出版社：Institute for Operations Research and the Management Sciences (INFORMS), Applied Probability Society
摘要：In this paper we consider approximate dynamic programmingmethods for ambulance redeployment. We first demonstrate throughsimple examples how typical value function fitting techniques, suchas approximate policy iteration and linear programming, may not beable to locate a high-quality policy even when the value function ap-proximation architecture is rich enough to provide the optimal policy.To make up for this potential shortcoming, we show how to use directsearch methods to tune the parameters in a value function approxima-tion architecture so as to obtain high-quality policies. Direct searchis computationally intensive. We therefore use a post-decision statedynamic programming formulation of ambulance redeployment that,together with direct search, requires far less computation with nonoticeable performance loss. We provide further theoretical supportfor the post-decision state formulation of the ambulance-deploymentproblem by showing that this formulation can be obtained through alimiting argument on the original dynamic programming formulation
