An Interval Prediction Approach of Wind Power Based on Skip-GRU and Block-Bootstrap Techniques
Hao Quan, Wei Zhang, Wenjie Zhang, Zixiong Li, Tao Zhou
Abstract
Quantifying the uncertainty in the wind power forecasting process is of great significance to the stability, security and reliability of the power system. The uncertainty can be effectively quantified by the interval prediction. In this paper, a novel interval prediction method based on skip-connection-based gate recurrent unite (skip-GRU) network and block bootstrap is proposed to predict short-term wind power. Both model parameters and data noise are concerned in uncertainty quantification. A large number of subsets are generated by different bootstrap methods firstly, and then skip-GRU models are trained by these subsets. The prediction intervals (PIs) on test datasets are generated with three confidence levels ranging from 85% to 95%. The results of the case studies demonstrate that the proposed method is able to obtain satisfactory PIs, and the overall quality of the PIs is higher than other three bootstrap methods and benchmark models.