Design and Implementation of an Edge Computing-Based Smart Energy Meter for Real Time Energy Monitoring and Management
Non-intrusive load monitoring infers the activity of individual appliances from a single measurement at the point of supply, avoiding the cost of instrumenting each device. The deep-learning approaches that dominate recent work place inference on a server, leaving the meter dependent on a network link. This paper reports a self-contained instrument in which inference runs on the meter itself. Galvanically isolated sensors are sampled against an analogue reference the device measures at every power-up, and six electrical quantities per acquisition frame are passed to a depth-three decision tree compiled into firmware, which identifies the connected load combination without any network connection. Validating across separate power cycles, rather than by randomly partitioning a single recording, proved decisive: it revealed that reactive power carried predictive value originating in converter offset rather than in the load, and discarding that feature raised accuracy on held-out sessions from 72% to 100% across four load states captured in eight independent acquisitions. Random partitioning conceals this failure entirely. Identified combinations are further decomposed into per-appliance power from learned resistances, agreeing with rated values to within 2.7%, and a shortfall exceeding 7% in combined resistance raises an over-draw alert. At this scale the limiting factor is therefore validation discipline rather than model capacity.
Authors
- Sharon Priya S
- Lavanya G
- Sahana N
- T S Ananya
- Varshitha G G
Publication Details
- Journal
- Iconic Research and Engineering Journals
- Published
- 2026-11-18
- DOI
- https://doi.org/10.64388/ire1725287
- Primary Topic
- Smart Grid Energy Management
- Type
- article
- Field-Weighted Citation Impact
- 0.00