Distribution drift-adaptive short-term wind speed forecasting

Accurate short-term wind speed forecasting is essential for wind power system scheduling optimization and profit maximization. However, the distribution of wind speed evolves over time. Stable and reliable forecasting results require that the wind speed forecasting methods be adaptive to the wind speed distribution drift. Thus, how to make the forecasting method adaptive to the wind speed distribution drift becomes a challenge. In this study, the distribution of future wind speed is predicted using a tiled convolutional neural network (TCNN) based-model. The distribution deviation between historical and future wind speed is minimized via weighting the loss contribution of historical data. A branch accumulation error decreasing (BED) rule is introduced to adaptively determine the optimal mode number for the variational mode decomposition (VMD) method. Two hybrid models which employ both the distribution drift correction process and BED rule-based decomposition process are proposed. The effectiveness of the proposed models is verified using data from two different wind farms in China. Compared with the traditional short-term wind speed forecasting models, the proposed models show considerably better robustness to the distribution drift of the wind speed and achieve significantly higher forecasting accuracy in both the one-step ahead and multistep ahead wind speed forecasting scenarios.