基于多智能体并行搜索的自适应软件建模方法与运行机制研究

Complex software systems have heterogeneous computing resources, dynamic and open architecture, and are continually constructed during the running time. Existing methods cannot simultaneously and effectively solve the problems (i.e. multilayer adjusting, conflicts of plans, online decision-making) of establishing the self-adaption of these complex software. The project combines the technologies of Search-Based Software Engineering (SBSE) and Multi-Agent Systems (MAS), and researches on a modeling method and running mechanism of self-adaptive software systems based upon multiagent-based parallel searching. Main research topics: (1) a modeling method for SASs based on multiple “Monitor - Analyze - Plan - the Execute (MAPE)” control loops and a structure of regional planning; (2) an running mechanism of SASs based on the parallel search of multi agents, consisting of a diverse modeling method of software changes, a mechanism of dynamic perception, an multilayer mechanism of adjusting software behaviors, a decision-making method based on search method of swarm intelligence, and a mechanism of the parallel search for the adaptive plans; (3) a supporting environment of SASs which has the above mechanisms and methods and supports SASs operating. Contributions of innovation: (1) a modeling approach of SASs based on multiple MAPE loops and a structure of regional planning is intended to establish the control structure of distributed self-adaption, achieving “multilayer adjusting”. (2) A decision-making method of self-adaption based on the search method of swarm intelligence is proposed to use the multi-objective optimization to deal with the self-adaption of software, solving the “conflicts of plans”. (3) A mechanism of the parallel search for the adaptive plans is established to achieve the “online decision-making” by the parallel task allocation and dynamical task scheduling.

复杂软件系统存在“资源异构、动态开放、持续构造”等新特征，现有方法无法有效同时解决建立该类软件自适应性所面临的“多维调整、策略冲突、在线决策”等问题。本项目将多智能体模型与基于搜索软件工程技术相结合，研究一种多智能体并行搜索的自适应软件建模方法与运行机制。主要研究：①基于多重MAPE区域规划的自适应软件建模方法；②基于多智能体并行搜索的自适应软件运行机制，包括软件变化动态感知机制、软件行为多层次自适应调整策略、群体智能搜索自适应决策方法以及自适应策略并行搜索机制；③研制支持上述机制的软件支撑环境。创新贡献：①提出基于多重MAPE区域规划的建模方法，构造分布式自适应控制结构解决“多维调整”问题；②提出群体智能搜索自适应决策方法，首次将多目标优化技术应用到自适应策略动态选择中以解决“策略冲突”问题；③建立自适应策略并行搜索机制，通过任务并行分配与动态调度以解决“在线决策”问题。

针对多维调整、策略冲突、在线决策等软件自适应领域新问题，本项目将多智能体技术与基于搜索的软件工程技术相结合，研究了复杂软件的自适应演化特征，提出了基于多智能体并行搜索的自适应软件建模方法与运行机制，开发了支持该机制的软件自适应支撑环境，已完成本项目预定目标。.具体成果如下：1）针对开放复杂环境特征，提出了面向复杂软件系统的自适应过程模型，并开发了基于智能体的软件自适应支撑环境。2）提出了一种基于多重MAPE区域规划的建模方法，包括基于联盟区域规划的多智能体组织模式，多层MAPE循环中的各类智能体模型以及自适应软件体系结构模型，满足软件在开放环境中的自适应动态演化需求。3）提出了面向软件内外环境变化、功能及非功能需求变化的感知对象多样化建模方法；建立了层次化动态感知框架，通过分层感知以准确感知来源多样的软件变化；建立了针对性感知机制，挑选重要感知对象，实现了对软件异常变化的精准感知；提出了“检测-推理”分析机制，可快速识别系统异常事件。4）建立了基于群体智能搜索的最优自适应决策机制，包括基于多目标优化的最优自适应决策方法和基于后优化理论的最优自适应策略选择方法，可在线产生最优自适应策略。5）实现了自适应决策优化机制，包括通过“任务间并行”与“任务内并行”模式提升决策效率的策略并行搜索机制，以及基于强化学习提升决策准确性的策略修正机制。6）实现了软件行为多层次自适应调整机制，针对不同影响范围的软件变化，分别建立了基于中介黑板的联盟层协商方法、基于OWL-S的角色层协作关系调整方法、基于优化理论的决策调整方法和基于节点筛选的资源层调整部署方法，对软件系统实现多维调整。.研究成果在《软件学报》等国内外核心期刊和ICSE、FSE等软件工程领域顶级国际会议上共发表（含收录）学术论文22篇，超出指标120%。申请国家发明专利4项，超出指标100%。软件著作权5项，超出指标66%。累计培养研究生26名，超出指标225%。相关成果为利用智能体技术与基于搜索的软件工程技术研究软件自适应性提供了支持，具有一定理论意义和应用价值。