Wavelet Neural Network Controller Optimized By Particle Swarm Optimization

Wavelet neural network is a novel network which combines wavelet analysis theory and artificial neural network. Therefore, it inherits the self-learning ability of neural network and the time-frequency localization of wavelet analysis, and it has great approximation ability and fault tolerant ability, which make it superior to traditional neural network in dealing with system with complicated nonlinearity and uncertainty. It has an expansively development and application prospects. This paper focuses on the learning algorithm and structure design of wavelet neural network, and develops a new excellent controller. Also simulation is conducted to test our method.This paper adopts particle swarm optimization to optimize wavelet neural network. As genetic algorithm, particle swarm optimization is also an iterative based optimization algorithm. Collaboration of individuals completes the work of finding optimal solution. Instead of crossing and variation in genetic algorithm, particle swarm optimization initializes a set of random solution, and then searches for the optimal particle in the solution space.The particle swarm optimization has a fast convergence speed and is not easy to fall into local minima. On these bases, this paper proposes an improved particle swarm optimaztion based on genetic variation factors. Crossing factors of genetic algorithm is added to particle swarm optimaztion and inertia weight factor is also added using linear regression strategy. Optimization results of the improved particle swarm optimization optimizing wavelet neural network indicate the proposed algorithm can effectively improve speed and accuracy of converge.Finally, the commonly used Morlet wavelet is selected as wavelet basis function of the wavelet neural network. According to the characteristics of double inverted pendulum, the controller module is designed. Simulation is also conducted, and the algorithm proposed in this paper is applied to control the double inverted pendulum. Experiment results verify the effectiveness of the wavelet neural network controller. In addition, the controller has great anti-interference ability.