A novel hybrid approach for hourly wind speed forecasting based on variational mode decomposition, data feature reconstruction, and machine learning

Short-term wind speed prediction plays a vital role in wind power generation, which has a significant impact on dispatching and operation decisions. However, the hourly wind speed time series exhibits the feature of high intermittency and high fluctuation, limiting the forecasting performance and timeliness. Therefore, the novel intelligent hybrid prediction approach was developed based on variational mode decomposition (VMD), fuzzy entropy test (FE), and Elman neural network. In the proposed approach, the chaos degree of wind speed time series was deceased by VMD to create a great environment for the following prediction work. Considering the timeliness, the reconstruction based on FE was designed, which reduced the number of times to run forecasting model. The proposed approach is applied to the five cases collected from two different wind farms. The obtained results indicated that the proposed approach owned the optimal performance and its average prediction accuracy was improved by 56.40% compared with that of other comparative models. Meanwhile, the timeliness of the proposed approach was doubled after the reconstruction based on FE. Hence, the proposed approach can meet the requirements of wind farm to obtain the accurate prediction of hourly wind speed timely.