EDGE APPROACH TO DETECTING ANOMALIES IN ELECTRICITY CONSUMPTION
Keywords:
machine learning, microcontrollers, ESP32, Smart Home Energy ConsumptionAbstract
The study presents an autonomous ESP32-based module with an INA219 sensor for early detection of energy consumption anomalies in household appliances. The key innovation is shifting machine-learning analytics to the network edge: a quantized TensorFlow Lite autoencoder learns the appliance’s normal profile within one week and then, in real time, compares the reconstruction error with a statistical threshold of μ + 2σ. Any excess instantly triggers an MQTT alert, eliminating the need to stream raw data to the cloud. Laboratory tests with a refrigerator, washing machine, and an electric kettle demonstrated 100% detection of abrupt surges (> 50%) and 92% detection of gradual drifts (~ 10% per week) with ~ 1 s alert latency. The modular firmware can be deployed on off-the-shelf smart sockets, enabling up to 15% annual energy savings and enhanced fire safety without compromising user privacy
References
Gopinath R. DeepEdge-NILM: a case study of non-intrusive load monitoring edge device in commercial building // Energy & Buildings. – 2023. – Vol. 294. – Art. 132263. – DOI: 10.1016/j.enbuild.2023.132263.
Miettinen M. A federated learning approach to anomaly detection in smart buildings // Proceedings of the 8th ACM International Conference on Systems for Energy-Efficient Built Environments (BuildSys 2021). – New York : Association for Computing Machinery, 2021. – P. 149–158. – DOI: 10.1145/3486611.3486651.
Rasheed A.; Tu M. TinyML: machine learning on ultra-low-power microcontrollers — state of the art // IEEE Access. – 2021. – Vol. 9. – P. 35980–36000. – DOI: 10.1109/ACCESS.2021.3061962.
Razavi S. E. Edge-based NILM for energy monitoring // Energy & Buildings. – 2024. – Vol. 297. – Art. 112506. – DOI: 10.1016/j.enbuild.2024.112506.
Smart Home Energy Consumption : [dataset]. – Version 08-2024. – Kaggle, 2024. – URL: https://www.kaggle.com/mexwell/smart-home-energy-consumption (дата звернення: 16.05.2025).
Kea K.; Han Y.; Kim T.-K. Autoencoder-based federated learning for power-system anomaly detection // PLOS ONE. – 2023. – Vol. 18, № 8. – e0290337. – DOI: 10.1371/journal.pone.0290337.
