Mwansa, M (2022) Modelling and Predicting Energy Usage from Smart Meter Data and Consumer behaviours in Residential Houses. Doctoral thesis, Liverpool John Moores University.

Preview

Text Modelling and Predicting Energy Usage from Smart Meter Data and Consumer behaviours in Residential Houses..pdf - Published Version Available under License Creative Commons Attribution Non-commercial. Download (6MB) | Preview

Abstract

Efforts of electrical utilities to respond to climate change requires the development of increasingly sophisticated, integrated electrical grids referred to as the smart grids. Much of the smart grid effort focuses on the integration of renewable generation into the electricity grid and on increased monitoring and automation of electrical transmission functions. However, a key component of smart grid development is the introduction of the smart electrical meter for all residential electrical customers. Smart meter deployment is the corner stone of the smart grid. In addition to adding new functionality to support system reliability, smart meters provide the technological means for utilities to institute new programs to allow their customers to better manage and reduce their electricity use and to support increased renewable generation to reduce greenhouse emissions from electricity use. As such, this thesis presents our research towards the study of how the data (energy usage profiles) produced by the smart meters within the smart grid system of residential homes is used to profile energy usage in homes and detect users with high fuel consumption levels. This project concerns the use of advanced machine learning algorithms to model and predict household behaviour patterns from smart meter readings. The aim is to learn and understand the behavioural trends in homes (as demonstrated in chapter 5). The thesis shows the trends of how energy is used in residential homes. By obtaining these behavioural trends, it is possible for utility companies to come up with incentives that can be beneficial to home users on changes that can be adopted to reduce their carbon emissions. For example consumers would be more likely prompted to turn of unusable appliances that are consuming high energy around the home e.g., lighting in rooms which are un occupied. The data used for the research is constructed from a digital simulation model of a smart home environment comprised of 5 residential houses. The model can capture data from this simulated network of houses, hence providing an abundance set of information for utility companies and data scientist to promote reductions in energy usage. The simulation model produces volumes of outliers such as high periods (peak hours) of energy usage and low periods (Off peak hours) of anomalous energy consumption within the residential setting of five homes. To achieve this, performance characteristics on a dataset comprised of wealthy data readings from 5 homes is analysed using Area under ROC Curve (AUC), Precision, F1 score, Accuracy and Recall. The highest result is achieved using the Two-Class Decision Forest classifier, which achieved 87.6% AUC.

Item Type:	Thesis (Doctoral)
Uncontrolled Keywords:	Smart Grid Systems, Smart Meters, Energy profiling, Greenhouse emissions, Residential electricity consumption
Subjects:	Q Science > QA Mathematics > QA75 Electronic computers. Computer science Q Science > QA Mathematics > QA76 Computer software T Technology > T Technology (General) T Technology > TD Environmental technology. Sanitary engineering
Divisions:	Computer Science & Mathematics
Date Deposited:	08 Mar 2022 13:41
Last Modified:	19 Dec 2022 15:31
DOI or ID number:	10.24377/LJMU.t.00016411
Supervisors:	Hurst, W and Yuanyuan, S
URI:	https://researchonline.ljmu.ac.uk/id/eprint/16411

View Item