【作者】 马静；

【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， 2012， 博士

【摘要】 随着网络通讯技术和计算机技术的飞速发展，大量的网络化控制系统随之应运而生。网络传输方式被广泛应用于国防建设和国民经济的各个方面，如军事、工业、医学、交通、航天等领域。但由于网络的承载能力和通信带宽有限，以及遮挡和干扰等因素的影响，网络系统中不可避免地存在时间延迟、数据包丢失和多速率采样等问题。这些问题在传统的控制系统中是不存在的，因此基于传统的估计理论给出的估计策略已不能直接应用到网络化控制系统中，必须针对网络化控制系统的复杂结构和所存在的上述特殊问题，研究适合网络环境的估计策略。在这样的背景下，本文综合考虑以上问题，基于射影理论和线性最小方差意义下的最优融合算法，研究网络化控制系统的建模和融合估计问题，主要内容和创新点如下：对观测数据包存在丢失的多传感器线性离散时变随机系统，提出了分布式和集中式信息融合估值器，包括滤波器、预报器和平滑器。推导了任两个局部估计误差之间的互协方差矩阵。对于线性定常系统，给出了分布式和集中式稳态估值器存在的一个充分条件。证明了任两个局部估计误差之间的互协方差阵的稳态解的存在性。进一步，对观测数据包和控制数据包同时存在丢失的多传感器线性离散时变随机系统，提出了集中式最优和稳态信息融合估值器。对带有随机乘性噪声不确定性和观测数据包丢失的多传感器离散时变随机系统，设计了局部单传感器子系统的估值器、分布式和集中式融合估值器。同时给出了稳态估值器存在的一个充分条件。对带有随机乘性噪声不确定性和不确定观测的多传感器离散随机时变系统，提出了集中式最优和稳态信息融合估值器。与目前文献所提出的鲁棒滤波器相比，不需要人为选择抗干扰噪声的系数矩阵和方差阵，并且所提出的滤波器在线性最小方差意义下是最优的。对同时带有随机传感器滞后、数据包丢失和不确定观测的多传感器线性离散时变随机系统，利用三个伯努利分布的随机变量，建立了统一的观测模型来描述这三种随机现象。进而基于所建立的观测模型对单传感器系统在现有文献的增广方法下提出了线性最优估值器。与文献中滤波器相比，由于考虑了增广的系统噪声是同时刻和相邻时刻都相关的有色噪声，因此在大滞后率下提高了估计精度。但有色噪声的存在导致滤波器的结构过于复杂，不便于实时应用。因此，我们提出一种新的增广方法，并基于这种增广方法，设计了一种新的线性最优估值器。与前面所提出的滤波器相比，由于建立了一个低阶的参数化系统并且避免了有色噪声的引入，因此所提出的估值器具有更简洁的形式和较小的计算负担。同时将目前所研究的问题推广到多传感器情形，基于新的增广方法，提出了分布式和集中式信息融合估值器。对多速率多传感器线性离散随机系统，其中状态具有最高的状态更新率，不同的传感器具有不同的较低的观测采样率，通过建立第i个传感器的观测采样点上的状态空间模型，给出了状态更新点上的局部状态滤波器。利用次优加权融合算法和协方差交叉融合算法，提出了状态更新点上的分布式次优融合滤波器。与状态增广方法相比，明显减小了计算负担。对带有不同观测数据包丢失率的多速率多传感器线性离散随机系统，通过引入伪观测序列将多速率系统的融合估计问题转化为等价的单速率系统的融合估计问题。进而，基于数据包丢失的实现过程，提出了分布式最优融合滤波器，推导了任两个局部估计误差之间的互协方差阵。与没有考虑互协方差阵的方法相比，可明显改善估计精度。通篇以Matlab仿真软件为平台，通过与已有文献结果的比较仿真研究，验证了所提算法的有效性。更多还原

