超宽带（UWB）技术作为新一代室内定位技术的典范，在实际应用中常结合惯性导航系统（INS）以解决定位中的非视距（NLOS）误差问题。但集中式信息处理方法无法有效区分NLOS误差来源，为保证定位精度需额外部署锚点，导致定位锚点出现冗余，进而造成信息浪费及成本增加。针对室内定位中的NLOS误差识别和剔除问题，该文提出了一种基于自复位遗传粒子滤波（SGPF）的UWB/INS室内定位方法。该方法以SGPF算法为核心，通过INS估计值对测量值中的NLOS误差进行溯源，以提高NLOS环境下的跟踪稳定性。该方法首先对物理锚点进行分组，并结合虚拟锚点划分似然区域；然后基于INS的初步估计，通过NLOS误差识别策略确定高概率区域，同时剔除NLOS锚点组及对应的测量值；最后结合有效粒子数判别粒子集状态，决定是否启用遗传重采样以优化粒子多样性，最终提升算法鲁棒性。SGPF算法融合了标准粒子滤波（PF）和遗传算法的结构优势，可有效缓解粒子退化与贫化问题，在更低的粒子数量与时耗下实现更高的鲁棒性。实验结果表明：在视距环境下，SGPF算法只需PF算法30%的粒子数即可达到同等定位效果，且其计算时耗远低于传统遗传粒子滤波算法；在非视距环境下，SGPF算法的平均定位误差为0.055 2 m，相比于传统粒子滤波与传统遗传粒子滤波算法分别降低了56.98%与48.94%。

As a paradigm of the new-generation indoor positioning technology, ultra-wideband (UWB) technology is often combined with the inertial navigation system (INS) in practical applications to solve the non-line-of-sight (NLOS) error issue in positioning. However, the centralized information processing method fails to effectively distinguish the sources of NLOS errors. To ensure positioning accuracy, additional anchor nodes need to be deployed, which leads to redundancy of positioning anchor nodes, and further results in information waste and increased costs. Aiming at the problems of NLOS error identification and elimination in indoor positioning, this paper proposed a UWB/INS indoor positioning method based on self-reset genetic particle filtering (SGPF). With the SGPF algorithm as its core, this method traces the source of NLOS errors in measured values using the estimated values of the INS system, so as to improve the tracking stability under NLOS environments. The method first groups physical anchor nodes and divides likelihood regions in combination with virtual anchor nodes. Then, based on the preliminary estimation of the INS, it identifies high-probability regions through an NLOS error identification strategy, while eliminating NLOS anchor node groups and their corresponding measured values. Finally, it judges the state of the particle set by combining the number of effective particles, determines whether to enable genetic resampling to optimize particle diversity, and ultimately improves the robustness of the algorithm. The SGPF algorithm integrates the structural advantages of the standard particle filter (PF) and genetic algorithms, and can effectively alleviate the problems of particle degradation and impoverishment and achieve higher robustness with a smaller number of particles and lower time consumption. Experimental results show that: under line-of-sight environments, the SGPF algorithm requires only 30% of the number of particles used in the PF algorithm to achieve the same positioning effect, and its calculation time is much lower than that of the traditional genetic particle filter algorithm; under NLOS environments, the SGPF algorithm has an average positioning error of 0.055 2 m. Compared to traditional particle filter and traditional genetic particle filter algorithms, the localization error is reduced by 56.98% and 48.94% respectively.