Optimal scheduling of low-carbon ethylene production system based on deep reinforcement learning: Considering flexible operation under furnace coking and renewable energy fluctuations

The integration of renewable energy and carbon capture technologies plays a vital role in energy saving and emission reduction of the ethylene industry. However, the joint deployment of various low-carbon technologies introduces significant challenges for the coordinated operation of the overall system. To address this, this study proposes a deep reinforcement learning-based scheduling approach for the low-carbon ethylene production system. High-fidelity models of the cracking furnace and post-combustion carbon capture process are first developed using physics-informed neural network, providing a physically consistent simulation environment for the learning process. Building on this, a coordinated scheduling framework is formulated that considers flexible operation under furnace coking and renewable energy fluctuations. The optimization problem is modeled as a Markov decision process and solved using the proximal policy optimization algorithm. Case study based on a 300 kt/a ethylene plant demonstrates that the proposed method can effectively coordinate the operation of multiple processes, maximizing economic profits while ensuring energy supply and demand balance. Furthermore, an in-depth discussion is conducted to understand the impacts of coking, flexible operation and solution methods on scheduling performance, offering broader insights into the operation strategies of the integrated low-carbon ethylene production system.