目标运动状态估计是雷达跟踪系统的核心任务之一。经典贝叶斯滤波器依赖准确已知的目标运动模型，面对目标机动、环境参数时变、存在未知扰动输入等模型不确定场景时，估计性能严重不足。平滑变结构滤波方法（smooth variable structure filter, SVSF）无需模型误差的先验分布，能够保证估计误差的有界性，同时具有复杂度低、与现有雷达跟踪系统兼容性好的优势，近年来倍受关注。论文针对模型不确定条件下的雷达目标鲁棒跟踪的迫切需求，深入研究了SVSF的状态估计性能提升方法，解决了抖振抑制、平滑层参数优化以及在多普勒雷达中应用的问题。论文主要贡献和创新点如下：1）针对SVSF的抖振问题，提出了一种基于修正切换函数的SVSF方法。理论分析了双曲正切函数相比于现有切换函数的抖振抑制优势。理论证明了所提方法的估计误差有界性。仿真和实测数据实验验证了所提方法显著提升了抖振抑制效果和雷达目标跟踪性能。2）针对SVSF的平滑层参数敏感性问题，提出了一种基于非线性广义时变平滑层策略的SVSF方法。该方法在修正切换函数基础上，采用自适应切换策略实时调整平滑层参数以适配动态变化的模型不确定度和噪声水平，推导了低模型不确定度条件下的最优平滑层参数，并保持了高模型不确定度条件下的鲁棒性和抖振抑制优势。仿真和实测数据实验验证了所提方法在前一项工作基础上进一步提升了雷达目标跟踪性能。3）面向多普勒雷达目标鲁棒跟踪的应用，提出了一种序贯SVSF方法，解决了现有SVSF无法应用于多普勒雷达非线性欠定观测模型的难题。在此基础上，针对特殊的匀速运动模型假设，提出了一种静态融合SVSF方法，降低了非线性模型近似误差，进一步提升了估计性能。仿真和实测数据实验验证了所提方法在多普勒雷达目标跟踪中的有效性和鲁棒性，以及所提静态融合SVSF相比于所提序贯SVSF进一步的性能提升。
Target state estimation is one of the core issues in radar tracking systems. The classical Bayesian filters rely on accurate target motion models and deteriorate seriously in case of the model uncertainty problem caused by maneuvering, time-varying environment parameters, or unknown disturbance inputs. The smooth variable structure filter (SVSF) guarantees the boundedness of estimation error without the prior distributions of modeling error and also features relatively low complexity and excellent compatibility in the existing radar tracking systems. Hence, the SVSF method becomes increasingly popular in recent researches. This thesis aims at the task of robust radar target tracking under model uncertainty and develops the enhanced smooth variable structure filters, addressing the problems of chattering suppression, parameter optimization, and the application to Doppler radar. The main contributions of this thesis are as follows:1) To solve the chattering problem, a novel SVSF method is proposed with a new switching function. Theoretical analysis validates the superiority of the hyperbolic tangent function in chattering suppression over the conventional switching functions. The error boundedness of the proposed method is proved. Both simulation and real-world radar data experiment substantiate that the proposed method improves the chattering suppression effect and the performance of target state estimation. 2) To solve the sensitivity to the smoothing boundary layer parameter, a novel SVSF method is proposed based on a nonlinear generalized variable boundary layer strategy. This method follows the first work and adjusts the smoothing boundary layer parameter adaptively to the modeling error and noise level. An optimal smoothing boundary layer parameter is derived in the low model uncertainty level case, while the robustness and the chattering suppression effect are maintained in the high uncertainty level case. Both simulation and real-world radar data experiment validate the improvement of the proposed method over the first work for radar target tracking.3) For Doppler radar application, a novel sequential SVSF is proposed to handle the nonlinear underdetermined measurement model which precludes the existing SVSF formulations. Furthermore, for the constant-velocity motion model, a statically fused SVSF is proposed to reduce the approximation error of the nonlinear measurement model and thus further improves the estimation accuracy. Both simulation and real-world radar data experiment validate the effectiveness and the robustness of the proposed methods for Doppler radar target tracking, and also substantiate the further improvement of the statically fused SVSF over the proposed sequential SVSF.
