Smart grid technology is developing rapidly around the world and is gradually applied to the operation and maintenance management of power systems, and its main advantage lies in its integration capability, which can effectively realize the high efficiency, security and reliability of power system operation and maintenance. This paper explores the integration of grid operation and maintenance by integrating computing and information theory using multidimensional data mining and analysis methods. The operation data of smart grid is first preprocessed, including resampling and PCA dimensionality reduction of multidimensional data signals. Then, a CNN-based power operation state prediction model and an R-CNN-based grid fault diagnosis model are constructed to ensure the stable operation and timely maintenance of the smart grid, and the predicted and actual values of the smart grid operation state of the CNN model are basically consistent with each other, with the MAE, MSE, and RMSE of 0.00104, 0.00014, and 0.012, respectively, and the prediction results are good. The effect is good. Compared with CNN and SVM, the performance of R-GNN model is better, and after PCA dimensionality reduction, the fault identification rate of R-GNN model is as high as 98.91%. And the delay of the R-GNN method for fault diagnosis is only 0.04s, while it can realize the comprehensive and accurate localization of the fault area. This paper provides methodological reference for the utilization of multidimensional data mining and analysis technology to realize the operation and maintenance integration of smart grid.