Probabilistic approaches to scheduling reserve selection
Probabilistic approaches to scheduling reserve selection
复制标题
DOI：
10.1016/j.biocon.2004.07.015
复制
发表时间：
2005-03
影响因子：
5.9
通讯作者：
M. Drechsler
中科院分区：
文献类型：
--
作者：
M. Drechsler
大多数现有的储备选择算法是静态的，因为它们假设一个储备网络是设计的，而补丁是由决策者在单个时间点上选择的。然而，在现实中，选择过程往往是动态的，补丁被逐个或分成几组选择，因为例如在过程开始时没有足够的资金将所有补丁置于保护之下。由于互补性原理，特定补丁的价值取决于网络中其他补丁的存在--包括那些尚未被选择的补丁--因此寻找最优的动态选择策略是棘手的。由于未受保护的补丁可能会丢失，例如通过开发，因此选择特定补丁的长期价值是不确定的。现有的动态选择算法要么目光短浅，只考虑那些已经受到保护的补丁，完全忽略了未来的不确定性，要么基于随机动态规划，提供了考虑不确定性的最优策略，但数值太复杂，无法用于实际选择问题。在这篇文章中，一个‘前瞻性’的选择策略以及一些变种是用概率论发展起来的。对于大量的选择问题，对不同的策略进行了比较。所有变种的表现都优于近视策略，表现接近最优策略。然而，所有策略的表现，包括最优策略和短视策略，都不是戏剧性的。
Most existing reserve selection algorithms are static in that they assume that a reserve network is designed and patches are selected by decision-makers at a single point in time. In reality, however, selection processes are often dynamic and patches are selected one by one or in several groups because for example there are insufficient funds at the beginning of the process to put all the patches under protection. Finding an optimal dynamic selection strategy is tricky since due to the complementarity principle the value of a particular patch depends on the presence of other patches in the network – including those that have not yet been selected. As unprotected patches may be lost, e.g., through development, the long-term value of selecting a particular patch is uncertain. Existing dynamic selection algorithms are either ‘myopic’ and consider only those patches that have already been protected, totally ignoring future uncertainty, or they are based on stochastic dynamic programming, which delivers the optimal strategy taking uncertainty into account but is numerically too complex to be employed in actual selection problems. In this paper, a ‘foresighted’ selection strategy as well as a number of variants are developed using probability theory. The different strategies are compared for a large number of selection problems. All variants outperform the myopic strategy and perform close to the optimal strategy. However, the performances of all strategies, including the optimal and the myopic one, are not dramatic.