分簇传感网络鲁棒融合Kalman滤波

发布时间：2018-03-17 21:01

本文选题：分簇传感网络　切入点：加权观测融合　出处：《黑龙江大学》2015年硕士论文　论文类型：学位论文

【摘要】：近年来,随着传感网络技术的发展,网络估计和控制问题引起了广泛关注。在网络系统中,不可避免的存在数据包丢失、观测滞后、模型参数和噪声方差不确定性等问题。周知,传统Kalman滤波要求精确已知系统的数学模型,这些问题使得基于传统的Kalman滤波的状态估计面临着较大的挑战,因此,在网络监视和控制系统中,不确定网络系统鲁棒状态估计问题显得尤其重要,特别是信息融合鲁棒估计问题越来越受到人们广泛关注。在传感网络系统中,按照未加工的数据是否被直接应用,信息融合方法分为集中式融合和分布式融合;集中式融合是将各传感器的观测数据直接送到融合中心进行处理,故获得的状态估值在无偏线性最小方差意义下是全局最优的,但该方法的缺点是通讯和计算负担较大。分布式融合可分为分布式状态融合和分布式观测融合。分布式状态融合是各传感器观测数据先通过相应的局部处理得到局部估计,再将各局部估计传送到融合中心得到全局的融合估计,其中按照去集中式组合方法获得的融合估计是全局最优的,按照加权方法(矩阵加权、标量加权和对角阵加权)获得的融合估计是全局次优的。分布式观测融合是通过加权局部观测方程得到一个融合观测方程,然后将其与状态方程联立,可获得加权观测融合估计,它是全局最优融合估计。与集中式融合方法比较,分布式观测融合方法的优点是减小了计算和通讯负担,减小能源消耗,具有全局最优性。传感网络系统是由一系列具有计算能力、存储能力和传输能力的传感器节点组成,在许多领域都有着非常重要的应用,如军事、环境监测、医疗、工业、农业、建筑、物流与交通等方面。为了节约能源,降低通讯负担,对传感器网络采用最邻近原则进行分簇,每个分簇都是由一个簇头和多个传感节点组成;传感节点首先进行局部估计,并将局部估计和观测数据传送到相应的簇头,然后每个簇头进行局部融合估计,并将所有局部融合估计和观测数据传送到基站,以获得全局融合估计。这是两级分簇融合结构。综上,本文针对带有不确定噪声方差的分簇传感网络系统,利用最小二乘法原理,根据极大极小估值原则,应用加权观测融合算法,提出了两级加权观测融合鲁棒Kalman滤波器和预报器,本文主要工作如下:采用矩阵满秩分解方法分别提出两级加权观测融合鲁棒时变Kalman滤波器和预报器的两种算法,并用取极限的间接方法分别提出两级加权观测融合鲁棒稳态Kalman滤波器和预报器,利用Lyapunov方程证明了它们的鲁棒性。通过信息滤波器形式证明了两级加权观测融合Kalman滤波器和预报器的鲁棒精度分别等价于整体集中式融合鲁棒Kalman滤波器和预报器的鲁棒精度,并高于每个局部鲁棒Kalman滤波器和预报器的鲁棒精度。通过动态系统误差分析方法(DESA)和动态方差系统分析(DVESA)分别证明了两级加权观测融合鲁棒时变和稳态Kalman滤波器和预报器的按实现的收敛性。基于稳态Kalman滤波的直接方法分别提出两级加权观测融合鲁棒稳态Kalman滤波器和预报器的两种算法,利用Lyapunov方程证明了它们的鲁棒性,用信息滤波器形式分别证明了它们与集中式融合鲁棒稳态Kalman滤波器和预报器的等价性。用若干仿真例子证明了上述理论的正确性与有效性。
[Abstract]:In recent years, with the development of sensor network technology, network estimation and control problem has aroused widespread concern. In the network system, the inevitable presence of packet loss, time delay, model parameters and noise variances uncertainty. As we know, the traditional Kalman filter requires a precise mathematical model of known system, these problems make the traditional Kalman filter the state estimation is facing greater challenges, based on the result, in network monitoring and control system, uncertain robust state estimation problem of network system is particularly important, especially robust information fusion estimation problem has attracted more and more attention. In the sensor network system, in accordance with the raw data is directly applied to information the fusion method for the fusion of centralized and distributed fusion; centralized fusion is the observation data of various sensors directly to the fusion center at Therefore, state estimates obtained is globally optimal in linear unbiased minimum variance sense, but the shortcomings of the method are the communication and computational burden is large. The distributed fusion can be divided into distributed state fusion and distributed observation. Distributed state fusion is the measurement data through local processing in the local estimation. Then the local estimation fusion transmitted to the fusion center are global, which according to the fusion estimation method is combined to the centralized optimal weighting method, according to the (weighted matrix, weighted by scalars and diagonal matrix weighting) obtained fusion estimation is globally suboptimal. Distributed measurement fusion is the weighted local observation equation a fusion of observation equations, and then the state equation can be obtained with simultaneous, weighted measurement fusion estimation, it is globally optimal fusion estimation and in Chinese. Comparison of fusion method, the advantages of distributed measurement fusion method is to reduce the computation and communication overhead, reduce energy consumption, with global optimality. Sensor network system is composed of a series of sensor nodes computing ability, storage capacity and transmission capacity, in many areas have very important applications, such as military, environmental monitoring, medical, industrial, agricultural, construction, logistics and transportation. In order to save energy, reduce the communication burden of using the principle of proximity sensor network into clusters, each cluster is composed of a cluster head and a plurality of sensor nodes; the sensor node first to estimate the local and local estimation and observation the data is transmitted to the corresponding cluster head, then each cluster head for local fusion, and all the local estimation and observation data are transmitted to the base station, in order to obtain the global fusion estimation. This is the two class Cluster fusion structure. In this paper, with uncertain noise variance clustering based sensor network system, using the principle of least square method according to the minimax valuation principle, application of weighted measurement fusion algorithm, this paper proposes two level weighted measurement fusion Kalman filter and robust predictor, the main work is as follows: the full rank decomposition of matrix method are put forward two kinds of algorithm two weighted measurement fusion robust time-varying Kalman filter and predictor, and indirect methods limit respectively proposed two level weighted measurement fusion robust steady-state Kalman filter and pre alarm, Lyapunov equation is used to demonstrate the robustness of them. Through information filter proved robust precision two weighted measurement fusion Kalman filter and predictor are equivalent to the overall accuracy of robust centralized fusion Kalman filter and robust predictor, and higher than that of each The accuracy of robust Kalman filter and robust predictor. Through dynamic error analysis method (DESA) analysis of variance and dynamic system (DVESA) are proved two weighted measurement fusion robust time-varying and steady-state Kalman filter and predictor according to the convergence achieved. Put forward two kinds of algorithm two weighted measurement fusion robust steady-state Kalman filter and predictor of the direct method based on steady-state Kalman filter respectively, using Lyapunov equation to prove the robustness of them, with the information filter form respectively proves their equivalence with the centralized fusion robust steady-state Kalman filter and predictor. Several simulation examples demonstrate the validity and effectiveness of the above theory.

【学位授予单位】：黑龙江大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TP212;TN713

【相似文献】