遗传规划多分类系统设计

近些年来,遗传规划用于预测、模式识别方面的应用越来越广泛,为了节省大量人力物力,设计一个专门用于解决分类问题的系统显得尤为重要。本文利用C++语言,编写出一个通用的多分类系统,适用与大部分分类问题。

基于VMD和改进多分类马田系统的滚动轴承故障智能诊断

为了有效提取滚动轴承的故障信号,选择合适的智能分类器识别故障状态,提出基于变分模态分解及多重马氏距离法的多分类马田系统的故障智能诊断系统。通过变分模态分解将振动信号分解为多个本征模函数并提取相关特征;并采用了多重马氏距离法的马田系统,以特征子集代替特征参与分类器的构建,以解决特征参数众多的问题;通过正交表和信噪比,筛选出各状态的敏感模态分量,并提出多分类马田系统,用于多类故障智能识别;将其应用于滚动轴承故障数据中,验证算法的有效性,并与其他算法对比分析。结果表明,基于变分模态分解及改进的多分类马田系统算法能简化诊断系统、训练耗时少,识别准确率高,是一种更为有效的故障智能诊断方法。

基于混合模型与改进多分类马田系统的控制图模式识别

为提高生产过程中产品质量的智能监控水平,提出基于时间序列混合模型及改进多分类马田系统的控制图模式识别算法。选用时间序列混合模型对控制图实时数据进行特征提取;改进马田系统的阈值计算方法并制定多类判别准则,将表征的特征向量代入改进多分类马田系统分类器中进行特征约减及模式识别。最后,将该识别算法应用于控制图公开数据集及生产案例中,以验证算法的有效性,并与其他算法对比了分析,结果表明,基于时间序列混合模型及改进多分类马田系统算法能简化识别系统,识别精度高,是一种更为有效的控制图模式识别方法。

Stacking Learning在高光谱图像分类中的应用

高光谱图像分类研究中,集成学习能够显著地提高分类效果。但是传统的并行多分类系统对基础分类器有较高要求,即要求差异性及分类均衡。为了解决这一问题,采用Stacking Learning的堆栈式学习方式,首先使用K-Fold和交叉验证的方式进行数据分割和训练,将原始特征进行特征变换,重新构建二级特征。再使用新特征进行对Meta分类器进行训练得到判决分类器,用于样本的最后分类判断。实验结果表明,采用的Stacking Learning方法不依赖基础分类器,且相比较于传统的多分类系统具有更高的精度和良好的稳定性。

基于人工智能技术的配电系统图像分类模型安全评估体系构建与测试研究

深度学习是利用大型数据集的高效训练算法,在新型智能电力系统中得到了广泛应用。然而,目前电力系统人工智能模型的深度神经网络训练阶段不完善,容易受到对抗性样本的影响,导致深度神经网络出现错误分类。因此,在新型电力系统中,确保基于配电系统图像多分类深度神经网络模型的安全性至关重要。本文首先确定了针对配电系统图像多分类的深度神经网络的对抗空间,并根据对模型的输入和输出之间映射的精确理解来制作对抗者样本。然后,定义了一个对抗攻击能力指标来评估不同样本类别对对抗性扰动的安全性。最后,通过定义良性输入和目标分类之间的距离的预防性措施,实现了对对抗性样本的防御。

Toxic effect of marine petroleum hydrocarb on pollution for polychaetes

With the development of economic activities, marine oil pollution is becoming an increasing serious problem. Instead of dissolved in seawater, petroleum hydrocarbons are more easily enriched in sediments, which poses a serious threat to benthos. In this paper, the research progresses on toxicity of marine petroleum hydrocarbon for polychaetes were reviewed in recent years, including the toxic mechanism of petroleum hydrocarbons, the feeding habits, the effects of petroleum hydrocarbons on the antioxidant defense system, the nervous system damages, the molecular damages of polychaetes under the stress of petroleum hydrocarbon, the accumulation and metabolism of petroleum hydrocarbons in polychaetes. The future research direction were also prospected.

