A Fusion Kalman Filter and UFIR Estimator Using the Influence Function Method

In this paper, the Kalman filter (KF) and the unbiased finite impulse response (UFIR) filter are fused in the discrete-time state-space to improve robustness against uncertainties. To avoid the problem where fusion filters may give up some advantages of UFIR filters by fusing based on noise statistics, we attempt to find a way to fuse without using noise statistics. The fusion filtering algorithm is derived using the influence function that provides a quantified measure for disturbances on the resulting filtering outputs and is termed as an influence finite impulse response (IFIR) filter. The main advantage of the proposed method is that the noise statistics of process noise and measurement noise are no longer required in the fusion process, showing that a critical feature of the UFIR filter is inherited. One numerical example and a practice-oriented case are given to illustrate the effectiveness of the proposed method. It is shown that the IFIR filter has adaptive performance and can automatically switch from the Kalman estimate to the UFIR estimates according to operating conditions. Moreover, the proposed method can reduce the effects of optimal horizon length on the UFIR estimate and can give the state estimates of best accuracy among all the compared methods.      

Adaptive compensation of persistent actuator failures of nonlinear systems

In this article, adaptive compensation designs are developed for nonlinear systems with uncertainties from the system functions and persistent actuator failures of characterizations that (i) some unknown system inputs are stuck at some unknown fixed or varying values at unknown time instants and (ii) the failure pattern always switches from one to another and the switching does not stop. Such a controlled plant is described by an uncertain time-varying nonlinear system, and some robust adaptive feedback linearization based failure compensation results are studied for closed-loop system stabilization and bounded output tracking for some specific conditions. To improve the tracking performance in the presence of persistent actuator failures, a new adaptive control scheme is developed, using the failure indicator function which contains the failure pattern and failure time in the formulation. Detailed stability and tracking performance are shown. Simulation results are shown to verify the effectiveness of the proposed adaptive actuator failure compensation method.     

The organization of biological sequences into constrained and unconstrained parts determines fundamental properties of genotype–phenotype maps

Biological information is stored in DNA, RNA and protein sequences, which can be understood as genotypes that are translated into phenotypes. The properties of genotype–phenotype (GP) maps have been studied in great detail for RNA secondary structure. These include a highly biased distribution of genotypes per phenotype, negative correlation of genotypic robustness and evolvability, positive correlation of phenotypic robustness and evolvability, shape-space covering, and a roughly logarithmic scaling of phenotypic robustness with phenotypic frequency. More recently similar properties have been discovered in other GP maps, suggesting that they may be fundamental to biological GP maps, in general, rather than specific to the RNA secondary structure map. Here we propose that the above properties arise from the fundamental organization of biological information into ‘constrained'' and ‘unconstrained'' sequences, in the broadest possible sense. As ‘constrained'' we describe sequences that affect the phenotype more immediately, and are therefore more sensitive to mutations, such as, e.g. protein-coding DNA or the stems in RNA secondary structure. ‘Unconstrained'' sequences, on the other hand, can mutate more freely without affecting the phenotype, such as, e.g. intronic or intergenic DNA or the loops in RNA secondary structure. To test our hypothesis we consider a highly simplified GP map that has genotypes with ‘coding'' and ‘non-coding'' parts. We term this the Fibonacci GP map, as it is equivalent to the Fibonacci code in information theory. Despite its simplicity the Fibonacci GP map exhibits all the above properties of much more complex and biologically realistic GP maps. These properties are therefore likely to be fundamental to many biological GP maps.     

具有光照鲁棒的图像匹配方法

针对现有的基于局部特征的图像匹配算法对光照变化敏感、匹配正确率低等问题，提出一种具有光照鲁棒性的图像匹配算法。首先使用实时对比保留去色（RTCP）算法灰度化图像，然后利用对比拉伸函数模拟不同光照变换对图像的影响从而提取抗光照变换特征点，最后采用局部强度顺序模式建立特征点描述符，根据待匹配图像局部特征点描述符的欧氏距离判断是否为成对匹配点。在公开数据集上，所提算法与尺度不变特征变换（SIFT）算法、加速鲁棒特征（SURF）算法、"风"（KAZE）算法和ORB算法在匹配速度和匹配正确率上进行了对比实验。实验结果表明：随着图像亮度差异的增加，SIFT算法、SURF算法、"风"（KAZE）算法和ORB算法匹配正确率下降迅速，所提算法下降缓慢并且正确率均高于80%；所提算法特征点检测较慢和描述符维数较高，平均耗时为23.47 s，匹配速度不及另外四种算法，但匹配质量却远超过它们。对实时性要求不高的系统中，所提算法可以克服光照变化对图像匹配造成的影响。     

一种新的多渐消因子容积卡尔曼滤波

将强跟踪思想引入容积卡尔曼滤波（cubature Kalman filter，CKF），建立强跟踪CKF能有效克服CKF在模型不确定、状态突变等情况下，滤波性能下降的问题。通过分析现有多渐消因子计算方法，发现它们均只利用了协方差矩阵的对角线元素，并没有考虑各个状态之间的相关性，不能充分发挥多渐消因子的优势。为此，本文提出渐消因子矩阵，基于正交原理推导渐消因子矩阵的求解方法，提出多渐消因子强跟踪CKF算法。多渐消因子强跟踪CKF算法突破了传统多渐消因子为向量的限制，也不再要求渐消因子取值要大于1。仿真验证了算法具有更好的滤波精度何鲁棒性，能更好的满足工程应用的要求。     

