Hybrid relay test auto-tuning with multi-objective K-Bandit: a FOPDT case

Abstract

The increasing complexity of industrial systems demands the use of more sophisticated control mechanisms capable of adapting to dynamic, nonlinear environments. This paper addresses the problem of tuning Proportional-Integral (PI) controllers in such conditions by integrating multi-objective reinforcement learning with traditional auto-tuning techniques. We propose a novel multi-objective version of the K-Bandit algorithm for PI controller tuning, which optimizes two control objectives simultaneously. By combining this with a relay test-based auto-tuning method, our approach significantly reduces the number of tuning episodes required, enhancing adaptability. The development and validation of this method are demonstrated through simulation, highlighting its effectiveness in maintaining system stability and achieving control performance.