高阶多模型状态估计算法及应用

发布时间：2018-04-28 00:38

本文选题：机动目标跟踪 + 高阶多模型算法　；参考：《哈尔滨工业大学》2017年硕士论文

【摘要】：在机动目标跟踪领域,由于无法获知目标每一时刻真实的运动情况以及何时发生了运动模型变化,模型的不确定问题就成为目标高精度跟踪的一大核心问题。目前用多模型方法来对机动目标的状态进行估计是处理模型不确定性问题最常用有效的方法,其中最经典的方法是交互式多模型算法。为了利用更多的先验信息,提高多模型算法的性能,本文研究了高阶多模型算法,主要围绕以下三个方面具体展开。第一:研究了基于混合转移分布的高阶交互式多模型滤波算法。与交互式多模型算法相比,高阶交互式多模型滤波算法利用更多连续多个时刻的信息,提高了估计精度;但是,高阶马尔科夫链所需设置的参数过多,缺少足够多的先验知识用以确定高阶模型转移概率矩阵。本文采用混合转移分布模型,用一阶马尔可夫模型转移概率的经验加权近似高阶模型序列的转移概率,大大减少了所需设置参数的个数,降低了合理确定高阶模型转移概率矩阵的难度。仿真验证了本算法的有效性;并且阶数越高,在模型不变区域的估计性能越好。第二:提出了模型切换受限的高阶多模型滤波算法。高阶马尔科夫链隐含了每一时刻都可能发生模型切换的假设,但在实际场景中这一假设不合理,目标通常不会在所有时刻都发生运动模型切换,为此本文在马尔可夫链基础上,增加模型切换次数有限的约束,即连续多个时间内最多只发生一次模型切换,从而给出了模型序列的更准确描述,带来估计精度的提高;同时由于删除了很多不符合条件的模型序列,也使得算法计算效率得到一定改善。另外,配合模型切换受限的多模型滤波,本文还设计了一种更为合理的高阶模型序列转移概率。推导了模型切换受限的高阶广义伪贝叶斯算法和高阶交互式多模型算法。仿真结果表明:该算法在模型不变区域的估计精度与模型序列已知的滤波算法极其接近,仅在模型跳变点处存在尖峰误差;与交互式多模型算法相比,所有区域的估计精度都得到提高;与同阶的普通高阶多模型滤波算法相比,模型跳变区域误差大大降低,过渡过程大为缩短,且节省了大量的计算量。第三:提出了模型切换受限的增广状态高阶交互式多模型平滑算法。该算法是在模型切换受限的高阶多模型滤波算法的基础上,进行状态增广,从而在滤波的同时实现平滑。仿真实验表明:该算法与增广状态的交互式多模型平滑算法相比,平滑效果得到进一步的提高;与模型切换受限的高阶多模型滤波算法相比,估计精度更好,且基本消除了尖峰误差;另外该算法可通过设置不同的固定延迟长度,实现不同程度的平滑效果。
[Abstract]:In the field of maneuvering target tracking, because it is impossible to know the real movement of the target at every moment and when the moving model changes, the uncertainty of the model becomes a core problem of target tracking with high accuracy. At present, the state estimation of maneuvering targets using multi-model method is the most commonly used and effective method to deal with the uncertainty of the model, among which the most classical one is interactive multi-model algorithm. In order to use more prior information and improve the performance of multi-model algorithm, this paper studies high-order multi-model algorithm, mainly focusing on the following three aspects. First, a high-order interactive multi-model filtering algorithm based on mixed transfer distribution is studied. Compared with the interactive multi-model algorithm, the high-order interactive multi-model filtering algorithm makes use of more information of continuous and multiple times, and improves the estimation accuracy. However, the high-order Markov chain requires too many parameters. There is a lack of enough prior knowledge to determine the transition probability matrix of higher order models. In this paper, the mixed transfer distribution model is used to approximate the transition probability of the high order model sequence with the empirical weighting of the transition probability of the first order Markov model, which greatly reduces the number of required parameters. The difficulty of reasonably determining the transition probability matrix of higher order model is reduced. The simulation results show that the algorithm is effective, and the higher the order is, the better the estimation performance is in the invariant region of the model. Second, a high-order multi-model filtering algorithm with model switching constraints is proposed. The higher-order Markov chain implies the assumption that model switching may occur at every moment, but this assumption is unreasonable in the actual scenario, and the target does not normally have a moving model switch at all times. Therefore, this paper based on Markov chain. By increasing the constraint of the limited number of model switching, that is, the model switching occurs only once at most in a continuous multiple time, thus the more accurate description of the model sequence is given, and the estimation accuracy is improved. At the same time, the computational efficiency of the algorithm is improved due to the deletion of many unqualified model sequences. In addition, a more reasonable transition probability of higher order model sequences is also designed with multi-model filtering with constrained model switching. High-order generalized pseudo-Bayes algorithm and high-order interactive multi-model algorithm are derived. The simulation results show that the estimation accuracy of the algorithm in the invariant region of the model is very close to that of the filter algorithm known by the model sequence, and the peak error exists only at the point of the model jump, and compared with the interactive multi-model algorithm, the proposed algorithm is more accurate than the interactive multi-model algorithm. Compared with the ordinary high-order multi-model filtering algorithm of the same order, the error of the model jump region is greatly reduced, the transition process is greatly shortened, and a large amount of computation is saved. Third, an augmented state high order interactive multi-model smoothing algorithm with constrained model switching is proposed. This algorithm is based on the high-order multi-model filtering algorithm which is limited by model switching, so that the filtering can be smoothed at the same time. The simulation results show that compared with the extended interactive multi-model smoothing algorithm, the proposed algorithm is more effective than the high-order multi-model filtering algorithm with limited model switching, and the estimation accuracy is better than that of the high-order multi-model filtering algorithm with limited model switching. In addition, the algorithm can achieve different degree of smoothing effect by setting different fixed delay length.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN713

【参考文献】