Reducing Feature and Temporal Complexity in High-Frequency NILM for Urban IoT Energy Monitoring

• Gabriel Arquelau Pimenta Rodrigues UnB https://orcid.org/0000-0002-4502-2153
• André Luiz Marques Serrano UnB https://orcid.org/0000-0001-5182-0496
• Geraldo Pereira Rocha Filho UnB / UESB https://orcid.org/0000-0001-6795-2768
• Rodolfo Ipolito Meneguette USP https://orcid.org/0000-0003-2982-4006
• Vinícius Pereira Gonçalves UnB https://orcid.org/0000-0002-3771-2605

DOI: https://doi.org/10.5753/courb.2026.22992

Resumo

High-frequency non-intrusive load monitoring (NILM) has achieved improved estimation accuracy with the use of harmonic electrical features in smart homes. However, the increased feature space produces significant computational, storage, and communication overheads, especially for scalable urban sensing deployments. Hence, this work investigates the reduction of feature dimensionality and of temporal complexity in high-frequency NILM for urban IoT energy monitoring while preserving predictive performance. A feature importance analysis is conducted using a Random Forest model, resulting in a reduced subset of 19 features, comprising those that together represent 90% of the aggregated importance for all appliances. The results show that models trained with the selected features achieve performance comparable to models trained with the complete feature set, despite a 48.65% dimensionality reduction, which reduces training and inference times and supports scalable deployment in urban sensing infrastructures. Assuming 32-bit floating-point representation, this reduction decreases the input feature vector from 148 bytes to 76 bytes per sample, reducing computational demands. Additionally, a temporal relevance analysis based on mutual information indicates that instantaneous measurements provide the most predictive information. Thus, these findings suggest that accurate high-frequency NILM can be achieved using a reduced feature set and minimal temporal context, promoting efficient deployment on IoT platforms to support scalable urban energy monitoring and more sustainable energy use.

Referências

AT, M. R., B, B., RR, S. A. P., Naidu, R. C., M, R. K., Ramachandran, P., Rajkumar, S., Kumar, V. N., Aggarwal, G., and Siddiqui, A. M. (2024). Intelligent energy management across smart grids deploying 6G IoT, AI, and blockchain in sustainable smart cities. IoT, 5(3):560–591.

Athanasiadis, C. L., Papadopoulos, T. A., and Doukas, D. I. (2021). Real-time non-intrusive load monitoring: A light-weight and scalable approach. Energy and Buildings, 253:111523.

Awal, M. A., Hossain, M. S., Debjit, K., Ahmed, N., Nath, R. D., Habib, G. M., Khan, M. S., Islam, M. A., and Mahmud, M. P. (2021). An early detection of asthma using BOMLA detector. IEEE Access, 9:58403–58420.

da Silva, M. V., Silva, G. G., and Gomes, R. L. (2020). Minimização do consumo de energia de dispositivos sem fio em ambientes industriais. In Workshop de Computação Urbana (CoUrb), pages 43–56. SBC.

Dinar, F., Paris, S., and Busvelle, E. (2025). Capturing high-frequency harmonic signatures for NILM: Building a dataset for load disaggregation. Sensors, 25(15).

Dong, Z., Zhang, X., Cai, S., Yang, Y., Jiang, W., Song, Y., Zhao, W., and Zhang, D. (2025). Real-time NILM: A lightweight and low power approach. Energy, page 138100.

Duan, J. (2025). Dynamic feature selection for silicon content prediction in blast furnace using BOSVRRFE. Scientific Reports, 15(1):20555.

Ewald F. Fuchs, M. A. S. M. (2023). Power Quality in Power Systems, Electrical Machines, and Power-Electronic Drives. Academic Press, 3 edition.

Filip, A. et al. (2011). BLUED: A fully labeled public dataset for event-based nonintrusive load monitoring research. In 2nd workshop on data mining applications in sustainability (SustKDD), volume 2012, page 5.

Fu, Y., Chen, S., Zhao, F., Shi, Y., Ye, S., Jiang, J., Wang, L., and Yang, X. (2024). Identification of electrical equipment based on VI trajectory, odd harmonic currents. In 2024 8th International Conference on Smart Grid and Smart Cities (ICSGSC), pages 164–169. IEEE.

