当前大规模网络化软件呈现“关联复杂、高度分布、去中心化”等特征，导致现有方法在解决该类软件在线优化问题时，存在“反射调整、集中控制、被动更新”等缺陷。本项目深度交叉融合自优化理论、群体智能思想与多智能体技术，研究基于群体智能的大规模网络化软件自优化机制。主要研究：①面向大规模网络化软件的自优化过程模型；②基于多智能体的软件自优化架构建模方法；③基于群体智能的软件自优化机制；④数据驱动的反馈优化机制；⑤研制支持上述机制的软件支撑环境。创新贡献：①建立分层群体感知与多维分析预测机制，预判系统态势，发掘潜在异常，解决“反射调整”问题，主动优化系统运行质量；②提出基于协同演化与博弈论的群体决策机制，指导群体发挥并行调整与自主协作能力，通过群体协同解决“集中控制”问题，提升决策效率与准确性；③提出数据驱动的反馈优化机制，深度挖掘反馈信息，动态制定调整方案，解决“被动更新”问题，支持机制实现主动演进。

The current large-scale networked software presents the characteristic as the complexity of relation, highly distribution, and decentralization. The existing methods have such problems as centralized control, reflective adjust, and passive update, therefore, it is difficult to solve the problem of runtime software optimization for the current large-scale networked software with the above specific characteristics. In this project, we deeply integrate self-optimization theory, swarm intelligence and multi-agent technology, and present a large-scale networked software self-optimization mechanism based on swarm intelligence. The research focus on (1) a large-scale networked software oriented self-optimization process model, (2) a modeling method of software self-optimization framework based on multi-agent theory, (3) a software self-optimization mechanism based on swarm intelligence, (4) a data driven feedback optimizing mechanism, (5) a prototype software to support and experiment above mechanisms. The contribution and innovation of this project lie in: (1) A hierarchical group perception mechanism and a multidimensional analysis, prediction mechanism are established to initiatively optimize the quality of the system operation, which solves the problem of reflex adjustment by pre-judging system status and exploring its potential anomalies. (2) A group decision-making mechanism based on co-evolution and game theory is intended to direct groups to play parallel adjustment and develop active collaborate ability, solving the problem of centralized control by collective collaboration, which improves the efficiency and accuracy of decision making. (3) A data-driven feedback optimization mechanism is proposed to solve the problem of passive update by mining feedback information deeply and formulating adjustment plan dynamically, which supports active evolution of the self-optimization mechanism.