Research On Ultra-short-term Wind Power Forecasting Based On Deep Learning

With the exhaustion of fossil energy,environmental pollution,global warming and other problems caused by the excessive exploitation and use of fossil energy have become increasingly prominent.Renewable clean energy has attracted extensive attention from all over the world.Wind energy resource has become an important part of renewable energy due to its abundant theoretical storage.In recent years,the development speed of wind power generation is accelerating year by year.With the increasing installed capacity and scale of wind power,the intermittency,instability and uncertainty of wind power itself bring numerous challenges to the safety and stable operation of the entire power grid after the integrated operation of wind power.Accurate forecasting of wind power is crucial to the development of wind power and the safety and stability of the power grid.This paper mainly studies ultra-short-term forecasting in wind power forecasting,including:1)Using the wind speed-power data in historical data of wind turbine operation,the changing point-quartile method is used to preliminarily separate normal data and abnormal data.Using the normal data separated and the abnormal data separated,a model based on generate against network is proposed to enhance the features of the abnormal data separated.Using the normal data separated and the abnormal data with enhanced features as the training data,an abnormal data identifying model based on XGBoost is proposed to identify the abnormal data in the original data of wind turbine operation.The difference of the abnormal identification effect of the proposed method when the generate against network generates different amounts of abnormal data is analyzed.2)The variational mode decomposition is used to extract features of the pre-processing data of wind turbine operation obtained by abnormal identification.Using the modal components extracted as the training data,an ultra-short-term forecasting model of wind power based on TCN-GRU-AT is proposed and its hyperparameters is optimized by using the tree growth algorithm.The root mean square error and the mean absolute percentage error and the stability for forecasting of the proposed model is proved to be better than that of the model compared when these models make predictions over the same time period.3)Using the pre-processing data of wind turbine operation obtained by abnormal identification,the Division-K-means is used to cluster multi-interval of the historical wind power over different numbers of the initial base clusters.Using the modal components mentioned before as the training data and the results of interval clustering as the data labels,an ultra-short-term interval forecasting model of wind power based on Mul Dense-LSTM whose hyperparameters is optimized by using Bayesian is proposed to output the forecasting intervals over different clustering results.The final forecasting interval outputted is selected based on the relative probability of the interval acquisition over different clustering results.It is proved that the prediction interval coverage probability and the prediction interval normalized averaged width of the model proposed can be taken into comprehensive consideration without relying on ultra-short-term forecasting which provides a new idea for ultra-short-term interval forecasting.