In order to improve the efficiency of power grid fault diagnosis in the power maintenance department, the SSAE-CNN fusion network fault diagnosis model of AC and DC power grid based on stack sparse automatic encoder (SSAE) and convolutional neural network (CNN) is proposed. The experimental results show that the fault line diagnosis accuracy of the fusion network diagnosis model is 99.86%, which is 0.62% higher than the SSAE-BP diagnosis model and two percentage points higher than the general fault diagnosis model. The accuracy rate of fault type diagnosis is 99.93%, which is 0.66% higher than SSAE-BP fusion model. The proposed power grid fault diagnosis model is superior to the general model in both diagnosis accuracy and speed. It can accurately diagnose and classify all kinds of power grid faults and provide data reference for power grid fault diagnosis by the power maintenance department, which has important practical significance.