A Photovoltaic Soiling Assessment Method Based on Irradiance Temporal Features and Power Spatial Consistency

Operational data from photovoltaic (PV) plants are widely used for soiling analysis, but their reliability is often reduced by weather disturbances, irradiance fluctuations, equipment heterogeneity, and measurement noise. This study proposes a baseline-referenced four-step filtering framework (S1–S5) to extract more physically consistent samples from inverter-level operational data. The framework combines irradiance-condition screening, irradiance and power temporal-trend constraints, and power–irradiance coupling consistency, and evaluates the retained samples using metrics of overall dispersion, intraday smoothness, interday continuity, spatial consistency, and sample retention. Based on 5 min data from the Lvhua PV plant in Shanghai, 280 parameter combinations were scanned under unified R2 and coupling-correlation (CC) thresholds. The results show that the first two filtering steps provide the dominant reduction in overall variability, improving the dispersion metric by approximately 97% and 89%, respectively, while the later steps further improve device-response and coupling consistency. The recommended configuration, W7_R75 with CC = 0.95, retains 8.57% of valid samples and achieves a balanced improvement in data quality and sample support. Stricter CC thresholds (0.97–0.98) provide additional smoothness and spatial-consistency gains but reduce retention to about 5–7%. The proposed framework provides a reproducible data-screening basis for PV soiling assessment and operational data-quality control.