Abstract: With the development of intelligentization and diversification of public buildings, a large number of heterogeneous loads (such as various types of air conditioners, water heaters, charging piles, etc. ) have been connected, resulting in a significant increase in the complexity of energy management. Traditional energy management systems lack the ability to accurately identify and dynamically integrate heterogeneous loads, facing problems such as coarse energy consumption prediction and unreasonable scheduling, causing energy waste and power supply risks for critical equipment. Based on this, a dynamic identification and monitoring system for heterogeneous loads in public buildings is proposed. Through multi-source sensor data fusion and deep learning technology, the real-time accurate identification and status monitoring of load types are achieved. Experimental results show that the system has an identification accuracy rate of 97.2% for centrifugal chiller units and 95.8% for VRF multi-split units, significantly enhancing the adaptability of the public building energy management system to complex load scenarios, providing a new solution to the problem of heterogeneous load management in public buildings, and having significant importance for improving energy utilization efficiency and promoting the low-carbon development of intelligent buildings.