Abstract: At present, the existing phase data in some areas can not support the work of fast fault location or effective analysis of line loss in some areas. When trying to identify the phase based on data driving, it is found that there are often high similarity of data sequence performance and gradual decrease of voltage amplitude of nodes along the line, which affect the accuracy of identification results. Therefore, a phase recognition algorithm based on multiple data feature selection is proposed. The correlation coefficient method and wavelet transform are mainly used to extract the voltage data with large difference characteristics to solve the above two problems. Then, the data set is clustered to realize the phase classification of users in the transformer area. Finally, the effectiveness of the algorithm is verified. Compared with K-means algorithm and Pearson correlation coefficient recognition method, the recognition accuracy is higher, and the sample adaptability is also strengthened.