Aiming at the difficulty for a large group of comprehensive energy users to choose when purchasing energy service packages, a comprehensive energy package recommendation method based on user feature clustering is proposed to improve user stickiness. First of all, the comprehensive energy user information collected is constructed by knowledge graph, the missing user information is supplemented and improved, and the relationship between users is analyzed. Then, spectral clustering method is used to cluster the constructed user knowledge map, calculate the similarity among users, and extract the interest features representing the diversity of energy use behavior of comprehensive energy users. Finally, the random forest model is used to calculate the predicted scores of comprehensive energy users for each energy service package. After sorting the predicted scores, the part of the package with the highest score is selected to present the package service content for users through the online platform, so as to achieve accurate recommendation for users. By comparing the package recommendation model proposed in this paper with the traditional recommendation model, the results show that the integrated energy package recommendation method based on user feature clustering can achieve effective user precision energy service recommendation for integrated energy service companies, which is conducive to improving the market competitiveness of energy service companies, and provides technical support for the transformation of power enterprises into integrated energy service providers.