Ultra-Short-Term Wind Power Forecasting And Available Wind Power Calculation

The achievement of ’ carbon peak,carbon neutralization ’ has become the common goal of all countries in the world,and it is the future energy trend.Increasing the proportion of renewable energy generation is an important way and decisive factor to continuously promote carbon emission reduction.The rapid development of high-proportion renewable energy has brought historic opportunities to the power system,but also new problems and challenges.Due to the inherent intermittency,randomness and volatility of wind energy,the consumption situation of wind farms nationwide is extremely severe.The existing regional active power dispatching control has not really supported the improvement of wind power consumption level.The accuracy of ultra-short-term prediction of wind power and the calculation of available generation power is not enough to support the optimization of regional active power dispatching control strategy.Therefore,how to improve the accuracy of wind power ultra-short-term prediction and available power calculation under the background of rapid development of wind power has become a key problem to be solved at this stage.The emergence of new technologies represented by deep learning provides new methods and new ideas for solving this practical problem.Based on the above background,this thesis first studies the problem of abnormal data identification,and takes the processed data as the input data of the subsequent model.Then,aiming at the problem of wind farm available power calculation,a wind farm available power calculation model based on long short term memory network(LSTM)is established after studying the influence of power generation conditions and station losses.Finally,for the ultra-short-term prediction of wind power,based on the study of the characteristics of wind power,a ultra-short-term prediction model of wind power based on LSTM network and graph convolutional network(GCN)network considering temporal-spatial correlation and station loss is proposed.The specific work of this thesis is as follows:(1)Through the analysis of wind speed and power data of wind turbines in stand-alone information system,according to the distribution characteristics and causes of abnormal data on the curve,the power abnormal data are divided into three categories: bottom accumulation type,middle accumulation type and decentralized abnormal data.In order to accurately identify various types of abnormal data,this thesis proposes a power abnormal data identification method based on the combination of logic(rule)judgment,Thompson tau and quartile method.Among them,the abnormal data identification method based on logic(rule)judgment is mainly applicable to identify the bottom accumulation type data,and the abnormal data identification method based on Thompson tau and quartile method is mainly applicable to identify the middle accumulation type and dispersed abnormal data.The proposed method is used to identify the abnormal data of an actual unit,and the abnormal power data is effectively identified.(2)Based on the preprocessed data,a single-machine theoretical power calculation model based on long-term and short-term memory network is established.On this basis,a calculation method of available power generation of wind farm considering power generation conditions and station losses is proposed.Firstly,based on Pearson correlation coefficient,the key historical moments of the theoretical power generation of wind turbines are selected,and a single-layer LSTM wind turbine theoretical power calculation model is designed.Then,the generation conditions of wind turbines are studied,and the available generation power of wind turbines under different conditions is analyzed in detail.Then,according to the primary system diagram of the wind farm and the geographical layout of the wind turbine,a simplified equivalent circuit model of the loss in the wind farm station is established.Finally,based on the above work,the calculation of available power generation of wind farm is realized.The case analysis results show that the root mean square error of the calculation method of available power generation of wind farm proposed in this thesis is only 1.32 %,which is significantly lower than that of the traditional calculation method.(3)Based on LSTM network and GCN network,an ultra-short term wind power prediction model considering spatial-temporal correlation and in-station loss is proposed.Firstly,the key influencing factors of wind power are screened by Spearman correlation coefficient method,and the temporal correlation of wind power and the spatial correlation between wind turbines in wind farms are analyzed.Then,LSTM is used to extract the temporal feature information of wind turbines,and GCN is used to extract the spatial feature information and solve the problem of in-station loss,and the ultra-short-term prediction model of wind power considering the temporal-spatial correlation and in-station loss is established.Finally,the case analysis shows that the model can effectively complete the ultra-short term prediction of wind power,and the accuracy and qualified rate are 88.38 % and 97.15 %,respectively,which are significantly higher than the back propagation neural network model.The related research results of this thesis will be conducive to the online scheduling of the dispatching department of SGCC and the optimization of the AGC system strategy of direct-regulated wind power,promote the consumption of new energy,reduce the carbon emission reduction of the power industry,and help realize the “double carbon” commitment in China.