在总结传统软件开发方法经验的基础上，严格遵守软件开发方法原则，从UML建模思想出发，对传统UML三层抽象建模结构进行了分析，并引入“抽象”和“分解”作为UML建模的核心思想；提出了UML分层抽象建模机制的构想，将计算机系统开发过程概括为对功能信元、结构信元、关联信元、实现信元和集成信元的提取过程，并以此作为UML建模的指导，有效了利用UML完整语义定义，克服了传统面向对象开发方法的缺陷，实现了建模过程从需求到分析的过渡以及功能和实现事实上的分离。

In the Mobile Decision Support System(MDSS), the key problems of the decision-making technology based on multiple analytical sources are how to articulateand represent the user's global decision-making requests, and then decompose and map them to local analyzing requests on each DSS analytical source, as well as how to integrate the heterogeneous analyzing results. This article develops an object-subject model (OSM) which is based on decision-making subjects. By means of subjects and their operations, the OSM model unifies the decision-making capabilities supplied by varied DSS analytical sources so as to facilitate the formulation of the user's global requests and the inte-gration of heterogeneous analyzing results. Then the article describes the OSM-based algorithms of the formulation and decomposition of the user's global decision-making requests, the dispatching of analysis agents and the integration of the analyzing results.

网格计算环境具有异构性、分布性、动态性、局域自治性等特点，针对以上特性，利用移动代理技术的网格计算方法，可以提供高效统一的应用接口，实现多编程环境下的协同应用求解，从而有效屏蔽其异构和分布等特性，动态适应其资源的变化。详细说明了基于移动代理的网格计算的体系结构和主要模块的功能，并阐述7移动代理的实现过程。

The existence, uniqueness, globally exponential stability and speed of exponential con-vergence for a class of cellular neural networks are investigated. The existence of a unique equilibrium is proved under very concise conditions, and theorems for estimating the global convergence speed approaching the equilibrium and criteria for its globally exponential stability are derived, Considering synapse time delay, by constructing appropriate Lyapunov functional, the existence of a unique equilib-rium and its global stability for the delayed network are also proved. The results, which do not require the cloning template to be symmetric, are easy to use in network design.

rincipal component analysis(PCA) by neural network is an adaptive statistical signal processing method which has many applications Since many PCA neural networks do not converge globally, it is nature to study the global convergence of PCA learning algorithm. In this paper, the previous works on the globally convergent PCA neural networks are presented first. Then based on the mismatch of previous PCA neural networks, we propose and analyze a PCA learning algorithm. This algorithm is convergent globally. And a rigorous mathematical proof is given. Simulation results show the efiiciency and effectiveness ofthis algorithm.