Short-term PV Power Generation Prediction Based On Data Reconstruction And Composite Residual Correction Model

As an environment-friendly and renewable green power source,photovoltaic power generation is playing a more and more important role in production and life.In order to make better use of this clean energy and coordinate the output balance between photovoltaic power generation and other power sources,it is of great value and significance to predict photovoltaic power generation.This paper firstly introduces the structure and principle of photovoltaic power generation,and analyzes the relationship between various influencing factors and photovoltaic power on this contact,which lays a foundation for the establishment of the model in the following paper.Next,for short-term photovoltaic power generation prediction,this paper improves the accuracy of prediction from three aspects.The first step is to preprocess the data.In view of the strong periodicity,randomness and instability of photovoltaic power data,a data reconstruction method based on Empirical Mode Decomposition(EMD)and correlation analysis was established in this paper.The original sample data of photovoltaic power generation is decomposed into reconstructed data and residuals under the premise of full consideration of the integrity of the data.A prediction model is established for the data subjects of the two parts respectively.The second part is to build the prediction model.This paper builds models for different data.For reconstructed data,this paper establishes the Long-Short-Term Memory Network(LSTM),Compared with Elman neural network model and Radial Basis Function(RBF)neural network model,the advantage of LSTM model in this calculation example is verified.For residual data,Extreme Gradient Boosting(XGBOOST)model was established to predict.Compared with the Support Vector Regression(SVR)model,the prediction accuracy of the proposed model is proved to be higher.The third step is to correct the results.In this paper,the results of the initial prediction are compared with the original data to get the secondary residuals and the Markov method is used to correct them,which improves the accuracy of the prediction.Finally,the validity of the proposed method is proved by the actual data.The paper has 52 pictures,19 tables and 81 references.