Abstract: Aiming at accurate identification of sensitive users in the process of power payment, this paper designs a new power payment user portrait method based on improved word vector model. First of all, the improved K-means algorithm is used to cluster the big data of power users' payment behaviors, so as to realize the clustering of power supply, electricity price, electricity fee and power failure demand data of power payment users. Then the user portrait method is designed based on the improved word vector model, and the clustering results are imported into the method, and the sensitive types of power payment users are identified by constructing the user portrait. Experimental results show that this method is accurate for the classification of four sensitive types of users, and the difference between the classification results is only 1. Moreover, this method has a good clustering effect on the power, electricity price, electricity fee and power failure demand data of power payment users in the power grid department. The recall rate, accuracy rate and F value are all greater than 0.95, which proves that this method has a good clustering effect and can accurately identify sensitive users in the process of power payment.