Distributional robustness and lateral transshipment for disaster relief logistics planning under demand ambiguity

This paper considers facility location, inventory pre-positioning and vehicle routing as strategic and operational decisions corresponding to preparedness and response phases in disaster relief logistics planning. For balancing surpluses and shortages, an effective lateral transshipment strategy is proposed to evenly distribute the relief resources between warehouses after the disaster occurs. To handle ambiguity in the probability distribution of demand, we develop a risk-averse two-stage distributionally robust optimization (DRO) model for the disaster relief logistics planning problem, which specifies the worst-case mean-conditional value-at-risk (CVaR) as a risk measure. For computationally tractability, we transform the robust counterpart into its equivalent linear mixed-integer programming model under the discrepancy-based ambiguity set centered at the nominal (empirical) distributions on the observed demand from the historical data. We verify the effectiveness of the proposed DRO model and the value of lateral transshipment strategy by an illustrative small-scale example. The numerical results show that the proposed DRO model has advantage on avoiding over-conservative solutions compared to the classic robust optimization model. We also illustrate the applicability of the proposed DRO model by a real-world case study of hurricanes in the southeastern United States. The computational results demonstrate that the proposed DRO model has superior out-of-sample performance and can mitigate the adverse effects of Optimizers' Curse compared with the traditional stochastic programming model.