An Application of Ensemble Prediction for Typhoon Intensity Based on MDS and PSO-ANN

Based on the samples of tropical cyclones in June-October from 2001-2010 over the western pacific sea. A nonlinear prediction models of tropical cyclones intensity has been presented by PSO-ANN method. It differs from traditional prediction modeling in the following aspects: (1)About the input factors of the PSO-ANN model, firstly using stepwise regression selected a combination of factors from 62 Climate continues factors, and then using multi-dimensional scale transformation to reduce dimension and extract information from the remaining factors of Climate continues factors.(2) Different from the traditional neural network model, the PSO-ANN model is able to objectively determine the structure of PSO-ANN model, and the model has a better generalization capability. In the prediction of the 30 independent samples test, the result from June to October months of 24-72h aging show that the PSO-ANN model are superior to the CLIPER models. In prediction accuracy, the average absolute error of the PSO-ANN model was less than the CLIPER models from 3% to 14%. It showed that the proposed nonlinear Pso-Ann West Pacific tropical cyclones intensity prediction model is valuable.