Scholarlink

Focus on the Carbon Peaking and Carbon Neutrality Goals, new energy such as solar and wind power generation developed rapidly. In 2023, the installation of solar energy in China exceeded 0.6 Terawatt, accounting for over 20% of the total installed electricity capacity, surpassing hydropower for the first time, becoming the second largest power supply in China. Annual photovoltaic (PV) power generation achieved nearly 583.3 TWh, gradually towards the main power supply. An accurate simulation and prediction of PV power generation is of great significance for the safe and economical operation of the new power systems. In this paper, on 15-min measured irradiance and power generation data of PV plants within one year and the reanalysis meteorological hourly data of ERA5 derived from ECMWF (European Centre for Medium-Range Weather Forecasts), Firstly, we discover the characteristics of PV power generation by analyzing the daily insolation hours and hourly mean power output. Then physical mechanism method is used through radiation model, inclined plane radiation correction model and photoelectric conversion model. PV power output is simulated based on grid-type reanalysis meteorological data. Finally, according to the deviation sequence of simulated and measured power output, a machine learning method extreme gradient boosting (XGBoost) is introduced. After dividing the deviation time series into training set and test set, the training set is applied to learn the patterns to correct the test set. And the test set is fed back to modify the prediction. The results show that by using machine learning method, the determination coefficient (R-squared) of hourly PV power output of a certain station for medium and long-term could reach 0.9, which contribute to improve the accuracy of PV power output generation prediction effectively.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Research on Simulation and Prediction of Photovoltaic Power Generation Based on Radiation Models and Machine Learning Method

  • 会议论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年

查看全文
  • Jie Gao,
  • Xu Wang,
  • Jianwei Gu,
  • Siwei Tang,
  • Fangliang Zhu,
  • Jingyi Li,
  • Yiming Zhu

摘要

Focus on the Carbon Peaking and Carbon Neutrality Goals, new energy such as solar and wind power generation developed rapidly. In 2023, the installation of solar energy in China exceeded 0.6 Terawatt, accounting for over 20% of the total installed electricity capacity, surpassing hydropower for the first time, becoming the second largest power supply in China. Annual photovoltaic (PV) power generation achieved nearly 583.3 TWh, gradually towards the main power supply. An accurate simulation and prediction of PV power generation is of great significance for the safe and economical operation of the new power systems. In this paper, on 15-min measured irradiance and power generation data of PV plants within one year and the reanalysis meteorological hourly data of ERA5 derived from ECMWF (European Centre for Medium-Range Weather Forecasts), Firstly, we discover the characteristics of PV power generation by analyzing the daily insolation hours and hourly mean power output. Then physical mechanism method is used through radiation model, inclined plane radiation correction model and photoelectric conversion model. PV power output is simulated based on grid-type reanalysis meteorological data. Finally, according to the deviation sequence of simulated and measured power output, a machine learning method extreme gradient boosting (XGBoost) is introduced. After dividing the deviation time series into training set and test set, the training set is applied to learn the patterns to correct the test set. And the test set is fed back to modify the prediction. The results show that by using machine learning method, the determination coefficient (R-squared) of hourly PV power output of a certain station for medium and long-term could reach 0.9, which contribute to improve the accuracy of PV power output generation prediction effectively.