Multi-Model Ensemble Forecast Method of PM2.5 Concentration Based on Wavelet Neural Networks

Abstract

Concerning the problem of uncertainty problem of different environment meteorological models and in order to improve the accuracy of PM2.5 concentration forecast, based on the forecast products of the three environment meteorological models including CUACE, BREMPS and WRF-Chem in Shanghai, the wavelet neural network combined wavelet theory with neural network was used to build the multi-model ensemble forecasting model of PM2.5 concentration. The experiment was carried out by data of Beijing station, and this model was compared with other four models (BP neural network, RBF neural network, Elman neural network and T-S fuzzy neural network). The results showed that PM2.5 forecasted by wavelet neural network was better than other models, the prediction deviation was reduced effectively. The PM2.5 daily average concentration forecasted by wavelet neural network was closest to the observation, the proposed model has comparatively high forecasting accuracies.