An adaptive genetic algorithm with parameter control based on dynamic fitness landscape analysis for operation sequencing in CAPP

In Industry 4.0, customised product volatility makes operation sequencing (OS) in CAPP frequently demanded. Solving the NP-hard OS problems requires an adaptive genetic algorithm (AGA) with robust performance and no tedious tuning, which is currently absent. Considering dynamic fitness landscape analysis (DFLA) enables measurement of GA’s parameters during search, we design an AGA with parameter control incorporating the feedback information from population fitness and DFLA. For the AGA, the direct encoding of feasible operation sequence and the precedence-constraint satisfaction method make it search in feasible solution space. In the AGA, the mutation rate is updated online by the population diversity, and four crossovers are online switched in a greedy fashion via the judgment combining population diversity and evolvability from DFLA. To improve efficiency and exploitation ability, the path-relinking local search is adaptively applied by the maximum unchanged generation of an individual’s fitness. The AGA is compared with state-of-the-art metaheuristic algorithms (MAs), including particle swarm optimisation, harmony search, two recent GAs and a self-adaptive GA, to verify its effectiveness and merits against five OS cases and six randomly generated OS problems. The comparative results show that the AGA outperforms existing MAs for solution quality, and illustrate its robust global search ability.