Developing neural network models for forecasting a sequence of events has been a challenging task in many domains, like, weather, finance, predictive control, etc. Wavelets neural network (WNN) has better approximation ability and error tolerance performance. It sufficiently utilizes the localization characteristics of wavelets combined with self-study and self-organizing functions of a common neural network. This paper presents a simple and definite training algorithm for tuning WNN parameters using Particle Swarm Optimization (PSO). The mean square error between network output and target was used as fitness function for PSO. The test results prove the faster convergence and better approximation of the PSO tuned WNN.
展开▼