The world is facing more energy crises due to extreme weather and the rapidly growing demand for electricity. Siting new substations and optimizing the location of existing ones are necessary to address the energy crisis. The current site selection lacks consideration of spatial and temporal heterogeneity in urban power demand, which results in unreasonable energy transfer and waste, leading to power outages in some areas. Aiming to maximize the grid coverage and transformer utilization, we propose a multi‐scene micro‐scale urban substation siting framework (UrbanPS): (1) The framework uses multi‐source big data and the machine learning model to estimate fine‐scale power consumption for different scenarios; (2) the region growing algorithm is used to divide the power supply area of substations; and the (3) location set coverage problem and genetic algorithm are introduced to optimize the substation location. The UrbanPS was used to perform siting optimization of 110 kV terminal substations in Pingxiang City, Jiangxi Province. Results show that the coverage and utilization rate of the optimization results under different power consumption scenarios are close to 99%. We also found that the power can be saved by dynamic regulation of substation operation.