Study On Application Of Time Series Models In Forecasting Crude Price

The paper, through the time series analysis, combined with the advantage of the data mining technology in the mass data processing, which combine the statistical model and intelligent model, proposes the prediction model of’ARIMA-Bernstein’, tries something in crude oil price prediction.First, the paper introduces the influence factors of crude oil prices, and how it affect the industry and economy of a country, the present situation of countries in the oil price prediction and some theory and method which is commonly used. A detailed derivation is given from regression model (AR), moving average model (MA) to regression moving average (ARMA) model and the evolution processing of ARIMA model, to lay the foundation for the new model. Second, based on the data characteristics of crude oil prices, the paper introduces the theory of Bernstein primary function in the data mining technology to forecast the oil price. Through comparing the method of prediction with ARIMA model and Bernstein primary function, we find the method of Bernstein primary function is better than ARIMA, but the results between prediction of oil prices and the real value exist the huge difference, we combine the two methods which use the fitting Bernstein primary function to fit the residual after using ARIMA model, then parallelogram law to predict the results of the residual, the two prediction results composed the combination function, which is the final result. The results of application show that the prediction of combination function is more close to real results.The paper combines the methods of time series analysis and the data mining technology, proves the better prediction results through the application. Time series model can make the tranquilization of unstable data and reflect the periodicity. Data mining reflects the function of filtering the mass data, so the convergence of the two methods have the vast potential for future development.