To solve the problem that the nonlinear suppression method requires prior statistical information to determine the optimal threshold and causes nonlinear distortion, an improved deep neural network(DNN) algorithm for power line impulse noise suppression and compensation is proposed. Firstly, the position of pulse noise is locked by DNN, and then the data of the position is cleared by zerosetting method. Finally, the nonlinear distortion is reconstructed and compensated for the processed signal, and the simulation and comparison experiments are carried out with other algorithms. The simulation results show that the proposed method improves the recognition rate of the DNN, reduces the bit error rate of the system, and has good robustness.