基于构件能力动态规划的自适应软件群体决策方法及其应用研究

In network and social environment which are complex, open and changing, adaptive collective systems should not only modify the structure and behavior to response the changing environment, but also satisfy the challenges of dynamic adaptation between heterogeneous and interacting components. Requirement aware approach could solve the problems of the running time behaviors are not optimized, or the deviation between the running time behavior of software and its requirements in current status. The changing trends of decision object and decision mode as precondition, the capability reorganizing and collaboration of heterogeneous intelligent component as point-cut, this project proposes an approach on intelligent component capability dynamic planning based self-adaptive systems collective decision-making. First, studying the semantic based formal description model of intelligent component capability and reasoning about capability interactions. The model could on one hand help adaptive goals match appropriate components; on the other hand making capability interaction is more lightweight and flexible than component interaction. Second, integrated planning and dynamic planning are fused to rebuild capability based decision making model in unpredictable states. At last, it is general and typical to use heterogeneous intelligent components collaboration control system as demonstration system. This project will proposed the research results of promoting and special theories, at the same time will provide the theoretical and technical support for development of intelligent industries and promotion of collective adaptation for intelligent systems in pervasive computing.

在复杂、开放并且动态变化的网络环境和社会环境之中，群体自适应系统不仅需要能够通过调整自身的结构和行为以响应环境变化，而且面临异构的、交互的群体动态自适应的新挑战。本项目在明确自适应动态决策对象和决策模式转变趋势下，以智能构件能力识别和协作为切入点，提出一种基于智能构件能力动态规划的自适应软件群体决策方法。首先，基于语义的智能构件能力的形式化描述和交互分析推理方法，既能帮助目标自适应匹配适合的构件又能使能力级协作比构件级协作更加轻量、灵活。其次，基于动态群体规划的自适应决策机制融合了集成规划和动态规划的优点，能够在未知状态下重新构建运行时动态决策模型。最后，以异构多智能构件协作控制平台作为上述模型和算法的实证系统，具有较强的概括性和典型性。本项目的研究将形成有推动作用、有特色的理论意义的研究成果，同时将对我国的智能产业发展、提高普适环境下的智能系统的群体自适应性提供关键理论支撑和技术保障。

本项目针对软件系统在偏离需求规约或处于非优化运行时呈现出由“被动应变”转变为“主动适应”的趋势，为协作自适应系统（CAS）提供一种可行实现框架以及高效的实现方法。本项目将异构构件的自动化理解、分析和推理作为异构构件能够相互识别和协作的基础，在此基础上提出自适应需求建模驱动的Meet-in-Middle的群体知识决策模式。首先，将自适应系统的需求抽象为可执行目标模型，将自适应构建的自身能力和协作能力抽象为能力和承诺，并且使用环境和系统上下文状态表示目标、能力和承诺的约束条件。然后，将自适应系统的需求和自适应构建的匹配问题转化为目标和能力之间的语义相似度问题，运用能力、承诺和目标间的语义相似度计算定量地描述他们三者之间的匹配关系，并能够择优选择能力或者承诺，为目标的完成选择出一条最适合的能力协议执行过程。最后，将中介适配器作为CAS需求以及环境层和多机器人（或多Agent）的行为执行层的中间件群体决策层，研发了基于可执行目标树的自动化协议生成算法（EWCC）的GCC mediation决策平台，并在此基础上研发了医疗垃圾智能AGV运输仿真平台。后续研究将在线演化系统的形式化验证方法和机器人在线验证方法相结合，以探索协作自适应系统的在线验证研究。