High-Reliability Load Recognition in Home Energy Management Systems

Cabral, Thales Wulfert; Lemes, Dimas Augusto Mendes; Fraidenraich, Gustavo; Neto, Fernando Bauer; Lima, Eduardo Rodrigues de; Meloni, Luís Geraldo Pedroso

doi:10.1109/access.2023.3262128

Cited by 7 publications

(37 citation statements)

References 72 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…For the REFIT dataset, the difference between RELM and ELM reveals a 0.36% accuracy advantage for the RELM model. This advantage is the same for RELM compared to the SVM of the state-of-the-art system proposed in Cabral et al [7]. For the REDD dataset, RELM's advantage over SVM is 0.21%.…”

Section: Major Contributionsmentioning

confidence: 64%

“…Our method provides an ultra-low training time of 0.082 s with the REFIT database, less than the SVM of the technique reported in Cabral et. al [7], which has a time of 0.469 s. This result means that the proposed approach is approximately 5.72 times faster than the competitor, representing a time saving of 82.52% compared to the competitor. Concerning ELM, the proposed approach is 2.33 times faster and saves approximately 57.07% of the time.…”

Section: Major Contributionsmentioning

confidence: 89%

“…Presently, the modern methods for load recognition use images generated from electrical signals such as voltage, current, active power, and others. The advantages of this approach are substantiated by the works as Faustine and Pereira [10], Matindife et al [11], Gao et al [17], De Baets et al [9], and Cabral et al [7]. In this regard, the motivation of this study is to propose enhancements to the load recognition system in HEMS.…”

Section: Introductionmentioning

confidence: 90%

“…The electricity demand of a household may exhibit seasonal variations, influenced by the type of appliances in operation, such as heaters and air conditioning devices. However, HEMS is capable of monitoring such appliance activities, incorporating additional functionalities such as load disaggregation techniques presented by Lemes et al [5], methods for anomaly detection in appliances as per Lemes et al [6], and load recognition mechanisms as reported by Cabral et al [7].…”

Section: Introductionmentioning

confidence: 99%

“…Furthermore, reducing the amount of information needed to identify the appliance in operation is a pertinent feature of the solutions. According to Cabral et al [7] and Soe and Belleudy [15], it is feasible to achieve this using ML techniques for feature extraction, such as Linear Discriminant Analysis (LDA), Principal Component Analysis (PCA), and others. Conversely, we can limit the amount of information required regarding household appliances by solely using a single type of electrical signal, such as voltage, current, reactive power, or active power.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Load Recognition in Home Energy Management Systems Based on Neighborhood Components Analysis and Regularized Extreme Learning Machine

Cabral,

Neto,

de Lima

et al. 2024

Sensors

Self Cite

View full text Add to dashboard Cite

Efficient energy management in residential environments is a constant challenge, in which Home Energy Management Systems (HEMS) play an essential role in optimizing consumption. Load recognition allows the identification of active appliances, providing robustness to the HEMS. The precise identification of household appliances is an area not completely explored. Gaps like improving classification performance through techniques dedicated to separability between classes and models that achieve enhanced reliability remain open. This work improves several aspects of load recognition in HEMS applications. In this research, we adopt Neighborhood Component Analysis (NCA) to extract relevant characteristics from the data, seeking the separability between classes. We also employ the Regularized Extreme Learning Machine (RELM) to identify household appliances. This pioneering approach achieves performance improvements, presenting higher accuracy and weighted F1-Score values—97.24% and 97.14%, respectively—surpassing state-of-the-art methods and enhanced reliability according to the Kappa index, i.e., 0.9388, outperforming competing classifiers. Such evidence highlights the promising potential of Machine Learning (ML) techniques, specifically NCA and RELM, to contribute to load recognition and energy management in residential environments.

show abstract

Section: Major Contributionsmentioning

confidence: 64%

Section: Major Contributionsmentioning

confidence: 89%