Towards Energy Disaggregation at Home: A Self-Hosted NILM System Integrated with Home Assistant

Details

Supervisors

Barbette, Tom ; Sadre, Ramin

Faculty

Ecole polytechnique de Louvain

Degree label

Master [120] en sciences informatiques, à finalité spécialisée

Abstract

enMost households have no visibility into which appliances drive their electricity bill, beyond a single aggregate figure. Non-Intrusive Load Monitoring (NILM) addresses this by disaggregating the whole-home power signal into per-appliance consumption from a single measurement point, without dedicated sub-metering hardware. Despite strong benchmark results, NILM systems remain largely absent from ordinary homes, held back by three obstacles : domain shift across households, the scarcity of appliance-level labelled data, and the complexity of deploying and operating such systems. The goal of this thesis is to make NILM deployable in a real household. It proposes a self-hosted system built around a compact causal Temporal Convolutional Network (TCN), sized for single-household training and inference on consumer-grade hardware, operating on low-frequency active power compatible with common smart plugs. The model is integrated into Home Assistant through a custom component and a Dockerised local backend, so that all fine-grained energy data stays on local hardware. Supervision is provided automatically by smart plugs, and a fine-tuning mechanism adapts an existing model when an appliance is replaced by a different instance.