Combined Method of Artificial Intelligence Regression Forecasting Models Under Fluctuating Errors

Abstract

Because precisions of different multiple-regression methods are different for the same samples, how to improve accuracy of forecasting has therefore generated wide interest. This paper attempted to improve precision of forecast by combining multiple linear regression and three artificial intelligence regressive models. In our study, a novel frame of model combination is proposed by fluctuating degree, complementarity and compatibility. Complementarity is available to judge which models can be combined to decease the errors and establish its sets. The assigned weights of each single model in complementary sets are calculated by fluctuating degree. Compatibility, superiors and inferiors of a combined model are evaluated by MAE, RMSE and MAPE. The empirical case of predicting electric demand demonstrated that the combined models based on fluctuating degree increase predicting precision of sample period and extrapolative forecast if there exists complement between different single-models.