Modeling Information Gathering Decisions in Systems Engineering Projects
Modeling Information Gathering Decisions in Systems Engineering Projects
复制标题
系统工程项目中信息收集决策的建模
DOI：
--
复制
发表时间：
2014
期刊：
影响因子：
--
通讯作者：
R. Malak
中科院分区：
文献类型：
--
作者：
Chuck Hsiao;R. Malak
系统工程项目中的决策通常是在大量的不确定性下做出的。这种不确定性可能存在于许多领域，如子系统的性能，子系统之间的相互作用，或项目资源要求，如预算或人员。系统工程师通常可以选择收集减少不确定性的信息，这可能会做出更好的决策，但以获取信息所花费的资源为代价。然而，我们对如何分析涉及收集信息的情况的理解是有限的，因此，通常使用逻辑学，直觉或最后期限来判断决策所需的信息收集量。系统工程师将受益于更好地了解如何确定所需的信息收集量，以支持decision.This本文介绍了部分可观察马尔可夫决策过程（POMDPs）作为一种形式主义建模信息收集决策系统工程。POMDP可以对不同的状态、备选方案、结果和结果的概率进行建模，以表示决策者对其情况的信念。它还可以以紧凑的格式表示顺序决策，避免决策树和类似表示的组合爆炸。POMDP的解决方案，以价值函数的形式，规定了基于决策者对他的情况的信念的最佳行动方案。价值函数还决定是否需要收集更多的信息。近年来，POMDPs的复杂计算求解器已经开发出来，可以直接分析不同的替代方案，并确定在给定情况下的最佳行动方案。本文演示了如何使用POMDP对系统工程问题进行建模，并将此方法与其他在决策过程中收集信息的方法进行了比较。© 2014 ASME
Decisions in systems engineering projects commonly are made under significant amounts of uncertainty. This uncertainty can exist in many areas such as the performance of subsystems, interactions between subsystems, or project resource requirements such as budget or personnel. System engineers often can choose to gather information that reduces uncertainty, which allows for potentially better decisions, but at the cost of resources expended in acquiring the information. However, our understanding of how to analyze situations involving gathering information is limited, and thus heuristics, intuition, or deadlines are often used to judge the amount of information gathering needed in a decision. System engineers would benefit from a better understanding of how to determine the amount of information gathering needed to support a decision.This paper introduces Partially Observable Markov Decision Processes (POMDPs) as a formalism for modeling information-gathering decisions in systems engineering. A POMDP can model different states, alternatives, outcomes, and probabilities of outcomes to represent a decision maker’s beliefs about his situation. It also can represent sequential decisions in a compact format, avoiding the combinatorial explosion of decision trees and similar representations. The solution of a POMDP, in the form of value functions, prescribes the best course of action based on a decision maker’s beliefs about his situation. The value functions also determine if more information gathering is needed. Sophisticated computational solvers for POMDPs have been developed in recent years, allowing for a straightforward analysis of different alternatives, and determining the optimal course of action in a given situation. This paper demonstrates using a POMDP to model a systems engineering problem, and compares this approach with other approaches that account for information gathering in decision making.© 2014 ASME