photovoltaic power station, fault diagnosis, random under-sampling, ensemble learning, Electrical engineering. Electronics. Nuclear engineering, TK1-9971

In response to the challenge of sample data imbalance in fault diagnosis methods for photovoltaic power plants based on machine learning， the paper proposes a fault diagnosis method leveraging an enhanced BP-Bagging algorithm. Firstly， a mapping relationship between photovoltaic data and fault types is established using a BP neural network to achieve fault diagnosis in photovoltaic systems. Subsequently， the Bagging algorithm is enhanced by utilizing random under-sampling （RUS） to address the issue of class imbalance in samples. Furthermore， to tackle the problem of overfitting in the BP network， the paper introduces a fault diagnosis model for photovoltaic power plants based on the enhanced BP-Bagging. This involves parallel training of multiple BP networks and determining fault diagnosis results through a voting method. Finally， the paper conducts comparative experiments with different algorithms， calculates evaluation metrics related to model accuracy， and validates that the proposed method demonstrates high overall performance. To a certain extent， it effectively mitigates the challenge of sample class imbalance in fault diagnosis of photovoltaic power plants， thereby improving the accuracy of fault diagnosis in such systems.