考虑动力学模型系统误差补偿的智能车GNSS/IMU组合定位算法

An Algorithm Considering Kinematic Model Systematic Error Compensation for Intelligent Vehicle GNSS/IMU Integrated Positioning

为了解决智能车动态组合定位过程中,因动力学模型与实际模型之间存在偏差导致滤波精度下降的问题,针对智能车全球导航卫星系统(GNSS)/惯性测量单元(IMU)组合定位系统,结合非线性预测滤波(NPF)和自适应滤波的优点,提出了一种考虑动力学模型系统误差实时估计和补偿的自适应非线性预测滤波(ANPF)算法。首先,根据NPF算法原理,通过最小化预测观测残差与系统误差的加权平方和,估计动力学模型系统误差;其次,结合自适应滤波原理,利用状态预测残差向量构造自适应因子,设计了一种自适应扩展卡尔曼滤波(AEKF)算法,用于估计系统状态向量,并通过自适应因子抑制动力学模型系统误差和线性化误差对系统状态估计精度的影响,克服NPF对系统状态估计精度有限的缺陷;再次,对动力学模型系统误差的估计误差和由动力学模型系统误差引起的系统噪声的等效协方差阵进行了分析和推导,以补偿动力学模型系统误差对系统状态估计的影响;最后,通过车载GNSS/IMU组合定位系统试验,从算法精度、鲁棒性和实时性方面对提出的算法和其他滤波算法的性能进行了验证和对比分析。研究结果表明:提出的自适应算法继承了NPF算法简易性和高实时性的优点,同时克服了NPF算法估计精度有限的缺陷,具有较好的滤波解算精度,水平定位精度小于1.0 m,算法单次平均执行时间约为0.013 9 ms,在精度和实时性的平衡方面显著优于其他滤波方法。

To solve the problem of decreasing filtering accuracy owing to the deviation between the kinematic and actual models during intelligent vehicle dynamic integrated positioning, by combining the advantages of nonlinear predictive filtering(NPF) and adaptive filtering, an adaptive nonlinear predictive filtering(ANPF) algorithm that considers the real-time estimation and compensation of the kinematic model systematic error is proposed for intelligent vehicle global navigation satellite system(GNSS)/inertial measurement unit(IMU) integrated positioning systems. First, according to the principle of the NPF algorithm, the systematic error of the kinematic model is estimated by minimizing the weighted square sum of the predicted observation residual and the system error. Second, combined with the principle of adaptive filtering, an adaptive extended Kalman filtering(AEKF) algorithm is designed to estimate the system state vector by using the state prediction residual vector to construct an adaptive factor. The influence of the kinematic model systematic error and linearization error on the system state estimation accuracy was suppressed by the adaptive factor to overcome the defect of NPF on the limited system state estimation accuracy. Thirdly, the estimation error of the kinematic model systematic error and the equivalent covariance matrix of the system noise caused by the kinematic model systematic error are analyzed and derived to compensate for the influence of the kinematic model systematic error on the system state estimation. Finally, by performing experiments using the vehicle GNSS/IMU integrated positioning system, the performance of the proposed algorithm and other filtering algorithms are verified and compared from the perspectives of accuracy, robustness, and real-time performance. The results show that the proposed adaptive algorithm inherits the advantages of simplicity and high real-time performance of the NPF algorithm, overcomes the disadvantage of the limited estimation accuracy of the NPF algorithm, and has good filtering accuracy. The horizontal positioning accuracy is less than 1.0 m, and the average time of a single execution is approximately 0.013 9 ms, which is significantly better than the other filtering methods in terms of the balance between accuracy and real-time performance.