Assessing the Effectiveness of Supervised and Semi-supervised NILM Approaches in an Industrial Context
Résumé
Non-Intrusive Load Monitoring (NILM) is a technique that aims to estimate the energy consumption and operational status of individual appliances in a building by analyzing only the aggregate power usage data. This technique plays a crucial role in demand-side management and energy conservation efforts by providing detailed information about the energy consumption patterns of individual appliances. Naturally, NILM is considered as a supervised problem, that is a Regression and Classification problem. Various Deep Neural Network models have recently been developed for NILM regression and classification tasks. However, training deep neural networks requires a significant amount of labeled data and collecting consumption data over a prolonged period exposes consumers to severe privacy risks. Hence semi-Supervised learning is also used for NILM. Furthermore, most NILM research uses datasets from the residential sector, and only a handful of research uses datasets from tertiary sectors. This paper comprehensively studies the performance of supervised and semi-supervised NILM algorithms based on a tertiary sector dataset named the GreEn-ER dataset. The semi-supervised deep learning NILM method performs classification tasks, whereas the supervised method performs classification and regression NILM tasks simultaneously. Based on the analysis, it is concluded that the presented NILM algorithms can yield satisfactory results using the tertiary sector's consumption dataset. Additionally, the ideal input parameters that facilitates the performance of the both NILM algorithms are also noted.
Origine | Fichiers produits par l'(les) auteur(s) |
---|