Research on prediction method based on ARIMA-BP combination model

Abstract

This paper puts forward a new weighting method for combined forecasting model-MSSE (Minimum Sum of Squared Errors), and combining ARIMA (Autoregressive Integrated Moving Average model) time series model with BP (Back Propagation) neural network, determining the weight of a single prediction model by MSSE. And the combined forecasting model is applied to the prediction of grain production in China. Through experiments, it is found that, prediction results of combined forecasting model based on MSSE, the sum of squares of errors is 40.62% lower than ARIMA time series model and is 29.99% lower than BP neural network. Therefore, the combined forecasting model obtained by this weighting method has good forecasting performance.