The Coal Price Forecasting Based On ARIMA And SVM Model

Coal is the basic resources of our country, playing an absolute dominant role in the energy markets for a long time. Sharp fluctuation of coal price will impact on macroeconomic and regional economy, resident’s life, especially for involved coal enterprises. Therefore, accurate predictions make decision makers of enterprises grasp short-term change trend and do strategic plan. The traditional ways to forecast coal price mainly be time series method or artificial intelligence method. The single model could not capture coal price’s comprehensive trend, because of unsteady and nonlinear time-series.1)The non-stationary sequence of coal price converts to smooth time series with the method of first order difference. Comparing the evaluation index and parameter estimation results, three possible models which conclude ARIMA(6,1,5), ARIMA(6,1,6) and ARIMA(6,1,9) are selected. Considering whether residual error sequence is pure randomness and the correlation coefficient is within the confidence interval, we choose ARIMA(6,1,6) to predict coal price during 2014.5 to 2015.2.The prediction results show that the value of MAPE is 1.65,achiving a high accuracy requirement;2)This paper build SVM prediction model, using PSO algorithm to optimize parameters of SVM. The result shows that MAPE is 1.68, gaining better effects.3)Coal price time series contains linear and nonlinear parts, which are not completely independent of each other. This paper builds ARIMA and SVM combined model, adopting the idea of parallel forecasting. Its MAPE is 1.36, which much smaller than single model, showing that it makes full use of the implicit information in original data and validates the feasibility and effectiveness.