A Novel Wind Power Prediction Approach for Extreme Wind Conditions Based on TCN-LSTM and Transfer Learning
Jifeng Song, Xiaosheng Peng, Zimin Yang, Peijie Wei, Bo Wang, Zheng Wang
Abstract
Reliable Wind Power Prediction (WPP) is significant to power system scheduling and safely stable operating. However, the WPP under extreme weather conditions such as extreme wind conditions is faced with the problem of data shortage. An effective approach to extract temporal and synoptic characteristics of extreme weather is required. In this paper, a novel wind power prediction approach for extreme wind conditions based on TCN-LSTM and transfer learning is proposed. Firstly, a pre-trained model is established based on TCN-LSTM structure. Secondly, a small amount of data under extreme wind conditions is utilized to retrain the fully connection layer at the tail of the neural network optimized by Sparrow Search Algorithm (SSA) to obtain a transfer learning model under extreme wind conditions. The results exhibit that the proposed approach efficiently improves the WPP accuracy. The RMSE and MAE under extreme wind conditions are reduced by 33.70% and 58.44% compared to the case without transfer learning.