Wind power interval forecasting based on adaptive decomposition and probabilistic regularised extreme learning machine

The uncertainty of wind power brings great challenges to large-scale wind power integration. The conventional point forecasting of wind power is difficult to meet the demand of power grid planning and operation. A novel two-stage short-term hybrid wind power interval forecasting model is proposed in this study. In the first stage, the original wind power data is automatically decomposed and divided into three different classes based on the data preprocessing method combining variational mode decomposition with sample entropy. In the second stage, the prediction model is established using the probabilistic regularised extreme learning machine (PRELM) and particle swarm optimisation (PSO). In view of the different characteristics of the subseries in the above three classes, prediction intervals (PIs) are constructed for each subseries. A novel interval evaluation index is used as the objective function of PSO to optimise the PRELM output weight matrix to find the optimal PIs. Also the prediction results of each subseries are reconstructed to obtain the final wind power prediction results. The numerical results based on actual wind power data show that the proposed model shows better performance compared with other methods and can effectively improve the prediction accuracy.