Hybrid HDBSCAN-FHMM Approach for Energy Disaggregation in Non-Intrusive Load Monitoring (NILM) Systems
Non-intrusive load monitoring (NILM) is emerging as a useful approach to improving the energy efficiency of buildings and households, particularly in the face of the growing challenges of environmental sustainability. Despite recent advances, the accuracy and reliability of disaggregation algorithms...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2025-01-01
|
| Series: | IEEE Access |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/11006080/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Non-intrusive load monitoring (NILM) is emerging as a useful approach to improving the energy efficiency of buildings and households, particularly in the face of the growing challenges of environmental sustainability. Despite recent advances, the accuracy and reliability of disaggregation algorithms remain limited by the diversity of household energy behaviors and the heterogeneous operation of domestic appliances. In this study, we propose a method based on the prior identification of appliance operating states using the HDBSCAN (Hierarchical Density-Based Spatial Clustering of Application with Noise) algorithm, prior to the disaggregation of overall energy consumption. Our study is based on the AMPds dataset (Almanac of Minutely Power Dataset), which provides detailed measurements of residential power consumption. Our methodology consisted of two main phases: an appliance operating state identification phase using the HDBSCAN algorithm, followed by an energy disaggregation phase based on a modified FHMM (Factorial Hidden Markov Model). The model’s performance was evaluated using three main metrics: F1 score, Mean Absolute Error (MAE) and Match Rate, in comparison with several reference models including the Adaptive-FHMM. The results obtained show that our HDBSCAN-FHMM model outperforms several models of energy disaggregation algorithms, but also the Adaptive-FHMM reference approach, with significant improvements of 6.25%, 46.24% and 12.04% for the F1, MAE and Match Rate metrics respectively. These performances reinforce the reliability of our method in accurately attributing the energy consumption of household appliances. Nevertheless, challenges remain, particularly with regard to model robustness in the face of variability in energy behavior and technological evolution. Future exploration of deep learning techniques combined with current methods could offer significant advances in energy disaggregation. |
|---|---|
| ISSN: | 2169-3536 |