Robust and Adaptive Load Frequency Control of Multi-area Power Systems with system parametric uncertainties via TDMLP

In this paper a robust and adaptive Temporal Difference learning based MLP (TDMLP) neural network for power system Load Frequency Control (LFC) is presented. Power systems, such as other industrial processes, are nonlinear and have parametric uncertainties that for controller design had to take the uncertainties into account. For this reason, in the design of LFC controller the idea of TDMLP neural network is being used. Some simulations with two interconnections are given to illustrate proposed method. Results on interconnected power system show that the proposed method not only is robust to increasing of load perturbations and operating point variations, but also it gives good dynamic response compared with traditional controllers. It guarantees the stability of the overall system even in the presence of generation rate constraint (GRC). To evaluate the usefulness of proposed method we compare the response of this method with RBF neural network and PID controller. Simulation results show the TDMLP has the better control performance than RBF neural network and PID controller. Introduction In power systems, one of the most important issues is the load frequency control (LFC), which deals with the problem of how to deliver the demanded power of the desired frequency with minimum transient oscillations [1, 2]. Whenever any suddenly small load perturbations resulted from the demands of customers occur in any areas of the power system, the changes of tie-line power exchanges and the frequency deviations will occur. Thus, to improve the stability and performance of the power system, generator frequency should be setup under different loading conditions. For this reason, many control approaches have been developed for the load frequency control. Among them, PID controllers [3], optimal [4], nonlinear [5] and robust [6-8] control strategies, and neural and/or fuzzy [9-11] approaches are to be mentioned. An industrial plant, such as a power system, always contains parametric uncertainties. As the operating point of a power system and its parameter changes continuously, a fixed controller may no longer be suitable in all operating conditions. In order to take, the parametric uncertainties into account, several papers have been published in the concept of variable structure systems [12], various adaptive control techniques [13] to the design of load frequency control. In this paper, because of the inherent nonlinearity of power system a new artificial neural network based intelligent controller, which has the advance adaptive control configuration, is designed. The proposed controller uses the capability of the MLP neural network based on Temporal Difference (TD) learning for the design of LFC controller. In this work, for the design of MLP neural network the idea of TD learning and applying it to nonlinear power system is being used. The motivation of using the TD learning for training of the MLP neural network is to take the large parametric uncertainties into account so that both stability of the overall system and good performance have been achieved for all admissible uncertainties. Moreover, the proposed controller also makes use of a piece of information which is not used in conventional controllers (an estimate of the electric load perturbation, i.e. an estimate of the change in electric load when such a change occurs on the bus). The load perturbation estimate could be obtained either by a linear estimator, or by a nonlinear neural network estimator in certain situations. It could also be measured directly from the bus. We will show by simulation that when a load estimator is available, the neural network controller can achieve extremely dynamic response. In this study, the TDMLP neural network is considered for control interconnected power system with two areas with power tie-lines to supply different consumers. The simulation results obtained are shown that the proposed controller not only has good performance in the presence of the generation rate constraint (GRC), but also gives good dynamic response compare to RBF neural network and PID controller. This paper is organized as follows: Section 2 describes the power system and its mathematical model. In section 3, the whole structure of the proposed TDMLP neural network is shown. Section 4 describes the application of TDMLP in LFC. Section 5 shows the simulation results that have been compared with RBF neural network and PID controller. Some conclusion and remarks is discussed in section 6. Mathematical Model of Power System Plant The power systems are usually large-scale systems with complex nonlinear dynamics. However, for the design of LFC, the linearized model around operating point is sufficient to represent the power system dynamics [1]. Fig.1 shows the block diagram of i-th area power system. Each area including steam turbines contains governor and reheater stage of the steam turbine. According to Fig.1, time-constants of the ri T , ti T and gi T are considered for the reheater, turbine and governor of the thermal unit, respectively. Wherever the actual model consists of the generation rate constraints (GRC) and it would influence the performance of power systems significantly, the GRC is taken into account by adding a limiter to the turbine and also to the integral control part all of areas to prevent excessive control action. The GRC of the thermal unit is considered to be 0.3 p.u. per minute ( 005 . 0 = δ ). All areas have governors with dead-band effects which are important for speed control under small disturbances. The governor dead-band is also assumed to be 0.06%. Based on the suitable state variable chosen in Fig. 1, the following state-space model will be obtained: i B