Domain Generalization-Based Evolutionary Algorithm for Hybrid Flowshop Dynamic Scheduling Problem

Hybrid flowshop dynamic scheduling problem (HFDSP) represents an important area of scheduling problems. In real life, machine status may change in the dynamic environment, impacting the actual processing time of jobs. Considering dynamic machine status, this paper introduces an HFDSP model including several discrete processing stages with parallel machines and a continuous packing stage with a buffer area. Before processing different batches of jobs, some machines may change into an abnormal status, in which they can still process jobs but cost more consumption. The objective of this paper is to optimize the makespan and total energy consumption. To solve this problem, we use domain generalization to extend the evolutionary algorithm (EA) and propose the DICA-EA based on domain-invariant component analysis (DICA). DICA-EA stores the Pareto set (PS) from historical environments and uses DICA to derive a feature transformation matrix that can minimize the distributional variance in the transformed space between domains. When a new environment is detected, the knowledge acquired by DICA is used to predict a high-quality population, thereby expediting the algorithm's search process. Finally, we compare DICA-EA with other DMOEAs through different dynamic scheduling instances to verify its effectiveness.