Multi-scale decision making: challenges in engineering and environmental systems
Multi-scale decision making: challenges in engineering and environmental systems
复制标题
多尺度决策：工程和环境系统的挑战
DOI：
10.1007/s10669-013-9469-y
复制
发表时间：
2013
影响因子：
--
通讯作者：
P. Beling
中科院分区：
文献类型：
--
作者：
P. Beling
许多系统分析涉及在不同时间、物理和组织尺度上定义的各种决策。规模问题可能会引入决策之间的依赖关系和影响，这些依赖关系和影响很难用优化、决策理论和风险管理的标准方法来预测或理清。这些问题由于技术（传感器、通信和计算）的快速发展而变得更加重要，这些技术支持以前所未有的规模（“大数据”决策）收集和处理新型数据。多尺度决策问题出现在各种领域，包括环境管理、制造和生产、服务系统和金融。多尺度决策与分布式决策问题密切相关。存在大量关于优化和规划的分布式模型的文献，其中的问题通常根据决策者之间存在的层次结构进行分类（例如，参见Deng和Papadimitriou 1999； Schneeweiss 2010）。人工智能社区也考虑了分布式问题，重点是根据智能体之间信息交换的范式定义的分类法中的机器学习问题（例如，Doran et al. 1997; Weiss 1999; Brafman and Tennenholtz 2011）。最近，Wernz和Deshmukh（2010, 2012）提出了一个统一的多尺度问题数学框架，该框架在博弈论模型中定义了决策层次和信息交换。为了与期刊的主题保持一致，本期《环境系统与决策》特刊从一个明显更实用的角度探讨了多尺度决策。在实践中，决策的背景是许多真正困难所在。多尺度问题经常跨越组织和利益相关者群体，这些环境的结构可能会违背数学模型中容易假设的那种平滑的信息交换和分层特征。作为多尺度决策是如何由系统上下文产生的问题驱动的一个例子，请考虑系统设计和操作中安全性日益增长的重要性。一个决策问题可能已经被正确地建模为协调组织内分布式元素的行为之一，现在必须重新建模，以解释智能对手的存在。这反过来又在许多新的涉众（包括对手本身）之间创建了决策问题的联系。下面概述了多尺度决策中的几个主要挑战。在每种情况下，都参考了特刊中最相关的论文。
Many systems analyses involve a variety of decisions defined over different temporal, physical, and organizational scales. Scale issues may introduce dependencies and influences between decisions that are difficult to anticipate or untangle with standard methods from optimization, decision theory, and risk management. These issues are made all the more important by the rapid proliferation of technology (sensors, communications, and computing) that supports the collection and processing of new types of data on unprecedented scales (‘‘big data’’ decision making). Multi-scale decision problems arise in a variety of domains, including environmental management, manufacturing and production, service systems, and finance. Multi-scale decision making is closely related to the topic of distributed decision making. There exists a large literature on distributed models for optimization and planning, with problems often classified in terms of the hierarchies that exist between decision makers (see, e.g., Deng and Papadimitriou 1999; Schneeweiss 2010). The artificial intelligence community also has considered distributed problems, with emphasis on machine learning problems in taxonomies defined in terms of paradigms for information exchange among agents (see, e.g., Doran et al. 1997; Weiss 1999; Brafman and Tennenholtz 2011). Recently, Wernz and Deshmukh (2010, 2012) have proposed a unified mathematical framework for multi-scale problems which defines decision hierarchies and information exchange in a game theoretic model. In keeping with the theme of the journal, this special issue of Environment Systems & Decisions explores multiscale decision making from a decidedly more applied perspective. In practice, the context for decision making is where many of the true difficulties lie. Multi-scale problems often cut across organizations and stakeholder groups, and the texture of these environments may defy the kind of smooth information exchange and hierarchical characterizations so easily postulated in mathematical models. As an example of how multi-scale decision making can be driven by issues that arise from systems context, consider the growing importance of security in systems design and operation. A decision problem that might have been properly modeled as one of coordinating the actions of distributed elements within an organization must now be recast to account for the presence of an intelligent adversary. This in turn creates linkages in the decision problem among many new stakeholders, including the adversary itself. Several of the principal challenges in multi-scale decision making are outlined below. In each case, references are made to the most relevant papers in the special issue.