考虑节点过载的碳排放空间关联系统级联失效模型

为提高突发事件级联失效对现实碳排放关联系统破坏程度的评估可信性，在传统复杂网络的"负载-容量"级联失效模型基础上，考虑个体成员对负载的冗余能力，提出一种过载失效概率，构建了考虑节点过载状态的级联失效模型，并基于节点特性提出了6种过载节点负载分配策略。仿真结果表明：在过载节点负载分配策略中，综合分配策略整体上较优，能够有效控制级联失效的规模，增加网络鲁棒性；在一定范围内提升过载参数有助于降低级联失效的影响，但提升到一定程度时改善效果不明显；在不同负载分配策略下，剩余系数存在一个最优值，容量可调参数存在最优区间，可以使碳排放关联网络保持较好鲁棒性的同时，花费较小的构建成本，其中紧密度分配策略对应的网络构建成本较高。     

基于非线性干扰观测器的自动弹仓终端滑模控制

针对某自动弹仓位置控制过程中的负载变化和不确定性干扰，设计了基于干扰观测器和非奇异快速终端滑模的复合控制策略，建立了自动弹仓的不确定性动力学模型。干扰观测器用于估计系统复合干扰，估计值用于补偿滑模控制器以增强控制器的鲁棒性和抗干扰能力。运用Lyapunov准则证明了系统闭环稳定。所提算法和PID算法的对比实验结果显示，设计的控制律实现了自动弹仓位置的精确控制，具有良好的鲁棒性，有效提高了自动弹仓的定位稳定性。     

基于Spider卷积的三维点云分类与分割网络

针对传统的卷积神经网络（CNN）不能直接处理点云数据，需先将点云数据转换为多视图或者体素化网格，导致过程复杂且点云识别精度低的问题，提出一种新型的点云分类与分割网络Linked-Spider CNN。首先，在Spider CNN基础上通过增加Spider卷积层数以获取点云深层次特征；其次，引入残差网络的思想在每层Spider卷积增加短连接构成残差块；然后，将每层残差块的输出特征进行拼接融合形成点云特征；最后，使用三层全连接层对点云特征进行分类或者利用多层卷积层对点云特征进行分割。在ModelNet40和ShapeNet Parts数据集上将所提网络与PointNet、PointNet++和Spider CNN等网络进行对比实验，实验结果表明，所提网络可以提高点云的分类精度和分割效果，说明该网络具有更快的收敛速度和更强的鲁棒性。     

一种半监督对抗鲁棒模型无关元学习方法

元学习期望训练所得的元模型在学习到的“元知识”基础上利用来自新任务的少量标注样本，仅通过较少的梯度下降步骤微调模型就能够快速适应该任务。但是，由于缺乏训练样本，元学习算法在元训练期间对现有任务过度训练时所得的分类器决策边界不够准确，不合理的决策边界使得元模型更容易受到微小对抗扰动的影响，导致元模型在新任务上的鲁棒性能降低。提出一种半监督对抗鲁棒模型无关元学习（semi-ARMAML）方法，在目标函数中分别引入半监督的对抗鲁棒正则项和基于信息熵的任务无偏正则项，以此优化决策边界，其中对抗鲁棒正则项的计算允许未标注样本包含未见过类样本，从而使得元模型能更好地适应真实应用场景，降低对输入扰动的敏感性，提高对抗鲁棒性。实验结果表明，相比ADML、R-MAML-TRADES等当下主流的对抗元学习方法，semi-ARMAML方法在干净样本上准确率较高，在MiniImageNet数据集的5-way 1-shot与5-way 5-shot任务上对抗鲁棒性能分别约提升1.8%和2.7%，在CIFAR-FS数据集上分别约提升5.2%和8.1%。     

Distribution-free double-sampling precedence monitoring scheme to detect unknown shifts in the location parameter

In most applications, parametric monitoring schemes are used to monitor the majority of industrial and nonindustrial processes in order to improve the quality of the outputs or services. However, parametric monitoring schemes are known to underperform when the normality assumption is not met or when there is not enough information about the symmetry or asymmetry nature of the process underlying distribution. Hence, in this paper, a new nonparametric Phase II Shewhart-type double-sampling (DS) monitoring scheme based on the precedence statistic is proposed in order to efficiently monitor quality processes when the underlying process distribution departs from normality. The performance is investigated using the average run length (ARL), standard deviation of the run length (SDRL), expected ARL (EARL) and expected average number of observations to signal (EANOS), and the average sample sizes (ASS) metrics. The latter metrics are computed using Monte Carlo simulation and exact formulae. In general, it is shown that the new DS precedence scheme outperforms the existing basic Shewhart precedence scheme with and without supplementary runs rules in many situations. A real-life illustrative example based on a filling process of milk bottles is provided to demonstrate the application and implementation of the new DS precedence monitoring scheme.