Current peak based device classification in NILM on a low-cost embedded platform using extra-trees

Non-Intrusive Load Monitoring (NILM) is a method for disaggregation of energy consumption of individual appliances in a household. This involves the classification of individual appliances, for which a number of electrical features in combination with machine learning algorithms have been used. Extraction of most of these features is a computationally demanding task, and use of complex machine learning algorithms further adds to this complexity. Although solutions to this problem exist, they tend to be expensive and unaffordable to consumers in developing countries. This necessitates a need for an inexpensive solution capable of running on low-cost embedded platforms. In this paper, the authors implement a machine learning approach on an embedded platform to address this problem using current-based features for device classification. The model was evaluated using the Building-Level fUlly-labeled dataset for Electricity Disaggregation (BLUED) which contains electrical measurements for a household in the US for one week. The classifier was trained on Raspberry Pi 3 in about 4 seconds and classification of an event was performed in under 400 ms, validating the feasibility of the classification model on such a platform.