Very Short-term Probabilistic Wind Speed Forecasting Method By Using Fusion LSTM

Wind power has become one of the most important renewable energy sources and plays an important role in renewable energy power systems.However,wind power has volatility,and large-scale wind power integration will affect the safety,stability and economic operation of the power system.A large number of studies have shown that high-precision wind power prediction is one of the important means to solve this problem.However,the predictive modeling limited to a single wind farm which only considers the local meteorological factors,and the lack of model input information.The traditional single wind farm modeling method has been difficult to meet the requirements of high-precision wind power prediction.When large-scale wind farms are connected to the grid,a large number of wind farms in the same area will be affected by the same weather system,and there is a significant spatiotemporal dependence between the wind speed and power of the adjacent wind farm.Predicting the wind speed of multiple wind farms at the same time,using the wind speed information of several adjacent wind farms to improve the prediction accuracy of wind speed is an important research topic in the study of wind speed prediction at present.The deep neural network has a strong nonlinear fitting ability,flexible network structure,as well as have broad application prospects in the field of renewable energy prediction.Among them,the long short-term memory network has a strong advantage in processing time series feature extraction.Based on the idea of multi-wind farm's wind speed feature fusion,this thesis establishes a multi-wind farm's ultra-short-term wind speed probability prediction model.The model uses long short-term memory network as the feature extraction module of wind speed sequence,and fully connected neural network to merge sequence features,and finally the wind speed probability prediction results of multiple wind farms are simultaneously obtained by quantile output.The model is verified by the measured wind speed data of three adjacent wind farms in South China.The verification results show that the multi-wind farm's ultra-short term wind speed probability prediction model proposed in this thesis can combine the wind speed information of adjacent wind farms to enhance all wind power in that area.The wind speed prediction accuracy of the farms in different seasons.