Short-Term Prediction of Wind Power Based on NWP Error Correction with TimeGAN and LSTM-TCN

The data quality of numerical weather prediction has a great influence on the accuracy of wind power prediction technology. In order to improve the short-term forecasting effect of wind power, we proposed a combined LSTM-TCN forecasting model based on NWP error correction model with TimeGAN. First, minimal redundancy maximal relevance (mRMR) algorithm is used to extract original NWP features and measured meteorological features, respectively. Then, TimeGAN is used to correct initial NWP data according to the internal hidden associations with the measured weather features. After that, the corrected NWP features combined with historical wind power are applied to establish a wind power short-term prediction hybrid model based on LSTM and TCN network. Finally, the performance of the hybrid model is tested by a wind farm dataset in China, which verified effectiveness of this method.