Research on Simulation and Prediction of Photovoltaic Power Generation Based on Machine Learning and Similar Day Optimization

Focus on the Carbon Peaking and Carbon Neutrality Goals, new energy experiences rapid development in China. New energy with intermittent features such as wind and photovoltaic (PV) power generation has brought great challenges to the power system. New energy, especially PV power generation, which currently ranks first, has reached 0.75 Terawatt by August 2024, urgently requiring an accurate and efficient simulation and prediction. Based on the mesoscale meteorological reanalysis data as input and physical methods such as radiation model with inclined plane radiation correction and photoelectric conversion model, the machine learning method extreme gradient boosting (XGBoost) with time variables such as month, day, hour and daily precipitation as features is introduced. A method for quantifying similar periods by moving optimization is proposed to realize an effective division of training set and test set. The method was applied to a PV power station in Qinghai Province, China and the average determination coefficient (R-squared) in the test set reaches 0.9187. XGBoost with similar period optimization method could increase the accuracy by <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$1\% \sim 9\%$</tex>, which may improve the simulation and prediction of PV power generation.