Multistage Energy Management of Coordinated Smart Buildings: A Multiagent Markov Decision Process Approach

Smart buildings provide an important opportunity for large-scale development of demand response, due to their existing flexibility that can be harvested through Internet-of-Things technologies with minimal cost of new equipment. However, after taking an energy management action, the resulting energy consumption of a building depends on several uncertain factors. Thus, the consumption of the smart building is not directly controllable and, contrary to the typical approach taken in the literature, it cannot be modeled as a decision variable in practice. In this paper, we consider the problem of coordinating the stochastic load control actions of multiple smart buildings under such endogenous uncertainties. We model the problem as a Multi-agent Markov Decision Process and, after reformulations, we bring it to a solvable decomposed form. Our simulations compare the proposed approach with a myopic approach that does not consider future uncertainties, and also quantify the trade-off between cost-effectiveness and computational time in terms of the look-ahead horizon length.