Emergency logistics network optimization with time window assignment

•A collaborative emergency logistics network with TWA is studied.•A tri-objective optimization model is proposed for emergency logistics delivery.•A multi-objective ALNS with split algorithm is devised to solve the model.•A case study is explored to show the applicability of proposed methods.•Comparative analyses with different TWA and sharing strategies are tested. During natural disasters or accidents, an emergency logistics network aims to ensure the distribution of relief supplies to victims in time and efficiently. When the coronavirus disease 2019 (COVID-19) emerged, the government closed the outbreak areas to control the risk of transmission. The closed areas were divided into high-risk and middle-/low-risk areas, and travel restrictions were enforced in the different risk areas. The distribution of daily essential supplies to residents in the closed areas became a major challenge for the government. This study introduces a new variant of the vehicle routing problem with travel restrictions in closed areas called the two-echelon emergency vehicle routing problem with time window assignment (2E-EVRPTWA). 2E-EVRPTWA involves transporting goods from distribution centers (DCs) to satellites in high-risk areas in the first echelon and delivering goods from DCs or satellites to customers in the second echelon. Vehicle sharing and time window assignment (TWA) strategies are applied to optimize the transportation resource configuration and improve the operational efficiency of the emergency logistics network. A tri-objective mathematical model for 2E-EVRPTWA is also constructed to minimize the total operating cost, total delivery time, and number of vehicles. A multi-objective adaptive large neighborhood search with split algorithm (MOALNS-SA) is proposed to obtain the Pareto optimal solution for 2E-EVRPTWA. The split algorithm (SA) calculates the objective values associated with each solution and assigns multiple trips to shared vehicles. A non-dominated sorting strategy is used to retain the optimal labels obtained with the SA algorithm and evaluate the quality of the multi-objective solution. The TWA strategy embedded in MOALNS-SA assigns appropriate candidate time windows to customers. The proposed MOALNS-SA produces results that are comparable with the CPLEX solver and those of the self-learning non-dominated sorting genetic algorithm-II, multi-objective ant colony algorithm, and multi-objective particle swarm optimization algorithm for 2E-EVRPTWA. A r