【作者】 李云；

【导师】 孙书利；

【作者基本信息】 黑龙江大学 ， 微电子学与固体电子学， 2018， 博士

【摘要】 随着现代控制系统规模不断扩大、复杂性不断提升,使得传统感知设备以及处理方式远远不能满足人们对系统精确、全方位认知需要的不断增加。信息融合理论和技术在此背景下应运而生,经过几十年的发展,线性多传感器系统信息融合估计逐渐形成一套比较完善的理论体系和研究方法。但对于非线性多传感器系统,由于非线性环节的复杂性和不确定性,使得非线性多传感器系统信息融合算法还没有得到系统的解决。非线性多传感器信息融合至今仍然是信息融合领域的主要问题和研究热点。本文研究了非线性多传感器系统的加权观测融合估计问题,主要研究内容如下:1.针对带有独立噪声的非线性多传感器系统,通过引入中介函数,使各个观测方程可由线性矩阵和中介函数相乘得到,再利用加权最小二乘法(Weighted Least Square,WLS),提出了一种非线性加权观测融合(Weighted Measurement Fusion,WMF)算法。该算法可降低集中式融合系统的观测方程维数,实现集中式融合系统的数据压缩,减少后续估计等环节的计算负担。本文通过Taylor级数构造了多项式形式的近似中介函数,使该算法得以实现。在此基础上,基于Taylor级数逼近的WMF算法和无迹Kalman滤波器(Unscented Kalman Filter UKF),设计了一种适用于非线性高斯系统的加权观测融合UKF算法,并证明了该算法的渐近最优性,即随着Taylor级数展开项的增加,该算法渐近等价于集中式融合UKF算法。进一步,基于Taylor级数逼近的WMF算法和粒子滤波器(Particle Filter,PF),又给出了一种加权观测融合PF算法,该算法可处理高斯或非高斯噪声情况下非线性多传感器系统的加权观测融合估计问题。2.针对带有独立噪声的非线性多传感器系统,基于Gauss-Hermite逼近方法,提出了另外一种具有普适性的加权观测融合算法。该算法利用Gauss函数和Hermite多项式构造中介函数。为了降低计算负担,本文采用了分段处理方法,将状态区间进行分段逼近,并离线计算每段的加权系数矩阵,形成数据库。与基于Taylor级数逼近的方法相比,该算法不需要在线计算加权系数矩阵,可减少在线计算负担。同时,本文采用加权观测融合算法对增广的高维观测进行压缩降维,有效降低了实时估计算法的计算量。基于Gauss-Hermite逼近方法的WMF算法和UKF,设计了加权观测融合UKF算法。基于Gauss-Hermite逼近方法的WMF算法和PF,又给出了一种加权观测融合PF算法。3.针对带有相关噪声的非线性多传感器系统,系统噪声和观测噪声在同时刻互相关,首先,利用去相关方法进行模型转换,将系统噪声和观测噪声相关的非线性系统转化成噪声不相关的非线性系统。然后,基于Taylor级数逼近方法,分别结合UKF滤波算法和PF滤波算法,设计了加权观测融合UKF算法和PF算法。再次,基于Gauss-Hermite逼近方法,分别结合UKF滤波算法和PF滤波算法,提出了加权观测融合UKF算法和PF算法。更多还原

【Abstract】 With the continuous enlargement of modern control system scale and the continuous improvement of system complexity,the traditional sensing devices and processing methods are far from meeting people’s demands for accurate and comprehensive cognitive needs.Information fusion theory and technology come into being in this background.After decades of development,some relatively complete theories and methods are gradually formed for linear multisensory systems information fusion estimation.However,for nonlinear multisensor systems,the nonlinear multisensor systems information fusion algorithm has not been resolved because of their complexity and uncertainty.Multisensor information fusion for nonlinear systems is still a major issue and research hotspot in the field.The weighted measurement fusion estimation for nonlinear multisensor systems is studied in this paper.The main research contents are as follows:1.For nonlinear multi-sensor systems with independent noise,the mediation function is introduced so that each measurement equation can be reformed by the linear matrix and the mediation function.Then a weighted measurement fusion(WMF)algorithm is presented by using weighted least squares(WLS)method.The proposed algorithm can reduce the dimension of the measuremnt equation of the centralized fusion system,and compress the data of the centralized fusion system,and reduce the computational cost of estimation and other works.In this paper,polynomial approximate mediation functions are constructed by Taylor series,which makes the algorithm implemented.On this basis,based on the Taylor series approximation WMF algorithm and Unscented Kalman filter(UKF),a weighted measuremnt fusion UKF algorithm is designed for nonlinear Gaussian systems,and the asymptotic optimality of the algorithm is proved,that is,with the increasing of Taylor series the algorithm is asymptotically equivalent to the centralized fusion UKF algorithm.Further,based on the Taylor series approximation WMF algorithm and particle filter(PF),a weighted measurement fusion PF algorithm is also presented.This algorithm can deal with the weighted measurement fusion estimation for nonlinear multisensor systems with Gaussian or non-Gaussian noise.2.For nonlinear multisensor systems with independent noise,based on the Gauss-Hermite approximation method,another universal weighted measurement fusion algorithm is proposed.The proposed algorithm uses Gaussian function and Hermite polynomial to construct the mediation function.In this paper,in order to reduce the computational burden,a segmentation method is used to segment the state intervals and calculate the weighted coefficient matrix of each segment offline,then a database is formed.Compared with the weighted measurement fusion algorithm based on Taylor series approximation,this algorithm does not need to calculate the weighted coefficient matrix online.So it can reduce the online computational burden.At the same time,the weighted measurement fusion algorithm is used to compress and reduce the augmented high-dimensional measurements,which effectively reduces the computational complexity of the real-time estimation algorithm.Based on the Gauss-Hermite approximation WMF algorithm and UKF algorithm,the weighted measurement fusion UKF algorithm is designed.Based on the Gauss-Hermite approximation WMF algorithm and PF,a weighted measurement fusion PF algorithm is also given.3.For a nonlinear multisensor system with correlated noise,the system noise and the measurement noise are correlated at the same time stamp.First,using a decorrelation method,a nonlinear multisensor system with correlated noise are transformed into a nonlinear system with independent Gaussian noise.Then,based on the Taylor series approximation method,combined with UKF algorithms and PF algorithms,the weighted measurement fusion UKF algorithm and PF algorithm are designed.Once again,based on the Gauss-Hermite approximation method,combined with UKF algorithms and PF algorithms,the weighted measurement fusion UKF algorithm and PF algorithm are proposed.更多还原