Consensus of linear multi‐agent systems with stochastic noises and binary‐valued communications

This article studies consensus problems of discrete‐time linear multi‐agent systems with stochastic noises and binary‐valued communications. Different from quantized consensus of first‐order systems with binary‐valued observations, the quantized consensus of linear multi‐agent systems requires that each agent observes its neighbors' states dynamically. Unlike the existing quantized consensus of linear multi‐agent systems, the information that each agent in this article gets from its neighbors is only binary‐valued. To estimate its neighbors' states dynamically by using the binary‐valued observations, we construct a two‐step estimation algorithm. Based on the estimates, a stochastic approximation‐based distributed control is proposed. The estimation and control are analyzed together in the closed‐loop system, since they are strongly coupled. Finally, it is proved that the estimates can converge to the true states in mean square sense and the states can achieve consensus at the same time by properly selecting the coefficient in the estimation algorithm. Moreover, the convergence rate of the estimation and the consensus speed are both given by O(1/t). The theoretical results are illustrated by simulations.     

Sampled‐data consensus of multiagent systems with switching jointly connected topologies via time‐varying Lyapunov function approach

Consensus problem of multiagent systems with switching jointly connected topologies under sampled‐data control is studied in this article. The main contribution is that the consensus problem for such system is solved without the assumption that the system matrices are stable or critically stable. For this purpose, a time‐varying Lyapunov function method is utilized to describe the state characteristics with switching jointly connected topologies. Based on the time‐varying matrix of Lyapunov function, the “decline” characteristics at the switching instants is derived to compensate the divergence among the agents with disconnected topologies. Utilizing the “decline” characteristics, the overall consensus of such system can be guaranteed in the framework of dwell time. Finally, the effectiveness of the proposed result is illustrated by two numerical examples.     

基于二阶相似度的即时学习软测量建模方法

针对即时(惰性)学习模型频率降低间接导致的精度下降问题,提出一种二阶相似性的即时学习方法。该方法综合顾及到样本集的整体分布特性,在传统一阶相似度准则的基础上建立二阶相似度准则,采用与测试样本具有绝大部分相同近邻的二阶相似样本建立当前时刻的模型;同时将累计相似度因子用于建立局部模型时样本量的确定,并采用相似度阈值的方式判断此刻模型是否需要重新建立。该方法在青霉素发酵过程产物浓度的预测实验中得到了有效的验证。     

Blind recognition of primitive BCH code based on average cosine conformity

In order to overcome the poor performance of existing algorithms for recognition of BCH code in low signal-to-noise ratio (SNR),a recognition algorithm based on average cosine conformity was proposed.Firstly,by traversing the possible values of code length and m-level primitive polynomial fields,the code length was identified by matching the initial code roots.Secondly,on the premise of recognizing the code length,the GF(2^m) domain was traversed under the m-level primitive polynomial and the primitive polynomial with the strongest error-correcting ability was the generator polynomial for the domain.Finally,the minimum common multiple corresponding to the minimum polynomial of code roots was obtained,and the BCH code generator polynomial was recognized.In checking matching,the statistic of average cosine conformity was introduced.The optimal threshold was solved based on the minimum error decision criterion and distribution of the statistic to realize the fast identification of the BCH.The simulation results show that the deduced statistical characteristics are consistent with the actual situation,and the proposed algorithm can achieve reliable recognition under SNR of 5 dB and code length of 511.Comparing with existing algorithms,the performance of the proposed algorithm is better than that of the existing soft-decision algorithm and 1~3.5 dB better than that of the hard-decision algorithms.     

