler and hence the distribution of heat absorption by various heat exchangers. The elevation of the fireball position angle in the boiler is varied from +30 o (100%) to -30 o (0%). The angle of +30o corresponds to maximum radiation and an angle of -30o to minimum radiation. 224
Fig 3 Training Curve for Trapezoidal Membership Function Fig 4 Training Curve for Gbell Membership Function Fig 1 Control Scheme for Superheated Steam Temperature Control Process. [7] V. DATA COLLECTION AND TRAINING Measurements were made on the process variables of the super heated steam temperature control process. The readings were taken at every minute. The 1440 readings corresponding to the data set for one day were used as the training data for the ANFIS. The data set was normalized before being utilized. By considering the number of membership functions to be three, for each process variable and by choosing back propagation algorithm in combination with a least squares type of method, the ANFIS system was trained. The performances of error and trained epochs of ANFIS for different types membership functions are shown below in Figures 2 to 9. Fig 5 Training Curve for Gauss Membership Function Fig 2 Training Curve for Triangular Membership Function Fig 6 Training Curve for Gauss2 Membership Function 225
Fig 7 Training Curve for Pi Membership Function Fig 8 Training Curve for Dsigmoid Membership Function Table 1 Comparison Of Performance Of Membership Functions. [7] Fig 9 Training Curve for Psigmoid Membership Function The performance of the various membership functions is compared with respect to errors, which occurred in different epochs, and is shown in Table 1.Among the membership functions, Gaussian2 membership function and trapezoidal membership function converge faster than others. The error in Gaussian membership function is the least and therefore is more accurate than that of others (error is 0.063038). The errors for dsigmoid membership function and psigmoid membership function are the same for each epoch. Trapezoidal membership function is less accurate than that of other membership functions. Epochs Trimf Tramf Gbellmf Gaussmf Gauss2mf Pimf Dsigmf Psigmf 1 0.07855 0.16512 0.086176 0.079215 0.13257 0.18627 0.13453 0.13453 13 0.076941 0.14346 0.081529 0.07585 0.095143 0.1656 0.11512 0.11512 25 0.075721 0.12395 0.077404 0.073739 0.080958 0.14397 0.095586 0.095586 50 0.074901 0.10235 0.070983 0.070537 0.072026 0.11404 0.08006 0.08006 75 0.074453 0.097958 0.068034 0.069285 0.067151 0.10284 0.076314 0.076314 100 0.073609 0.09645 0.066527 0.068406 0.066432 0.097366 0.072067 0.072067 125 0.073146 0.09645 0.065475 0.067762 0.066432 0.094842 0.070066 0.070066 150 0.073146 0.09645 0.064815 0.067023 0.066432 0.093693 0.06924 0.06924 175 0.073146 0.09645 0.064327 0.065547 0.066432 0.074919 0.069051 0.069051 200 0.073146 0.09645 0.063768 0.064206 0.066432 0.073252 0.069051 0.069051 225 0.073146 0.09645 0.063223 0.063346 0.066432 0.072458 0.069051 0.069051 250 0.073146 0.09645 0.063223 0.06317 0.066432 0.071965 0.069051 0.069051 275 0.073146 0.09645 0.063223 0.063061 0.066432 0.07085 0.069051 0.069051 300 0.073146 0.09645 0.063223 0.063038 0.066432 0.07085 0.069051 0.069051 226
VI. ANFIS CONTROLLER The temperature of the steam at the outlet of the super heater, which runs the turbine, is the controlled output of the process. It has to be maintained within narrow limits around the set point value of 5400 o C. To achieve this, the spray water level is to be determined. This parameter is taken as the output of the controller. The super heater inlet steam temperature, steam flow rate, burner tilt angle are taken as the other inputs to the controller [8]. The FIS automatically generates 243 rules for the determination of the spray water level depending on the value of the inputs so as to minimize the error. The surface plot of the rules for various membership functions taking the final super heater outlet temperature and final super heater inlet temperature as inputs and spray water level as output and with keeping the other process variables at constant values corresponding to optimum response are given in Figures 10 to 17. The plots are obtained by keeping the elevation of the fire ball in the boiler (Burner Tilt Master) as 73.75% and steam flow is 666.68 kg/sec. Plots can be generated in the same manner for other combinations of input and output process variables. They can also be generated in a straightforward manner [9]. Fig 12 Surface Plot for Gbell Membership Function Fig 13 Surface Plot for Gauss Membership Function Fig 10 Surface Plot for Triangular Membership Function Fig 14 Surface Plots For Guass2 Membership Function Fig 11 Surface Plot for Trapezoidal Membership Function Fig 15 Surface Plot for Dsigmoid Membership Function 227