Genetics and phylogeny of genus Coilia in China based on AFLP markers

The taxonomy of Coilia has been extensively studied in China,and yet phylogenetic relationships among component taxa remain controversial.We used a PCR-based fingerprinting technique,amplified fragment length polymorphism(AFLP) to characterize and identify all four species of Coilia in China.We examined the genetic relationships of the four species of Coilia and a subspecies of Coilia nasus with AFLP.A total of 180 AFLP loci were generated from six primer combinations,of which 76.11% were polymorphic.The mean genetic distance between pairs of taxa ranged from 0.047 to 0.596.The neighbor-joining tree and UPGMA dendrogram resolved the investigated species into three separate lineages:(1) C.mystus,(2) C.grayii and(3) C.brachygnathus,C.nasus,and C.nasus taihuensis.Phylogenetic analysis of the AFLP data is inconsistent with current morphological taxonomic systems.The AFLP data indicated a close relationship among C.brachygnathus,C.nasus taihuensis,and C.nasus.Therefore,the two species described under Coilia(C.brachygnathus and C.nasus taihuensis) are treated as synonyms of C.nasus.

DISTRIBUTED TRACKING CONTROL OF SECONDORDER MULTI-AGENT SYSTEMS UNDER MEASU-REMENT NOISES

This paper considers a leader-following tracking control problem for second-order multiagent systems(MASs) under measurement noises and directed communication channels.It is assumed that each follower-agent can measure the relative positions and velocities of its neighbors in a noisy environment.Based on a novel velocity decomposition technique,a neighbor-based control law is designed to realize local control strategies for these continuous-time agents.It is shown that the proposed consensus protocol can guarantee that all the follower-agents track the active leader.In addition,this result is extended to a more general case with switching topologies.Finally,a numerical example is given for illustration.

Method for extracting geometrical characteristics of joint probability density based on contour lines

This paper presents a method for extracting geometrical features of the joint probability density function(PDF)of two-dimensional systems based on its contour lines,with particular interests given to the number and position of peaks and craters.In order to detect those two types of structures,a series of horizontal planes are applied to truncate the joint PDF with contour lines generated.Starting with the analysis of contour lines in a single plane,shape characteristics of the peak and the crater can be reflected on the contour lines in the aspects of gradient direction and inclusion relationship.Aided by the properties of PDF,the information about gradient direction and inclusion relationship of contour lines can be obtained simultaneously if the contour tree is built.According to the contour tree,the contour lines can be classified as two groups.Then the corresponding relation between contour lines in different planes is discussed.Based on the corresponding relation,clustering analysis about contour lines belonging to the same group but having different heights is performed.Two sets of contour lines are finally obtained as the simplest expression of geometrical characteristics of a joint PDF.They can be used to obtain the number and position of each peak and crater.Three oscillators of different types are chosen to test this method,which shows that this method can pave the way for numerical calculation about the stochastic P-bifurcation of multi-dimensional systems.

A generic construction of low correlation zone sequences based on interleaved technique

Low correlation zone (LCZ) sequences are useful in quasi-synchronous code-division multiple access (QS-CDMA) communication systems. In this paper, a generic construction of LCZ sequences based on inter-leaved technique is investigated. Firstly, the shift sequence is shown to correspond to two-tuple balanced d-form function essentially, which results in new shift sequence. Secondly, an optimal design of p2-ary sequences over the integer residue class ring Zp2 is proposed, which improves the previous construction when p is an odd prime.

On enumeration of polynomial equivalence classes

The isomorphism of polynomials （IP）, one of the hard problems in multivariate public key cryptography induces an equivalence relation on a set of systems of polynomials. Then the enumeration problem of IP consists of counting the numbers of different classes and counting the cardinality of each class that is highly related to the scale of key space for a multivariate publi9 key cryptosystem. In this paper we show the enumeration of the equivalence classes containing ∑n-1 i=0 aiX^2qi when char（Fq） = 2, which implies that these polynomials are all weak IP instances. Moreover, we study the cardinality of an equivalence class containing the binomial aX2qi ＋ bX2qj （i ≠ j） over Fqn without the restriction that char（Fq） = 2, which gives us a deeper understanding of finite geometry as a tool to investigate the enumeration problem of IP.