Activity-Based Recommendations for Demand Response in Smart Sustainable Buildings

TL;DR

This paper presents an activity prediction-based recommendation system for demand response in smart buildings, aiming to reduce CO2 emissions and energy costs by suggesting activity shifts without extensive sensor data.

Contribution

It introduces a novel utility-based multi-agent framework that predicts activities using energy data alone, enabling cost-effective and user-friendly demand response strategies.

Findings

12% CO2 emissions reduction when focusing on emissions savings

20% energy cost savings achievable with recommendations

System integrates seamlessly into daily routines for long-term adoption

Abstract

The energy consumption of private households amounts to approximately 30% of the total global energy consumption, causing a large share of the CO2 emissions through energy production. An intelligent demand response via load shifting increases the energy efficiency of residential buildings by nudging residents to change their energy consumption behavior. This paper introduces an activity prediction-based framework for the utility-based context-aware multi-agent recommendation system that generates an activity shifting schedule for a 24-hour time horizon to either focus on CO2 emissions or energy cost savings. In particular, we design and implement an Activity Agent that uses hourly energy consumption data. It does not require further sensorial data or activity labels which reduces implementation costs and the need for extensive user input. Moreover, the system enhances the utility option of saving energy costs by saving CO2 emissions and provides the possibility to focus on both dimensions. The empirical results show that while setting the focus on CO2 emissions savings, the system provides an average of 12% of emissions savings and 7% of cost savings. When focusing on energy cost savings, 20% of energy costs and 6% of emissions savings are possible for the studied households in case of accepting all recommendations. Recommending an activity schedule, the system uses the same terms residents describe their domestic life. Therefore, recommendations can be more easily integrated into daily life supporting the acceptance of the system in a long-term perspective.