Research And Application Of Combined Model Based On Multi-Objective Theory

The combined model theory proposed by Bates and Granger has been widely used in the forecasting of time series problems.At present,the forecasting model based on the combined model theory can predict some time series data better and overcome the deficiency of poor forecasting accuracy of single model.However,most of combined models in the previous studies are based on the combination model theory and most of them only focus on improving accuracy or stability.For an effective forecasting model,considering only a single index can not effectively describe the complexity of data,thus affecting the accuracy and stability of forecasting.Therefore,it is necessary to study the multi-objective combined forecasting model in depth.It has important significance to promote and improve the development of combined forecasting theory and forecasting.In this paper,a combined forecasting model with three objective functions based on multi-objective optimization is proposed,and then it is evaluated with several popular measurements.The model consisting of weights from a single model and three objective functions are optimized by non-dominated sorting genetic algorithm ?(NSGA-?)algorithm which determines the optimal weights of the combined model.The model presents the characteristics of high accuracy and stability.The experimental results show that compared with the traditional single models,the forecast accuracy and stability of the combined forecasting model based on multi-objective optimization have been greatly improved.It provides strong technical support for practical problems such as power system scheduling and wind power evaluation.