多智能体仿真模型演化特征关系挖掘及可信度评估方法研究

Evaluating the credibility of a simulation model is of the most importance as it establishes that the model is trustworthy to represent the system under study. With the development of multi-agent simulation, traditional expertise-based techniques have become inapplicable for evaluating the dynamic evolutionary process of multi-agent model. This project established a two-level cognitive map to represent the implicit relationships among different characteristics of multi-agent model. Based on the configurable intelligent optimization technique, a parallel self-organizing evolutionary algorithm is designed to generate the optimal two-level cognition map and predict the future evolutionary states of the multi-agent model. To further understand the long-term behavioral principle of the model, a temporal and spatial pattern discovery algorithm is also established to analyze its temporal and spatial evolutionary pattern. Combined with the current simulation data and the predicted evolutionary pattern, an automatic evaluation method is constructed to realize automated “evaluation feature mapping, dynamic pattern discovering and high-efficient model validation” for the whole evolutionary process of multi-agent simulation. The methods proposed in this project can be applied in different multi-agent systems to guide their model construction, dynamic calibration and value-added extension with high credibility.

仿真模型的可信度评估是仿真系统构建及有效应用的基石，是实现真实世界复杂对象正确分析的重要保障。随着基于多智能体的协同仿真技术发展，传统基于专家经验的评估方法已难以适应由多智能体交互组成的复杂模型仿真推演，成为制约多智能体仿真模型扩展和应用的瓶颈问题。因此，本项目拟开展多智能体仿真模型隐性特征关系挖掘及模型演化可信度评估方法的研究，基于可配置智能优化算法建立最优二阶认知图模型，揭示多智能体仿真模型多层次特征对模型可信度的影响机理，并从时域和空间域角度推导模型演化规律，对比模型的全局演化过程与真实对象的动态特性，进而建立起一套全新的基于“多维特征映射、全局演化推导”的评估方法，实现多智能体仿真模型的可信度有效判定和快速修正。为基于多智能体技术的复杂系统建模中，仿真模型是否正确可信，如何建立符合真实系统的多智能体仿真模型，如何实现多智能体仿真模型的动态校准及增值扩展提供理论指导依据。

多智能体仿真模型评估过程中，由于智能体交互所带来的仿真模型动态演化，使得传统基于专家经验的评估方法难以适应评估精度和速度的要求。本项目针对该问题，开展了多智能体仿真模型隐形特征关系挖掘及模型演化可信度评估方法的研究，并在研究的基础上开发了一套基于增量学习和模式匹配的多智能体仿真模型评估方法；基于可配置的智能优化算法建立最优二阶认知图模型，揭示了多智能体仿真模型多层次特征对模型可信度的影响，并进一步研究了优化算法的并行化对于求解质量和求解速度的提升，开发了一种并行迁移进化算法，基于该算法优化、加速了模型评估过程。项目最终研究建立了一套“多维特征映射、全局演化推导”的评估方法，实现了多智能体仿真模型演化过程的动态评估，通过并行迁移进化算法优化了仿真模型评估过程。为多智能体仿真模型是否正确可信提供了判断依据，为实现多智能体仿真模型的智能快速动态校准及增值扩展提供了理论指导。