Study Of Synchronous Generator Excitation Control Based On Neural Network Identification

At present, with the increasing of single generator capacity and scale of power system, improving the stability of synchronous generator operation is the one of essential requirements for guaranteeing grid security and ensuring power system economical efficiency. Using modern control theories and advancing generator excitation system control effects are recognized as one of the economic and effective means in numerous measures of improving synchronous generator stable operation.It is difficult to get the exact mathematical model during the operation process of the synchronization generator because of its characteristics, such as nonlinear, multi-variable, strong coupling and so on so force. The traditional PID control method based on the exact model cannot satisfy these requirements. This article is to advance the performance of excitation control of the synchronous generator by improving control strategies.Artificial neural network algorithm's merit is not dependent on accurate models. This article puts forward a control mode which based on cosine basis functions RBF neural network identify in order to solve these problems mentioned above. Firstly, it illustrates the traditional neural network deficiency, lists several commonly neurons excitation functions and discusses Variable Learning Rate advantages. Secondly, to adjust the PID three weight coefficient by using the gradient descent algorithm. Thirdly, the partial derivative referring to generator output voltage to generator input voltage variation cannot accurate value by conventional mathematical calculation because it is hard to gain accurate model of the generator during PID three weight coefficient's adjustments. There are usually two ways to solve this problem. One is to use the changes of generator voltage output and the changes of input to replace the partial derivative; the other is to use neural network identification output to calculate the partial derivative. The research shows that using neural network identification output to calculate the partial derivative can improve control accuracy. Fourth, in view of the neural network identification problem carried on the detailed research, studies show that using cosine basis function as neural network excitation function would help to improve the whole excitation control system effect. Finally, this paper compares the fixed weight coefficient ( k p,k i,k d) PID excitation control, non-network identification PID excitation control, Gaussian function RBF neural network identify PID excitation control and based on cosine basis functions RBF neural network identify PID excitation control system respectively according to the excitation control simplified transfer function model by using MALTAB simulation. From the simulation results can be seen, the based on cosine basis functions RBF neural network identify PID excitation control effect is better than the other way.