Approximate dynamic programming for network recovery problems with stochastic demand

Immediately after a disruption, in order to minimize the negative impact inflicted on society, its imperative re-establish interdicted critical services enabled by infrastructure networks. In this paper, we study stochastic network recovery problem that tackles planning of restoration activities (considering limited resources) links so pre-disruption service flows can be re-established as quickly possible. As an illustrative case study, consider disaster scenario road obstructs flow relief-aid commodities and search-and-rescue teams between providing facilities locations need these services. many realistic applications, amount demand for stochastic. First, present Markov decision process (MDP) formulation (SRNRP), then propose approximate dynamic programming (ADP) approach heuristically solve SRNRP. We develop basis functions capture important complex interactions used cost-to-go values MDP states. conduct computational experiments set small-scale randomly generated instances demonstrate ADP provides near-optimal results regardless distribution topology. practical suitable solving real world sized instances, framework where first model derive policy spatially aggregated large scale instance. Next, show performance through testing disaggregated network. Moreover, provide managerial insights assessing importance each function contributing policies. test based Boston observe that, urgency re-establishing increases or resources become more scarce, information gained from characteristics short-term decisions should main driving factors The all strongly evidence significance utilizing inherent attributes generate sets models yield high-quality