【作者】 王喆；

【导师】 王红卫；

【作者基本信息】 华中科技大学 ， 系统工程， 2012， 博士

【摘要】 各类突发事件的频繁发生严重地威胁着社会的和谐稳定,并对政府的应变能力提出了更高的要求。由于突发事件应急响应具有突发性、非常规性、层次性、资源稀缺性和时效性的特征,通过理想化的数学模型难以应对复杂的应急决策问题。层次任务网络(HTN)规划通过任务分解形成应对方案的思路类似于应急管理组织实际进行应急决策的过程,并且能够通过逻辑语言描述复杂的应急领域知识,从而成为了应急决策中广泛采用的人工智能方法。但是HTN规划采用任务网络来描述动作推理过程,难以直接控制资源状态的变化以有效的处理资源管理问题。而应急决策中又存在着方案制定过程和资源调度过程高度耦合的特征。方案制定过程中选择不同的任务执行方式会产生不同的资源需求,反过来资源因素又会影响应急管理组织选择应急目标、战略、战术和操作等决策行为。因此将HTN规划和资源调度独立成两个连贯过程的传统方法在应急决策中并不适用,应急决策需要在HTN规划应急行动方案制定和执行过程中处理各种资源管理问题。本文紧密围绕应急响应中方案制定和资源调度高度耦合的过程,以HTN规划为核心,研究了应急行动方案制定与执行中的资源管理方法,主要工作体现在以下四个方面：识别各参与单位中异构的应急资源信息是在HTN规划中进行应急资源管理的基础工作。为此,提出了面向HTN规划的应急资源本体建模方法。本方法中,基于OWL本体(Web Ontology Language)语言建立了应急资源本体模型以识别劳动力资源信息和物力资源信息。基于应急资源本体模型,设计了面向HTN规划的应急资源本体管理方法将OWL本体语言描述的应急资源信息转换为HTN规划语言描述的领域知识。层次资源推理是在HTN规划中进行应急资源管理的关键技术。针对应急决策对层次资源推理的表达能力和处理速度的要求,设计了资源增强型HTN规划方法REHTN。 REHTN设计了资源时间轴来描述各种资源变量和约束。通过自上而下的资源推理将上层任务的资源约束分解给下层任务。同时,通过设计因果链来处理不同分支任务之间的资源推理。在原子任务层,通过不同的资源分配过程验证了消耗性资源和可重用性资源的资源状态的可行性。并进一步设计了资源约束传播加速算法来加快层次资源推理过程。应急资源缺项问题是在充分利用现有资源尽可能地完成应急目标的同时,分析那些由于资源缺项而暂时无法完成的应急目标及其所需补充的资源信息。应急领域专家可以依据这些资源缺项信息,通过请求增援等方式最终完成这些应急目标。针对应急资源缺项问题,设计了基于REHTN的应急资源缺项问题求解方法REHTNRS。 REHTNRS通过扩展REHTN语法来描述应急资源缺项问题,其中将规划目标编码为包括可达成目标、不可达成目标和有条件达成目标在内的软任务目标,并通过带资源缺项标记的任务来处理有条件达成目标。应急资源缺项问题的规划处理算法中,提出了资源缺项识别方法,通过资源缺项标记来识别规划过程中出现的资源缺项。在此基础上,设计了带资源缺项标记的Anytime启发式算法。围绕应急行动方案执行过程中的资源执行异常问题,在构建混合规划与执行集成框架的基础上设计了应急资源执行异常的管理方案,其中包括应急资源执行异常的监控、分析和处置。更多还原

【Abstract】 Various frequent emergencies, which threaten the harmony and stability of society severely, request the high reaction ability of government. Because of the sudden, unconventionality, hierarchy, resource scarcity and timeliness of emergency response, complex emergency decision-making issues cannot be solved by establishing idealized mathematical model. Hierarchy Task Network (HTN) planning, which obtains the plan by task decomposition, can simulate the whole process of emergency decision-making from strategical levels to operational levels. Furthermore, HTN planning can express complex emergency domain knowledge by logical language. Therefore, it is widely used in emergency decision-making. Unfortunately, the task network is adopted in HTN planning to describe action reasoning process. For this reason, HTN planning is difficult to control the resource state to deal with resource management issues. Emergency decision-making contains the coupling relationship of task planning and resource scheduling. In task planning, different task execution methods require different resources. Meanwhile, emergency resource profiles affect the decision-making behaviors of emergency responders, such as choosing emergency goals, strategies, tacticses and operations. Hence, HTN planning and resource scheduling in emergency decision-making cannot be treated as two independent processes. On the contrary, in order to support emergency response efficiently, the resource management ability must be enhanced for HTN planning.Focusing on the coupling relationship of task planning and resource scheduling of emergency response, resource management approach based on HTN planning in emergency decision-making is studied. The contributions of this thesis are as follows:In order to identify heterogeneous emergency resource information from multiple participating units, an emergency resource ontology modeling approach is designed. Based on OWL ontology language (Web Ontology Language), emergency resource ontology model is established in the approach to obtain human resource information and material resource information. Based on emergency resource ontology model, emergency resource ontology management is established to transform emergency resource information from OWL ontology language to HTN logical language.Hierarchical resource reasoning is one of the key issues to successfully apply HTN planning into emergency decision-making. This thesis proposes a Resource Enhanced HTN (REHTN) planning approach for emergency decision-making with the objective to enhance the expressive power and improve the processing speed of hierarchical resource reasoning. In the approach, resource timelines are defined to describe various resource variables and constraints. Top-down resource reasoning is used for decomposing the resource constraints of upper-level tasks into those of lower-level tasks. Meanwhile, resource and temporal constraints of tasks in different branches are processed by causal links. In primitive task layer, resource profiles of consumable resources and reusable resources are checked by separate resource allocation processes. Furthermore, a constraint propagation accelerator is designed to speed up hierarchal resource reasoning.Emergency resource shortage issue includes two aspects. On the one hand, emergency goals must be achieved as far as possible by full using of existing resources. On the other hand, temporary unachieved emergency goals must be analyzed to identify the resource shortage information. Emergency responders can achieve those emergency goals by requesting reinforcements according to the resource shortage information. An REHTN approach for resource shortage (REHTNRS) is proposed for emergency resource shortage issue. The issue is described in REHTNRS by expanding the syntax of REHTN. In the describing process, plan goals are modeled as soft goals, including achieved goals, unachieved goals and conditional achieved goals. Meanwhile, conditional achieved goals can be accomplished by the tasks with resource shortage stamps. In the processing algorithm of the resource shortage issue, resource shortage identification method adopts resource shortage stamps to identify the resource shortage in planning. Meanwhile, anytime heuristic algorithm with resource shortage stamps is designed.Emergency resource exception happens frequently in emergency action plan implementing. Based on the establishing of integrated planning and execution framework, an emergency resource exception treatment scheme is designed. The scheme includes emergency resource exception monitoring, analysis and disposing.更多还原