The freight allocation problem with lane cost balancing constraint

We consider a problem faced by a buying office for one of the largest retail distributors in the world. The buying office plans the distribution of goods from Asia to various destinations across Europe. The goods are transported along shipping lanes by shipping companies, many of which have collaborated to form strategic alliances; each lane must be serviced by a minimum number of companies belonging to a minimum number of alliances. The task involves purchasing freight capacity from shipping companies for each lane based on projected demand, and subject to minimum quantity requirements for each selected shipping company, such that the total transportation cost is minimized. In addition, the allocation must not assign an overly high proportion of freight to the more expensive shipping companies servicing any particular lane, which we call the lane cost balancing constraint.This study is the first to consider the lane cost balancing constraint in the context of freight allocation. We formulate the freight allocation problem with this lane cost balancing constraint as a mixed integer programming model, and show that even finding a feasible solution to this problem is computationally intractable. Hence, in order to produce high-quality solutions in practice, we devised a meta-heuristic approach based on tabu search. Experiments show that our approach significantly outperforms the branch-and-cut approach of CPLEX 11.0 when the problem increases to practical size and the lane cost balancing constraint is tight. Our approach was developed into an application that is currently employed by decision-makers at the buying office in question.

► Introduced a new “lane cost balancing constraint” that emphasizes allocation fairness. ► Showed that finding a feasible solution to the resultant new problem is NP-hard in the strong sense. ► Designed a customized tabu search approach to solve problem instances of practical size. ► Generated benchmark instances for the problem based on typical values encountered in practice. ► Demonstrated our approach is superior to CPLEX for large instances.