Graph-To-Sequence Approach for Job Shop Scheduling Problem
摘要
Reinforcement learning has recently been frequently researched for solving the Job Shop Scheduling Problem (JSSP). Typically, JSSP is approached through a dynamic scheduling perspective, applying the Markov Decision Process. This study breaks the mold and provides a static perspective for solving JSSPs. To this end, a graph-to-sequence (graph2seq) architecture incorporating graph neural networks, multi-head attention, and pointer networks is proposed. Experiments are conducted on JSSP benchmarking data to validate the proposed method.