Abstract: The incipient fault is the manifestation before a permanent fault. Timely and accurate identification of incipient fault is of great significance for realizing fault early warning and reducing the occurrence of permanent faults. Incipient fault characteristics in non-solidly grounding system are weak, and it’s quite difficult to be detected and identified. Therefore, an incipient fault identification algorithm is proposed based on least squares support vector machine(LS-SVM). Firstly, the shallow feature sets of disturbances are extracted based on physical characteristics and statistical characteristics respectively, and the energy entropy and singular entropy of different frequency bands are obtained based on S-transformation. Then, the maximum relevance and minimum redundancy method is adopted to reduce the data dimension of the feature set while retaining strongly correlated features, and the optimal feature set is constructed. Finally, PSCAD/EMTDC simulation system is used to obtain various disturbance samples, and the incipient fault identification model is obtained based on LS-SVM, then the particle swarm optimization algorithm is used to optimize LS-SVM parameters to improve the efficiency. Simulation examples show that the proposed algorithm can accurately identify incipient faults in non-solidly grounding system, which verifies the correctness and effectiveness of the proposed method.