Learning from Random Solutions: Data-Mining-Guided Heuristic Search for Permutation Flow Shop Scheduling

The permutation flow shop scheduling problem (PFSP) is a fundamental scheduling problem for which heuristic methods are widely used due to the vast size of its solution space. This study investigates whether solution populations generated entirely from random permutations contain structural information capable of guiding heuristic search. Independent random pools were generated for each problem instance, and pairwise precedence (P) and relative-position region (R) information was extracted from the Elite and Poor groups defined according to their objective function values. The Elite-only approach captures structural concentration among better-performing random solutions, whereas the Contrast approach captures frequency differences between Elite and Poor solutions. The extracted information was integrated into the NEH, NEH-TBKK2, KK2-style-NEH, FRB4-p1, and INEH-inspired methods while preserving Cmax as the primary decision criterion in all cases. In the constructive methods, the information is used only to distinguish among insertion alternatives tied in terms of Cmax, whereas in the reinsertion-based methods, it also guides the selection of search candidates. The proposed approach was evaluated on 140 benchmark instances, comprising 110 Taillard and 30 VFR instances. The results show that structural information extracted from random populations provides stronger and more consistent improvements, particularly in reinsertion-based methods, and that the algorithmic decision point at which the information is incorporated is as critical as the information content itself.