| 面向DVL缺失工况的水下导航观测补偿方法 |
| Underwater Navigation Observation Compensation Method for DVL Outage Conditions |
| |
| DOI：10.3969/j.issn.1003-2029.2026.03.001 |
| 中文关键词: 水下自主航行器 多普勒测速计 随机森林 因子图优化 |
| 英文关键词:AUV DVL Random Forest FGO |
| 基金项目:中北大学第二十一届研究生科研科技计划项目 （20252158）；山西省自然科学基金资助项目 （202303021211152）；东海实验室资助项目（L25QH003） |
|
| 摘要点击次数: 131 |
| 全文下载次数: 71 |
| 中文摘要: |
| 在水下自主航行器(Autonomous Underwater Vehicle，AUV) 导航过程中，通常依赖惯性导航系统 (Inertial Navigation System,INS) 和多普勒测速计 (Doppler Velocity Log, DVL) 融合定位。受水声环境复杂性和设备工作状态影响，DVL易出现数据缺失。针对DVL缺失条件下传统融合方法鲁棒性不足的问题，本文提出了一种数据驱动观测补偿与统一因子图融合的系统性思路，通过训练随机森林回归模型，学习航行器局部运动状态与DVL速度观测之间的非线性映射关系，进而预测得到“虚拟DVL观测”，并将其作为速度先验输入统一的因子图优化(Factor Graph Optimization，FGO)。在不改变因子图后端结构的前提下，通过提升软约束的观测质量改善导航系统在DVL缺失工况下的稳定性与精度。基于真实水下试验数据验证，在缺失后外部观测极为稀疏的最不利工况下，随机森林方法将终点误差从千米级缩小至百米量级，相较于不补偿方案，终点误差降低约98%。同时，随机森林在终点误差、全程均方根误差(Root Mean Squared Error，RMSE)和航程归一漂移率上较线性插值分别进一步降低约29%、14%和30%，表明其在抑制误差累积和保持全航程轨迹一致性方面具有更强的鲁棒性。 |
| 英文摘要: |
| During the navigation of an Autonomous Underwater Vehicle (AUV, positioning typically relies on the fiusion of an InertialNavigation System (INS) and a Doppler Velocity Log (DVL). Due to the complexity of the underwater acoustic environment and theoperational status of the equipment, DVIs are prone to data loss. To address the insufficient robustness of traditional fusion methods underDVL outage conditions, this paper proposes a systematic approach combining "data-driven olservation conpensation with unified factorgraph fusion." By training a Random Forest regression model, the model leams the nonlinear mapping relationship between the vehicle'slocal motion state and DVL velocity observations, This predicts a "virtual DVL observation" that seres as a prior velocity input for unifiedFactor Graph Optimization (FCO). Without allering the backend structure of the factor gtaph, this approach enhances navigation systenstability and accuracy during DVL outage conditions by improving the quality of sof -constnained observations. Based on real underwatertest data, under the most adverse scenario with extremely sparse external observations during DVL outages, the Random Forest methodreduces the terminal eor from the kiloneter scale to the hundred-meter scale, achieving a reduction of approximatdy 98% compared tothe uncompensated scheme. Additionally, compared to linear interpolation, the Random Forest approach further reduces the teminal error,full-course RMSE, and nomalized drift rate by approximately 29%, 14% and 30%, respectively. This demonstrates its superior robustnessin suppressing error accumulation and maintaining trajectory consistency throughout the entire voyage. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
|
|
|
