Reinforcement Learning for Mixing Loop Control with Flow Variable Eligibility Trace

Mixing Loops are often used for proper pressurization and temperature control in building thermal systems. Optimal control of the mixing loop maximizes comfort while minimizing cost. To ensure optimal control for mixing loops in a wide range of different buildings with different load conditions, a self learning controller is here proposed. The controller uses Reinforcement Learning with flow variable eligibility trace. The controller is shown to improve performance of the mixing loop control compared to state of the art reinforcement learning and industrial grade controllers. The controller is tested on a hardware in the loop setup for rapid testing of mixing loop control used in building heating.