CRISP Type 1: Multi-agent Modeling Framework for Mitigating Distributed Disruptions in Critical Supply Chains

这一关键的弹性相互依赖的基础设施系统和流程(CRISP)类型1赠款侧重于新的数学模型和分析方法，用于提高复杂的人工管理系统的弹性。相互依赖的关键基础设施系统中的利益相关者传统上专注于在面对单一的大型破坏性事件时提供弹性的战略。然而，这种有限的关注使供应链容易受到由多个较小的分布式事件级联而产生的大规模中断的影响，而且这些事件单独太小，不足以引起利益相关者的注意，直到为时已晚。作为这些特征的例证，该奖项支持对美国国内药品短缺危机的调查，这是在制药和医疗保健基础设施内发生的不同时间段和不同水平的事件以及人类对这些事件的反应的组合，这些事件往往会导致连锁效应和长期的药品短缺。目前缓解短缺的努力并不成功。研究结果将为制定新的政策提供信息，并有助于制定新的政策，以解决这一当代至关重要的问题，并有助于从总体上了解如何创建一个有弹性的系统。对于因环境变化、恐怖主义和全球化而面临许多逆境的21世纪来说，理解弹性系统是关键。除研究部分外，还将开发教育活动和内容，并公开分发，以便整合到供应链和高等教育机构的复原力课程中。这些课程将展示如何使用跨学科和变革性的方法来处理复杂的当代问题。这项研究试图推动基础设施弹性和供应链风险管理方面的范式转变，关键是将具有信息共享和有限理性的现实人类决策部分纳入其中。这是通过开发一种独特且具有变革性的多代理建模框架来实现的，该框架将数学建模、逼真的人类行为建模和游戏设计相结合，以便更恰当地考虑关键基础设施中底层物理、信息和社会系统的人类行为和动态。这需要在部分可观测马尔可夫决策过程(POMDP)和随机博弈中开发新的建模框架，包括开发报酬函数结构，该结构是：(1)尽管有大的动作和状态空间定义，但在计算上容易处理，(2)能够捕获具有分布式中断的供应链中的基本权衡，以及(3)适合于从自动生成和专门设计的计算机游戏获得的真实人类决策数据。这项研究的结果将是提取和定义新的弹性度量和分类，结合对稳定系统和未中断的多层网络系统中的代理行为的稳健表征，并将其与分布式和单一极端事件后的系统性能和代理行为进行比较。

This Critical Resilient Interdependent Infrastructure Systems and Processes (CRISP) Type 1 grant focuses on new mathematical models and analysis methods for improving resiliency in complex human-managed systems. Stakeholders within interdependent critical infrastructure systems traditionally focus on strategies to provide resiliency in the face of single, large disruptive events. This limited focus, however, leaves supply chains vulnerable to large disruptions that arise from cascades of multiple smaller distributed events and which individually are too small to warrant stakeholders attention until it is too late. Exemplifying these features, this award supports the investigation of the drug shortage crisis within the United States for which it is the combination of events, occurring at distinct time periods and distinct levels within the pharmaceutical and health care infrastructures, along with the human responses to these events, that often leads to cascading effects and prolonged drug shortages. Current efforts to mitigate the shortages, have not been successful. The research findings will inform and contribute to the development of new policies to address this contemporary life-critical problem and to understanding in general how to create a resilient system. Understanding resilient systems is key for a 21st century that faces much adversity due to environmental change, terrorism, and globalization. In addition to the research components, educational activities and content will be developed, and publicly distributed for integration in supply chain and resiliency courses in higher education institutions. These courses will demonstrate how complex, contemporary problems can be dealt with using a transdisciplinary and transformative approach.This research seeks to drive a paradigm shift with regard to infrastructure resiliency and supply chain risk management, accounting for the critical inclusion of realistic human decision making components with information sharing and bounded rationality. This is achieved by developing a unique and transformative multi-agent modeling framework that combines mathematical modeling, realistic human behavior modeling, and game design in order to more appropriately consider human behavior and dynamics of the underlying physical, information, and social systems in critical infrastructures. This requires development of new modeling frameworks in partially observable Markov decision process (POMDP) and stochastic games, including the development of a reward function structure that is: (1) computationally tractable despite large action and state space definitions, (2) able to capture fundamental trade-offs in a supply chain with distributed disruptions, and (3) amenable to being fitted with real human decision making data obtained from automatically generated and specifically designed computer games. The outcomes of this research will be to extract and define new resiliency measures and classifications incorporating a robust characterization of stable systems and agent behavior in undisrupted multilayered networked systems with multiple realistic agent-based models of human decision makers and contrast these to system performance and agent behaviors after distributed and single extreme events.