An Event-Driven Convolutional Neural Architecture for Non-Intrusive Load Monitoring of Residential Appliance

Abstract

Nowadays, the advancement of non-intrusive load monitoring (NILM) is hastened by the everincreasing requirements for smart power utilization and demand side management. Thus, an intelligent event-driven non-intrusive load monitoring method based on convolutional neural network (CNN) is proposed in this article to profile residential consumer behavior, which focused on limited applicability and low recognition accuracy of existing NILM techniques. The deep convolutional neural architecture improves the performance of load monitoring procedure, and extends its application range to monitoring process in terms of various and combined characteristics. Meanwhile, the event-driven procedure is conducted to find the start-stop time and state change of the appliance accurately, including zero-cross detection, current similarity detection, threshold evaluation, and event current acquisition. Subsequently, the current-to-image conversion is carried out to represent the characteristics of residential appliance as the input of CNN. And the convolutional neural architecture is proposed to identify the power consumption load and its working condition. Selected experimental results are performed that the average recognition accuracy is over 93%, which demonstrates the effectiveness and superiority of the proposed methodology for home energy management.