Answers are grounded in this paper and cite their sources.
Despite the effort of converting batch processes into continuous processes through process intensification, batch processes are still the dominant mode of production in many industrial sectors such as food, pharmaceuticals and specialty chemicals. Batch processes pose challenges in their control and optimization due to a number of factors such as nonlinearity, unsteady state operation, and batch-to-batch variations. Recent developments in reinforcement learning (RL) have provided new tools for the optimization and control of batch processes. This paper provides a review of recent applications of RL in batch process optimization and control. The review starts with conventional approaches for batch process control and optimization. RL is then presented as a complementary data-driven approach that formulates batch process control as a sequential decision-making problem. Different RL methods are discussed, along with their applications in batch processes, including recipe optimization, batch-to-batch learning, trajectory tracking, constraint handling, and multi-agent reinforcement learning for plant-wide coordination. The limitations of previous review papers on RL for process control are also examined, highlighting how the present review addresses existing gaps. In addition, key knowledge gaps and barriers to the industrial adoption of RL are explored. Finally, conventional and RL-based methods are compared across different levels of the batch control hierarchy. Based on this comparison and the identified knowledge gaps, research directions are proposed and linked to specific studies and unresolved research gaps.
No takes yet. Share an insight, caveat, or question.
Drew et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: