Most existing studies on photovoltaic(PV) fault diagnosis focus on single faults.To accurately identify composite faults in PV modules

this paper proposes a fault diagnosis method based on feature extraction and machine learning.A mathematical model of a PV array is established based on its series-parallel configuration

and a simulation model for PV faults is constructed within a PV generation simulation framework.The voltage-current(V–I) output charac-teristics under both single and composite fault conditions are analyzed.Feature parameters are refined based on observable surface feature variations

and new feature parameters are introduced to form fault descriptors.A mapping relationship between the extracted features and corresponding fault types is then established.A diagnostic model is developed using the CatBoost(categorical boosting) algorithm

optimized by the harris hawks optimization(HHO) method.Experimental results from a custom-built test platform demonstrate that the proposed method achieves diagnostic accuracy exceeding 98% for both single and composite faults.