Precision maintenance and operation of building systems are crucial for energy efficiency and carbon reduction in building sector. Fluctuations in people flow play a significant role in building cooling load. Traditional air handling unit (AHU) operating under constant air volume tends to waste energy, as it lacks the capability to dynamically adjust to varying demand. This study introduces an intelligent digital twin AHU system considering the people flow variation, and the energy-saving potential is evaluated by a commercial building in Shanghai. The IoT technology is used to monitor the parameters of the AHU, using MobileNetSSD and NMS algorithms to extract people flow information from a camera. Through sensitivity analysis of the collected data, the key parameters for load prediction are identified, forming the basis for establishing a dynamic load prediction model using ANN. A digital model of the AHU is set up by TRNSYS, and the optimal control strategy is determined to minimize fan and pump energy consumption under different cooling loads. The digital twin system has been successfully implemented in a shopping mall in Shanghai, and 29.5% reduction of fan energy consumption can be achieved by dynamically adjusting air supply rate in response to the changing cooling loads.