Comparative Study of Deep Reinforcement Learning Algorithm for Optimization of Hydrodynamic Characteristics in Multiphase Reactors

Suchita Walke, Jagdish W. Bakal

2025

Abstract

The proper design of multiphase reactors to optimize their hydrodynamic properties is still one of the most important challenges in chemical and process engineering due to the complexity of fluid interactions and dynamic behaviours of these systems. This challenge is addressed in this study, which presents a comprehensive comparative study of state-of the-art Deep Reinforcement Learning (DRL) algorithms, namely Deep Q-Networks (DQN), Proximal Policy Optimization (PPO), and Soft Actor-Critic (SAC). Diligently designed virtual simulation environment mimics the complex functionalities of multiphase reactors for an accurate assessment of gas holdup, liquid velocity profiles, and bubble size distribution, which are significant parameters in terms of reactor performance. This involves developing a custom reward function, that weights these against energy consumption, to allow the reactor to perform ideally. Experimental results indicate that SAC converges faster to solutions, as well as is more accurate in the optimization of hydrodynamic parameters and energy casting. DQN is limited by its discrete action space preventing it from being applied to continue reactors and PPO has a relatively moderate performance. This approach not only highlights the promise of DRL for optimizing reactor dynamics, but also offers tangible guidance on algorithm selection for practical engineering implementations. The results open doors to implementing state-of-the-art DRL algorithms in industrial environments, greatly improving energy-efficient management of industrial systems. Future studies will be targeted at implementing in the real world and hybrid DRL-CFD frameworks for multiphase reactor systems.

Paper Citation

in Harvard Style

Walke S. and Bakal J. (2025). Comparative Study of Deep Reinforcement Learning Algorithm for Optimization of Hydrodynamic Characteristics in Multiphase Reactors. In Proceedings of the 1st International Conference on Research and Development in Information, Communication, and Computing Technologies - ICRDICCT`25; ISBN 978-989-758-777-1, SciTePress, pages 362-368. DOI: 10.5220/0013930000004919

in Bibtex Style

@conference{icrdicct`2525,
author={Suchita Walke and Jagdish Bakal},
title={Comparative Study of Deep Reinforcement Learning Algorithm for Optimization of Hydrodynamic Characteristics in Multiphase Reactors},
booktitle={Proceedings of the 1st International Conference on Research and Development in Information, Communication, and Computing Technologies - ICRDICCT`25},
year={2025},
pages={362-368},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0013930000004919},
isbn={978-989-758-777-1},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 1st International Conference on Research and Development in Information, Communication, and Computing Technologies - ICRDICCT`25
TI - Comparative Study of Deep Reinforcement Learning Algorithm for Optimization of Hydrodynamic Characteristics in Multiphase Reactors
SN - 978-989-758-777-1
AU - Walke S.
AU - Bakal J.
PY - 2025
SP - 362
EP - 368
DO - 10.5220/0013930000004919
PB - SciTePress