Uncovering hidden demand flexibility using non-intrusive load monitoring (NILM) : a case for Southern Africa – Namibia

Mollel, Rachel Stephen and Elombo, Andreas and Stankovic, Lina and Stankovic, Vladimir and Hambata, Jona and Thiel, Gunther; (2024) Uncovering hidden demand flexibility using non-intrusive load monitoring (NILM) : a case for Southern Africa – Namibia. In: 12th International Conference on Energy Efficiency in Domestic Appliances and Lighting (EEDAL’24). UNSPECIFIED, JPN. (In Press)

[thumbnail of EEDAL-2024-Uncovering-hidden-demand-flexibility-using-non-intrusive-load-monitoring] Text. Filename: EEDAL-2024-Uncovering-hidden-demand-flexibility-using-non-intrusive-load-monitoring.pdf
Accepted Author Manuscript
Restricted to Repository staff only until 1 January 2099.
Download (1MB) | Request a copy

Abstract

The energy system is undergoing an unprecedented transformation, with increasing integration of emerging energy technologies challenging its capacity. This paper explores operational aspects (demand composition in particular) of the energy system and proposes a data-driven methodology that could enhance active network management by uncovering unseen demand flexibility through characterising energy consumption profiles of key electrical loads. We address the challenges of the lack of curated energy datasets, focusing on the Southern Africa region, specifically Namibia. We also address the challenging very low-frequency non-intrusive load monitoring (NILM) problem. Since there was no sub-metering or labelled data for model training, we propose an unsupervised K-means-based methodology, leveraging the user-survey information on appliance wattage and the time the appliance was used to gain granular insights on disaggregated energy consumption patterns, thereby providing appliance-level visibility into which electrical loads can be considered for demand-side response by utilising flexible usage of such loads. To validate our approach, we evaluated the proposed methodology on the public resampled REFIT (UK) dataset, benchmarked against state-of-the-art very low-frequency NILM methods, viz. Graph Signal Processing (GSP), Convolution Neural Network (CNN), Factorial Hidden Markov Model (FHMM), Combinatorial Optimization (CO), Discriminative Disaggregation Sparse Coding (DDSC), and Optimisation-based (OPT) approaches. We demonstrated successful disaggregation of appliances with a duty cycle of 30 minutes or more in REFIT Houses 4 and 8. Upon evaluating several appliances, such as TV, washing machine, and washer-dryer, for REFIT Houses 4 and 8, we achieved match rates as high as 80% and classification accuracies of up to 89%. These results are consistent with state-of-the-art approaches, establishing confidence in our load disaggregation methodology, which we then applied to the unlabelled residential energy dataset from Namibia.

ORCID iDs

Mollel, Rachel Stephen ORCID logoORCID: https://orcid.org/0000-0001-8591-9830, Elombo, Andreas, Stankovic, Lina ORCID logoORCID: https://orcid.org/0000-0002-8112-1976, Stankovic, Vladimir ORCID logoORCID: https://orcid.org/0000-0002-1075-2420, Hambata, Jona and Thiel, Gunther;
CORE (COnnecting REpositories)