Short-term Wind Power Cluster Power Point Prediction And Probability Prediction Based On Deep Learning

Wind power is intermittent,random and volatile.In recent years,new energy power generation technologies were developed vigorously worlwide,among which the dominant position was tanken by wind power.However,after large-scale grid connection of wind power,a greater disturbance is caused to the grid,and the safe and stable operation is affected of the grid.The ability of power grid to absorb wind energy can be effectively improved and the guarantee for the safe and stable operation of the power grid systems can be provided by the research on the wind power cluster short-term power prediction and probabilistic prediction technology,and also by the prediction accuracy and prediction reliability of wind power cluster power.In this paper,wind power prediction is made from four aspects: the volatility of wind power cluster output,the short-term power prediction of wind power clusters based on cluster dynamic division,the short-term power probability prediction of wind farms based on QRLSTM,and the short-term power probability prediction of wind power clusters based on QRMGM.The fluctuation characteristics of wind power cluster output are analyzed in this paper.The wind power cluster data was studied in this paper from two aspects of data preprocessing and data volatility.Firstly,the causes of abnormal data in wind power clusters and the cleaning method were studied,and the standard normalization of the cleaned data were performed;secondly,by separating the 15min-level components of wind power cluster power,the volatility of wind power cluster output is studied,which provides theoretical support for wind power cluster power prediction.A short-term power prediction model for wind power clusters based on cluster dynamic division was proposed in this paper.In order to effectively improve the power prediction accuracy of wind power clusters,based on the K-means clustering algorithm,a short-term power prediction model for wind power clusters based on cluster dynamic division was proposed,which was compared with the selection of different network training models and selection of different strategy of cluster predictions on the accuracy of cluster prediction.The results show that compared with static division,accumulation,statistical upscaling,and overall forecasting,the average RMSE of cluster dynamic division strategy within 24 hours is reduced by 0.13%,0.87%,0.78%,and 0.97%,respectively.And the average RMSE of cluster dynamic division strategy withins 96 hours is reduced by 0.21%,0.62%,0.53%,and 0.89%,respectively.A short-term power probability prediction model for wind farms based on QR-LSTM was established in this paper.The quantile regression principle,long and short-term memory network and the kernel density estimation method were studied in this paper,and a short-term power probability prediction model for wind farms based on QR-LSTM was proposed combined with the quantile regression principle and long-short-term memory network.The interval prediction results of wind power and the probability prediction research combined with the principle of kernel density estimation was studied.A short-term power probability prediction model for wind power clusters based on QRMGM was established in this paper.Compared with wind farms,wind power clusters have higher input feature dimensions and is with larger amount of data.Due to the more complex structure of LSTM,the processing speed of high-dimensional data is slower.Therefore,a simpler structure of deep learning network MGM was proposed.And combined with quantile regression,a short-term power probability prediction model for wind power clusters based on QR-MGM was proposed.First,the feature selection of cluster input features was studied,then the interval prediction results of wind power clusters by QR-MGM were compared with different models,and finally the probability density curves of wind power clusters by QR-MGM were compared with different models.