【作者】 金学波；

【导师】 孙优贤；

【作者基本信息】 浙江大学 ， 控制科学与工程， 2003， 博士

【摘要】 随着传感器技术、通信技术的发展，各种面向复杂应用背景的多传感器系统的研究越来越受到人们的关注。多传感器系统中，信息表现形式的多样性、信息数量的巨大性、信息关系的复杂性以及要求信息处理的及时性，都大大超出了人脑的信息综合处理能力，多传感器信息融合理论应运而生。 状态融合估计是多传感器信息融合理论的一个非常重要的研究领域，主要研究如何利用多个传感器的信息更准确地估计系统状态，在跟踪系统及需要精确估计的其他领域应用十分广泛。 本文的研究内容为多传感器信息融合理论中的状态融合估计理论，主要针对精确估计的实际应用中，状态融合估计理论存在的一些问题提出了解决方法。 针对分布式次优融合方法迭代计算复杂、无法用于多于三个传感器的系统的现状，提出了消除相关估计方差的分布式算法。该算法保持了分布式系统在结构上与计算上的优点，仿真实验表明，算法的性能与传统分布式次优融合算法的估计性能相近。 丰富和发展了目前的最优分布式、集中式融合方法，将最优融合估计算法推广到更一般的非标准多传感器系统中。针对存在控制输入、过程噪声与观测噪声相关且为非零均值高斯白噪声的多传感器系统，推导了二级、三级融合算法。算法形式更具一般性，很容易推广到更高级的多传感器系统。 定义并提出了多传感器系统的方差性能函数，从理论上严格证明了其与融合估计方差的关系，结论用仿真实验得到了验证。 研究了不确定多传感器系统的两种不确定描述模型及相应的集中式鲁棒融合估计方法，并通过仿真详细地比较了两种融合方法的频域、时域性能。从理论上严格证明了将集中式鲁棒融合估计转化为相同估计性能的分布式融合估计算法的条件。 论文系统的研究了各个测量传感器相关的多传感器融合系统。对于测量噪声相关矩阵是确定的，且该矩阵可以通过相似变换变成对角阵的多传感器系统，给出了最优集中式、分布式融合估计。对于其他的系统，给出浙江大学博士学位论文了分解一合并的融合估计方法。仿真实验表明，针对测量噪声相关的多传感器系统，分解一合并的融合估计方法可以得到比一般的参数不确定鲁棒融合估计方法更好的估计结果，具有更强的鲁棒估计性能。 关于多传感器信息融合理论在工业过程中的应用实际研究，论文将状态融合估计理论应用于精密纸机的成纸定量估计中。指出了使用多个传感器可以提高状态估计的性能，在部分传感器出现故障时仍然可以保证具有较好的估计效果。但如果多个传感器使用不当，出现了测量噪声相关的情况，若不能使用正确的状态估计方法，也会使系统估计性能下降。更多还原

【Abstract】 Due to the advent of the sensor technology and communication technique, multi-sensor systems have recently attracted considerable attention, especially, those with diversified complex using background. The emerging interest of research into multi-sensor information theory is viewed as timely since the multiformity, massiness, complex and real-time processing of information in multi-sensor system has gone beyond the human brain capacity of processing information.State fusion estimation is an important study field in the information fusion theory, mainly dealing with how to estimate the system state exactly by multi-sensors. It is usually applied in tracking system and other exact estimating systems.This dissertation considers state fusion estimation of multisensor information fusion theory. The main work of here is to solve the problems when fusion estimation theory is applied in practice.In details, the major contributions of this thesis are as the following:As ,we know, distributed suboptimal method need complex compute processing and can’t be used in the system containing more than 3 sensors. The optimized algorithm is developed, which avoids computing correlated estimate covariance and has the advantages of the distributed structure. Meanwhile, the simulations show the developed algorithm has the similar performance as the classical distributed suboptimal fusion method.The present optimal distributed and centralized fusion methods are enriched and expanded. Two-levels and three levels algorithms are discussed in the more general system, which has control input and in which the processing noise and measurement noise are correlated Gaussian white noise with nonzero mean. The discussed algorithms have more universal formats and are easilygeneralized to the more-levels system.The dissertation defines and develops the covariance performancefunction and proves that it can decide the fusion estimate covariance. The simulations identify the conclusions.The dissertation studies the fusion estimation of uncertainty multisensor system. Two uncertain models and corresponding centralized robust fusion estimate methods are discussed. The simulations compared different fusion methods detailedly in frequency and time fields. Moreover., it’s proved that with the exact transforming condition, robust centralized estimate can be transformed to the distributed fusion method with the same fusion estimate performance.The dissertation systemically studies the multisensor system with correlated measurement noise. When the measurement noise covariance is certain matrix that can be transformed to a diagonal matrix by matrix resemble transform, the dissertation develops optimal centralized and distributed fusion estimate. For the other systems, the decomposed-combined fusion estimation method is discussed. The simulations show that the developed algorithm can obtain a better performance than the general robust fusion estimation methods for the multiserisors system with correlated measurement noise.As to the application of information fusion theory, the state fusion estimation methods are applied in the basis weight estimation of exact paper machine. The paper studies several practical situations and points out that the estimate results can develop if more sensors are used, even though when some of them fail. But if the measurement noises are correlated, the estimation methods must be chosed correctly, or else the estimation performance may decline.更多还原