Ultra-short-term Probability Prediction Of Wind Power Cluster Power Considering The Spatial And Temporal Dependence Of Prediction Error

At present,wind energy is an important source of clean energy,but the randomness,volatility and intermittence of wind energy lead to high uncertainty of wind power generation.This disadvantage of wind energy poses a serious challenge to the attempt to integrate large-scale wind energy into modern power systems.Accurate wind power prediction is one of the bases to ensure the stable operation of power system and promote wind energy consumption.This paper studies the modeling methods and forecasting strategies on the basis of the factors that affect the ultrashort-term wind power forecasting.After obtaining the deterministic prediction results,the prediction error scene division,spatio-temporal dependence and probability prediction are studied.First of all,starting from several aspects that affect the accuracy of ultra-short-term wind power forecasting,this paper systematically analyzes and summarizes the impact of data processing,forecasting strategy,prediction model and predictability on ultra-short-term wind power forecasting.Based on the data of a wind farm in Inner Mongolia Autonomous region of China,a quantitative and qualitative study was carried out.Secondly,in order to solve the problem of increasing error caused by ultra-short-term power multi-step prediction,a new prediction strategy is proposed.The existing wind power forecasting model can not effectively extract the effective information from high-dimensional input,and the prediction accuracy will be affected by the increase of prediction step size.In order to solve the problem that the prediction model can not effectively present the information in high-dimensional input data and reduce the error caused by multi-step prediction,a new prediction strategy combining integrated learning and machine learning prior knowledge is proposed in this paper.The prediction strategy is applied to wind farms in three different provinces of China to verify the effectiveness of the prediction strategy.Thirdly,because the error of deterministic prediction of wind power can not be avoided,probability prediction can fully describe the uncertainty of wind power,and then provide further guidance for the decision-making of the dispatching department.Especially for the ultra-shortterm this specific time scale,the current interval prediction method to mine the physical change process of wind power is still incomplete.Therefore,a new framework for ultra-short-term wind power interval prediction based on power fluctuation process is proposed in this paper.In this paper,a fluctuation process of wind power series is defined,and a clustering division method of wind power fluctuation process is constructed.Aiming at multiple fluctuation processes,a quantile regression forest interval prediction model for ultra-short-term time scale is built.Finally,in order to further quantify the uncertainty information of ultra-short-term wind power cluster power prediction,explore the spatio-temporal characteristics and analyze the error characteristics.The prediction error itself has the characteristics of time series evolution,and it is also related to the geographical location of the new energy stations,which makes the prediction error more complex and changeable and difficult to analyze qualitatively.Based on the deterministic prediction results,by analyzing the dominant factors of power prediction,a highdimensional probability distribution model considering the spatio-temporal dependence of prediction errors is established to explain the propagation mechanism of prediction errors among multi-field stations.reveal the spatio-temporal characteristics of multi-station prediction errors.Then a graph convolution network is designed by using spatio-temporal dependence to extract spatio-temporal features and make probability prediction.Numerical examples show that the method proposed in this paper can more effectively reflect the uncertainty of wind power.