A data-driven digital twin approach to optimize continuous production environment with deep reinforcement learning

Date

2023-9-6

Author

Sivri , Ali Haydar

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

296
views

152
downloads

Cite This

Today, the world mainly strives to minimize carbon emissions and maximize efficiency in energy production to lower energy costs and greenhouse effect. Therefore, optimizing hydrogen production processes becomes a crucial objective in organizations, especially for manufacturers that consume significant energy in production. Organizations should operate their production processes under the most optimum conditions to ensure their survival and attain competitive advantage. However, they face substantial challenges in optimizing their energy utilizations in their production processes. Those challenges consequently necessitate utilizing a data-driven approach and dynamic framework for managing and optimizing their process workflows. This thesis study consists of two main phases. The first phase focuses on developing a data-driven simulation environment (DDSE) of the hydrogen unit for Deep Reinforcement Learning (DRL) that can adapt to the complexity and variability of processes by utilizing the machine learning approaches. The DDSE comprises the intricacies of the hydrogen production process, including critical parameters such as temperature, pressure, and catalyst variables. At the second phase, advanced DRL algorithms are developed to support continuous learning and making decisions based on long-term outcomes rather than immediate rewards. The DRL model effectively interacts with the DDSE in adjusting controllable production parameters to optimize the hydrogen production yield. The proposed model's validation and efficacy are demonstrated through extensive simulations and real-world data. The test results reveal that we can improve energy efficiency by approximately 4%.

Subject Keywords

Machine learning, Deep reinforcement learning, Data driven simulation, Hydrogen production, Optimization

URI

https://hdl.handle.net/11511/105436

Collections

Graduate School of Informatics, Thesis