activIn: A novel Non-Intrusive Activity Inference Tool
Asimina Dimara, Stelios Krinidis, Dimitrios K. Tzovaras · 2020
The first step towards energy awareness is to gain the knowledge on how much energy each household device consumes. This paper introduces a new methodology for activity inference (activIn) estimating the action of a resident that may be directly associated with the use of a specific electric device. The devices' energy load is distinguished in low power consumption and high power consumption using an unsupervised Hidden Markov Model establishing the devices power consumption base-load. The device's operation mode is inferred using a Random Forest Classifier that extracts the device's daily operation pattern while inferring the device's associated activity. Finally, the proposed activIn tool is evaluated against real-life and realtime devices' consumption data, while the experimental results show that it is a powerful tool for user's activity, device profiling, energy consumption awareness and device-driven applications.