Research And Application Of Combination Forecasting Model

Combination forecasting is a kind mode,which considers each model as a fragment to store different information and then integrate each fragment with some criterion to maximize the information integration.Scattered information is combined to minimize the uncertainty and randomness of information in order to meet the need of prediction accuracy.The combination forecasting mainly includes two important aspects,one is the choice of individual model,and the other is the combination mode of the two model.In this paper,the differential autoregressive moving average model(ARIMA model),adaptive filtering model,BP neural network model,gray GM model and exponential curve model are studied.In the way of combining,this paper combines the weighted combination and the error correction combination to study the two individual models.In weighted combination,this paper presents a combined weight forecasting model based on the percentage of the mean absolute error(MAPE)and the least square.First,the ARIMA model and exponential curve regression model are established.Then the weight coefficients of the combined model are determined by MAPE and least square method.Finally,we use MAPE and least squares comprehensive weight combination forecasting model to predict GDP in Hubei Province.The simulation results show that the synthetic weights can improve the accuracy of combination forecasting compared to the single weight.In the aspect of error correction combination,based on the complementary and common principle of the model,two combinations of grey adaptive filter and ARIMA-BP error correction model are established and verified by an example.The results show that the combination method can effectively improve the prediction accuracy of the model,and can also be a better combination forecasting model.The innovation of this paper is reflected in the following two points.1.At present,the weight of weighted combination forecasting model is mainly focused on obtaining weights with one weight standard.In this paper,a comprehensive weight combination forecasting model based on MAPE and least squares is proposed.Compared with a standard weight coefficient,the comprehensive weight can effectively improve the prediction accuracy.2.the study of the combination model of the combination forecasting model is concentrated in the form of weighting,and this paper puts forward another way of combining error correction.The example proves that the combination forecasting model of error correction is better than the weighted combination forecasting model,and it shows that the error correction is also a better combination mode.