Abstract: It has been one of the hot topics for years in the fields of electrical engineering and academia how to improve the comprehensive diagnostic performance of fault diagnosis software as much as possible, which is the core software of integrated intelligent alarm module in the dispatch center. This paper proposes a new method of artificial intelligent (AI) fault diagnosis based on the idea of information fusion. First, a fault diagnosis fusion model based on the neural network(NN) is established, which can determine the weights of three AI basic fault diagnosis models automatically by using NN's capability to extract features, so that the fault diagnosis model can achieve higher accuracy. Then, the model is checked with test samples: If the accuracy meets the requirements, which means that the amount of the historical samples is sufficient, and the trained NN is satisfactory and can be directly used for the fault diagnosis of the new information of the corresponding equipment. Otherwise, it indicates that the amount of current samples is not sufficient, which indicates the method of information fusion based on NN is no longer applicable in this situation. In this case, an information fusion method for fault diagnosis is adopted to determine adaptively the weight of the three AI basic fault diagnosis models through fault diagnosis evaluation system and scatter degree method to achieve the more satisfactory performance of the fault diagnosis, which more fits the scene with fewer historical samples. Examples of different samples demonstrate that the new fusion method proposed in this paper has better comprehensive performance of fault diagnosis in any case, compared with the other fault diagnosis methods.