Short Term Prediction Of Equivalent Load In Wind Farm Area Based On Deep Learning Algorithm

Power load forecasting is an indispensable part of power system,which plays an important role in stable operation and energy saving of power system.Under the new situation,with the increase of wind power grid capacity year by year,the fluctuation and intermittence of wind power generation raise a great interference to the voltage and frequency of power system.At the same time,because the fluctuation of wind power is not synchronous with the fluctuation of power load,the load forecasting and system scheduling in the wind farm area are facing more challenges.Therefore,by introducing the concept of regional equivalent load with wind farm,the equivalent load prediction considering the grid connected wind power plays an important role in the stable and economic operation of the grid.In this context,this paper proposes a power load forecasting model based on the improved echo state network and a wind power forecasting model based on the threedimensional convolution gating unit network,and then builds a equivalent load forecasting model for regional with wind farms.A robust echo state network model is proposed for power load forecasting.The echo state network has the advantages of good prediction ability and low cost of computing power,but it also has the disadvantages of unstable training,difficult initialization and over fitting.Through research and analysis,the initialization and training process of the network is optimized by using the phase space reconstruction theory.At the same time,three kinds of regularization algorithms are proposed to improve the training speed and overcome over fitting problem of the model.Finally,in order to meet the requirements of more robust power load forecasting,this paper combine the echo state network with quantile regression and build echo state quantile regression network to achieve power load probability forecasting.For the task of wind power prediction,a three-dimensional convolution gate recurrent unit network based on the variational mode decomposition is proposed as the short-term wind power prediction model.Firstly,the time series of wind power and wind speed are decomposed into several modal components by using the variational mode decomposition.The modal components are reconstructed into three-dimensional feature map sequence,and then the prediction is completed by using the three-dimensional convolution gating gate recurrent unit network.Finally,an equivalent load forecasting model is built based on the proposed power load forecasting model and wind power forecasting model,and the effectiveness of the proposed model is proved by numerical simulation and comparison with various models.