The Research Of Forecasting Method Based On Combined Algorithm

Prediction is closed to our life, which has an important position in real life. Suchas weather forecasts, stock market prediction, etc., it is related to all aspects of our lifeand our production. The prediction of time series is the important area in the researchof forecasting. By analyzing the link between the data, to identify their law, toestablish a mathematical model to fit such a law, is the basic method for thequantitative prediction. Traditional time series forecasting method is relatively simple,lacks of flexibility, and only to linear the prediction of stationary time series. However,a large number of data existed in our real life is non-linear data, which requires peopleto search better methods of prediction.Gray theory and neural networks have feasibility in the prediction of non-lineardata. Also, combined forecasting can predict the best result. Combined forecasting canovercome the shortcomings of a single model, and get to achieve higher predictedaccuracy. This article describes the traditional time series forecasting methods, graymodel, neural networks, the basic theory of the combination of model and dataanalysis and forecasting.Three forecasting methods are proposed at fist, AR model, the gray GM (1,1)model and gray neural network model. The neural network is a dynamic Elman neuralnetwork model. By comparing error and the square error, we finding that the graymodel have better advantage in predicting nonlinear data than the AR model, andcombination forecasting method predicted results better than a single predictionmethod.Finally, it presents the advantages and disadvantages of all methods, andsummarizes the content of this study and the direction and focus of research in thefuture.