Utility-based context-aware multi-agent recommendation system for energy efficiency in residential buildings

A significant part of CO 2 emissions is due to high electricity consumption in residential buildings. Using load shifting can help to improve the households’ energy efficiency. To nudge changes in energy consumption behavior, simple but powerful architectures are vital. This paper presents a novel algorithm of a recommendation system generating device usage recommendations and suggests a framework for evaluating its performance by analyzing potential energy cost savings. As a utility-based recommender system, it models user preferences depending on habitual device usage patterns, user availability, and device usage costs. As a context-aware system, it requires an external hourly electricity price signal and appliance-level energy consumption data. Due to a multi-agent architecture, it allows for easy integration of new agents, enabling seamless functionality expansion, or the disabling of existing agents to tailor the system to specific needs. Empirical results show that the system can provide energy cost savings of 18% and more for most studied households.