Non-intrusive load monitoring is a new type of load monitoring technology, which saves a lot of installation costs of monitoring and sensing equipment. In this paper a non-invasive load monitoring system for the edge computing framework is designed, which includes five parts: data acquisition, data preprocessing, event detection, feature extraction and load decomposition. The data acquisition is realized by the smart meter, using STM32 embedded controller to communicate with the server through WIFI. Based on the instantaneous peak value and instantaneous change of power data aimed at different types of load equipment, the load event detection algorithm is studied. The effective detection of switch and state change of load equipment is realized. Then, the harmonic wavelet transform is used to decompose the power data of different charge devices. The experimental platform is built, and the separate operation and mixed operation experiments of different types of load equipment are carried out. The results shown that the non-invasive power monitoring algorithm is effective. On the basis of distinguishing the operating states of different load equipment, the power data from an electric heater, an hair drier and an electric drill can be decomposed, which provides a basis for the realization of non-invasive load monitoring.