A framework for enabling energy-aware facilities through minimally-intrusive approaches

摘要: This thesis presents a framework for leveraging easy-to-obtain data sources in a residential building to infer the operational schedule and electricity consumption of the appliances present in it. The framework, which utilizes and extends techniques from the Non-Intrusive Load Monitoring (NILM) domain, is designed around the end-user experience with a particular focus on attempting to automate the process of training and calibrating the algorithms. To simplify the training I developed an approach for generating electrical signatures that can generalize the state transition transients for a particular type of appliance. These new generalized signatures, the eigen-transients, reduce the need of the user to train the system on the appliance class in question. A continuous calibration process for the appliance models learned by the system is also developed. This approach can leverage user input as well as triggers based …