Comparion Of The Short-term Prediction Methods Based On Machine Learning For Photovoltaic Power Geneation

With the acceleration of the modernization process,large-scale mining and use of fossil energy not only lead to increasingly scarce energy resources,but also accompanied by a large number of pollutant emissions.Renewable energy,especially the development of solar energy with its unique advantages,is attached great importance by governments of all countries.However,the power of solar photovoltaic system is affected by meteorological factors and has random properties,which hinders the normal operation of power grid.Therefore,the accurate prediction of photovoltaic power can help the electric power department to make a scientific electricity dispatching and planning arrangement.Based on relevant research,this paper proposes a method for predicting photovoltaic power generation based on genetic algorithm and machine learning model.Through a real case,dividing the data into different seasons,establishing forecasting models in different seasons,and comparing the error values between the forecast results and the true values of several different forecasting methods under different seasonal conditions,It is verified that the GA-SVM model used in this paper can effectively improve the accuracy of photovoltaic power generation prediction,and the prediction effect is best in winter.The main research contents and results of this paper are as follows:(1)Analyze the impact of meteorological factors on photovoltaic power output.According to the obtained historical data of photovoltaic power stations,the irradiance,temperature and humidity are the main meteorological factors affecting photovoltaic power through correlation coefficient method and Pearson Chi-square test.(2)Preprocess the missing and outliers of the original data.The data is divided into four seasons by month.The output power of the test set was predicted by Random Forest(RF)model,Support Vector Machine(SVM)and Long Short Term Memory(LSTM)neural network model.At the same time,the RMSE,MAPE and five-fold cross validated RMSE values are output,and the prediction capabilities of the three models is compared.(3)The parameters of the Random Forest model are adjusted by the grid search method,the parameters of Support Vector Machine model and Long Short Term Memory neural network model are optimized by Genetic Algorithm,and the GA-SVM model and the GA-LSTM model are established.Finally,the prediction effect of five models in four seasons was comprehensively evaluated,and it was concluded that the GA-SVM model had the best prediction effect on photovoltaic power generation and the model in winter had better prediction effect on photovoltaic power generation than other seasons.