This paper aims to classify the growing number of real-time traffic during the secure access process of the power business system more effectively and to improve the speed of business processing of the power system. A real-time traffic classification method of the power business based on improved random forests algorithm is proposed. On the basis of analyzing characteristics of real-time traffic in secure access of the power business, traditional random forests algorithm is improved. This paper prunes random forests based on margin weight to improve real-time performance of classification and does data-editing for the new sample data to improve accuracy performance of classification. Based on this improved algorithm, a process of real-time traffic classification in secure access of the power business is designed. At last, an instance of a province’s real-time traffic classification in secure access of the power enterprise is used to validate the feasibility and efficiency of the method proposed.