Scholarlink

<p>Reliable estimation of photovoltaic (PV) power generation in arid regions demands integrated understanding of radiative, meteorological, and aerosol influences. This study examines environmental drivers of Photovoltaic (PV) variability at Bhadla Solar Park, India, combining explainable machine learning with PCMCI causal discovery. Copernicus Atmosphere Monitoring Service (CAMS) irradiance is validated against ground observations (R² = 0.81), with reduced performance during monsoon and a systematic overestimation bias during high-irradiance periods. Leave-one-out climatological baseline isolated seasonal variability; Random Forest and XGBoost better captured atmospheric PV variability (R² = 0.89–0.90), though monsoon-season showed substantially degraded performance (JJA R² &lt; 0) due to enhanced cloud and aerosol variability. Seasonal SHAP analysis reveals that near-surface temperature and clear-sky irradiance dominate PV variability during dry periods, while dust aerosols, black carbon, column water vapor, and relative humidity exert stronger control during monsoon. PCMCI causal networks identify persistent negative effects of absorbing aerosols, seasonally varying dust impacts, and indirect wind-driven aerosol redistribution. Generalized Additive Models further expose pronounced nonlinear PV responses to radiation, aerosols, and moisture. Overall, Bhadla PV generation is primarily radiation-driven but substantially modulated by aerosols and atmospheric moisture, particularly during monsoon months. This framework demonstrates the value of integrating explainable AI with causal discovery for disentangling solar power variability drivers, with direct relevance to solar energy planning.</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

Quantifying drivers of photovoltaic power generation at Bhadla using explainable machine learning and causal discovery

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月3日

查看全文
  • Sachin Budakoti

摘要

Reliable estimation of photovoltaic (PV) power generation in arid regions demands integrated understanding of radiative, meteorological, and aerosol influences. This study examines environmental drivers of Photovoltaic (PV) variability at Bhadla Solar Park, India, combining explainable machine learning with PCMCI causal discovery. Copernicus Atmosphere Monitoring Service (CAMS) irradiance is validated against ground observations (R² = 0.81), with reduced performance during monsoon and a systematic overestimation bias during high-irradiance periods. Leave-one-out climatological baseline isolated seasonal variability; Random Forest and XGBoost better captured atmospheric PV variability (R² = 0.89–0.90), though monsoon-season showed substantially degraded performance (JJA R² < 0) due to enhanced cloud and aerosol variability. Seasonal SHAP analysis reveals that near-surface temperature and clear-sky irradiance dominate PV variability during dry periods, while dust aerosols, black carbon, column water vapor, and relative humidity exert stronger control during monsoon. PCMCI causal networks identify persistent negative effects of absorbing aerosols, seasonally varying dust impacts, and indirect wind-driven aerosol redistribution. Generalized Additive Models further expose pronounced nonlinear PV responses to radiation, aerosols, and moisture. Overall, Bhadla PV generation is primarily radiation-driven but substantially modulated by aerosols and atmospheric moisture, particularly during monsoon months. This framework demonstrates the value of integrating explainable AI with causal discovery for disentangling solar power variability drivers, with direct relevance to solar energy planning.