Historic residences often suffer from low energy efficiency due to inadequate management and obsolete infrastructure. To ensure these buildings operate in an environmentally friendly manner, intelligent energy control and management are necessary. Accurate indoor user positioning is fundamental to realizing occupant-centric controls. However, historic dwellings, with architectural heritage preservation requirements and diverse operational modes, present multiple challenges in terms of occupant monitoring and energy management. This research proposes a BLE positioning system based on the SAE-CNN algorithm to realize the analysis and prediction of occupant locations in a cost-effective and minimally-intrusive manner. The process begins by dividing spaces based on the mapped BIM model and cluster analysis. The RSSI maps of wireless signals collected from the Bluetooth beacons in each space are then preprocessed by Gaussian filter and sliding window techniques. Lastly, a machine learning algorithm refines the prediction model, which is integrated into a digital twin platform designed for energy consumption management. The system was implemented in a traditional Chinese dwelling in Wufu town, Fujian Province. The results are promising, showing that the positioning accuracy reached 1.5–2 m (95 % confidence interval), while the platform effectively presented the real-time positioning results as well as realized occupant-centric energy management. Moreover, the collected location information shed light on occupants' spatial usage habits, offering valuable insights for devising decarbonization retrofit strategies.