ElectriSense: Single-Point Sensing Using EMI for Electrical Event Detection and Classification in the Home

Abstract

Imagine an energy feedback system that displays not only your total power consumption, but also continuously shows real-time usage while breaking it down categorically by electrical appliances. In addition, such a system provides personalized and cost-effective energy saving recommendations, for example, it could report, "Based on your usage patterns, you could save $215 per year by switching to a more efficient heating unit, which will pay for itself in 27 months". The challenge in this scenario is to sense end-uses of energy to provide feedback at the appliance level. Though an individual measurement device could be installed on each and every appliance to obtain such fine-grained energy usage patterns, such an approach is not only expensive but requires tedious maintenance and installation. In this thesis I formalize and evaluate a new sensing techniques that requires minimal instrumentation in the form of a single plug-in sensor for measuring disaggregated electrical energy usage in the home. In particular, this sensor leverages Electromagnetic Interference (EMI) produced by most consumer electronics as signatures to detect and classify their use in real-time. In addition, I highlight the complementary and exclusive machine learning features can be extracted from this EMI which can profoundly strengthen prior algorithms used for non-intrusive energy disaggregation. I also show that EMI from appliances can also be used to estimate their operating states or mode of operation yielding implications for not only better energy disaggregation but also human activity sensing.