Scholarlink

<p>Solar energy is a fundamental kind of renewable energy and is expected to be pivotal in electricity generation in the future smart grids. Conversely, the very intermittency and uncertainty of SE will bring further challenges for the steady operation and management of power infrastructure. Should one successfully penetrate the obstacles of photovoltaic (solar power) integration into the grid, great forecasting of the PV power is required. This study pertains to the practical, detailed forecast of photovoltaic performance for three prime sites in the USA. With the help of the recent ML schemes, hybrid schemes for CatBoost-GS and HGBoost-GS optimized by the SSA are compared here with their respective standalone versions. Initial outcomes show that CatBoost has predominantly performed well for standalone scenarios, whereas HGBoost provides better services whenever SSA is chosen as the optimizer in the hybrid scenario. Most importantly, temperature is one of the environmental factors that largely influences PV output. The R<sup>2</sup> values obtained from the analysis for the train data were 0.9565 and 0.9786, and for the test data, they&#xa0;were 0.9117 and 0.9615 for the&#xa0;CatBoost-SSA and HGBoost-SSA hybrid schemes, respectively. Also, the HGBoost-SSA model outperformed the CatBoost-SSA model with R<sup>2</sup> values of 0.9739 and 0.9725 in the&#xa0;JDMT and Malmstrom sites, respectively. These add useful knowledge to the PV sector, highlighting that hybrid schemes may increase accuracy in predicting photovoltaic outputs using machine learning techniques.</p>

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

A robust hybrid machine learning model for predicting short-term photovoltaic power output: integrating CatBoost and HGBoost

  • 期刊论文

  • 发表日期：

    2026年1月20日

查看全文
  • Ping Xia

摘要

Solar energy is a fundamental kind of renewable energy and is expected to be pivotal in electricity generation in the future smart grids. Conversely, the very intermittency and uncertainty of SE will bring further challenges for the steady operation and management of power infrastructure. Should one successfully penetrate the obstacles of photovoltaic (solar power) integration into the grid, great forecasting of the PV power is required. This study pertains to the practical, detailed forecast of photovoltaic performance for three prime sites in the USA. With the help of the recent ML schemes, hybrid schemes for CatBoost-GS and HGBoost-GS optimized by the SSA are compared here with their respective standalone versions. Initial outcomes show that CatBoost has predominantly performed well for standalone scenarios, whereas HGBoost provides better services whenever SSA is chosen as the optimizer in the hybrid scenario. Most importantly, temperature is one of the environmental factors that largely influences PV output. The R2 values obtained from the analysis for the train data were 0.9565 and 0.9786, and for the test data, they were 0.9117 and 0.9615 for the CatBoost-SSA and HGBoost-SSA hybrid schemes, respectively. Also, the HGBoost-SSA model outperformed the CatBoost-SSA model with R2 values of 0.9739 and 0.9725 in the JDMT and Malmstrom sites, respectively. These add useful knowledge to the PV sector, highlighting that hybrid schemes may increase accuracy in predicting photovoltaic outputs using machine learning techniques.