A combined forecasting system based on modified multi-objective optimization and sub-model selection strategy for short-term wind speed

Abstract

Forecasting models have been widely used in wind-speed time series forecasting that are often nonlinear, irregular, and non-stationary. Current forecasting models based on artificial neural network can adapt to various wind-speed time series. However, they cannot simultaneously and effectively forecast the entire wind-speed time series of a wind farm. In this paper, a novel combined forecasting system is developed for a wind farm that includes that SSAWD secondary de-noising algorithm is used to pre-process original wind speed data, and then the sub-model selection strategy is used to select five optimal sub models for the combined model. Meanwhile, a modified multi-objective optimization algorithm optimizes weight of the combined model, and the experimental results show that this forecasting system outperforms other traditional systems and can be effectively used to forecast wind-speed time series of a large wind farm.