One innovative solution to the last-mile delivery problem is the self-service locker system. Motivated by a real case in Singapore, we consider a POP-Locker Alliance who operates a set of POP-stations and wishes to improve the last-mile delivery by opening new locker facilities. We propose a quantitative approach to determine the optimal locker location with the objective to maximize the overall service provided by the alliance. Customer’s choices regarding the use of facilities are explicitly considered. They are predicted by a multinomial logit model. We then formulate the location problem as a multi-ratio linear-fractional 0–1 program and provide two solution approaches. The first one is to reformulate the original problem as a mixed-integer linear program, which is further strengthened using conditional McCormick inequalities. This approach is an exact method, developed for small-scale problems. For large-scale problems, we propose an alternating algorithm, i.e., Quadratic Transform with Linear Alternating (QT-LA). The numerical experiment indicates that QT-LA is an efficient approach that yields high-quality solutions. Finally, we conducted a case study. The results highlighted the importance of considering the customers’ choices. Under different parameter values of the multinomial logit model, the decisions could be completely different. Therefore, the parameter value should be carefully estimated in advance.

自助式储物柜系统是解决最后一英里交付问题的一种创新解决方案。受新加坡一宗真实案例的启发，我们考虑了一个POP储物柜联盟，该联盟经营一套POP站，并希望通过开设新的储物柜设施来改善最后一英里的运送。我们提出一种定量方法来确定最佳的储物柜位置，目的是使联盟提供的整体服务最大化。明确考虑了客户在设施使用方面的选择。它们由多项式logit模型预测。然后，我们将位置问题公式化为多比例线性分数0-1程序，并提供两种解决方案。第一个是将原始问题重新表述为混合整数线性程序，并使用条件McCormick不等式进一步加强了该程序。这种方法是针对小规模问题开发的一种精确方法。对于大规模问题，我们提出了一种交替算法，即具有线性交替的二次变换（QT-LA）。数值实验表明，QT-LA是产生高质量解决方案的有效方法。最后，我们进行了一个案例研究。结果强调了考虑客户选择的重要性。在多项式logit模型的不同参数值下，决策可能完全不同。因此，应事先仔细估计参数值。数值实验表明，QT-LA是产生高质量解决方案的有效方法。最后，我们进行了一个案例研究。结果突出了考虑客户选择的重要性。在多项式logit模型的不同参数值下，决策可能完全不同。因此，应事先仔细估计参数值。数值实验表明，QT-LA是产生高质量解决方案的有效方法。最后，我们进行了一个案例研究。结果突出了考虑客户选择的重要性。在多项式logit模型的不同参数值下，决策可能完全不同。因此，应事先仔细估计参数值。