A Particle Swarm Optimization-Neural Network Prediction Model for Typhoon Intensity Based on Isometric Mapping Algorithm

Abstract

In terms of the Particle Swarm Optimization-Neural Network (PSO-NN), a new prediction model has been developed using the stepwise regression method combined with the feature extraction technique of Isometric Mapping (ISOMAP) algorithm to treat the Climatology and Persistence (CLIPER) predictors. The model is validated with forecasts of ten years of typhoon intensity formed and numbered in the Western Pacific Ocean over May-October, 2001-2010. Using identical sample cases, predictions of the PSO-NN model based on ISOMAP algorithm are compared with the CLIPER model widely used in China and overseas, and it has been proven experimentally that the former is more accurate.