Load Disaggregation: Towards Energy Efficient Systems

Abstract

Electricity is one of the valuables and widely used forms of energy. However, with the fast-paced technological development in the electrical and electronics market, the electricity demand is on a constant rise. To tackle energy and sustainability issues, two paths can be followed by the world community. Either new generation plants to be established with an expense of millions of dollars or to explore the existing system by integrating innovative techniques that can lead to energy efficiency and conservation. The latter one is a more viable solution, and the research community is extensively working to propose and develop innovative techniques towards energy efficiency and conservation. In this context, energy monitoring is one of the key techniques that play a significant role in the field of sustainable energy.

Load disaggregation is one of these promising energy monitoring techniques, where a non-intrusive load disaggregation technique commonly referred to as non-intrusive load monitoring (NILM) is widely adopted to provide individual load profiles to the stakeholders. Appliance-level energy monitoring is not only beneficial for the consumers in terms of having valuable information regarding the operation status of their loads and corresponding consumption but also benefit the system operators, policymakers, and manufacturers in terms of analyzing the network’s energy flow, creating policies/tariffs, and manufacturing of smart appliances, respectively.

This research work contributes to the existing research and development of non-invasive load disaggregation systems, by proposing and developing a robust event-based non-intrusive load disaggregation approach for low sampling data granularity. As a way forward, this research work contributes to different aspects of a NILM system. For NILM event detection, three new low complexity and computationally fast algorithms based on statistical parameters are proposed and validated on real-world datasets. In terms of electrical load features, a set of nine distinct load features based on statistical, geometrical, and power features is proposed. The extracted load features are further investigated in terms of significance using different feature selection methodologies and the extracted results are validated in the context of classification performance. For load classification, this research work investigated different supervised machine learning models towards an optimal learning model for the given conditions. In addition to standalone machine learning models, this research also presents a combinatorial learning model, i.e., ensemble learning, for load classification in the context of the NILM. Further, a comprehensive comparative evaluation of these techniques is also part of this thesis.

The entire digital simulations and corresponding analysis presented in this research work are based on real-world electricity datasets, originating from different geographical regions, i.e., New Zealand and the United States of America. Based on the low data granularity of the employed databases, three different appliances/circuits, i.e., air conditioning unit, electric vehicle charging, and water heating are successfully disaggregated using the proposed non-intrusive load disaggregation approach. Moreover, a proof of concept in terms of real-world deployment, i.e., the application of the proposed non-intrusive load disaggregation, is also proposed and validated in this research work.

Due to low data granularity nature, this research work is more relevant for the existing metering infrastructure. Therefore, the proposed methodologies and corresponding simulation studies presented in this research work will significantly contribute to the existing state of the art on low sampling NILM systems particularly in terms of event detection, electrical load features, and learning model selection. The study presented in this thesis will also facilitate future research in terms of real-world deployment of NILM systems and its broader applications. Concisely, this research work based on a non-invasive load disaggregation approach is a way forward for energy efficient systems.