【作者】 黄金才；

【导师】 陈文伟；

【作者基本信息】 国防科学技术大学 ， 管理科学与工程， 2001， 博士

【摘要】 论文以网络环境下决策资源的共享和集成为背景，针对基于数据开采的决策支持技术、网络环境下DSS的体系结构和运行协议、广义模型服务器系统、DSS可视化集成环境和DSS集成语言等方面进行了深入研究。作者完成的工作和取得的创新成果如下： 为了实现网络环境下决策资源的有效共享和快速集成，作者提出了DSS层次模型(DSSLM，包括表现层、方案层、实例层、服务器层)、决策资源接口规范、决策资源集成和运行协议以及能够集成多种决策支持技术的综合决策支持系统体系结构，它明确了各种决策支持技术之间的集成关系，明确了网络环境下资源运行协议、接口规范与服务器之间的关系，对于网络DSS的开发和应用具有积极的意义。 数据开采是一种新的决策支持新技术，它是模型驱动决策的重要补充。作者在文中重点论述了用于数据开采的非线性神经网络。目前的神经网络研究基本上是基于超平面的线性神经网络，通常这种网络存在着学习时间长，网络结构不容易理解等问题。为此作者提出了新的用于分类学习的非线性神经元模型--CC模型以及相应网络结构。CC模型采用超圆划分样本空间。由于超圆具有良好的几何特性，网络权值可由解析方法获得，因此学习效率会大大提高。非线性神经网络模型不仅可以保证了样本学习时的效率，而且学习后的结果可理解性比较好。这是一种新型的网络结构形式。 广义模型服务器实现了网络环境下决策资源的共享和存取，它能够对各种决策资源，如模型(包括数学模型、数据开采方法等)、方案、知识、实例等，进行统一组织、管理和运行；支持多种决策资源之间的交互；支持多客户的同时访问；管理命令语言(RML)提供了远程客户访问服务器的手段。系统采用开放的多层客户/服务器结构，网络上的远程客户通过RML请求资源服务，获取决策信息。目前在国内外，模型服务器的概念和产品还未见到报道。作者研制的广义模型服务器系统已经得到成功的应用。它实现了真正的资源共享，是DSS资源管理的一种崭新模式。 作者开发了DSS可视化快速集成环境(DSSE)，它通过构造决策方案框架流程，连接网络上模型服务器和数据库服务器上的共享决策资源，快速生成、修改、运行和评价实际问题的解决方案。DSSE支持模型驱动的决策、基于数据开采的数据驱动决策以及二者结合的决策过程，从而能够达到更好的辅助决策效果。作者中还设计和完成了网络环境下DSS集成语言系统，集成语言是可视化快速集成环境的重要补充。中国科学院遥感所利用DSSE环境，开发了全国农业投资决策支持系统(SDSS/AIC)。该系统证实DSSE环境的有效性和技术水平。DSSE环境通过了中国科学院组织的技术鉴定，鉴定意见为国内领先、国际先进。 本文的研究工作对于推动决策资源的共享和集成，实现网络决策支持系统，促进决策支持技术水平的提高，并最终达到管理决策的网络化、科学化、规范化，都具有一定的理论和实践意义。更多还原

【Abstract】 Aimed at decision resources sharing & integration in network’s environment,the thesis focuses on the decision support technologies based on data mining,the DSS architecture and running protocol,generalized model server system,DSS integration language. DSS rapid development environment and so on. The main achievements are as fellow:For effective sharing and rapid integration of decision support resources in network environment,DSS Layer Model (DSSLM,including representation,plan,instance and server). decision resources interface specification,the integration & running protocol of decision resources and the architecture of integration DSS have been achieved. The architecture composites many decision support technologies,which defines the relation among the integration protocol,the interface specification and decision resources servers. It has positive values to the development and application of network-DSS.Data mining is arising as a new decision support technology. Here the author emphasizes non-linear neural networks used to data mining. The neural networks currently studied are almost linear based on super-flat. Usually they need long training time,and are hardly understood. In this case,the author puts forward to new aon-linear neural networks for classing-CC model and its network architecture. CC model utilizes super-circle to divide the example’s space. Thanks to the good geometry characteristic of super-circle,the weights and architecture of networks are easily obtained by math solution. Therefore,the networks can provide satisfying efficiency in training and satisfying generalization after training,and the trained networks are easily understood. They are a new type of neural networks.Generalized Model Server provides the sharing and access abilities to decision resources in network environment. The server unifies organization,management and running of generalized models (model,algorithm,plan,knowledge,instance and so on),which provides intercommunion between different resources,supports the running of data mining algorithms,support synchronous access of clients system,and provides remote accessing mechanism by management language (RML). The system owes open multi client server architecture. The remote client utilizes RML to request model services through network,acquiring decision information. By now,the conception and software of model server has not been represented and reported. This is the first time that the author puts forward to and develops the Generalized Model Server. It’s a new management mode of DSS resources,and the true sharing of DSS resources comes true.DSSE Rapid Development Environment,developed by the author,can quickly generate,modify,run and evaluate the solution plan of real-world problem,by editing decision frame. linking DSS resources on the different servers. DSSE can support model-driven decision,data-driven decision and the combination of both,providing better and richer decision-aided information. At the same time,the author designs and implements DSS Integration Language innet environment,which reinforce the aid-decision functions of Rapid Integration Environment. In the thesis,the DSS application developed in DSSE:Spatial Decision support System for Agricultural Investment in China (SDSS AIC) is presented. The successful implementation and application of DSSE reward a high score of Information Center of State Development Planning Commission and Remote Sensing Application of Chinese Academy of Sciences (RSA CAS).The achievements of the thesis have great theoretic and realistic significance in promoting the sharing of decision resources,implementing network decision support system,advancing the decision support technologies,achieving the networking,scientific and standardization of management and decision.更多还原