In view of the current risks faced by electric power companies in the recovery of special transformer users’ electricity bills, a special transformer users’ electricity bill recovery based on Stacking model fusion is proposed Collect risk identification methods. Carry out feature processing, feature construction and feature screening for special transformer user data, and optimize the model from sample distribution and feature attributesThe generalization performance of; The Stacking model is used to fuse multiple base learners to build a risk identification model for electricity fee recovery of special transformer users. The experimental results show that the phaseCompared with other commonly used classification algorithms, the proposed method has better accuracy, recall, P-R harmonic mean, AUC value and model generalization performance,The recognition rate of users with special transformer risk is also higher.

This article starts with the importance of electric power marketing systems, introduces the technical characteristics of data mining and its application status in electric power marketing systems, thereby providing decision-making basis for the economic operation of power grids. And propose using C5.0 decision tree algorithm to deeply analyze the marketing data of the electric power marketing management system. The original C5.0 decision tree algorithm is improved by introducing information entropy, which improves its classification speed and accuracy. Experimental results on UCI machine learning dataset and power marketing dataset show that the proposed improved C5.0 decision tree algorithm has good classification performance and can meet the classification and prediction requirements in power marketing work.