Evacuate or Not? A POMDP Model of the Decision Making of Individuals in Hurricane Evacuation Zones
Evacuate or Not? A POMDP Model of the Decision Making of Individuals in Hurricane Evacuation Zones
复制标题
DOI：
--
复制
发表时间：
2019
期刊：
影响因子：
--
通讯作者：
A. Sankar;Prashant Doshi;Adam Goodie
中科院分区：
文献类型：
--
作者：
A. Sankar;Prashant Doshi;Adam Goodie
在图9（见下一页）中，我们展示了DID的两个完整时间片。以“HN”为前缀的节点为隐变量，以“O”为前缀的节点为观测变量。注意，DID反映了POMDP的结构——前一个状态变量和动作影响下一个状态变量，前一个动作和当前状态变量影响观测值。一个时间片有6个与时间无关的人口统计状态变量（跨时间片共享），7个与时间相关的经验变量，9个观察变量，3个隐藏节点和评估风险节点，此外还有决策和效用节点。我们在四个时间步骤中展开这个DID。除了有限的数据量所带来的明显约束之外，我们还讨论了模型的一个限制。我们的调查没有关于参与者决定在飓风登陆之前或之后撤离的时间表的问题。例如，一个参与者可能在疏散命令下达后立即撤离，但另一个参与者可能等到情况变得更糟后才撤离。我们的调查并没有区分这两种情况，尽管我们的决策模型显示了一个时间维度。
In Fig. 9 (see next page), we show two complete time-slices of the DID. Nodes prefixed with ‘HN’ are the hidden variables while those prefix with ‘O’ are the observation variables. Observe that the DID reflects the structure of a POMDP – previous state variables and action influence next state variables, and previous action and current state variables influence the observations. A time-slice has 6 time-independent demographic state variables (shared across time-slices), 7 time-dependent experiential variables, 9 observation variables, 3 hidden nodes and the assessed risk node in addition to the decision and utility nodes. We unroll this DID over four time-steps. We discuss one limitation of our model in addition to the obvious constraint imposed by a limited amount of data. Our survey did not have questions about the timeline of when the participant decided to evacuate prior or post hurricane making landfall. For example, one participant might have evacuated immediately after an evacuation order but another participant might have waited until the situation became worse and then evacuated. These two cases are not differentiated by our survey, though our decision-making model exhibits a temporal dimension.