Optimal Bound Based Ensemble Approach for Probabilistic Wind Power Forecasting

Due to the intermittency and fluctuation of wind power, a prediction interval (PI) with high stability, reliability and low sharpness is utmost important in system economic dispatch. This paper proposes a novel optimal bound based ensemble approach (OBE) where the weights for upper and lower prediction bounds are optimized by heuristic method respectively. Three base probabilistic model, K-means based conditional error Distribution (KCED), Quantile regression neural network (QRNN), and Pretraining-NN based direct interval forecast (PNNDI) are introduced where the KCED and PNNDI have been modified to obtain higher performance. A numerical case of a wind farm in Northeast China proves the superior PI-quality by the proposed ensemble method and the effectiveness of the modification of the base probabilistic models.