Optimize RBF Neural Network Controller Based On The Genetic Algorithm

Radial Basis Function (Radial Basis Function, RBF) Neural Network is a kind of three-layer feed-forward network, which has a lot of charcateristics, such as fast learning ability, simple network structure, excellent performance of approximation and nonexistence of local minimality, etc, and has been widely used in domains of functional approximation, pattern recognition, signal processing and system identification, and so on. The RBF neural network, which has a neural network architecture of simulating partial adjustments that occur in human-beings'brain, and of mutully covering receiving fields, can approach any continuous function with any precision. In this paper, mainly, the RBF neural network's learning algorithm and structural design methods have been researched, and design a controller with good performance based on genetic algorithm.Genetic algorithm is a kind of adaptive optimized algorithm, which is formed by simulating creatures'heredity and evolution in the natural environment, and this ideology is originated from biological genetics and the survival fittest in nature, thereby this is an algorithm with the iterative process of survival and detection. But the parameter selecting of genetic algorithm is subjective, which has slower speed of convergence and more easily tend to fall into premature. In order to solve these problems, this paper presents a RBF neural network's learning method based on improved genetic algorithm.The improved genetic algorithm overcome the premature phenomenon by introducing elite preservation, adaptive crossover probability and the theroy of position of particle swarm optimization, consequently improve the network's convergence precision. By making the controller search the optimal solution in the global, ideal result will be brought.Finally, we choose double inverted pendulum as a controlled object, and use a improved genetic algorithm to optimize RBF neural networks, moreover, construct and design control module of RBF neural network. we verify the controller's performance by practical control and Matlab simulation.