Study Of Short Term Climatic Prediction Method Of Neural Network Based On Phase Space Reconstruction

Short-term climate prediction is compared to long-term climate prediction, mainly is the prediction of monthly, quarterly, and annual. Short-term prediction has great influence on human life and production, and also takes important role on national economic construction and disaster prevention and mitigation. According to the characteristic nonlinear and chaos of climate system, this article adopt the method of phase space reconstruction theory and ANN analysis to establish the mode of short-term climate analysis. Through this method, we can improve the accuracy of prediction and provide the guarantee for national economic and social production.The research of this article are as follows:Introduce the basic theory of artificial neural network and the application status in short-term prediction. Elaborate the basic theory and training process of BP neural network, ELM and do the improvement for ELM. Take the monthly prediction of Hangzhou station as example to observe the prediction result of BP neural network, ELM and AD-ELM.Introduce the development and application of Chaos theory, the concept of phase space and phase space reconstruction, and the selection of time delay in phase space reconstruction method and embedding dimension. Analysis of the time Chaos of monthly precipitation and use the phase space reconstruction method on monthly precipitation prediction and analyse the feasibility.Combining the Chaos theory with neural network theory, to establish phase space reconstruction for neural network prediction mode. Use empirical mode decomposition method to process data for the complicated and non-stationary time series. Both the phase space reconstruction technology and artificial neural network are the effective tool to solve the nonlinear system problems. We can combine them to excavate the information of time Chaos and solve the nonlinear problems effectively, and make the result of prediction consistent with reality. Take the climate changed data around50years of Hangzhou, Nanjing, Pukou in Nanjing, Suzhou in subtropical monsoon climate zone as example, to do the training and test for phase space reconstruction neural network predication mode, and then compare it with actual data, the predication result has highly consistency with actual condition. The accuracy of the prediction method is reasonable and it can be a good reference for short-term climate prediction.