Research on combined model based on multi-objective optimization and application in time series forecast

Shenghui Zhang, Jiyang Wang, Zhen-hai Guo

Abstract: Combined model theory has been widely used to forecast the time series problems that are often nonlinear, nonstationary and irregular. Current forecasting models based on combined models theory could adapt to some time series data and overcome some disadvantages of the single models. However, in previous studies, most forecasting models have just focused on improving the accuracy or stability. Nevertheless, for an effective forecasting model, considering only one criterion or rule (stability or accuracy) is insufficient. In this paper, a novel forecasting system, called non-dominated sorting genetic algorithm III combined system with three objective functions, was proposed and successfully employed to solve the predicament of electricity load forecasting which demands to obtain both high accuracy and strong stability as an example. Both stability and accuracy of our proposed combined system are superior to the model compared which showed in the experiment results.