The Research Of Forecasting Method Based On Intelligent Algorithm

Forecasting is the deduction and expectation of the future which can be forecasted according to the history data behavior. Scientific forecast is the precondition and guarantee to make right policy. Forecasting, study and practice are called the three pillars in the activity that human explore the nature. There are many time series data in business, finance, engineering, medicine and social science fields. For example, daily sales volume in supermarket, the stock opening price, closing price, trading volume, they are all be regard as the time series data. The rule about how is the data evolved can be getted using the time series analysis method, at the same time the next behavior can be forecasted. Because of this, the forecasting becomes very important in our life.Three models are proposed, Wavelet-AR(WAR),Wavelet-AR-Elman neural network and Wavelet-nonparametric model. In this paper, the time series is decomposed into two parts using the method of multi-scale decomposition by the algorithm of Mallat and the wavelet of Daubechies based on the analysis of multi-scale theory. One is the detail part and the other is the approximate part. Then the detail part and approximate part are restructed to the original scale by the reconstmctible method of multi-scale of Mallat and the Daubechies. And new predicted data could be got from the three models. The sum of the results of every scale time series is the predict result of the original time series.Two time series are analyzed in this paper. First, the AR model, the wavelet-AR model and the wavelet non-parameter model are used to analysis and forecast the same time series about the quarterly data of a certain country gross national product. The forecasting results are analyzed and compared. The results show that the wavelet non-parameters model has feasibility in the prediction of time series.Second, the wavelet-AR model and the wavelet-AR-Elman neural network model is used respectively to analyze and forecast the periodic time series about gross industrial output value in a certain city. And the two forecasting results are analyzed and compared. It shows that the wavelet-AR-Elman model has feasibility in the prediction of time series with periodicity.