A cooperative multi-agent system to accurately estimate residential energy demand

In recent years much attention has been devoted to understanding and predicting when and how occupants perform their daily routines and use electric appliances within buildings. The purpose here is to better estimate energy consumption. Despite efforts to capture individual behavior with precision, models often neglect how occupants interact with one another. Concretely, they do not accurately reproduce how occupants perform joint activities, such as having a meal or watching TV together, or sole activities, such as self-caring. Such inaccuracies, in turn, influence energy demand estimation, as joint and sole activities involve sharing and non-sharing of electrical appliances. Therefore, in this extended abstract, we propose a cooperative multi-agent system, where interaction of occupants is explicitly modeled by Interactive Markov chains. A preliminary study using data from five households in Osaka, Japan suggests this technique is better suited to capture interaction between occupants than the traditional Markov chain approach.