【作者】 王国庆；

【导师】 夏国清；

【作者基本信息】 哈尔滨工程大学 ， 控制理论与控制工程， 2017， 博士

【摘要】 导航系统是水面无人艇(USV)在各类复杂海况下,安全、自主、有效的完成不同类型作业任务的重要保障。USV相对较低的研发成本和有限的载荷能力,决定了其导航系统可配置的传感器需要在成本、重量、体积、功耗和精度等指标之间做权衡。微机电惯性传感器(MEMS-IMU)具有低成本、微重量、低功耗等特点,但是其精度相对较低。研究基于MEMS-IMU技术的INS/GNSS组合导航状态估计滤波方法对于低成本、低载荷、高导航精度要求的USV具有重要意义。对基于MEMS-IMU的USV导航系统状态估计,存在以下需要解决的问题:1.由于MEMS-IMU相对较低的测量精度进一步加重了INS/GNSS紧耦合组合导航系统的非线性程度,需要针对具有强非线性特性的MEMS-IMU/GNSS导航系统设计更加有效的非线性状态估计算法;2.GNSS测量噪声统计通常无法精确已知,且存在野值等受污染情况,需要滤波算法实时对系统测量噪声统计特性做出反应,适时调节系统参数,保障滤波算法收敛;3.现有的状态估计方法大都是基于贝叶斯理论或其改进的滤波方法,结构复杂,形式单一,且其稳定性和收敛性的证明。需要寻求计算法复杂度低、运算效率高,且稳定性能够保障的状态估计方法。基于以上分析,本文主要研究工作有以下几个方面:针对基于MEMS-IMU的INS/GNSS紧耦合组合导航系统的强非线性问题,提出一种基于贝叶斯估计理论的改进的混合粒子滤波算法框架。将贝叶斯后验状态估计作为粒子滤波采样的重要性概率密度函数,粒子无需重复使用,避免了粒子退化问题和粒子重采样过程,在保障了滤波精度的同时,有效减少同等精度下所需粒子个数。针对INS/GNSS紧耦合组合导航系统测量噪声协方差统计特性存在不准确、突变等问题,提出一种基于残差序列的自适应无迹粒子滤波算法。通过仿真,验证了当GNSS伪距和伪距率测量噪声方差存在突变时,AUPF算法能够较为准确的跟踪噪声方差变化趋势,从而获得较高的滤波精度。针对GNSS测量噪声协方差矩阵存在未知、时变、受扰动和污染等问题,基于变分贝叶斯估计理论和Huber鲁棒估计理论,提出一种变分贝叶斯鲁棒自适应无迹粒子滤波算法。通过逆Wishart分布近似逼近系统测量噪声协方差矩阵分布,推导出基于高斯分布假设条件下的变分贝叶斯自适应无迹卡尔曼滤波算法(VBAUKF)。考虑到GNSS测量值存在野值等受污染情况,将Huber估计与VBAUKF算法相结合,提出一种VBRAUKF算法,并结合本文所提出的的混合粒子滤波算法框架,提出了VBRAUPF算法。仿真结果表明,当系统测量噪声统计特性存在时变、受扰动或污染等情况时,VBRAUPF能够有效解决INS/GNSS紧耦合组合导航系统滤波发散问题,同时增强了系统的鲁棒性,具有较高的估计精度。将USV多传感器组合导航状态估计问题转换为非线性内反馈级联系统状态估计问题,提出了一种基于辅助虚拟高程测量的变增益非线性观测器。基于非线性观测器设计理论,由Riccati方程计算观测器增益矩阵,通过Lyapunov分析证明了所设计的观测器的稳定性。仿真结果表明,姿态观测器提供四元数姿态向量估计的同时,能够提供陀螺仪偏差估计;辅助虚拟高程测量的引入改善了USV垂向的位置和速度估计精度。基于非线性观测器理论的估计方法,为USV多传感器组合导航状态估计提供了一种新的解决思路。当USV低速运行时,针对一阶波浪力引起的波频运动经控制回路反馈易对执行机构带来不必要耗损的问题,提出了一种不基于船舶动力学模型的自适应海浪滤波观测器。设计海浪遭遇频率估计算法,并基于该估计设计自适应陷波滤波器。由陷波滤波器的输出和导航系统非线性观测器输出重构USV水平方向低频运动状态量。更多还原

