Compared with the traditional data of electricity consumption, the data of smart meter is more volatile and less predictable. In order to realize the reasonable planning of power generation and distribution, the energy industry needs to quantify the uncertainty of power demand through the probability prediction of smart meter data. In this paper, an additive quantile regression model is proposed to estimate the future distribution of smart meter data by using the gradient boosting (GB) algorithm. Firstly, the method gives the quantile regression and quantile correction algorithm for the probability prediction. Based on the quantile algorithm, the GB algorithm of the additive quantile considering the external factors is given, and the selection method of main parameters in additive quantile model based on the GB algorithm is studied, thus the high-performance probability prediction model of smart meter data is established. Finally, compared with others, the case study shows that the proposed method is more accurate and effective in the probability prediction of smart meter data of integrated and single users, especially in the probability prediction of electricity meter data of single user.