Deep reinforcement learning with shallow controllers: An experimental application to PID tuning. (April 2022)

• Record Type: Journal Article
• Authors: Lawrence, Nathan P.; Forbes, Michael G.; Loewen, Philip D.; McClement, Daniel G.; Backström, Johan U.; Gopaluni, R. Bhushan
• Is Part Of: Control engineering practice. Volume 121(2022)
• Publication Date: 2022-04
• Subjects: Reinforcement learning -- Deep learning -- PID control -- Process control -- Process systems engineering; Automatic control -- Periodicals; 629.89
• DOI: 10.1016/j.conengprac.2021.105046
• Languages: English
• ISSNs: 0967-0661

Abstract

Abstract: Deep reinforcement learning (RL) is an optimization-driven framework for producing control strategies for general dynamical systems without explicit reliance on process models. Good results have been reported in simulation. Here we demonstrate the challenges in implementing a state of the art deep RL algorithm on a real physical system. Aspects include the interplay between software and existing hardware; experiment design and sample efficiency; training subject to input constraints; and interpretability of the algorithm and control law. At the core of our approach is the use of a PID controller as the trainable RL policy. In addition to its simplicity, this approach has several appealing features: No additional hardware needs to be added to the control system, since a PID controller can easily be implemented through a standard programmable logic controller; the control law can easily be initialized in a "safe" region of the parameter space; and the final product—a well-tuned PID controller—has a form that practitioners can reason about and deploy with confidence. Graphical abstract: Highlights: Reinforcement learning (RL) is used to tune a real-world PID controller. The RL policy is a PID controller, for compatibility with many current systems. Good tuning is achieved in roughly 40 min of training time. Full implementation details and thorough lab results are presented. A multi-criterion scorecard compares RL with several known auto-tuning methods.