NILM : energy monitoring, modelling, and disaggregation

Student thesis: Doctoral Thesis


Smart meters have begun to replace traditional energy meters across the world. These meters will be able to give near real-time energy usage to consumers, industry and suppliers, the benefits of this are marketed as; easier energy management, saving money and reducing emissions. However by themselves smart meters are unlikely to be able to achieve this as only a small sub-set of users are likely to remain engaged with their smart meter long term. Non-intrusive Load Monitoring (NILM) aims to provide a method to explain to consumers in more detail about their energy usage without need for their input or attention, be it explaining which appliances are causing high energy consumption within the home or in an industrial setting, explaining appliance/machinery usage to maximise scheduling with time of use tariffs. We demonstrate the steps and methodology to produce meaningful and explainable results which could as part of an energy suppliers service to provide enhanced billing information, similar to some credit card statements, showing a breakdown of appliance usage and statistics. This thesis provides steps from data collection to results and visualisation as part of a complete NILM workload. We demonstrate data management, pre-processing, appliance modelling for analysis of individual appliances, neural network model creation and evaluation for both commercial and residential premises, the need for transfer learning to work at scale, and the explainability of the networks and results, necessary to provide accurate information and ensure customers can understand any errors they might see from a NILM system.
Date of Award5 Jun 2023
Original languageEnglish
Awarding Institution
  • University Of Strathclyde
SponsorsUniversity of Glasgow
SupervisorLina Stankovic (Supervisor) & Ivan Andonovic (Supervisor)

Cite this