Scholarlink

Solar yield prediction is essential for optimizing electricity generation and sizing photovoltaic (PV) power systems. As the number of utility-scale PV installations rises, the industry is shifting the focus on solar yield prediction due to its inherently variable nature. Accurate solar output prediction enhances energy market trading, optimizes solar power system performance, and supports grid stability, efficient energy management, financial planning, investment decisions, and maintenance of solar installations. In this study, we assessed the effectiveness of various Support Vector Machine (SVM) models for solar radiation prediction, utilizing different kernels and configurations. Our results show that according to the Pearson correlation feature selection method, the most important features are Direct Normal Irradiance (DNI), Diffuse Horizontal Irradiance (DHI), temperature, humidity, and wind speed. The Support Vector Regressor (SVR) with a polynomial kernel outperformed the other SVR models, as well as the persistence model and linear regression model, in terms of R-squared (R2), Mean Absolute Error (MAE), Mean Square Error (MSE), Root Mean Square Error (RMSE), and Mean Absolute Percentage Error (MAPE). This model achieved an R2 value of 0.97, with MSE, RMSE, and MAPE values of 2601.99, 51, and 1.80, respectively. These findings suggest that methods like SVR are highly effective for solar radiation forecasting, particularly in capturing complex, non-linear patterns in the data.

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

Solar Radiation Prediction Using Support Vector Machines with Spearman Correlation Feature Selection

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Khouili Oussama,
  • Hanine Mohamed,
  • Louzazni Mohamed

摘要

Solar yield prediction is essential for optimizing electricity generation and sizing photovoltaic (PV) power systems. As the number of utility-scale PV installations rises, the industry is shifting the focus on solar yield prediction due to its inherently variable nature. Accurate solar output prediction enhances energy market trading, optimizes solar power system performance, and supports grid stability, efficient energy management, financial planning, investment decisions, and maintenance of solar installations. In this study, we assessed the effectiveness of various Support Vector Machine (SVM) models for solar radiation prediction, utilizing different kernels and configurations. Our results show that according to the Pearson correlation feature selection method, the most important features are Direct Normal Irradiance (DNI), Diffuse Horizontal Irradiance (DHI), temperature, humidity, and wind speed. The Support Vector Regressor (SVR) with a polynomial kernel outperformed the other SVR models, as well as the persistence model and linear regression model, in terms of R-squared (R2), Mean Absolute Error (MAE), Mean Square Error (MSE), Root Mean Square Error (RMSE), and Mean Absolute Percentage Error (MAPE). This model achieved an R2 value of 0.97, with MSE, RMSE, and MAPE values of 2601.99, 51, and 1.80, respectively. These findings suggest that methods like SVR are highly effective for solar radiation forecasting, particularly in capturing complex, non-linear patterns in the data.