The Meteorological Prediction Model Study of Neural Ensemble Based on PSO Algorithms

This paper presents the evolving neural network architecture and beginning connection weights based on particle swarm optimization algorithms in order to produces the better network architecture and beginning connection weights, trains again the traditional back propagation by training samples and ensembles results by mean. This method be established the forecast model. The applied example is build with monthly mean rainfall the whole area of Guangxi. The result show that method represses neural network dependence on beginning connection weights, acquires compact neural network architecture, can effectively improves convergence speed and generalization ability of neural network