The accuracy of power costumer baseline load (CBL) prediction will greatly affect the evaluation of demand response.In this paper,taking multidimensional electricity behavior and its influencing factors into consideration,a user differentiated baseline load prediction method based on load subdivision is proposed by elaborating user behavior characteristics.This method uses Ward-FCM cluster analysis and combines the load characteristic index to improve the effect of user load curve cluster analysis.Based on the analysis of multi-dimensional influencing factors such as meteorology and time,a refined analysis method of electricity consumption behavior is put forward considering urban micro meteorological factors such as temperature,humidity and temperature cumulative effect and social behavior factors during holidays.The temperature-sensitive,holiday-sensitive and both-insensitive differential power behaviors are obtained.On this basis the customer baseline load forecasting analysis of differentiated power consumption is performed and a comprehensive evaluation model is established to analyze the prediction accuracy.The analysis results show that the proposed algorithm can further improve the accuracy of baseline load forecasting.