Research On Real-time Prediction Of Output Power Of Large Wind Farm Based On Data Driven

In recent years,with the gradual exhaustion of traditional fossil energy and the problem of environmental pollution,new energy has been favored by all countries.The development and utilization of wind energy resources in China began later,but the speed of development was fast,and the installed capacity of wind power and the capacity of grid connected increased year by year.Because of the randomness and volatility of the near surface wind in the nature,the wind power is stochastic and fluctuating.After the wind power is connected to the grid,its volatility will have an adverse effect on the safe operation of the power system.By studying the time distribution characteristics of wind power,we can improve the accuracy of wind power prediction and analyze the prediction error of wind power,which is of great significance for large-scale development and utilization of wind power and increasing the capacity of wind power grid connection.In view of the phenomenon of missing data in real-time collection of wind power,the characteristics of missing data are analyzed,and then the appropriate method is selected to make up for missing data to ensure the quality of the data and prepare for the follow-up work.The index of evaluating the missing data is also introduced.In view of the characteristics of the output power fluctuation of large scale wind farms,a mixed distribution model is proposed to describe the probability distribution of wind power output power fluctuation characteristics.From the perspective of probability,the size and variation rule of wind power output fluctuation under different time intervals and time windows are studied,which is of great significance for wind power output prediction.In view of the prediction of wind power,this paper use a comprehensive correlation index based on entropy-weight to evaluate nonlinear mapping relationship between different historical periods of wind power sample and reference samples quantitatively,and solve the correlation redundancy between input and output variables of prediction model.And compare with the indicators of Pearson,Kendall,Spearman correlation coefficient and correlation coefficient of mutual information.Then,through intimate-samples selection,hidden layer structure optimization and network weights assignment,a modified model of real time wind power prediction is used to overcome the defect of the redundant degree training samples and slow convergence in traditional neural network training process,and improve the generalization ability and computational efficiency of the prediction model.In view of the wind power prediction error,the amplitude and wave characteristics of real-time prediction error of wind power output are analyzed firstly,and a probability density distribution model is selected to describe its probability density distribution.The confidence interval of wind power output prediction is obtained by using the distribution model under different confidence levels,and then the prediction error is stratified according to the calculation results of confidence interval combined with the actual power value.The error amplitude of different error layers is different,and corresponding measures are taken to compensate the error according to the amplitude characteristic of the error.