In recent years, the major Smart Grid problem is that users are unable to determine the usage conditions of electronic appliances at home, hence, users, network service developers, and service providers are all hesitant to use or provide services. In order to provide specific power utilization information, effective recognition of the type of appliances is a topical subject. At present, most studies on appliance recognition focus on single appliance recognition, however, in most homes, users switch multiple appliances on and off at the same time. Thus, a database is required for multiple recognition features in order to achieve multiple appliance recognition. This study aims to explore methods to create samples for recognition, reduce the amount of computation, and make it applicable for the operation of embedded systems with a small amount of computation. This study creates a database mechanism, appliance recognition classification, and a waveform recognition method, in order to solve the large data volume problem in current appliance recognition systems.