BGP安全对于互联网可靠稳定运行至关重要。目前解决方案难以切实保障BGP安全。课题从域间路由系统开放、分布、自治和商业性的本质特性出发，假设自治系统是“理性”的。在该假设下，探索研究自治系统路由行为的评估与预测技术，以期通过抑制异常路由通告攻击的发生来改善BGP安全。研究包括拟提出基于多平面证据关联的异常检测技术和基于攻击效力的行为评估方法，以期准确、全面评估路由行为；拟提出基于时间域单源信息融合理论的行为预测方法，以期通过融合历史行为来预测未来行为；针对自治系统发起前缀可达路由行为的“幂律”动态性，拟提出基于机器学习的行为动态性建模方法和基于自适应窗口调节的行为预测演化机制，针对邻居自治系统间在竞争-合作关系约束下动态交互的特点，拟提出基于演化博弈的邻居路由转发行为动态性建模方法和基于交互策略演化的行为预测演化方法，以期实现行为预测的快速收敛。研究将丰富BGP安全问题的解决方法与手段。

It is highly important for Internet to ensure BGP security. Unfortunately, current BGP security solutions can’t actually protect BGP from prefix hijacking and path forging attacks. Based on the observation that the inter-domain routing system is open, distributed, autonomous and business-driven, the project will assume that the autonomous system in the inter-domain routing system is rational. Under this assumption, the project will conduct the thorough research on autonomous system routing behavior evaluation and prediction technology, to improve BGP security by suppressing prefix hijacking and path forging attacks. Firstly, to accurately and comprehensively evaluate routing behavior, the project will provide the anomaly detection technique based on multi-dimensions evidences correlation and the routing behavior evaluation method based on attack consequences. Secondly, to fuse historical routing behaviors to reflect prospective behavior, the project will provide the routing behavior prediction method based on the fusion theory of single source information in time domain. Thirdly, to make routing behavior prediction quickly converge, considering the “power-law” dynamic characteristic of the originating prefix routing announcement behavior, the project will provide the originating routing behavior dynamics modeling method based on machine learning and the behavior prediction evolution mechanism based on self-adaptive adjustment window, considering the dynamic interactions between neighbor autonomous systems under the restriction of the competition-cooperation relationship, the project will provide the propagating routing announcement behavior dynamics modeling method based on evolutionary game and the behavior prediction evolution method based on interaction policy evolution. This topic research will enrich methods to solve BGP security problem.