Two-echelon location-routing optimization with time windows based on customer clustering

This paper develops a three-step customer clustering based approach to solve two-echelon location routing problems with time windows. A bi-objective model minimizing costs and maximizing customer satisfaction is formulated along with an innovative measurement function to rank optimal solutions. The proposed methodology is a knowledge-based approach which considers customers locations and purchase behaviors, discovers similar characteristics among them through clustering, and applies exponential smoothing method to forecast periodic customers demands. We introduce a Modified Non-dominated Sorting Genetic Algorithm-II (M-NSGA-II) to simultaneously locate logistics facilities, allocate customers, and optimize the vehicle routing network. Different from many existing version of NSGA-II, our algorithm applies partial-mapped crossover as genetic operator, instead of simulated binary crossover, in order to properly handle chromosomes. The initial population is generated through a nodes' scanning algorithm which eliminates sub-tours. Finally, to demonstrate the applicability of our mathematical model and approach, we conduct two empirical studies on generated benchmarks and the distribution network of a company in Chongqing city, China. Further comparative analyses with multi-objective genetic algorithm (MOGA) and multi-objective particle swarm optimization (MOPSO) algorithm indicate that M-NSGA-II performs better in terms of solution quality and computation time. Results also support that: (1) the formation of clusters containing highly similar customers improves service reliability, and favors a productive customer relationship management; (2) considering product preference contributes to maximizing customer satisfaction degree and the effective control of inventories at each distribution center; (3) clustering, instead of helping to improve services, proves detrimental when too many groups are formed. Thus, decision makers need to conduct series of simulations to observe appropriate clustering scenarios.