Short term wind power probability prediction based on support vector regression

An increased level of convenience for grid connection and assistance with scheduling, operation, and maintenance of the power system can be obtained by the power sector with high-accuracy wind power prediction. However, a single prediction model has limited ability to handle wind power sequence data, and often crucial elements go unused. To address this problem, a combined multivariate neural network prediction model based on support vector regression (LSTM&CNN-SVR) was proposed. First, the model combined the processing characteristics of the two neural network models, LSTM and CNN, to obtain three sets of predicted values with their respective advantages. Then the model utilizes the support vector regression (SVR) model's processing ability for nonlinear data to optimize the outputs of the two neural networks. In order to correct the data’s inhomogeneity, SVR was combined with the kernel density estimation (KDE) method to obtain the probability density distribution curves of the predicted values and their power prediction intervals at a certain confidence level, which could more intuitively represent the trend of wind power. The experimental validation of the model is based on the measured data of a wind farm, and the results show that compared with the traditional prediction model, the proposed model achieves higher prediction accuracy, smaller error and better overall prediction performance