The Effect of Model Input on Forecast Results of the Neural Network Ensemble Model

摘要

A new calculation method for the input of the neural network ensemble prediction (NNEP) model has been developed based on the data mining technology using the feature extraction method of Empirical Orthogonal Function (EOF) and the stepwise regression method, for investigating the effect of different model input with the same dimension on the prediction capacity of the NNEP model. Taking typhoon intensity in summer (June, July and August) in the Northwest Pacific in China as the prediction object, a new NNEP model for typhoon intensity was established. Using identical sample cases and input dimension, predictions of typhoon intensity with multi-model and large sample size were performed. Results show that the methodology of EOF combined with stepwise regression method can mine the useful prediction information from all the predictors, so the prediction accuracy of the NNEP model is clearly improved.