一种改进的CURE聚类算法

聚类分析是数据挖掘领域的一个重要研究方向。已经有多种用于大规模数据库的聚类算法,CURE就是一个典型的代表。本文对CURE进行了改进,新方法用多点表示一个类,但舍弃了代表点收缩的过程;通过对类内最邻近距离统计特征的分析,提出了自动分离子类的方法,因而不用预先给定聚类个数;在CURE对原始数据进行随机采样和分区聚类的基础上,增加了划分网格一步,能降低噪声影响并缩短聚类时间。对二维数据的测试表明：改进的CURE能正确识别大多数类,速度上优于原算法。     

改进k-means聚类算法多模型建模的一种新的评价函数

pH中和过程的建模与控制一直是过程控制领域的难题。针对pH过程，提出了一种基于新性能评价函数的k-means聚类算法的多模型建模方法。针对k-means聚类算法中普遍存在的k值已知以及对初始点依赖严重的问题，在k-means聚类算法的基础进行改进，并且引入一个自定义的聚类效果评价函数确定聚类个数，然后采用偏最小二乘PLS算法建立相应的局部线性化模型。通过仿真研究，利用本文算法建立的多模型，获得到了良好的跟踪效果，验证了该改进算法的可行性和有效性。     

油水界面测量过程中自适应阈值聚类优化算法

精确检测原油储罐内油水界面及液位高度是保证净油外输含水率控制精度及联合站盘库系统计量精度的前提条件，是石油化工过程系统工程中的重要环节。鉴于油水界面测量过程中传统分类统计算法和经典K-means聚类算法存在依赖人工选取典型值和初始聚类质心、计算结果不确定性以及精度难于保证等问题，本文提出了一种改进的K-means自适应阈值聚类优化算法。该算法能自动获取最优初始阈值，并改进了油水界面测量传统分类统计算法和经典K-means聚类算法的思想，可实现最优数据分类。首先采用自适应阈值查找算法自动查找油水界面最优初始阈值，其次采用改进K-means聚类优化算法对油水界面数据进行最优划分，最后根据最优化聚类结果计算油水界面及液位高度。实验结果表明：相对于油水界面测量的传统分类统计算法和经典K-means聚类算法, 该算法无需人工选值、能够完全保证计算结果的准确性，且比经典K-means聚类算法和其他改进K-means聚类算法所需的迭代次数更少、运行时间更短。     

基于特征聚类与特征选择算法的SOFC系统故障定位

提出一种基于DBSCAN特征聚类的改进随机森林特征选择算法对SOFC系统的故障进行定位。该方法通过DBSCAN聚类算法挖掘变量间的相关关系,将变量聚类,再挑选出特征选择结果。实验表明:该方法不仅可以筛选出与故障强相关的特征,还能尽可能地减少特征间的冗余,可以高效、快速、准确地对故障进行定位。     

陶瓷砖特征颜色提取算法研究

针对特征颜色提取这一陶瓷砖色差分级的关键问题,提出了应用灰色系统理论的改进特征聚类算法。首先在将图像转换到HSV颜色空间,并根据实际情况确定初始聚类中心,然后计算图像中每个像素点到各聚类中心的灰色关联度,最后应用改进的聚类运算完成特征颜色的提取。对不同纹理的陶瓷砖进行了特征颜色提取实验。实验结果表明,改进算法可准确有效的提取较为复杂的随机纹理陶瓷砖的特征颜色,与人的视觉判断相一致,且在计算效率方面比传统的模糊聚类算法有较大提高。     

BCLCP聚类算法及其在批过程控制中的研究

针对传统的K-均值、FCM和PCA等聚类算法的缺陷,提出了一种连续更新的过聚类和合并相结合的竞争学习(BCLCP)的聚类算法.该算法依据OPTOC的原则,构建原型点的动态邻接区域和动态背离区域,通过自分裂和聚合的策略,识别数据集中的天然聚类并给出聚类的数目.采用该算法对批过程中的历史数据进行分析,原型点的特征与对应的质量等级相符,有效地预测了该批次产品的质量,为后续批次的决策提供依据.     

