The simple design principle of IoT (Internet of Things) is to ubiquitously connect different end devices and sensors, and build a network with the surrounding environment to provide intelligent recognition and management. The appliance recognition is a technology to distinguish electric appliances by the operating features of each appliance. It can allow the users to learn information regarding the power consumption of appliances during usage, and provide an easy and effective way of recognition for home the underlying network objects. However, the huge and complex computing required by the recognition of features of various appliances, a central server will be needed in order to implement the technology in most cases. The additional cost will reduce the user intentions and create difficulty in promotion. In view of this, this study proposes a distributed energy management system implemented within family heterogeneous network architecture. Through the communication and coordination of things in IoT, the system automatically search nodes with comparatively low computing load within the micro-processors in neighboring appliances, and implements distributed computing by using multiple nodes and the recognition algorithm. This way, the appliance recognition can utilize the multi-node recognition results, and fully exploit the computing power of the micro-processors with comparatively lower working load. The acquired features are recognized to reduce the load of appliance recognition and context power saving. Meanwhile, using the information of the surrounding sensors, it further provides context sensing and measuring services to allow the user immediate access to the information regarding types of appliances functioning in the house, and analyze the power consumption behavior of the user by a variety of sensing nodes distributed in the house. The system will recommend the most appropriate power consumption pattern or switch off unwanted appliances to achieve the effectiveness of energy management.