Scholarlink

To address the limitation of Kalman Filtering (KF) in handling non-Gaussian deviations, this study proposes one-step ultra-wideband (UWB) localization correction models based on General Regression Neural Network (GRNN) and Back-Propagation Neural Network (BPNN). These models refine UWB coordinates after KF processing, using Global Navigation Satellite System Real-Time Kinematic (GNSS RTK) data as the reference for dynamic scenarios. The models’ effectiveness is evaluated using Bias, standard deviation, and root mean square error. Results indicate that artificial neural networks offer a fast and effective approach for UWB error correction. This method is both quicker and more straightforward, as it doesn’t need to eliminate distances deviations and serve to realize KF in hardware way, utilize the attributes of High concurrency and fast performance of hardware. The correction models considerably enhance the accuracy of both dynamic and static positioning coordinates. Notably, the BPNN-based model outperforms the GRNN-based model, achieving an average dynamic positioning accuracy of 10.018 cm and centimeter-level precision for static positioning.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Kalman Filtering and Artificial Neural Network Combined Model for UWB Optimization

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Zeen Cai,
  • Zhenmin Li

摘要

To address the limitation of Kalman Filtering (KF) in handling non-Gaussian deviations, this study proposes one-step ultra-wideband (UWB) localization correction models based on General Regression Neural Network (GRNN) and Back-Propagation Neural Network (BPNN). These models refine UWB coordinates after KF processing, using Global Navigation Satellite System Real-Time Kinematic (GNSS RTK) data as the reference for dynamic scenarios. The models’ effectiveness is evaluated using Bias, standard deviation, and root mean square error. Results indicate that artificial neural networks offer a fast and effective approach for UWB error correction. This method is both quicker and more straightforward, as it doesn’t need to eliminate distances deviations and serve to realize KF in hardware way, utilize the attributes of High concurrency and fast performance of hardware. The correction models considerably enhance the accuracy of both dynamic and static positioning coordinates. Notably, the BPNN-based model outperforms the GRNN-based model, achieving an average dynamic positioning accuracy of 10.018 cm and centimeter-level precision for static positioning.