Comparison of Optimal Control Techniques for Building Energy Management

Optimal controllers can enhance buildings’ energy efficiency by taking forecast and uncertainties into account (e.g., weather occupancy). This practice results in savings making better use of systems within the buildings. Even though benefits advanced optimal have been demonstrated several research studies some demonstration cases, adoption these techniques built environment remains somewhat limited. One main reasons is that novel control algorithms continue to be evaluated individually. hampers identification best practices deploy widely building sector. paper implements compares variations model predictive (MPC), reinforcement learning (RL), reinforced (RL-MPC) same problem for management. Particularly, controllers’ hyperparameters like step, prediction horizon, state-action spaces, algorithm, or network architecture value function are investigated. The optimization testing (BOPTEST) framework used as simulation benchmark carry out study it offers standardized scenarios. reveal that, contrary what stated previous literature, model-free RL approaches poorly perform when tested environments with realistic system dynamics. a available simulation-based implemented, MPC outperforms an equivalent formulation problem. performance gap between both reduces using RL-MPC algorithm merges elements from families methods.