Flexible flow shop scheduling problems with sequence-dependent setup times are among the most challenging scheduling problems. To address this, we proposed a Deep Reinforcement Learning (DRL) framework based on a Multi-Layer Percep- tron (MLP). The model utilizes an instance-size-independent state representation and a discrete action space constituted by Priority Dispatching Rules (PDRs), en- abling training and inference on instances with varying numbers of jobs and ma- chines. We also implemented a multi-instance training strategy to reduce overfit- ting and improve generalization to unseen instances. The experiments show that the DRL model with in-sample inference outperforms the individual PDRs for minimiz- ing makespan and total tardiness. The multi-instance training enhances generaliza- tion, especially when the training and inference instances share structural objective- relevant characteristics. The findings suggest that using multiple representative pro- duction scenarios for training can generate an agent capable of scheduling a range of problems.     «

Flexible flow shop scheduling problems with sequence-dependent setup times are among the most challenging scheduling problems. To address this, we proposed a Deep Reinforcement Learning (DRL) framework based on a Multi-Layer Percep- tron (MLP). The model utilizes an instance-size-independent state representation and a discrete action space constituted by Priority Dispatching Rules (PDRs), en- abling training and inference on instances with varying numbers of jobs and ma- chines. We also im...     »