Ensuring precise prediction of community heating loads is crucial for optimizing the performance of district heating systems. However, the currently prevalent methods, like the simultaneous usage coefficient method and the area index method (AIM), which heavily rely on practical engineering experience, may not be suitable for ultra-low energy residential buildings (UERBs). Applying these methods to UERBs may result in overcapacity of the heating system, which increases energy consumption and costs. This study aims to enhance the accuracy of heating load calculations for UERB communities. A probabilistic model for heating load calculation was proposed, considering the “part time, part space” spatiotemporal distribution characteristics of residents. Typical occupancy patterns in cold zones of China were identified through questionnaire surveys and K-means cluster analysis, and typical models of UERBs were established, with both the building models and survey data validated. The residential building energy model (RBEM) was applied to a case community in Dalian, analyzing the impact of occupancy patterns on heating loads. The results indicate that the RBEM method increased the system's heating load factor by 6.9 % and reduced its design capacity by 32.9 % compared to the AIM. The proposed method might guide the development of management strategies in the energy planning stage of district energy systems.