广义未知扰动下上限滤波研究及其在目标跟踪中的应用

In complex battlefield environment, the coexistence of different factors such as system biases, deceive jamming, model approximation and many others, always leads to model mismatch between the prior nominal measurement equation and the ground truth. Such mismatch is depicted as the generalized unknown disturbance added to the nominal measurement model. Meanwhile, the system modeling has to face with different complex features, such as parameter multi-modal, additive/multiplicative noise, model nonlinearity, unknown noise parameters, and so on. In this proposal, a series of adaptive upper-bound filtering methods are developed, by considering data inaccuracy caused by model mismatch and modeling uncertainties due to complex features simultaneously. Firstly, the novel recursive minimum upper-bound filtering mechanisms are established for Markovian jump nonlinear systems and discrete-time nonlinear systems with multiplicative noises, in the case of generalized unknown disturbances. Secondly, considering the coexistence of noise parameter uncertainty and unknown disturbances, the joint framework of state estimation and unknown noise parameter learning is designed, along with the idea of upper-bound filtering. Thirdly, in multi-sensor cooperative tracking, a new centralized fusion estimation is put forward through multiple parameters optimization in the structure of upper-bound filtering. Then, the distributed information filtering-type implementation, which is based on maximizing the constructed lower bound of information matrix, is further proposed via average consensus to pursue the final consistent estimate. In this proposal, it is expected to effectively decrease the peak tracking error caused by multiple uncertainties through seeking the minimum upper-bound covariance of estimate error with the help of fast parameter optimization, in order to provide new principle and method for reliably, accurately and rapidly sensing and tracking.

复杂战场环境下，系统偏差、干扰欺骗、模型近似等多因素并存，导致量测的先验标称模型与实际模型失配，表征为量测标称模型上附加广义未知扰动。同时，参数多模态、加/乘性噪声共存、模型非线性、噪声参数未知、分布式架构需求等使系统建模面临多复杂特性影响。考虑模型失配导致的数据不准确与系统复杂特性导致的模型不确定往往共存，本项目拟研究并发展一系列自适应上限滤波机制。在广义未知扰动下，探索跳变马尔可夫非线性系统和带乘性噪声非线性系统的递推上限滤波新机制；针对噪声参数不确定与未知扰动共存，在递推上限滤波途径下实现状态估计与模型未知参数学习的联合处理；考虑多传感器协同跟踪，设计多参数凸优化下基于上限滤波的集中式融合，并给出极大化信息矩阵下限的类信息滤波分布式共识实现。本项目通过快速参数寻优以最小化估计误差协方差上限，从而抑制多不确定引起的跟踪峰值误差，为可靠、精确、实时的目标感知与跟踪提供新原理和新途径。

复杂战场环境下，系统偏差、干扰欺骗、模型近似等多因素并存，导致量测的先验标称模型与实际模型失配，表征为量测标称模型上附加广义未知扰动。同时，参数多模态、加/乘性噪声共存、模型非线性、噪声参数未知、分布式架构需求等使系统建模面临多复杂特性影响。考虑模型失配导致的数据不准确与系统复杂特性导致的模型不确定往往共存，本项目研究并发展了一系列自适应上限滤波机制。（1）围绕数据不准确与模型不确定耦合共存，提出面向跳变马尔可夫非线性系统的自适应上限滤波及基于多模型框架的上限滤波方法，实现未知扰动下机动目标跟踪；设计乘性噪声下非线性系统的高精度扩展上限滤波，完成广义未知扰动与乘性噪声共存下径向距雷达目标跟踪；建立有色量测噪声下随机动态系统的上限滤波，克服雷达目标跟踪过程中由于雷达高频采样导致的量测噪声多拍相关。（2）针对实际作战环境中模型噪声或量测噪声参数多变现象，构建了噪声参数未知下线性、非线性系统的自适应变分贝叶斯滤波，在有效处理广义未知扰动的同时，实现了状态估计与系统参数的联合优化。（3）围绕多平台多源雷达目标跟踪，考虑多源未知扰动的存在，分别提出上限滤波机制下多传感器单速率采样和多速率采样的融合估计；针对同步采样情况，设计了多参数凸优化下集中式融合估计，并利用信息滤波形式给出了分布式估计实现。（4）此外，围绕模型非线性问题，以L1范数取代L2范数，构造了基于L1范数的递推在线滤波，实现了非高斯噪声环境下点目标的高精度跟踪；基于量测非线性坐标转换，提出了结合随机矩阵和变分贝叶斯的机动扩展目标跟踪状态估计方法。本项目通过快速参数寻优以最小化估计误差协方差上限，从而抑制多不确定引起的跟踪峰值误差，为可靠、精确、实时的目标感知与跟踪提供新原理和新途径。

内容获取失败，请点击重试