Adequate capacity planning of substations and feeders mainly depends on accurate estimation of future peak power demand. However, the traditional future peak demand estimation is based on the empirical measurement, which is conducted after the maximum diversity demand, and represents the individual peak consumption level and the diversity of multi-resident demand. According to the characteristics of the smart meter in smart city, a data-driven peak probability power consumption estimation framework based on fine-grained smart meter data and user social demographic data is proposed in this paper. In particular, four major phases are integrated into the proposed approach, including load modeling and sampling, the proposed variable truncation R-vine linkage approach, correlation-based customer grouping, normalized maximum diversification demand estimation, and probabilistic peak demand estimation for new customers. The superiority of the proposed method in point estimation and probability results is proved quantitatively by means of mean absolute percentage error and marble loss function.