Real-time optimization (RTO) is crucial for modern industrial process operations, enabling quick adaptation to dynamic conditions and delivering substantial economic and operational benefits. However, implementing RTO faces significant challenges, such as the high cost of developing accurate models and the risk of model mismatch in complex, dynamic systems. Traditional model-based approaches often struggle to adapt effectively, leading to suboptimal or unsafe decisions.
This workshop provides an in-depth tutorial on data-driven RTO, which leverages real-time data to overcome the limitations of model-based optimization. Targeted at PhD students, researchers, and practitioners, it explores how data-driven methods reduce reliance on high-fidelity models, addressing modeling costs and mismatches to facilitate effective decision-making in complex systems.
The program includes an introduction to data-driven RTO, a detailed exploration of constrained Bayesian Optimization (BO) for real-time control, and methods to manage constraints without probabilistic models (D. Krishnamoorthy). It also covers feedback-based approaches to optimize partially unknown systems and introduces self-learning capabilities for industrial plants (M. Zagorowska). Topics include scalability, safety-critical constraints, and incorporating domain knowledge through advanced frameworks such as ensemble uncertainty estimation and bilevel programming (M. Mercangoz). The workshop will also include how meta-learning leverages prior experience to enhance real-time learning and adaptation across optimization, modeling, and control tasks (V. Breschi). Additionally, the workshop addresses measurement-based gradient estimation (L. F. Bernardinho), feedback optimization for interconnected systems, and multi-agent frameworks for distributed optimization (D. Krishnamoorthy). Participants will gain insights into cutting-edge techniques for adaptive, scalable, and decentralized process optimization, equipping them with practical knowledge for real-world applications.