Self-Adaptive Energy Saver

Currently one of the main areas of improvement of the buildings energy performance is the optimization of regulation systems and controlling the flow of energy. To this end, we propose an approach based on multi-agent systems in which the optimization is performed without prior knowledge about the dynamics of the building. We evaluate the developed multi-agent system on its learning ability and optimization of the set point during the night. We show that the result converges efficiently towards the optimum, previously determined by a professional building simulator. This approach is generic enough to be extended to many observable and checkpoints building without modification of the algorithms decision agents.