Fig 16 Surface Plot for Pi Membership Function [3] Talaq J. and Al-Basari F. Adaptive Fuzzy Gain Scheduling for Load Frequency Control, IEEE transactions on power systems, Vol.14, No.1, pp.145-150. [4] Wang L.X. (1994) Adaptive fuzzy systems and control: Design and stability analysis, Prentice Hall. [5] Rizzi A., Mascioli F.F.M. and Martinelli G. Automatic training of ANFIS Networks, Proceedings of the IEEE international fuzzy systems conference, Vol.3. Pp.1655-1660. [6] Sugeno M. and Kang G.T. Structure identification of fuzzy model, IEEE transactions on fuzzy sets and systems, Vol.28, No. 4, pp.15-33. [7] Li Y., Tan K.C., Ng K.C. and Murray-Smith D.J. ( Performance based linear control systems design by genetic evolution with simulated annealing, Proceedings of the 34th IEEE conference on decision and control, Vol.1. pp. 731-736. [8] Krolikowski A. Sequential identification and control for bounded noise ARX signals, IEEE transactions on automatic control, Vol.26, No.2, pp.325-331. [9] Kulessky R., Hain Y. and Nudelman G. Conception of PID Robust Control for Power Station Processes, IEEE Transactions on Power Systems, Vol.15, pp.1073-1080. Fig 17 Surface Plot for Pisigmoid Membership Function VII. CONCLUSION In this paper, Adaptive Network based Fuzzy Inference Systems (ANFIS) is effectively used for generation of membership functions to model the process variables of a superheated steam temperature control process. The inputs to the system are obtained by making real time measurements of process variables. The performance of the various membership functions in tracking the input output data set is compared. The design of an ANFIS based controller is carried out. The rule base for various membership functions is automatically generated and the surface plot is shown for a combination of inputs and output. Here a detailed description of the various schemes for the control of the superheated steam temperature controls process is discussed and the transfer functions identified from the real time data were used as the models on which the various controllers can be designed. REFERENCES [1] Christopher Foslein W. and Samad T. Fuzzy controller synthesis with neural network process models Proceedings of the IEEE international symposium on intelligent control, pp.370-375. [2] Lima C.A.M., Coelho A.L.V., Fernando J. and Zuben V. Fuzzy system design via ensembles of ANFIS, IEEE Proceedings of the IEEE International conference on fuzzy systemsvol.1, pp. 506-511. AUTHOR S PROFILE Mr. Subhash Gupta has received his B.E in Electrical Engineering from BIT Durg (India) and M.E. in Power Electronics from SGSITS,Indore (India) in 2000 and 2001 respectively. He has published many papers in the IEEE and IET and some other referred journals.he is a research scholar in Shri Venkateshwara University,India where he is currently pursuing PhD. His area of interest includes power quality improvement in distribution networks, electric machine modeling, power systems control, control system and integration of renewable into the power delivery system.. Mr. L.Rajaji is working as a Professor in the Department of Electrical & Electronics Engineering, PB College of Engineering, Chennai. He received his BE (Electrical & Electronics) and ME (Electrical Power Engineering) degree in the year 1997 and 2000 from University of Madras and The Maharaja Sayajirao University of Baroda Vadodara, Gujarat, India. He received his PhD in the area of Electrical Energy Conservation during the year 2010 from Sathyabama University, Chennai. He has guided 10 BE projects, 5 ME projects.he is guiding 5 students towards PhD program. He published 25 research articles in various referred international journals, national journals, international conferences and national conferences. He is review committee member for various journals. He acted as a Chair Person for various national level conferences. His area of interest includes power quality improvement in distribution networks, electric machine modeling, power systems control, and integration of renewable into the power delivery system. Ms. Kalika has received her Bachelor of Technology degrees in Electrical Engineering from Jamia Millia Islamia, New Delhi (India) and Master of Engineering in Power Electronics from RGPVV, Bhopal (India), in 2001 and 2003 respectively. She has attended many conferences and published paper in the IEEE and TET. and She is a research scholar Shri Venkateshwara University,India where she is currently working towards her PhD. Her current research interest includes renewable energy system, power electronics and solar energy system. 228