Developing Purely Data-Driven Multi-Mode Process Controllers Using Inverse Reinforcement Learning

In recent years, process control researchers have been paying close attention to Deep Reinforcement Learning (DRL). DRL offers the potential for model-free controller design, but it is challenging to achieve satisfactory outcomes without accurate simulation models and well-designed reward functions, particularly in multi-mode processes. To address this issue, this paper presents a novel approach that combines inverse RL (IRL) and multi-task learning to provide a purely data-driven solution for multi-mode control design, allowing for transfer learning and adaptation in different operating modes. The effectiveness of this novel approach is demonstrated through a CSTR continuous control case using multi-mode historical closed-loop data. The proposed method offers a promising solution to the challenges of designing controllers for multi-mode processes.