直觉模糊多核聚类算法及其在乙烯原料属性聚类中的应用

随着乙烯裂解原料种类的日益增多,原料分析仪价格昂贵,因此根据乙烯裂解原料属性进行在线聚类,对实现乙烯收率建模,优化乙烯产率、节能减耗具有重要现实意义。为了提高原料在聚类的准确性,提出了一种基于直觉模糊集理论的核聚类算法。即在定义直觉模糊集隶属度时通过引入犹豫度来表征数据的不确定信息,同时利用直觉模糊熵对多核聚类算法的损失函数重新定义,使类簇中的数据点最优化;进一步地,使用随机森林对裂解原料属性进行特征选择,依据对乙烯产率的贡献度选取聚类的主要特征属性。最后根据实际工业裂解的石脑油数据验证了所述算法的有效性。     

直觉模糊多核聚类算法及其在乙烯原料属性聚类中的应用

随着乙烯裂解原料种类的日益增多,原料分析仪价格昂贵,因此根据乙烯裂解原料属性进行在线聚类,对实现乙烯收率建模,优化乙烯产率、节能减耗具有重要现实意义。为了提高原料在聚类的准确性,提出了一种基于直觉模糊集理论的核聚类算法。即在定义直觉模糊集隶属度时通过引入犹豫度来表征数据的不确定信息,同时利用直觉模糊熵对多核聚类算法的损失函数重新定义,使类簇中的数据点最优化;进一步地,使用随机森林对裂解原料属性进行特征选择,依据对乙烯产率的贡献度选取聚类的主要特征属性。最后根据实际工业裂解的石脑油数据验证了所述算法的有效性。     

基于灰色系统理论的陶瓷制品颜色提取研究

针对陶瓷制品陶瓷墙地砖颜色特征提取的关键问题,提出了灰色聚类的改进算法。并同其他已有的模糊C均值算法(FCM)、MOP_FCM、鲁棒性的FCM、改进的鲁棒性模糊C均值算法,传统的灰色聚类算法,通过实验进行算法性能分析,颜色特征提取效果比较。实验结果表明,本文提出的改进算法,算法性能更高,颜色特征提取效果更明显,更有利于进行色差比较。     

基于在线聚类和关联向量机的多模型软测量建模

针时软测量样本具有按工况点聚类的特性,提出一种基于在线聚类和关联向量机的多模型软测量建模方法.聚类算法通过设定各辅助变量的权重、按引力原理聚类以及合并子聚类,可把样本按照不同的工作点进行聚类.子模型通过关联向量机实现概率化预测,并采用一种更加有效的核参数选择算法提高算法速度.该建模方法在加氢裂化分馏塔装置的轻石脑油终馏点在线预测系.统中取得了良好的效果.     

改进的无线传感器网络ACE分簇路由协议

网络层的路由技术在无线传感器网络体系结构中至关重要,分簇路由具有能量利用高效、数据融合简单、拓扑管理方便等优点,成为当前研究的重点。ACE是一种具有良好反馈机制的自适应分布式成簇算法,可是分簇完成后网络中簇间重叠情况比较严重,通信开销比较大。为此提出一种基于ACE的改进算法,在ACE完成网络分簇后,通过对网络中冗余簇进行合并和未分簇节点进行合并来降低簇间重叠,使网络中的簇拓扑结构更加合理。仿真实验结果表明,改进的算法显著降低簇内重叠区域的大小,能有效降低网络中冗余簇的数量。     

Atomic Cluster Generation with an Atmospheric Pressure Spark Discharge Generator

A new method for generating metal clusters in the gas phase is described and characterized in this work. The method is based on material evaporation by spark ablation at atmospheric pressure. The characterization of atomic clusters was done by measuring their electrical mobility. The measured mobilities were compared with values found in literature in order to identify the cluster species. We show that silver clusters consisting from one up to 25 atoms can be produced in helium at atmospheric pressure. In addition, the effect of oxygen concentration on the resulting cluster mobility distribution was investigated. Results show that at higher oxygen level, the mobility distribution is dominated by the abundance of stable clusters (i.e., magic number clusters). This can be attributed to an oxidation etching effect.

