SoD: A Normative Theory for the Design of Discrete Event Dynamic Systems for Supervisory Control

这个项目解决了一个与离散事件动态系统（DEDS）设计相关的问题领域。该项目的目标是确定控制这些系统的监督政策的可处理的综合。DEDS的例子包括：空中交通管制系统；自动化制造系统；计算机网络;综合指挥、控制、通信和信息（C3I）系统；具有事件驱动动态的多组件组织的运营管理，如造船厂、机场、医院等。尽管DEDS有无数不同的应用，但在DEDS监督控制的核心方面取得的进展并没有在学术界以外的领域得到应用，主要是因为合成过程需要对系统的预期行为进行完整的规范，这是一项困难的任务，如果不是不可能的话。即使有了完整的规范，适当的监督策略的综合也可能是棘手的。本项目开发了一种规范理论，说明如何构建DEDS，以便从执行不完全指定行为的监督政策开始，逐步应用监督纠正，从而最终执行（完整）期望的规范。本项目倡导的监督政策是基于对观察到的系统的“可取”和“不可取”状态的分析。该项目侧重于DEDS的三个领域：(1)学习阶段：开发用于监控系统的学习算法，其中可以使用示例/反例轻松获得/识别期望行为；(2)监管政策综合阶段：识别学习易于处理的DEDS结构；(3)渐进改进阶段：改进学习行为的方法的发展，这种方法可以在获得额外信息时，用很少的计算努力逐步改进。与本项目相关的开放研究课题的成功解决可能意味着可以消除上述每个DEDS示例中的特殊问题（空中交通瓶颈、自动化制造中的死锁/活锁、网络中的低质量服务、医院急诊室中的差分类服务等）。

This project addresses a problem domain related to the design of Discrete Event Dynamic Systems (DEDS). The goal of the project is to identify tractable synthesis of supervisory policies that control those systems. Examples of DEDS include: air-traffic control systems; automated manufacturing systems; computer networks; integrated command, control, communication and information (C3I ) systems; operations-management of multi-component organizations with event-driven dynamics like shipyards, airports, hospitals, etc. Despite the myriad and diverse applications for DEDS, advances in core aspects of supervisory control of DEDS have not found application beyond academia, primarily due to the fact that synthesis procedures require complete specification of the desired behavior of the system, a difficult, if not impossible task. Even when complete specifications are available, the synthesis of the appropriate supervisory policy can be intractable. This project develops a normative theory of how DEDS should be structured so that, starting from a supervisory policy that enforces an incompletely specified behavior, progressive supervisory corrections can be applied such that the (complete) desired specification is eventually enforced.The supervisory policy advocated by this project is based on analysis of observed "desirable" and "undesirable" states of a system. This project focuses on three areas of DEDS: (1) Learning phase: the development of learning algorithms for supervisory systems where desired behavior can be easily earned/identified using examples/counter-examples, (2) Supervisory policy synthesis phase: the identification of DEDS structures where learning is tractable, and (3) Progressive improvement phase: the development of methods for improving learned behavior that can be progressively improved with little computational effort as additional information is made available. The successful resolution of the open research topics related to this project may mean that idiosyncratic problems in each of the examples of DEDS described above (air-traffic bottlenecks, dead-lock/live-lock in automated manufacturing, poor quality of service in networks, poor triage services in the emergency room of a hospital, etc.) could be eliminated.