DDQN-AJCM Algorithm for Flexible Job Shops Problem with Random Job Arrivals

In real-world production systems, unpredictable dynamic events frequently occur, giving rise to the Dynamic Flexible Job Shop Scheduling Problem (DFJSP). As jobs arrive asynchronously and compete for multiple machines, complex interdependencies emerge among them. To address this issue, this study proposes a Double Deep Q-Network (DDQN) algorithm enhanced with an attention mechanism. This method utilizes an Attention-based Job Competition Model (AJCM) to capture the degree of influence between jobs and selectively extracts job information to facilitate representation learning. Numerical experiments conducted on randomly generated production configurations demonstrate that the DDQN-AJCM outperforms other methods in overall performance and exhibits robust generalization capabilities across production environments of varying scales.