Copyright 2015 American Association for Aerosol Research     

A New Paradigm for KIM-PTP Drug Discovery: Identification of Allosteric Sites with Potential for Selective Inhibition Using Virtual Screening and LEI Analysis

The kinase interaction motif protein tyrosine phosphatases (KIM-PTPs), HePTP, PTPSL and STEP, are involved in the negative regulation of mitogen-activated protein kinase (MAPK) signalling pathways and are important therapeutic targets for a number of diseases. We have used VSpipe, a virtual screening pipeline, to identify a ligand cluster distribution that is unique to this subfamily of PTPs. Several clusters map onto KIM-PTP specific sequence motifs in contrast to the cluster distribution obtained for PTP1B, a classic PTP that mapped to general PTP motifs. Importantly, the ligand clusters coincide with previously reported functional and substrate binding sites in KIM-PTPs. Assessment of the KIM-PTP specific clusters, using ligand efficiency index (LEI) plots generated by the VSpipe, ascertained that the binders in these clusters reside in a more drug-like chemical–biological space than those at the active site. LEI analysis showed differences between clusters across all KIM-PTPs, highlighting a distinct and specific profile for each phosphatase. The most druggable cluster sites are unexplored allosteric functional sites unique to each target. Exploiting these sites may facilitate the delivery of inhibitors with improved drug-like properties, with selectivity amongst the KIM-PTPs and over other classical PTPs.     

EMMS曳力模型及其颗粒团模型的构建和检验

准确描述颗粒团聚特性是发展完善基于多尺度最小能量原理（EMMS）的曳力模型的重要方向之一。提出描述颗粒团聚特性的数学模型,不仅符合物理判断,而且与实验结果吻合。采用颗粒团模型,改进EMMS曳力模型,与实验及直接数值模拟结果吻合较好。改进的曳力模型与欧拉-欧拉双流体方法耦合,实现了不同工况下A、B类颗粒流化床流动特性的数值模拟。成功预测了颗粒非均匀分布特性、局部滑移速度、局部非均匀度以及噎塞状态。     

Use of percolation clusters in nucleation theory

The capillarity approximation of classical nucleation theory assumes that the concept of surface area and surface tension can be used even for very small droplets (clusters); this surface energy enters the average number of clusters of a given size and thus the nucleation rate. Numerical evidence from these average cluster numbers in a simple model (percolation theory) suggests that this concept of a well-defined surface contribution can be extrapolated down to very small clusters and even single molecules, provided one uses an effective surface tension. We also estimate numerically the pre-exponential factor in the cluster size distribution and find it to vary as (cluster size)^−θ with 1 θ 2 in two dimensions.     

Numerical analysis of Pt utilization in PEMFC catalyst layer using random cluster model

A random cluster model was proposed to simulate the catalyst layer of PEMFC. The cluster model consists of a random distribution of three kinds of particles, i.e., Pt/C catalyst, Nafion and poly-tetra-fluoro-ethylene (PTFE), which were generated on a computer by means of Monte Carlo method. Based on such a cluster model, the catalyst utilization was calculated through counting the number of Pt/C clusters and Nafion particles clusters. It was assumed that the effective Pt/C clusters are those that not only have electron channels via carbon particles to current collector but also have proton channels via Nafion polymer particles to the Nafion membrane. The factors influencing catalyst utilization was thoroughly discussed. For the case of high catalyst utilization, numerical results showed that there is a threshold for ratio of Pt/C catalyst loading to Nafion. Beyond this threshold, the catalyst utilization may drop dramatically. The results also showed that the Pt/C catalyst with higher Pt content could allow a larger range of the ratio of the Pt/C catalyst loading to Nafion. Generally, there is high catalyst utilization around the ratio of 1. Results also showed that the lower the Teflon loading in the catalyst layer, the higher the catalyst utilization will be. However, the Pt/C catalyst with higher Pt content can tolerate relatively high Teflon loading than that with a lower Pt content.     

Numerical study on the cluster flow behavior in the riser of circulating fluidized beds

In this paper, the cluster flow behavior in the riser of circulating fluidized beds (CFBs) was predicted by means of the Euler–Lagrange approach. Gas turbulence was modeled by means of Large Eddy Simulation (LES). Particle collision was modeled by means of the direct simulation Monte Carlo (DSMC) method. An extended cluster identification method was used to obtain the solid concentration and velocities of clusters. The flow behavior of falling clusters in the near wall region and the up-flowing and down-flowing clusters in the core of riser were analyzed respectively. At the same time, some cluster transient flow characteristics observed in the CFBs experimental studies were obtained in present simulation. Simulated results showed that there exists a low-pressure zone between the tails of a down-flowing cluster in the core of riser and a new cluster can be formed in the near wall region and transferred from the near wall region to the core of riser under the effect of upgoing gas. Simulated results have a reasonable agreement with the previous experimental findings.