Design And Research Of Controller Based On RBF Neural Network

Radial Basis Function (Radial Basis Function,RBF)Neural Network for its strong physiological basis, fast learning ability, simple network structure, excellent approximation performance, has obtained a wide range of fields of application in function approximation, pattern recognition, signal processing, system identification. RBF neural network's simulation of human brain partial adjustment, mutual coverage received neural network architecture, is a kind of universal approximation properties of three-layer feed-forward network. In this paper, the RBF neural network's learning algorithm, parameter initialization, Gaussian function and structural design methods have been studied to design a good performance of the controller.Firstly, the particle swarm optimization is through the collaboration between individuals to find the optimal solution. PSO is similar with the genetic algorithm,and is a tool based on iterative optimization. The system is initialized to a set of random solutions, through the iterative search for the optimal value. But there is no cross-use of genetic algorithms and variation, the particles in the solution space of the particles follow the optimal search. For the particle swarm algorithm is easy to fall into local minima, convergence speed and other characteristics, this paper presents a genetic variation factor based on improved particle swarm RBF neural network learning algorithm.Secondly, PSO algorithm is added to the inertia weight factor, which is using the linear regression of the right strategy to enhance the convergence precision. At the same time, combined with genetic algorithm crossover and mutation factor to optimize the hidden layer and output layer connection weights in each step of the genetic algorithm to optimize operations and combined with the elite preservation strategy can play a very good optimization result. Finally, on the basis of previous studies, according to two characteristics of an inverted pendulum,this paper optimize RBF neural network using inproved particle swarm optimization algorithm, build the design module for RBF neural network and and verify the controller's performance.