Research Of PMSM Control System Based On GA ANN

The character that neural network(NN) is able to approximate arbitrarily any nonlinear function makes NN widely applied in control field as a excellent tool to handle nonlinear and uncertain systems.Permanent magnet synchronous motor (PMSM) control system is a typical nonlinear system, its dynamic performance changes great because of the time varying parameter, nonlinear factor and change of load. All these make it difficult to grasp the running environment and to build correct mathematic model. Considering the characters of NN, such as independence of model, self-studying, self-adaption, nonlinear function approximating, parallel computing, fault-tolerant, high performance control strategy is expected by applying neural networks to PMSM control system.Firstly, this paper analyses the fundamental of neural network and separately gives learning algorithm of feed-forward neural network and feedback neural network, taking BPNN, RBFNN and Elman NN for example. Then the PMSM control system and its identification based on neural network are researched. The simulation results of BPNN and RBFNN show that neural network can approximate PMSM control system characteristic with high precision and has satisfying generalization ability.Aiming at problems that the structure of RBF network is uncertain and its initial parameters have great influence on control performance in application, this paper makes further research and gives a method which applies genetic algorithm(GA) to optimize the structure and parameters of RBFNN. GA is a global search and optimization algorithm which simulates the genetic and long-term evolvement process of biology. This paper detailedly analyzes the development, theory foundation and characteristics, inscape, operation, the current situation in optimizing the neural network and realization of genetic algorithm. Then applying GA to optimizing RBF network is researched. The method's feasibility and utility have been proved in theory and practice, and the simulation results show scheme has the virtues of neural networks and GA.This paper presents speed loop internal model control of PMSM based on the RBFNN optimized by GA. The simulation shows the method effectively overcomes the influence caused by nonlinear factor and time varying parameters, improves the robustness, dynamical and steady performance, and reaches the expected goal.