Short-Term Wind Power Probabilistic Prediction for Newly Built Wind Farms Scenarios Based on W-GAN and CNN-LSTM

Accurate wind power probabilistic prediction is of great significance to the safe and stable operation of the power system. In order to make accurate wind power probability prediction for newly built wind farms with insufficient historical training samples, this paper proposes a short-term wind power probability prediction method for new wind farms based on the combined model of W-GAN and CNN-LSTM. First, the training data is sampled based on the W-GAN network, and then CNN and LSTM are used in the prediction modeling to learn the information contained in the generated samples and the original samples, and finally the probability prediction of power is performed based on the quantile regression deep learning method. The example analysis shows that: 1) Using W-GAN for sample expansion can effectively make up for the problem of insufficient training samples for new wind farms. Compared with direct modeling, the ACE and Winkler indicators are reduced by 0.07 and 0.18 respectively under the 90% confidence interval 2) Using the combined model to learn the information of the generated samples and the original samples respectively can effectively improve the prediction accuracy. Compared with using the generative model to train the network alone, the ACE, AW, and Winkler indicators are reduced by 0.12, 0.08, 0.16.