The Improvement Of Algorithm For Controller Parameters Self-tuning Based On BP Neural Network

The controller of PID is widely used in many fields, but there are some setting problems of parameters in the process of using. As a general rule, the setting of the parameters are according to the experience of the engineers, but it is a hard work for the ones who have no experience. With the BP neural network, the problem of the setting of parameters can be solved easily by its self-learn ability. With the problem of the complex of the BP neural network, the evolution of the structure is always the core one. The research of the BP neural network model has significant meanings.BP neural network is widely used in the classification pattern recognition, image processing, and controller systems as a nonlinear system. It has a feed forward structure, includes the input layer, the hidden layer and the output layer. The improvement is mainly about the hidden layer, because the input layer and the output layer have the relationship with the system itself, can’t do big improvement. The hidden layer include in two problems, the nodes number and the layer number. Generally, only one layer is enough to solve all the problems, so the node number is the key problem we want to solve. When the structure can be sufficient for the industry needs as it is the most simple, the neural network is the best.This article is mainly about the problems of the hidden layer nodes and proposes some improve methods, a form up to down optimization method. With the fully introduction of the method of design and algorithm, using the matlab platform, the method is verification. Two differential operators are designed for the PIαDβ controller, with the BP neural network and the controlled object, an integral system is composed. With the platform, we can sec whether the algorithm is reliable. From the result we can see that after deleting or combining some non-important nodes, the performance is better, the time of the self-learning and PIαDβ controller self-turning are shortened, the robustness of the system is enhanced.