In order to solve the problem of low accuracy of devices fault diagnosis in intelligent grid dispatching control system, this paper proposes a fault diagnosis method based on deep learning theory. First, building deep learning diagnosis models for various devices by extracting the characteristics of historical fault samples and training them. Then, when the dispatching control system of smart grid detects the risk and the online diagnosis function will integrate real-time fault signals for calling the diagnosis model. Finally, the probability value of this faulty device tripping will be obtained. In this paper, an online diagnosis system based on deep learning is built and its effectiveness in improving the accuracy of trip fault identification.