Community logistics : a new and efficient strategy for solving e-commerce last mile delivery

Abstract

Last mile delivery refers to design delivery services from a public delivery node to customers’ specified destinations, which is one of the most money-consuming parts of the whole logistic chain. The rapidly increasing business-to-customer (B2C) e-commerce orders significantly complicate the fulfillment of last mile delivery. Logistics service providers are facing challenges in handling the numerous and fragmented B2C e-commerce orders. Hence, an efficient delivery strategy is strongly required to manage the decision making of e-commerce last mile delivery. This thesis develops a new delivery strategy for e-commerce last mile delivery, namely community logistics (CL). Its solution format is “delivery community”, which is a compact subarea served by a vehicle within a time slot. In a delivery community, the subarea and time slot are dynamically interchanged according to the real-time information. This solution format can simplify the decision process and guarantee high delivery solution quality. The CL is described and evaluated by the following four studies: The study I presents a formal CL definition, including its application context, application procedure, criteria for decision making and solution format. Based on this new delivery strategy, two basic policies, namely spatial and temporal policy, are proposed to generate delivery communities. These two CL-based policies are compared with a routing policy given several real-life delivery instances. Numerical results demonstrate the CL’s advantages over conventional route-based delivery strategies in solution time, vehicle traveling distance and route compactness. The study II applies the CL in a scenario of e-commerce last mile delivery, in which all orders should be delivered to smart lockers. The delivery time requirement here is regarded as “delivery as soon as possible”. With the CL, the original delivery problem is transformed into a dynamic community partitioning problem (DCPP). The DCPP is formulated as a Markov decision process (MDP) model, and a solution framework based on myopic cost function approximation is proposed to solve the resulting model. Numerical results reveal the influences from different order features and the significant benefits of the proposed solution framework. The study III adopts the CL in another scenario of e-commerce last mile delivery, where each customer requires to receive the order at home in a designated time window. The original delivery problem is transformed into a dynamic community partitioning problem with time window (DCPPTW). A formal MDP model is presented, and a heuristic solution framework based on policy function approximation (PFA) is developed to solve the MDP model. Numerical experiments show the merits of the proposed solution framework compared with conventional solution methods. The study IV presents a synchronization-based online batching strategy for last mile delivery warehouses under the CL. A new problem is investigated, namely dynamic order batching problem with delivery decisions (DOBP-DD), which aims to minimize the time gap between receiving delivery decisions with actual vehicle departures. A MDP model is presented and solved by the PFA, in which a convolution neural network is integrated to forecast future delivery decisions. Numerical experiments exhibit that the proposed strategy can reduce the time gap and alleviate orders’ overstock.