一种基于分组的区块链共识算法

区块链作为一门新技术,因为其去中心化、不可篡改等性质而被广泛研究。但由于公有链共识算法存在效率低、浪费资源等缺陷,研究者们转向对节点规模较小的联盟链的研究。目前联盟链使用的共识算法为传统的分布式共识算法,受制于节点的规模。当节点数目上升时,系统中的通信量也会上升。提出一种分组的共识算法,第一阶段通过盲签名投票选取胜利节点,第二阶段使用PBFT算法进行主节点的选取,有效缓解单纯使用PBFT算法带来的节点数目增多通信量过大的问题。最后使用该共识算法提出一种物联网系统的架构。     

基于迭代学习的多智能体系统分布式跟踪控制

文章考虑了具适多智能体系统的分布式跟踪控制问题。通过设计带有初始学习机制的$P$型和$PD^{\alpha}$ 型迭代学习控制策略求解跟踪问题。具适导数具有良好的性质且可以刻画不同步长的实际数据采样情况。初始学习机制放松了初始值条件且提高了算法实现趋同跟踪的性能。在可重复操作环境和有向通信拓扑的假设下,提出了一种分布式迭代学习方案,通过重复同一轨迹的控制尝试和用跟踪误差修正不满意的控制信号来实现有限时间趋同。严格证明了随着迭代次数增加,提出的$P$型和$PD^{\alpha}$ 型迭代学习控制策略使得所有智能体能渐近跟踪上参考轨迹。两个代表性数值仿真验证了算法的有效性。     

一种新的L1度量Fisher线性判别分析研究

Fisher线性判别分析（Fisher Linear Discriminant Analysis,FLDA）是一种典型的监督型特征提取方法,旨在最大化Fisher准则,寻求最优投影矩阵。在标准Fisher准则中,涉及到的度量为[L2]范数度量,此度量通常缺乏鲁棒性,对异常值点较敏感。为提高鲁棒性,引入了一种基于[L1]范数度量的FLDA及其优化求解算法。实验结果表明：在很多情形下,相比于传统的[L2]范数FLDA,[L1]范数FLDA具有更好的分类精度和鲁棒性。     

数据丢失下多智能体系统迭代学习跟踪控制

考虑数据丢失下非线性多智能体系统的一致性跟踪问题。假设多智能体系统使用固定网络通信拓扑结构,由于通信网络自身限制导致多智能体系统中存在数据丢失现象。将数据丢失现象描述为取值0/1的随机伯努利序列,设计分布式一致性跟踪误差,提出该系统在数据丢失下的P型迭代学习控制算法。采用压缩映射的方法给出收敛性条件,并在理论上分析了跟踪误差的收敛性。仿真结果表明,提出的算法可以实现该系统在有限时间区间上对期望轨迹的完全跟踪,验证了算法的有效性。     

事件触发控制多智能体网络点对点一致性

针对具有领导层与跟随者层的多智能体网络点对点的一致性问题进行了研究。为了有效降低多智能体一致性控制协议的更新频率,同时为避免芝诺现象的产生,采用了周期采样与事件触发控制相结合的方案,实现了多智能体网络的点对点一致性。进一步,根据李雅普诺夫稳定性理论,通过建立线性矩阵不等式,给出了领导层与跟随者层网络达到点对点一致性的充分条件。最后,数值仿真进一步验证了理论结果的有效性。     

基于学习的核偏最小二乘法优化扩展卡尔曼滤波

为了解决非线性系统参数不确定和噪声信息不准确导致EKF随时序估计误差较大的问题,利用核偏最小二乘法与系统方程参数和噪声信息无关的特点优化EKF。先将量测数据和EKF的收敛估计作为学习样本,建立KPLS预测模型,然后,融合KPLS和EKF的预测值进行状态估计;同时,若状态估计的收敛判据为真,将估计值作为学习样本,并利用滑动窗口更新KPLS核矩阵,使KPLS能时序预测;若收敛判据为假,则更新量测协方差。最后,通过实验仿真的方法,分析KPLS-EKF算法的收敛性和性能。实验结果表明：KPLS-EKF能够有效地解决非线性系统参数和噪声信息不准确导致的EKF误差较大的问题。