Abstract: In order to improve the accuracy of high resistance fault diagnosis of distribution network lines, this paper proposes a high resistance fault diagnosis model based on reinforcement learning.The intrinsic mode function (IMF) of the local features of the original signal at different time scales is used as the input variable of the reinforcement learning fault classifier for iterative learning. The corresponding faults are identified based on the vector values of the input samples, and the fault type is matched to the output. The q increment is adjusted through the reward and punishment mechanism, and the fault classification of distribution network lines is obtained after training and testing a large number of sample data. Finally, compared with the wavelet analysis method, the accuracy of high-resistance fault classification by reinforcement learning in this paper is significantly improved.