【Abstract】 As a most vitial component of Unmanned Surface Vehicle(USV),the navigation system is a guarantee for the USV to efficiently accomplishdifferenttasks on their own saftly in various complex sea conditions.The navigation system of the USV has to balance cost,size,mass,power and accuracy for its low development costs and limited capability of payloads.Micro-electromechanical systemInertial measurement Units(MEMS-IMU)are marked by micro-weighte,low power consumption and cost-effective,but low accuracy at the same time.The research of INS/GNSS filtering algorithm based onMEMS-IMU plays an important role in USVs with low-cost,low-load,and high accuracy requirements.For the state estimation based on MEMS-IMU of USV navigation syste,there are following problems need to be solved: Firstly,the low measurement accuracy intensify the nonlinearity of INS/GNSS integrated navigation,more powerful nonlinear state estimation algorithm is needed to solve such problem;Secondly,the statistic characteristics of the GNSS measurements have contaminated disstribution or outliers,and cannot be accurately obtained,an online statistic characteristics estimation algorithm for GNSS measurements is needed;Thirdly,commonly used methods for state estimation of navigation system are Bayesian theory based and specially tailored variants.While the Bayesian methods has found wide applicability,it does not come without some drawbacks.This includes relatively high computational cost and a rather implicit and not easily verifiable convergence properties.Based on the analysis above,the main contents of this paper are as follows:As to the nonlinear non-Gaussian characteristic of the INS/GNSS tightly coupled system,a framework of Bayes estimation theory based hybrid particle filter algorithm is proposed.This algorithm first obtains posterior state estimates based on Bayes filtering,then regard these distribution as importance density function for generating particles.As theparticles are not reused,it will avoidthe problems such as particle degeneration and re-sampling.The algorithm guarantees high estimation accuracy and less particles.The covariance of measurement noise has inaccurate statistical properties,and may has sudden change,an adapting unscented particle filtering algorithm based on residual sequence is proposed.By simulation,we proved that when the variance of measurements noise of pseudorange and pseudorange rate have sudden changes,AUPF algorithm may track and capture the trend of the noise variance,thus obtaining higher filtering accuracy.When the measurement noise covariance of GNSS is unknown,time-varying,disturbed or there are outliers in the measurements,variational Bayesian robust adaptive unscented particle filter algorithm based on variational Bayesian estimation theory and Huber robust estimation theory is proposed.By using inverse Wishart distribution to approximate the covariance distribution of the measurement noise,Gaussian distribution assumation-based variational Bayesian adaptive unscented Kalman filtering algorithm(VBAUKF)is proposed.Consider that GNSS measurement may be disturbed by outliers,we propose VBRAUKF algorithm by combining VBAUKF algorithm with Huber estimation.Then by combining the Hybrid particle filter algorithm framework proposed before,the algorithm VBRAUPF is derived.Simulation results show that,when the statistical property of the measurement noise is time-varying,or the noise is disturbed or there are outliers in the measurements,VBRAUPF can efficiently solve the problem of filtering divergence in INS/GNSS tightly coupled integration systems,enhancing the robustness of the system.As to the multi-sensor INS/GNSS/Magnetometer integrated navigation state estimation problem,aiding virtual vertical measurement(AVVM)based nonlinear observer is proposed.First transfer the state estimation of multi-sensor integration problem into the problem of state estimation in nonlinear feedback-interconnectedsystem.Then calculate observer gain matrix with Riccati equation based on nonlinear observer theory.The stability of the observer can be analyzed through Lyapunov theory.The simulation results show that the attitude observer can provide both quaternion estimation as well as the bias estimation of the gyroscope.By introducing AVVM,the estimation accuracy of vertical position and speed of a USV can be improved.When a USV sails at a low speed,the first order wave force may cause the ware down of the actuators.In order to solve this,this thesis proposed an adaptive wave filter observer which is not based on the dynamic model of a ship.First derive estimation algorithm for the wave encounter frequency,then design an adaptive notch filter based on the estimation.The output of the notch filter,accompanied with the output of nonlinear observer of the navigation system,can then be used to reconstruct the states of low frequency in the horizontal plane.更多还原