The Forecasting Combine Method With Wavelet And Optimization Theory And Its Application

Predicting is the process of studying historical events, building system models onprevious results and making predictions. With the developing of science and technology,the prediction methods have been greatly improved. In the practice of prediction,time-series often present in highly sophisticated dynamics nonlinear states. For this reason,fitting predictions can not be conduct well by single prediction method. Based on thissituation, to improve the accuracy of fitting, various ways of prediction methodsmodification were put forward in this dissertation. On the other hand, if combined in aproper way, different prediction methods will become combination forecasting methods.In this way, information collected in different ways could be put together, thus improvesthe accuracy of fitting. Compared with single prediction method, the method gets betterresults and has been widely used.The combination prediction model is based on ARIMA model, BP neural networkand gray systematic theory. Meanwhile, the conjugate gradient methods and waveletanalysis theory are applied to improve the original single model, and results proved thatthe methods mentioned are better. In this thesis, the improvement and combinationprediction model are applied in the stock market, demographic data and other fields. Theeffectiveness of the improved method is verified by actual data fitting results.Wavelet Transform are used to preprocess stock history time series data, afterwards,preprocessed series are modeled and predicted by combination neural network models andvarious prediction models.First, wavelet noise reduction processing method, nonlinear conjugate gradientmethod and the basic idea of combination forecasting model, system analysis of timeseries model parameter optimization method are studied. Chapter three and chapter fourpresents two modified nonlinear conjugate gradient method and successfully applied to thetime series model parameter optimization in the final of chapter four, namely the use of animproved conjugate gradient method to optimize the parameters of the model. Withexamples demonstrate the effectiveness of proposed method, advantages. By the end ofthe chapter, examples prove the feasibility of this method. After that, an introduction of gray model is given. New formulas are applied to formadjacent generate sequence while forming combination prediction by using artificialneural network. By the end of the chapter, examples prove the feasibility of this method.Based on wavelet denoising model, ARIMA model, neural network model andcombination methods, on the basis of, some prediction methods are given according towavelet transform combination models. And predicting process are presented in flowcharts.Finally, the paper gives several forecasting methods compared with traditional timeseries analysis method and a single model. Then stock prediction has been greatlyimproved.