Scholarlink

<p>The increasing deployment of bifacial photovoltaic (bPV) modules introduces additional complexity in power prediction due to the dependence of rear-side irradiance on surface reflectance. This study investigates bPV power forecasting under varying surface albedo conditions in agrivoltaic (AV) systems at two geographically distinct locations, Delhi and Jodhpur, India, representing contrasting agro-climatic conditions. A hybrid stacked ensemble machine learning framework based on eXtreme Gradient Boosting (XGBoost) and Categorical Boosting (CatBoost) is developed to predict bPV power output under different albedo levels and crop-influenced ground reflectance scenarios. Results show that increasing surface albedo from 0.2 to 0.5 and from 0.2 to 0.8 improves annual bPV energy yield by 15.52% and 30.86% in Delhi and by 12.68% and 25.22% in Jodhpur, respectively. The proposed model achieves coefficients of determination (R<sup>2</sup>) above 0.999 with relative RMSE close to 1% and negligible prediction bias across all cases. This research highlights the challenges of accurately predicting power output in bifacial systems across diverse locations due to variability in surface albedo, particularly in AV systems where crop type and growth stages influence ground reflectance. The integration of surface albedo management with bPV systems is shown to be an effective approach for optimizing solar energy generation while simultaneously supporting sustainable agricultural practices under different agro-climatic conditions. SHAP-based interpretability analysis further confirms the dominant influence of irradiance variables and the positive contribution of surface albedo through rear-side irradiance enhancement.</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

Forecasting agrivoltaic power using a hybrid model accounting for crop and ground effects

  • 期刊论文

  • 发表日期：

    2026年3月23日

查看全文
  • Lipika Datta,
  • Astitva Kumar,
  • M. Rizwan

摘要

The increasing deployment of bifacial photovoltaic (bPV) modules introduces additional complexity in power prediction due to the dependence of rear-side irradiance on surface reflectance. This study investigates bPV power forecasting under varying surface albedo conditions in agrivoltaic (AV) systems at two geographically distinct locations, Delhi and Jodhpur, India, representing contrasting agro-climatic conditions. A hybrid stacked ensemble machine learning framework based on eXtreme Gradient Boosting (XGBoost) and Categorical Boosting (CatBoost) is developed to predict bPV power output under different albedo levels and crop-influenced ground reflectance scenarios. Results show that increasing surface albedo from 0.2 to 0.5 and from 0.2 to 0.8 improves annual bPV energy yield by 15.52% and 30.86% in Delhi and by 12.68% and 25.22% in Jodhpur, respectively. The proposed model achieves coefficients of determination (R2) above 0.999 with relative RMSE close to 1% and negligible prediction bias across all cases. This research highlights the challenges of accurately predicting power output in bifacial systems across diverse locations due to variability in surface albedo, particularly in AV systems where crop type and growth stages influence ground reflectance. The integration of surface albedo management with bPV systems is shown to be an effective approach for optimizing solar energy generation while simultaneously supporting sustainable agricultural practices under different agro-climatic conditions. SHAP-based interpretability analysis further confirms the dominant influence of irradiance variables and the positive contribution of surface albedo through rear-side irradiance enhancement.