An unsupervised hourly weather status pattern recognition and blending fitting model for PV system fault detection

•The performance of PV systems is improved by hourly fault detection, compared to daily.•The monthly correction method proposed during the data preprocessing can correct the measurement errors and improve data quality.•The unsupervised hourly weather status pattern recognition method established provides basis for online fault detection under hourly weather types.•The blending fitting model of sub-weather status developed can ensure accuracy of PV fault detection by weather scenario.