Adaptive Convolution Long-Short Memory Network Short-Term Wind Power Prediction Based on Transitional Weather Classification

The current wind power prediction scheme still has a large error in the transitional weather period. In the case of large-scale wind power integration, it will affect the safe operation of the entire power grid. In order to solve the above problems, an adaptive prediction model based on transitional weather classification is proposed. Firstly, the quartile method is used to clean and interpolate the abnormal data of the wind farm, and then the parameters of the extreme gradient lifting tree (XGB) are optimized by the improved snake swarm algorithm (CBAMSO). The scene classification model is established to divide the transitional weather, and the sensitive meteorological factors of typical transitional weather are selected to construct the input feature sequence. The convolutional neural network (CNN) fusing spatial pyramid pooling (SPP) is used to extract variable dimension features. Finally, the final wind power prediction value is obtained by using the attention mechanism (ATT) to redistribute the weight to the output of long and short memory network (LSTM). The results show that CBAMSO-XGB accurately divides all kinds of transitional weather, and the average absolute error and root mean square error of adaptive prediction model are about 42.49% to 72.91% and 65.34% to 91.2% compared with CNN-LSTM model.