惯性推算误差抑制的神经网络自学习模型设计与验证

惯性推算误差抑制是提升复杂场景下组合导航定位性能的关键，现有采用运动约束或系统误差高阶建模的方法从运动学模型及传感器误差模型出发，通过经验确定参数及模型的最优解。深度学习隐式模型能够挖掘数据之间的隐含关系，进行自主化参数寻优，并在提升惯导误差建模精度方面具有一定优势。总结了现有主流网络模型设计的优缺点，通过对比不同的输入输出方案进行优选，最终利用卷积神经网络构建了一套惯性误差抑制的轻量化神经网络自学习模型，并利用实测车载数据验证了该模型的有效性。实验结果表明，在GNSS信号失锁300 s的路段I和失锁285 s的路段II，网络模型速度约束的算法相较于纯惯性推算和传统NHC算法均有一定提升，融合NHC及网络模型速度约束的算法在水平定位精度上分别改善了41.7%~47.4%和26.7%~36.6%，一定程度上抑制了惯性推算误差。

A neural network self-learning model design and validation for inertial dead reckoning error suppression

Inertial dead reckoning error suppression is key to improving the performance of integrated navigation and positioning in complex scenarios. Most existing motion constraint or system error high-order modeling methods rely on kinematic models and sensor error models, with optimal model parameter solutions determined empirically. Deep learning implicit models can uncover implicit relationships within data, autonomously optimizing parameters, and offering advantages in enhancing the accuracy of inertial navigation error modeling. The article summarizes the advantages and disadvantages of the existing mainstream network model design, and by comparing different input and output schemes for preference, a set of lightweight neural network self-learning model for inertial dead reckoning error suppression is finally constructed using convolutional neural network. The model''s validity is verified using measured vehicle data. Experimental results demonstrate that the network model speed constraint algorithm has certain improvements compared with inertial dead reckoning and traditional non-holonomic constraint (NHC) algorithm. Specifically, when the GNSS signal is lost for 300 seconds in Section I and 285 seconds in Section II, integrating NHC with the network model''s speed constraints enhances horizontal positioning accuracy by 41.7% to 47.4% and 26.7% to 36.6%, respectively, effectively suppressing inertial dead reckoning errors.