Non-Intrusive Load Monitoring (NILM) enables the disaggregation of the global power consumption, measured from a single smart electrical meter, into appliance-level details. State-of-the-Art is based on Machine Learning methods and on the fusion of time- and frequency-domain features. Running compute-demanding and low-latency NILM on low-cost MCU-based meters is currently an open challenge. This paper addresses the optimization of the feature spaces and the computational and storage cost reduction for SoA NILM algorithms on memory- and compute-limited MCUs. We compare four supervised learning techniques on different classification scenarios and characterize the overall NILM pipeline's implementation. Experimental results demonstrate that optimizing the feature space enables edge-based NILM with 95.15% accuracy, resulting in a small drop compared to the most-accurate technique (96.19%), while achieving up to 5.45x speed-up and 80.56% storage reduction. Furthermore, we show that low-latency NILM relying only on current measurements achieves 80% accuracy, allowing cost reduction by removing voltage sensors.
Tabanelli, E., Brunelli, D., Acquaviva, A., Benini, L. (2022). Trimming Feature Extraction and Inference for MCU-based Edge NILM: a Systematic Approach. IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, 18(2), 943-952 [10.1109/TII.2021.3078186].
Trimming Feature Extraction and Inference for MCU-based Edge NILM: a Systematic Approach
Tabanelli E.
;Brunelli D.;Acquaviva A.;Benini L.
2022
Abstract
Non-Intrusive Load Monitoring (NILM) enables the disaggregation of the global power consumption, measured from a single smart electrical meter, into appliance-level details. State-of-the-Art is based on Machine Learning methods and on the fusion of time- and frequency-domain features. Running compute-demanding and low-latency NILM on low-cost MCU-based meters is currently an open challenge. This paper addresses the optimization of the feature spaces and the computational and storage cost reduction for SoA NILM algorithms on memory- and compute-limited MCUs. We compare four supervised learning techniques on different classification scenarios and characterize the overall NILM pipeline's implementation. Experimental results demonstrate that optimizing the feature space enables edge-based NILM with 95.15% accuracy, resulting in a small drop compared to the most-accurate technique (96.19%), while achieving up to 5.45x speed-up and 80.56% storage reduction. Furthermore, we show that low-latency NILM relying only on current measurements achieves 80% accuracy, allowing cost reduction by removing voltage sensors.File | Dimensione | Formato | |
---|---|---|---|
trimming feature extraction post print.pdf
accesso aperto
Tipo:
Postprint
Licenza:
Licenza per accesso libero gratuito
Dimensione
1.34 MB
Formato
Adobe PDF
|
1.34 MB | Adobe PDF | Visualizza/Apri |
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.