Generalization Capacity Analysis of Non-Intrusive Load Monitoring Using Deep Learning

Abstract

Appliance Load Monitoring is a technique used to monitor devices existing in homes, industry or naval vessels. Acquisition of device-level data can provide great benefits in many areas such as energy management, demand response, and load forecasting. However, the monitoring process is often provided with a costly installation, as it requires a large number of sensors and a data center. Non-Intrusive Load Monitoring (NILM) is an alternative and cost-efficient load monitoring solution. Simply put, NILM is the process of obtaining device-level data by analyzing the aggregated data read from the main meter that measures the electricity consumption of the whole house. Before NILM analysis is performed, the load patterns of the appliances are usually modeled individually. In general, one model for each appliance is modeled even if the appliance has more than one operating program such as washing machine and oven. Therefore, when the appliance operates in other programs, the accuracy of NILM analysis decreases. In this paper, an appliance-based NILM analysis has been made considering the appliances having multiple operating programs. In order to increase the accuracy of NILM analysis, several deep learning methods, which are the most important data-driven technique of recent times, are used. Developed models were tested in IoT Microgrid Laboratory environment.