Abstract: The power distribution equipment is easily affected by natural disasters, human activities, tree barriers, bird damage, etc. during operation. Information reusing of the historical inspection records of power distribution equipment can support for equipment risk early warning and patrol strategy optimization. Consequently, the multi-scale recurrent convolutional neural network (MRCNN) is proposed to classify inspection text, and the maximum information coefficient-Chebyshev graph convolutional network (MIC-ChebNet) is proposed to predict the environmental risk coefficient in this paper. First, the MRCNN was employed to classify the inspection records based on the character embedding to obtain the environmental risk coefficient of various power distribution equipment corresponding to the inspection text. Then, the inspection area was divided into multi grids, the environmental risk in the grid was counted, and the graph was built according to the maximum information coefficient matrix. Moreover, the environmental risk in the grid was calculated and the time series was generated, of which the prediction was conducted based on the MIC-ChebNet. Finally, we evaluated the classification effect of the MRCNN and the prediction results of MIC-ChebNet. Verified by industry data, the results show that the classification accuracy of the MRCNN reaches 92.40%, which is 1.03% and 0.76% higher than that of BiLSTM and RCNN, respectively. In environmental risk prediction, the average relative error and root mean square error of MIC-ChebNet are 0.06 and 0.12, respectively, which decrease by 73.91% and 82.86% compared to those of XGBoost. This method can be adopted to effectively extract the regional risk information in the inspection record, and accurately predict the spatial distribution of the equipment operating environment risk in the next 6 days, providing reference for optimizing the inspection cycle and route.