PhD student Jon A. R. Liisberg
“Data-driven models for energy advising leading to behavioural changes in residences”
Supervisors:
Principal supervisor: Associate Professor Jan Kloppenborg Møller, DTU Compute
Co supervisor: Associate Professor Peder Bacher, DTU Compute
Co supervisor: Professor Henrik Madsen, DTU Compute
Co supervisor: Projectleader Eva Sass Lauritsen, SEAS-NVE
Co supervisor: Data Scientist Anders Spur Hansen, SEAS-NVE
Co supervisor: Daniel Cermak-Sassenrath, ITU
Examiners:
Associate Professor Lasse Engbo Christiansen, DTU Compute
Senior lecturer Jesper Rydén, Department of Energy and Technology, SLU
Assistant Professor Simon Rouchier, Université Savoie Mont-Blanc
Chairperson at defence:
Associate Professor Uffe Høgsbro Thygesen, DTU Compute
Popular Science Summary:
Given the smart-meter roll-out in all of Europa and the implementation of the supplier-centric DataHub in Denmark since 2016, there is a great potential for increasing awareness among residential electricity consumers on their consumption. The smart-meter data is usually in hourly temporal resolution, available up-to one year in the past.
This thesis deals with feedback/advice to residential electricity consumers, based on data-driven models using the consumers own individual smart-meter data. Beside data-driven feedback a game was developed as a study on gamified interaction of electricity consumption.
Models for medium to long-term prediction of daily individual residential electricity consumption were developed. The models were applied in large-scale as a monitoring tool and presented to the consumers through the app Watts. The effect on electricity consumption of the feedback provided by the prediction model through the app Watts was evaluated, showing a decreasing effect for active users.
Methods for disaggregating hourly electricity consumption were also developed using Hidden Markov models (HMM). The states of the HMM are described in accordance to appliance-related activities and shows potential for probabilistic short-term load forecasting. These methods are yet to be applied in large-scale.