Research On Power Prediction And Simulation Of Photovoltaic Power Generation Based On Improved CNN-LSTM

With the growth of global demand for electricity,the share of renewable energy sources is increasing and promoting the development of high-quality photovoltaic energy has become a global common goal.The uncertainty of PV power outputs will threaten the stable operation of the power system,so it is necessary to focus on the study of photovoltaic power generation forecasting.The accurate forecasting of PV power generation has become an important part of the renewable energy dispatching and operation,which strongly guarantees the stability of the power system.This thesis focuses on PV power forecasting technology,aiming at the problem of data redundancy influence model building to extract features and reduce dimensionality from the data provided by photovoltaic stations from the perspective of data mining.Based on the principle of feature selection,Pearson Correlation Coefficient（PCC）and Gradient Boosting Decision Tree（GBDT）are used to calculate the feature importance,only meteorological factors have strong correlation with the PV power output are retained to achieve data dimensionality reduction.Then,based on the principle of feature transformation,Principal Component Analysis（PCA）is used to reduce the dimension of data while keeping the sample data information to the maximum extent.This thesis combines feature selection and feature transformation to propose a new dimensionality reduction method,Pearson-GBDT-PCA dimensionality reduction method,In the case of retaining important factors of photovoltaic power output,the dimension of data is again reduced to improve data quality.Based on the obtained dimensional-reduction data samples,Long short-term Memory（LSTM）and Convolutional Neural Networks（CNN）,which introduce attention mechanism,are studied.The bayesian optimization method is used to optimize the prediction model of photovoltaic power generation,and the superiority of bayesian optimization method is proved through experiments.Finally,combined with the characteristics of different neural networks,an improved CNN-LSTM pv power prediction model was proposed,and the model was optimized by Bayesian optimization algorithm.