分布式信息融合相关算法研究

发布时间：2018-05-29 08:43

本文选题：分布式目标跟踪 + 传感器管理　；参考：《杭州电子科技大学》2017年硕士论文

【摘要】：信息融合技术通过对从单个和多个信息源获取的数据和信息进行关联、相关和综合,获得对被观测目标精确的位置和身份估计。其中分布式融合结构可以在局部传感器上处理量测数据,相比集中式融合结构,降低了融合中心的计算负荷,提高了系统的可靠性和生存能力。为提高分布式目标跟踪的估计精度并降低传感器的使用率,本文对分布式目标跟踪及传感器管理算法进行了研究,主要内容如下:首先,对分布式目标跟踪以及传感器资源管理技术的国内外研究现状进行了综述。介绍了分布式目标跟踪的基本理论,阐述了分布式目标跟踪的结构和基本算法。其次,针对线性系统、观测噪声为高斯噪声的分布式目标跟踪问题,提出了基于预测信息加权一致性算法的分布式卡尔曼滤波算法。该算法首先将一致性算法与卡尔曼滤波算法结合,对目标的局部估计值进行一致化处理,然后利用邻居节点前一时刻的目标状态估计值,修正当前时刻目标的局部状态预测值,并对局部预测值进行加权处理。提高了局部预测值和估计值的一致性,从而提高了全局信息的一致性。仿真实验表明,所提算法减少了平均一致性估计误差,提高了目标的跟踪精度。再次,针对线性系统、观测噪声为非高斯噪声的分布式目标跟踪问题,实验比较分析了基于分布式高斯混合模型的粒子滤波算法的性能(DGMMPF)。该方法利用高斯混合模型来近似加权粒子的后验概率分布,利用一致性算法来交换邻居节点间的高斯混合模型的参数,使每个节点可以获得近似粒子后验分布的高斯混合模型。由于该算法只通过交换模型参数,减少了计算负荷。为降低分布式结构下传感器的使用率,设计了一种结合DGMMPF和分布式后验克拉美罗-下界(d PCRLB)的传感器管理算法。该算法以d PCRLB为传感器选择准则,每一时刻选出使得d PCRLB最小的传感器组合,然后采用DGMMPF算法估计目标状态。仿真实验表明,设计的算法相比DGMMPF算法在使用较少传感器的情况下,提高了目标跟踪精度。最后,对本文的研究工作进行了总结与展望。
[Abstract]:By correlating, correlating and synthesizing the data and information obtained from single or multiple information sources, the information fusion technology can obtain accurate position and identity estimation of the observed target. The distributed fusion structure can process the measurement data on the local sensor. Compared with the centralized fusion structure, the computing load of the fusion center is reduced, and the reliability and survivability of the system are improved. In order to improve the estimation accuracy of distributed target tracking and reduce the utilization rate of sensors, the distributed target tracking and sensor management algorithms are studied in this paper. The main contents are as follows: first, This paper summarizes the research status of distributed target tracking and sensor resource management at home and abroad. This paper introduces the basic theory of distributed target tracking, and expounds the structure and basic algorithm of distributed target tracking. Secondly, a distributed Kalman filter algorithm based on weighted consistency algorithm of predictive information is proposed for the distributed target tracking problem in which the observed noise is Gao Si noise in linear systems. The algorithm combines the consistency algorithm with the Kalman filter algorithm to process the local estimation of the target, and then modifies the local state of the target at the current moment by using the state estimation of the target at the previous moment of the neighbor node. The local prediction value is weighted. The consistency of the local prediction value and the estimated value is improved, thus the global information consistency is improved. Simulation results show that the proposed algorithm reduces the error of average consistency estimation and improves the tracking accuracy of the target. Thirdly, the performance of particle filter algorithm based on distributed Gao Si hybrid model is compared and analyzed for the distributed target tracking problem in which the observed noise is non-Gao Si noise in the linear system, and the performance of the particle filter algorithm based on the distributed Gao Si hybrid model is compared and analyzed. In this method, the Gao Si hybrid model is used to approximate the posterior probability distribution of weighted particles, and the consistency algorithm is used to exchange the parameters of the Gao Si mixing model between neighbor nodes, so that each node can obtain the Gao Si mixed model with approximate particle posteriori distribution. Because the algorithm only exchanges model parameters, the computational load is reduced. In order to reduce the utilization rate of sensors in distributed structure, a sensor management algorithm combining DGMMPF and distributed posterior Claramero-Lower bound (PCRLB) is designed. In this algorithm, d PCRLB is used as the sensor selection criterion, and the combination of sensors to make d PCRLB minimum is selected at every moment, and then the target state is estimated by DGMMPF algorithm. Simulation results show that compared with DGMMPF algorithm, the proposed algorithm improves the target tracking accuracy with fewer sensors. Finally, the research work of this paper is summarized and prospected.
【学位授予单位】：杭州电子科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP202

【参考文献】