A Prediction of the Monthly Precipitation Model Based on PSO-ANN and its Applications

Abstract

A nonlinear prediction model has been presented of PSO-ANN of monthly precipitation in rain season. It differs from traditional prediction modeling in the following aspects: (1) input factors of the PSO-ANN model of monthly precipitation were selected from a large quantity of preceding period high correlation factors, and they were also highly information-condensed by using the empirical orthogonal function (EOF) method; which effectively condensed the useful information of predictors. (2) Different from the traditional neural network modeling, the PSO-ANN modeling is able to objectively determine the network structure of the PSO-ANN model, and the model constructed has a better generalization capability. The model changes the prediction of climate field to that of the principal component of that field. According to the approximate invariability of eigenvectors of climate field, the prediction of climate field is obtained by return computation, together with the principal component. A test example is predicting the flood period rainfall for the 37 basic stations in Guangxi. The prediction of field for June to September in 2009 is made and comparisons with the field of observations. The results show that the predictive efficacy is remarkable.