Aiming at the limitation of traditional detection model for electricity stealing detection only by a single method and the class imbalance in electricity consumption data, this paper proposes an electricity theft detection model based on the entropy weight method fusing heterogeneous classifiers from the perspective of ensemble learning. Firstly, the problem of imbalance in electricity consumption data is handled by synthetic minority oversampling technique (SMOTE). Secondly, the diversity of individual classifiers and their respective detection performance and training mechanism are considered to optimize the base classifier. Finally, the concept of information entropy is introduced to calculate the weight share of each base classifier based on the dispersion of its classification results, and the output of each base classifier is integrated with this weight share. The experimental results show that compared with the traditional electricity stealing detection model, the model proposed in this paper performs better in multiple evaluation indicators and has good detection performance.