Data-driven characterization of photovoltaic power generation under multi-factor coupling effects

Abstract To address the inefficiency of analyzing massive photovoltaic (PV) data under rapid capacity expansion, this study proposes a multi-time-scale data-driven framework integrating meteorological factors. First, Spearman correlation analysis reduces feature dimensions by 42%, identifying irradiance (ρ=0.89) and humidity (ρ=-0.62) as dominant factors. Second, a linear regression-based cleaning method achieves 96.3% accuracy in reconstructing missing data. Finally, experimental validation using a 100 MW PV station in Northwest China (2023) reveals distinct output patterns: peak generation under cloudy days (78.5% capacity) versus rainy days (38.2% capacity), with autocorrelation confirming daily (R=0.91) and seasonal (R=0.76) periodicity. The proposed method enhances data utility by 31% compared to traditional approaches.