A Multi-Stage Stochastic Facility Routing Model for Humanitarian Logistics Planning
MetadataShow full item record
This thesis presents a humanitarian logistics decision model to be used in the event of a disaster. The operations under consideration span from opening of local distribution facilities and initial allocation of supplies, to last mile distribution of aid. An introduction of the field of disaster management is given, which forms the basis for the following description of the disaster response problem faced by the decision maker. Two mathematical models are developed aiming to enable efficient decision making. The mathematical models solve the disaster response problem and seek to maximize the utility of distribution of aid amongst beneficiaries. Utility is expressed in terms of amount of satisfied demand and cost-effectiveness. The main mathematical model is formulated as a multi-stage mixed-integer stochastic model to account for the difficulty in predicting the outcome of a disaster. The model will be applied to earthquakes in particular for reasons of concreteness. Accessibility of new information implicates initiation of distinct operations in the humanitarian supply chain, be it facility location and supply allocation, or last mile distribution planning and execution. The realized level of demand, in addition to the transportation resources available to the decision maker for execution of last mile aid distribution, are parameters treated as random due to uncertainty. Complete information regarding these variables is revealed in stage two. As a direct consequence of treating demand as an uncertain parameter, marginal utility will also be subject to stochasticity. Also, the state of the distribution network is treated as a random parameter due to uncertainty arising from the vulnerability of the local infrastructure. Reception of complete information concerning the state of the infrastructure indicates transition from stage~2 to stage~3. The mathematical models are applied to an illustrative example to demonstrate their application as decision-making tools in practice. An assessment of the applicability and validity of the stochastic program is made, based on several test instances generated by the authors. Results show that instances of considerable size are challenging to solve due to the complexity of the stochastic programming model. Still, optimal solutions may be found within a reasonable time frame. Moreover, findings prove the value of the stochastic programming model to be significant as compared with an deterministic expected value approach.