Research On Wind Power Prediction Method Based On Deep Learnin

In order to alleviate the current non-renewable energy consumption and pollution problems,many countries are vigorously developing the wind power industry.However,due to the strong volatility of wind power generation,as large-scale wind power is connected to the grid,the safe and stable operation of the grid is threatened.Accurate wind power forecasting can provide assistance for grid dispatching planning to ensure safe and reliable power supply.In order to improve the effect of wind power forecasting,the work done in this paper from the aspects of wind power data preprocessing and forecasting model optimization are as follows:Considering that abnormal data will have adverse effects on wind power prediction,this paper uses a combination of density clustering algorithm DBSCAN and optimal interclass variance(OIV)to identify abnormal values in the data,and the outliers are corrected by random forest algorithm(RF),and the correlation between wind speed and power data is significantly improved after the correction.Aiming at the problem that the nonlinearity and volatility of the wind power sequence will increase the difficulty of prediction,this paper uses the complete ensemble empirical mode decomposition with adaptive noise(CEEMDAN)to decompose the original wind power sequence,and reconstructs the sub-series by Kmeans clustering according the similarity degree of the sample entropy(SE)to characterize the wind power series properties,which reduces the impact of wind power series volatility on prediction.The sub-series and the meteorological data reduced by principal component analysis(PCA)are input to the CNNLSTM model to complete the wind power prediction.The analysis shows that CNN can effectively extract the local features of the input data and improve the prediction accuracy,CEEMDAN-SE-Kmeans decomposition can capture the trend characteristics of wind power series and reduce the volatility of the series,which improves the prediction accuracy significantly,and PCA dimensionality reduction has a small improvement on the prediction effect and effectively reduces the computational complexity.Aiming at the problem that it is difficult to extract the characteristics of wind power sequence,this paper adopts non-local neural network(Non-Local,NL)to obtain the global characteristics of input data to enhance data correlation,and the Conv LSTM prediction model with Encoding-Forecasting structure is further in-depth mining feature connections between hidden layers.The NL-Conv LSTM-EF prediction model is used for wind power single-step and multi-step prediction experiments.The experiments show that the model structure can effectively improve the prediction accuracy in single-step prediction,and can also be competent for multi-step wind power prediction tasks.