Automated Dynamic Resource Allocation for Wildfire Suppression
Automated Dynamic Resource Allocation for Wildfire Suppression
复制标题
用于野火扑灭的自动动态资源分配
DOI：
--
复制
发表时间：
2017
期刊：
影响因子：
--
通讯作者：
D. Bertsimas
中科院分区：
文献类型：
--
作者：
J. Griffith;Mykel J. Kochenderfer;Robert J. Moss;V. Mišić;Vishal Gupta;D. Bertsimas
2002年至2012年间，美国林务局和内政部平均每年在荒地防火上花费31.3亿美元。其中近10亿美元专门用于灭火工作。目前，分配灭火资源是一个主观过程，在许多情况下依赖于个别指挥官基于不充分信息的决定。自动化系统将通过提供可操作的建议来帮助指挥官科普情况的不确定性，时间压力和有限的资源。战术荒地火灾管理是一个例子，涉及在高度不确定的，动态的环境中的资源分配问题。类似的问题包括城市搜索和救援、洪水监测和管理以及地震反应。这样的问题涉及高维度（例如，大量的位置或资源）、不确定的动态、资源分配的许多组合以及多个竞争目标的平衡。动态资源分配（Dynamic Resource Allocation，简称MRP）问题是一类更一般的问题的特殊情况，称为马尔可夫决策过程（Markov Decision Processes，简称MDP）（参见下一页标题为“马尔可夫决策过程”的边栏）。MDP是一个数学公式的问题，其中系统动态是部分随机的，部分可控的决策者。MDP是一个合理的模型，野火抑制以及许多其他自治系统的防御问题。例如，MDP还可以用于模拟舰队保护、军事后勤、地雷对策、监视任务、战斗管理、通信、火灾对生命和基础设施构成重大威胁，而灭火工作在人力和资源上都是昂贵的。尽管态势感知工具和通信技术的重要进展极大地帮助了事件指挥官指挥压制工作，但仍然缺乏有效的自主动态决策支持系统，该系统提供资源分配建议，以帮助指挥官科普不断变化的火灾动态和时间压力的不确定性。J.丹尼尔格里菲斯，Mykel J. Kochenderfer，Robert J. Moss，Velibor V. Mišić，
Between 2002 and 2012, the U.S. Forest Service and the Department of the Interior spent on average $3.13 billion per year on wildland fire protection [1]. Nearly $1 billion of that funding was devoted solely to fire suppression efforts. Currently, allocating resources for fire suppression is a subjective process, relying in many cases on individual commanders’ decisions that are based on insufficient information. An automated system would help commanders cope with situational uncertainty, time pressures, and limited resources by providing actionable recommendations [2]. Tactical wildland fire management is one example of problems involving resource allocation in highly uncertain, dynamic environments. Similar problems include urban search and rescue, flood monitoring and management, and earthquake response. Such problems involve high dimensionality (e.g., large number of locations or resources), uncertain dynamics, many combinations of resource assignments, and the balancing of multiple competing objectives. Dynamic resource allocation (DRA) problems are special cases of a more general class of problems called Markov decision processes (MDP) (see sidebar titled “Markov Decision Processes” on the following page). MDPs are a mathematical formulation of problems in which the system dynamics are partially random and partially controllable by a decision maker. MDPs are a reasonable model for wildfire suppression as well as for many other autonomous systems problems in defense. For example, MDPs can also be used to model fleet protection, military logistics, mine countermeasures, surveillance missions, battle management, comWildland fires pose a significant threat to life and infrastructure, and suppression efforts are costly in manpower and resources. Although important advances in situational awareness tools and communication technologies have greatly aided incident commanders in directing suppression efforts, there is still a lack of effective autonomous dynamic decision support systems that provide resource allocation recommendations to help commanders cope with the uncertainty of evolving fire dynamics and time pressures. » J. Daniel Griffith, Mykel J. Kochenderfer, Robert J. Moss, Velibor V. Mišić,