【Abstract】 With the rapid development of network communication technology and computertechnology, a large number of networked control systems come into being. Networktransmission is widely used in various aspects of national defence and nationaleconomy such as military affairs, industry, medical science, traffic, aerospace and otherareas. However, in networked control systems, time delays, packet dropouts andmultirate sampling are unavoidable due to the limited carrying capacity,communication bandwidth of the network, the shelter and interference factors. Intraditional control systems, all the issues do not exist, which results that the estimationstrategies based on the conventional estimation theory can not be directly applied tothe networked control system. For the complicated structure and above special issuesof the networked control system, we must explore the estimation strategies that aresuitable to the network environment. Under this background, based on the projectiontheory and the optimal fusion algorithms in the linear minimum variance sense, thispaper is concerned with the above problems and investigates the modeling and thefusion estimation problem for networked control systems. The main contents andinnovations are as follows:For multi-sensor linear discrete time-varying stochastic systems with measurementpacket dropouts, the distributed and centralized information fusion estimatorsincluding filter, predictor and smoother are presented. The cross-covariance matricesbetween any two local estimation errors are derived. For linear time-invariant systems,a sufficient condition for the existence of the steady-state estimators is given. Theexistence for the steady-state solution of the cross-covariance matrices between anytwo local estimation errors is proven. Further, the centralized optimal and steady-state information fusion estimators are proposed for multi-sensor linear discretetime-varying stochastic systems with measurement and control packet dropouts.For multi-sensor linear discrete time-varying stochastic systems with randommultiplicative noises uncertainties and measurement packet dropouts, the local filtersfor single sensor system, distributed and centralized fusion estimators are designedrespectively. A sufficient condition for the existence of the steady-state estimators isgiven. Furthermore, the centralized optimal and steady-state information fusionestimators are obtained for multi-sensor discrete time-varying stochastic systems withrandom multiplicative noises uncertainties and uncertain observations. Compared tothe robust filter of the existing reference, our filter does not need to select artificiallythe coefficient matrices and variance matrices of the anti-interference noises. Also, theproposed filter is optimal in the linear minimum variance sense.The unified measurement model is established to describe the three stochasticphenomena by using three Bernoulli distributed random variables for linear discretetime-varying stochastic systems with multiple sensors subject to random sensor delays,multiple packet dropouts and uncertain observations. Based on the establishedmeasurement model, the linear optimal estimators are presented for a single sensorsystem by using the augmentation approach of the existing references. Compared tothe estimators of the existing literatures, the estimation accuracy is improved underlarge delay rate since the correlation of the augmented process noise at the same andadjacent moments is taken into account. However, the obtained filter is complicatedand unsuitable for real-time application due to the existence of the colored noises.Hence, a new augmentation approach is given. Based on it, the new local linearoptimal estimators are designed. The proposed estimators have the simpler forms andthe lower computational cost compared to the above estimators since a lowerdimension parameterized systems is constructed and the colored noise is avoided.Moreover, we generalize the results to the case of multi-sensor systems and the distributed and centralized information fusion estimators are given based on the simpleaugmentation approach.By establishing the state space model at the measurement sampling points of theith sensor, the local state filters are given at the state update points for multiratemulti-sensor linear discrete time stochastic systems, where the state has the highestupdate rate and different sensors have different lower measurement sampling rates. Inaddition, the distributed suboptimal fusion filter at the state update points is proposedby applying the suboptimal weighted fusion algorithm and covariance intersectionfusion algorithm. Compared to the state augmentation approach, the computationalcost is obviously reduced. For multirate multi-sensor linear discrete stochastic systemwith different packet dropout rates, the fusion estimation problem for multirate systemsis transformed into the equivalent fusion estimation problem for a single rate system byintroducing the pseudo measurement sequences. Further, based on the realization of themeasurement packet dropouts, the distributed optimal fusion filter is proposed. Thecross-covariance matrices between any two local estimation errors are derived.Compared to the non-state augmentation approach, the estimation accuracy issignificantly improved since all the measurement information is used at each stateupdate point.Throughout this paper, the MATLAB software is used to do the simulationresearch. The simulation studies show the feasibility and effectiveness of the proposedalgorithms by comparing to the existing results.更多还原