Modes decomposition forecasting approach for ultra-short-term wind speed

Abstract

The accurate forecasting of ultra-short-term wind speed is of great significance in theory and practice. This paper proposes a modes decomposition forecasting approach based on adaptive variational mode decomposition and weighted combination models for ultra-short-term wind speed. First, an adaptive variational mode decomposition algorithm is used for decomposing the original ultra-short-term wind speed time series into several modal components. Second, auto regressive integrated moving average, support vector machine and the improved long short-term memory are determined as forecasting models of different components by Hurst exponent analysis. Then, an improved particle swarm optimization algorithm is proposed to optimize the weight coefficient of each forecasting model. Finally, the final forecasted value is obtained by multiplying the forecasted value of each sub-forecasting model by their respective weight coefficient. Four groups of measured ultra-short-term wind speed data with 5-minute, 10-minute, 20-minute and 30-minute sampling periods are taken as the research object. Compared with other single or combination forecasting models, the proposed forecasting approach has higher prediction accuracy and more promising prediction performance for ultra-short-term wind speed.