Multi-State Appliances Identification through a NILM System Based on Convolutional Neural Network

Electrical loads have a unique energy consumption pattern, often referred to as a "signature", which allows disaggregation algorithms to distinguish and recognize the operations of different users from aggregate load measurements. Both in the case of civil users and in that of commercial and industrial users, the presence of multi-state appliances is extremely common. The disaggregation algorithms must be able to correctly identify the consumption patterns of such devices from the aggregate load measurements. In this paper we will illustrate a NILM algorithm, based on a convolutional neural network, able to identify the electrical loads connected to a house, starting only from the measurement of the total electrical current. Furthermore, the algorithm provides information on events in real time through a process of simultaneous detection and classification of them, without having to perform a double processing, thus reducing calculation times. Then we will focus on the algorithm's ability to manage multi-state devices, analyzing a specific case and commenting on the possibilities of the system.