Reinforced model predictive control (RL-MPC) for building energy management

Buildings need advanced control for the efficient and climate-neutral use of their energy systems. Model predictive (MPC) reinforcement learning (RL) arise as two powerful techniques that have been extensively investigated in literature application to building management. These methods show complementary qualities terms constraint satisfaction, computational demand, adaptability, intelligibility, but usually a choice is made between both approaches. This paper compares approaches proposes novel algorithm called reinforced (RL-MPC) merges relative merits. First, complementarity RL MPC emphasized on conceptual level by commenting main aspects each method. Second, RL-MPC described effectively combines features from approach, namely state estimation, dynamic optimization, learning. Finally, MPC, RL, are implemented evaluated BOPTEST, standardized simulation framework assessment algorithms buildings. The results indicate pure cannot provide satisfaction when using formulation equivalent same controller model new can meet constraints similar performance while enabling continuous possibility deal with uncertain environments. • compared merged optimal problem. an machine theory. benchmarked BOPTEST framework. uses violates constraints. enables meets MPC.