Due to the rich and diverse functions of smart meters, the equipment fault types and failure rates is gradually increasing. It is very important to ensure the safe and stable operation of smart meters that how to accurately determine the fault types of smart meters and improve the maintenance efficiency of fault meters. In this paper, integration algorithm model of multi-classification is proposed in fault pretition. Firstly, the multi-dimensional analysis and fault type selection are carried out for the intelligent meter fault, and the problem of class imbalance in the data set is solved by the combination of undersampling and oversampling, and the data needed for the classified prediction model is constructed; Using the combination of the basic classification algorithm to obtain the optimal fusion algorithm, the accuracy of the proposed algorithm is proved on the public data set, and the accuracy after fusion is steadily improved by comparing with the basic classification model. Finally, based on the real-time fault data of smart meters collected in recent years in the power grid system, the experimental comparison between the basic model and the fusion algorithm model is carried out, and the results show that the proposed algorithm is effective. The accuracy and reliability of fault prediction are improved obviously by using the multi-classification integration algorithm model.