基于冲突的应急救灾高性能计算资源优化配置研究

Complexity in nature systems, China is prone to natural disasters. Insufficiency of disaster information sharing and resource allocation, however, has underperformed the national emergency management to alleviate hazardous impacts. Big data, especially spatial data, compared to historical data or expert experience, has proved to be more useful in covering a wider and clearer picture of highly dynamic disaster situation, thereby enabling a prompter and better disaster response support. Practically, given the constraint of available resources that can be mobilized during disaster response phase, complicated technical computation has prevented the resources from being timely processed and readily utilized, resulting in the issues, such as data overload, information unsaturation. To address these issues, the study, will first employ an integrated method of the grounded theory and knowledge mapping to summarize all proxy extraction algorithms used in disaster situation. This is to construct the evaluation system of data processing performance. A simulation will be performed in clusters thereafter in order to evaluate the scalability of the algorithms in a high performance computing environment. After this, the rough set theory and cluster analysis will be adopted for coordinating the conflicts in the prioritization and sequence of different data processing tasks. Finally, Data Envelopment Analysis (DEA) and bi-level programming method will be utilized for computation resource allocation. This study is proposed to address two leading research questions: (1) how to systematically evaluate the performance under a high performance computing environment; and (2) how to allocate the computation resources from the data processing conflicting perspective. It is expected that the project outputs can contribute to the improvement of data-driven disaster management by providing more effective information support for disaster emergency management.

近年来，我国自然灾害频发，灾情信息共享与资源统筹不足等问题严重影响了应急管理工作的顺利开展。相对历史数据、专家经验，大数据涵盖灾情信息更充分全面，能有效指导自然灾害的应急管理。然而，在有限资源的紧急救援过程中，过于庞杂的技术资料导致了数据处理量过载、有效信息不足等问题。本项目拟通过扎根理论和知识图谱的文献归纳、计算机集群仿真的高性能计算、粗糙集和聚类的冲突协调、数据包络和双层规划的资源优化等方法，构建大数据挖掘灾情特征因子的算法集及算法效用评价体系，评估高性能计算介入下算法运行效能的提升，协调灾情特征因子的选取与挖掘次序冲突，探索特定自然灾害下资源的优化配置。项目实施后可建立高性能计算介入下大数据挖掘灾情特征因子算法运行效能评估模型及冲突视角下大数据挖掘灾情信息的资源优化配置方案，以丰富、完善数据驱动下的灾害管理理论与模型，为重大自然灾害及时有效的应急管理提供决策依据。

越来越多的灾害大数据被收集，然而，多源、异构和海量的数据已经超出了传统计算资源的处理能力，导致只有一小部分数据被真正用于应急管理决策支持中。为了解决这个问题，本项目基于Apache Strom的分布式流处理架构与计算资源分配模型相结合，探索在时间和计算资源的双重冲突下提高灾害大数据算力的有效解决方案。.本项目完成了申请书设定的研究内容，主要进行了四个子课题的研究：1）灾害大数据在应急管理中的实际应用的系统回顾；2）高性能计算介入下的灾害大数据处理算力提升模型；3）灾害数据处理任务的需求测度及冲突协调；4）高性能计算资源算力优化配置模型，并作了一些有益的尝试。项目的研究成果基于语义群决策的效率模型对于化解应急管理决策中，各信息需求主体间的矛盾冲突有较好的理论贡献，而基于Apache Storm流处理架构的分布式计算的实现，对于提高数据处理的效率有较好的实践参考意义。此外，研究成果为政府提高危机应对效能，增强危机决策的数据支持提供了决策参考。.项目的实施，产生了一些有益的效果。项目负责人以第一或通讯作者共发表或录用中英文论文12篇，其中SSCI/SCI检索论文11篇。部分论文发表在诸如Information Technology & People、International Journal of Disaster Risk Reduction等领域顶尖期刊。项目执行期间，共有4名硕士研究生，6名本科生参与其中，其中3名已前往或拟前往境外高校深造。