ABSTRACT
With the increasing applications of new information technology represented by big data in electricity customer marketing, how to accurately recognize the characteristics of customers from massive data, especially group, has become the key to improving the quality and efficiency of marketing. However, the random missing of electricity data affects the extraction of user feature and creates greater difficulties for electricity customer group recognition. In this paper, we propose an electricity customer grouping method to reduce the influence of missing data. By introducing Gaussian distribution as the kernel function into a Bayesian estimation model, the proposed method can learn the probability distribution of user portrait from data, and estimate the missing labels of customers. With the estimated user portrait, the group recognition method based on Gaussian mixture model is optimized. The experiment results show that the accuracy of proposed method is 6% higher than existing method.
