数字流域框架下的水环境本体自动构建与自适应组织方法研究

Massive amounts of information sources and domain knowledge pertaining to river basin can't be rapidly accessed and effectively applied in current water environment regulation applications due to various heterogeneity problems among multi-data recourses and the diversity of the knowledge representation. To meet this challenge, this research aims to propose the methodologies for automatic construction and adaptive organization of water environmental ontologies under the framework of digital river basin.This goal will be achieved by systematically and innovatively applying all kinds of methodologies and technologies in the realm of ontology, intelligent data mining and optimization.. The mechanism for data-driven ontology learning will be proposed and the ontology learning approaches towards different heterogeneous data sources will be implemented by the integrated application and optimized regrouping of various ontology learning algorithms, thereby improving the coverage rate, consistence and accessibility of water environmental ontologies. The water environmental ontologies will be build up to accurately represent the characteristics of water environment system at multiple levels such as structure, semantics and knowledge. The improved and highly flexible ontology organization methodologies will be proposed in order to precisely illustrate the complicated hierarchies and semantic mapping relationships of all the spatial-temporal factors characterizing the water environment of the river basin, and to deftly response to the dynamic update of the requirement of water environmental regulations, as well as to explicitly track the process of ontology evolution. The adaptive optimized ontology mapping mechanism will be realized by implement the strategies for optimizing selection and hybrid-matching of multi-ontology mapping algorithms with the purpose of enhancing intelligent semantic integration and adaptive cooperation among the data of multi-dimension and of multi-sources, and satisfying the requirement of water environmental regulations and the ontology model. To support for vivid presentation, navigation and query of large scale ontologies, the ontology editing and browsing engine will be developed by using the component technologies. . This research will provide a novel methodology system and more effective solutions to address the common issues on multi-sources heterogeneous information integration, knowledge sharing and system interoperability existing in the water environmental system of digital river basin. Moreover, The research are expected to greatly improve and innovate the methodologies related to digital river basin and to significantly facilitate the interdisciplinary research between ontology and water environmental sciences. To demonstrate the feasibility and validity of the research achievements, the plain river network located in the southeast of China are selected as a representative research area.

针对信息海量异构与知识表示差异导致难以服务于流域水资源与环境调控的困难，超越现有数字流域研究范式，通过对本体论、智能数据挖掘等先进技术的系统性和创新性应用，构建基于数据驱动的本体学习模式，研究基于智能数据挖掘算法组合优化的异构数据源本体学习方法，自动建立能在结构、语义和知识多层次准确描述流域水环境系统的本体模型, 提高水环境本体的覆盖率、一致性、可访问性；提出改进的多本体组织方法，灵活响应水环境调控多变需求和显式跟踪本体演化进程；研究基于多策略优选和混合匹配的异构本体自适应优化映射方法，实现多源异构数据、调控需求与本体模型的语义集成和自主协同；构建本体可视化展示引擎，支持大规模水环境本体的生动展示、导航与检索。研究成果是对数字流域有关方法的完善和重要创新，对促进本体论与环境水利学科交叉研究起到极大推动作用，为解决流域多源海量信息管理、知识共享与系统互操作提供新的方法体系和更为有效的途径

为了增强数字流域框架下海量异构数据检索、领域知识共享和系统互操作的敏捷性、准确性和协同性，以流域水环境涉及的领域知识为研究对象，基于对本体论、自然语言处理、统计学、智能数据挖掘等多学科理论和技术方法进行交叉和创新性应用，提出具有很强逻辑性、可操作性及可拓展性的水环境领域本体自动构建和自适应组织方法。研究工作取得以下研究成果：（1）针对现有本体构建方法费时费力、结构固化难以更新等问题，研究基于数据驱动的本体学习策略，提出基于TF-IDF改进法、种子概念-TFIDF混合算法、种子概念-遗传-TFIDF混合算法的概念提取算法，解决传统TF-IDF方法存在低频词抽取较少、概念提取精确率较低等方面的局限性;研究层次聚类法、K-Means聚类、遗传算法和K-Means混合聚类法、关联规则法等概念关系学习方法。混合聚类方法利用遗传算法的高效全局优化搜索能力有效克服K-Means算法人工选取K值的随机性和无法自动生成本体层次结构的局限性，提高了聚类分析的精确度。建立水环境本体自动构建工作流机制，按照动态循环的模式进行本体学习，通过多种算法的有机融合指导本体演化过程，有效消除单个算法在某些方面的偏置和缺点，减少对领域专家的依赖性，增强本体构建的自动化程度和性能。建立能在结构、语义和知识层次上准确描述水环境领域的共享概念及语义映射关系的本体模型。采用组件化技术开发了一个整合的针对中文水环境领域知识的通用本体构建系统，用于完成文本、Web网页、主题词表、数据库等数据源的本体学习。（2）针对现有本体组织方法结构固化、环境适应性差等不足，提出面向流域水环境调控的多本体自适应组织架构，既便于集成领域视角不同的分布式异构数据源，又支持数据源的动态更新和本体重用，增强本体组织的柔性和可扩展性。提出基于综合相似度计算、神经网络、贝叶斯学习的多策略本体自适应映射优化方法，提高异构本体映射的精度和效率。开发了一个基于本体驱动的水环境信息获取系统，通过本体映射和语义推理建立领域本体、异构数据源、调控查询需求的语义映射关系，消除语义异构，实现从分布式海量异构水环境数据源中快速定位、获取调控所需数据，并在国内外流域进行了实例研究。研究工作可填补水环境领域本体构建方法体系的研究空白，为解决大数据时代下数字流域极度异质、高度动态的海量水环境信息集成与知识共享问题提供新的途径和强有力的技术支撑。