In 2017, the National Development and Reform Commission took the lead in launching the construction of the "Xinyi +" alliance, and the construction of the social credit system was fully rolled out. Among them, corporate credit assessment is an essential part, and the state attaches great importance to the construction of corporate credit. The construction of corporates’ credit assessment system on power use can effectively support the power industry in early warning of abnormal power consumption behaviors. Under the dual background of the state"s vigorous promotion of the construction of a social credit system and the construction of a "three-type, two-network, world-class" enterprise by the State Grid Corporation of China, various business departments and units of the State Grid Corporation of China have actively developed their credit system. This paper uses the power industry upstream and downstream customers" electricity consumption information, compliance status, safety supervision, and other data, combined with the improved fuzzy clustering algorithm to classify the power corporate users and construct group portraits. This paperalso utilizes the convolutional neural network (CNN) model and analyzes corporate electricity rise. The improved fuzzy clustering algorithm can adapt to different data distribution and model the corporate credit assessment system. The design of the multi-scale convolution kernel CNN model can overcome the shortcomings of the traditional CNN algorithm, such as a large calculation amount and easy to overfit. Experiments show that multi-dimensional power data sets can reflect corporate credit information well, and the proposed classification model has high computing efficiency and accuracy, which help to realize the Corporate Electricity Risk Assessment system.