基于协同式情景进化的高频地波雷达目标跟踪算法高效测评方法研究

Tracking the targets above the sea surface is the major function of any HFSWR systems, which is vital to national maritime security. Therefore, it is necessary to test and assess the performance of the target tracking algorithm. However, challenges have been observed, which include that experiments on the sea is very expensive, simulation-based testing fails to cover the vast uncertainty space of the target tracking algorithm, and the target tracking performance is significantly affected by the parameter settings of the target tracking algorithm. As a result, there exists a lack of effective research in testing and assessment to the target tracking algorithm. In this proposal, we focus on the HFSWR target tracking algorithm and study on an effective testing and assessment approach for HFSWR target tracking algorithm based on scenario and parameters coevolution. Concise, diversified and reliable scenarios are firstly generated as the testing inputs using a novel PTAG-guided event-based hybrid scenario population generation method. Vulnerable scenarios can then be searched from the randomly generated scenarios using a novel TAG3P-based scenario evolution method. Moreover, a novel adaptive competitive co-evolutionary algorithm is built to detect the fine flaw and space of improvement of the HFSWR target tracking algorithm, with a testing and assessment report provided. The testing and assessment approach in this proposal can be verified by conducting simulation comparison experiments and real HFSWR target detection data verification experiments. This research will provide an effective testing and assessment approach as well as a decision support tool for building reliable HFSWR target surveillance and tracking system.

海上目标跟踪是高频地波雷达探测的主要功能，对于我国海上安全有至关重要的作用，因此对目标跟踪算法进行高效的测试评估势在必行，但目前仍因实测成本高、仿真测试不全面且跟踪性能受参数配置影响大，导致目标跟踪算法的测评研究存在较大不足。本项目以高频地波雷达目标跟踪算法为对象，对其测评方法开展系统性的研究，探索基于协同式情景进化的目标跟踪算法高效测评方法。提出基于PTAG语法的事件型混合式目标跟踪情景种群产生方法，实现自动产生简洁、多样且可信的目标跟踪情景，巧妙运用基于TAG3P的目标跟踪失效情景的进化方法，从随机产生的情景中发现目标跟踪算法的失效情景，并在现有竞争式协同进化算法中创新性地引入自适应协同策略，从失效情景中发现算法的本质缺陷并确定算法的优化配置集，形成测评报告。通过仿真对比实验及实测数据验证理论方法的有效性。本研究将为建立完善的高频地波雷达目标监测系统提供高效的测评方法及决策支持工具。

高频地波雷达（HFSWR）能够实时、连续地探测和跟踪超视距的舰船目标。当舰船在某一海域航行时，其在某一时间段内的运动形成一幅情景。该运动情景的多样性和复杂性使得目标跟踪系统难以准确地跟踪到这些舰船，常见的跟踪错误有航迹断裂等。然而，目前还不清楚舰船的运动和其他干扰因素对跟踪性能的影响，尤其是哪些运动模式会导致跟踪失败。本项目针对这一问题，提出了一种基于协同式情景进化的方法，旨在为舰船目标跟踪系统提供详尽的定量性能评估和脆弱性检测方案。使用基于语法的模型生成多个机动舰船目标的运动情景作为测试输入，利用进化计算引入闭环反馈，有效地收集到能导致跟踪性能越来越差的情景，为采用数据挖掘技术分析并发现跟踪漏洞提供了多样化的测试数据集。在对领域内典型的常用高频地波雷达目标跟踪算法进行了详尽的测试和对比实验后，发现不同算法在跟踪性能上的差异，获得了目标跟踪算法大量多样化且有针对性的测试数据集，以及可量化的细致化的本质缺陷。实验结果表明，舰船目标运动时的聚集和交汇模式越多、聚集和护航模式的持续时间越长，目标跟踪算法越容易出现航迹断裂的情况，跟踪效果越差。