Abstract: With the accelerated transformation of traditional energy companies such as power companies to integrated energy service providers, the original extensive power management model of users has gradually become difficult to meet the needs of power marketing management. A stratified clustering method of electricity consumption patterns and a model of user cluster identification are proposed for the scenarios including massive users. Based on the identification results of user clusters, this paper proposes a probability forecasting model of conditional residual simulation load. The load-stratified probability forecasting is carried out to realize the refined power management of users. A typical case verifies the feasibility and superiority of the proposed method.