A Comparison of Classical and Reinforcement Learning-based Tuning Techniques for PI controllers

Abstract

This study compares two tuning techniques for Proportional-Integral (PI) controllers. The first strategy uses the Pole-Zero Cancellation method, a well-established technique in the field of dynamic systems control. The second strategy introduces an innovative approach by using a reinforcement learning-based technique to adaptively tune a PI controller. To compare these tuning methodologies, a Heating, Ventilation and Air Conditioning (HVAC) control system was selected as a case study to guarantee users’ thermal comfort. In both cases, the HVAC process was modelled as a first-order system without time delay. In addition, the performance of the proposed controllers analysed by evaluating temperature set-point tracking and disturbance rejection caused by the occupancy of people in the room. The results demonstrate that classical methods are efficient and quick to implement, while the use of RL also enables optimization of energy consumption and reduction of operating costs.