Abstract: To address the challenges in substation line fault diagnosis, including protection maloperation and failure-to-trip, interference from secondary circuit abnormalities, scarcity of fault samples and insufficient reliability of diagnosis relying on a single criterion, this paper proposes an intelligent fault diagnosis method for substation lines based on the coefficient of variation method combining multiple criteria, Least Squares Generative Adversarial Network (LSGAN) data augmentation, and the Light Gradient Boosting Machine (LightGBM). Firstly, three types of fault features, namely action sequence, power direction and differential current, are extracted to construct the corresponding feature matrices. Secondly, the coefficient of variation method is adopted to assign objective weights to the three groups of criteria and form a comprehensive protection action criterion, which compensates for the drawbacks of single-criterion diagnosis. Thirdly, LSGAN is introduced to augment fault samples and alleviate the problem of insufficient field fault recording data. Finally, features of the comprehensive criterion are fed into the LightGBM model to realize fault section location and fault type identification. Simulation results demonstrate that the proposed method can achieve accurate location and identification of substation line faults, providing technical support for substation operation, maintenance, and emergency repair.