Gao, J., Giri, S., Kara, E. C., and Bergés, M. (2014). Plaid: a public dataset of high-resoultion electrical appliance measurements for load identification research: demo abstract. In proceedings of the 1st ACM Conference on Embedded Systems for Energy-Efficient Buildings, pages 198–199.

Goudarzi, A., Ghayoor, F., Waseem, M., Fahad, S., and Traore, I. (2022). A survey on IoT-enabled smart grids: emerging, applications, challenges, and outlook. Energies, 15(19):6984.

Hart, G. W. (1992). Nonintrusive appliance load monitoring. Proceedings of the IEEE, 80(12):1870–1891.

Huber, P., Calatroni, A., Rumsch, A., and Paice, A. (2021). Review on deep neural networks applied to low-frequency NILM. Energies, 14(9):2390.

Kamenska, L., Moiseenko, V., Shendryk, V., Kamenskyi, S., and Shendryk, S. (2022). Implementation of distributed information systems in solving problems of energy consumption monitoring. In International Conference “New Technologies, Development and Applications”, pages 758–767. Springer.

Kang, H., Kim, H., et al. (2020). Household appliance classification using lower odd-numbered harmonics and the bagging decision tree. IEEE Access, 8:55937–55952.

Kelly, J. and Knottenbelt, W. (2015). The UK-DALE dataset, domestic appliance-level electricity demand and whole-house demand from five uk homes. Scientific data, 2(1):1–14.

Kolter, J. Z. and Johnson, M. J. (2011). REDD: A public data set for energy disaggregation research. In Workshop on data mining applications in sustainability (SIGKDD), San Diego, CA, volume 25, pages 59–62.

Li, K., Yin, B., Du, Z., and Sun, Y. (2020). A nonintrusive load identification model based on time-frequency features fusion. IEEE Access, 9:1376–1387.

Mishra, P. and Singh, G. (2023). Energy management systems in sustainable smart cities based on the internet of energy: A technical review. Energies, 16(19):6903.

Mylona, D. N. and Bouhouras, A. S. (2025). A digital twin-based framework for load identification using odd harmonic current plots. Applied Intelligence, 55(7):1–15.

Nour, M., Le Bunetel, J.-C., Ravier, P., and Raingeaud, Y. (2023). Data augmentation strategies for high-frequency NILM datasets. IEEE Transactions on Instrumentation and Measurement, 72:1–9.

Papageorgiou, P., Mylona, D., Stergiou, K., and Bouhouras, A. S. (2023). A time-driven deep learning NILM framework based on novel current harmonic distortion images. Sustainability, 15(17):12957.

Papageorgiou, P. G., Christoforidis, G. C., and Bouhouras, A. S. (2025). NILM in high frequency domain: A critical review on recent trends and practical challenges. Renewable and Sustainable Energy Reviews, 213:115497.

Picon, T., Meziane, M. N., Ravier, P., Lamarque, G., Novello, C., Bunetel, J.-C. L., and Raingeaud, Y. (2016). COOLL: Controlled on/off loads library, a public dataset of high-sampled electrical signals for appliance identification. arXiv preprint arXiv:1611.05803.

Rodrigues, G. A. P., de Oliveira, M. N., Serrano, A. L. M., Rocha Filho, G. P., Vergara, G. F., Mosquéra, L. R., and Gonçalves, V. P. (2025). MELISSA: An LLM-powered smart home energy consumption monitoring framework. In Simpósio Brasileiro de Computação Ubíqua e Pervasiva (SBCUP), pages 11–20. SBC.

Rodrigues, G. A. P., Serrano, A. L. M., Filho, G. P. R., Gonçalves, V. P., and Meneguette, R. I. (2026). An information-theoretic analysis of high-frequency load disaggregation. Entropy, 28(3):334.

Schirmer, P. A. and Mporas, I. (2022). Non-intrusive load monitoring: A review. IEEE Transactions on Smart Grid, 14(1):769–784.