基于XGBoost算法的车载场景识别辅助组合导航研究

Research on integrated navigation assisted by XGBoostalgorithm for vehicle scene recognition

现有组合导航算法在卫星信号复杂的环境中存在误差随时间发散的问题。针对不同的车载场景增加不同的辅助策略,以提升导航精度,提出了一种基于XGBoost算法的车载场景识别辅助组合导航技术。首先,根据行车过程中的卫星导航数据和车辆状态数据构建特征,并通过Kruskal-Wallis检验比较不同车载场景下特征的分布差异;其次,使用XGBoost算法拟合经过预处理的数据,得到车载场景识别模型;最后,当识别到地库场景时,通过力学编排计算航向角的变化量,也使用轮速运动学模型计算航向角的变化量,然后对两种方式计算的航向角变化量求均值,重新计算当前时刻的姿态,再用新的姿态更新速度和位置。实验结果表明,在地库场景下,相较于不增加轮速运动学辅助的算法,增加辅助的组合导航算法的航向精度标准差平均提升55.27%。

The existing integrated navigation algorithms suffer from the problem of error divergence over time in complex satellite signal environments.To improve the navigation accuracy by adding different auxiliary strategies for different driving scenarios,an integrated navigation supported by XGBoost algorithm for vehicle scene recognition is proposed.Firstly,features are constructed based on satellite navigation data and vehicle status data during driving,and the distribution differences of features in different vehicle scenarios are compared by Kruskal-Wallis test;secondly,the XGBoost algorithm is used to fit the preprocessed data and obtain a vehicle scene recognition model;finally,when the underground storage scenario is recognized,the change in heading angle is calculated by mechanical arrangement,the wheel speed kinematic model is also used to calculate the change in heading angle,and then the average of the changes in heading angle calculated by the two methods is calculated.The current attitude is recalculated,then the speed and position are updated with the new attitude.The experimental results show that in the underground storage scenario,compared to the algorithm without adding wheel speed kinematic assistance,the standard deviation of the heading accuracy of the integrated navigation algorithm with added assistance increases by an average of 55.27%.