A fault diagnosis model is proposed by using improved SOM neural network for the purpose of improving fault and safety monitoring, especially when it lacks accurate positioning and correlation analysis in power dispatching automation system. Firstly, based on the analysis of the historical data of the dispatching system, the feature vector of the fault is extracted and the learning sample is established. And then the connection with input and output for the subsequent test is trained for verification through the algorithm. Finally, the experiment which tests the data and verifies the effectiveness of its fault diagnosis is in the network with the inherent mapping of the data. The final results show that this model is an effective artificial intelligence diagnosis method for different types of fault recognition and diagnosis.