Analysis of Energy Disaggregation for Sustainable Building Applications

Abstract

Energy disaggregation approaches have been employed for various smart living applications to keep the consumers aware of their everyday power consumption behaviors, and its overall impact on their utility bills and their living environments. These techniques have been widely investigated at a fine grain to improve the energy efficiency related to Heating Ventilation and Air-conditioner (HVAC) operations, residents' comfort management, and occupancy detection in built environments. In this thesis, we investigate the relevance of energy disaggregation for energy analytics applications with the availability of the additional information related to human activities of daily living (ADLs), acoustic signature of the appliances, metadata of the buildings, thermostat setpoints, and external weather conditions. First, we present an iterative noise separation based approach to perform energy disaggregation using sparse coding-based methodologies which works at the single ingress point of a home, i.e., at the meter level. We extend this approach by embedding deep learning techniques based on de-noising autoencoders to investigate energy disaggregation across a large set of homes across different granularities. Next, we present two independent studies showing the feasibility of exploiting ambient sensing and contextual information for energy analytics. We postulate a human activity augmented method to identify individual appliances based on their unique transient and steady-state power signatures. In the second case study, we show a correlation between acoustic noise of appliances with their power consumption and design a probabilistic approach for acoustic-based appliance identification. Finally, we showcase the application of energy disaggregation in building retrofit. We present a systematic study of Bayesian approaches to the modeling of buildings' thermal dynamics. We postulate a generalized Bayesian State Space Model (BSSM) that can combine physics-based thermal models into a probabilistic framework. We further embed prior intuition and knowledge regarding buildings into the model based on subjective beliefs, such as the age of the building and area size.