基于RBF神经网络的水处理系统建模与仿真

徐中 叶希贵

利用径向基函数（RBF）神经网络的特点，采用神经网络对活性污泥法水处理系统进行建模，提出了一种RBF网络学习的新算法，即将改进的减聚类算法和动态最近邻聚类算法相结合的算法。仿真结果表明，该算法先进有效，用其建立的模型具有较强的实用性，为实现水处理系统的在线实时预测控制提供了可行的途径。[著者文摘]

一种改进的RBF神经网络学习算法

王洪斌 杨香兰 等

提出了一种改进的RBF神经网络学习算法，分别通过减聚类和监督学习算法对网络参数和权值进行训练，既可以根据样本合理地聚类、确定RBF径向基函数的个数和相应参数，又具有较强的网络映射能力，从而不仅使RBF神经网络结构得以优化，性能也得到了提高。仿真结果表明了该学习算法的实用性和有效性。

基于摄动思想的传递闭包聚类法的分析与应用

Analysis and Application of the Method of Aggregation Based on Perturbation and Transitive Closure

程鹤 耿双军

分析了传统的模糊聚类方法,基于摄动思想,将传递闭包聚类法与目标函数法相结合,得出了一种既满足聚类效果要求又减小聚类失真的简便算法,同时给出了该聚类方法的步骤.最后通过对水体的水质进行聚类的例子对算法加以应用,显示了算法的可靠性、有效性.[著者文摘]

The traditional fuzzy aggregation method is analysed. Based on the mind of perturbation,a method which combines transitive closure method with the objective function method is introduced,and the steps of the method is presented. The method satisfies the need of the effect of aggregation and minimizes the disfeasible, verified and simple.[著者文摘]

中国各省市能源结构的聚类分析

Clustering Analysis on Energy Structure of Provinces in China

王成瑞 赵达薇

中国能源安全问题不仅是供给总量与需求总量的矛盾，还有清洁能源供给不足引发的结构性矛盾；石油短缺是中国国内能源结构性矛盾中的主要方面，这导致中国各省市难以调整能源消费结构。以中国各省市199-2003年的数据为基础，采用模糊C-均值聚类方法，把各省市的能源消费结构划分成5类，分别讨论了各种类型消费结构的特点，并对决定能源消费结构的主要因素（资源禀赋、科技与人力资本素质、人均收入、产业结构、区位条件等）进行分析。[著者文摘]

For energy security problems and energy structure contradiction, this paper classes the structure of energy consumption into five types based on the data from all provinces during 1995-2003 with fuzzy C-clustering method. The characteristics of various types of consumption structure were discussed. The main factors of energy consumption structure were analyzed （natural resources, technology and human capital quality, industry structure, geographic conditions, etc.）.[著者文摘]

语音库裁剪的一种不定长递阶聚类方法

A Non-Uniform Clustering Synthesis Instances Pruning Approach for Corpus-Based TTS

张巍[1,2] 吴晓如[3] 刘江[3] 王仁华[2]

大量使用不定长是大语料库语音合成质量的一个重要保证，而语音库裁剪方法通常会导致不定长的损失．针对这一关键性问题，该文构建了NuClustering-VPA算法：对不同粒度的不定长变体进行聚类，根据高阶聚类结果调整低阶变体的聚类，使得低阶聚类中心有所偏向．NuClustering-VPA算法保留了最重要的不定长，从而有效减小了裁剪对不定长的破坏．测听实验表明，利用NuClustering—VPA算法，即使在语音库裁减率为39．63％时，合成自然度下降较小，仍然保持在较高的水平．这一技术已被应用在科大讯飞公司的实际语音产品中．[著者文摘]

The employment of non-uniform does great help for Corpus-based TTS to synthesize high natural speech. But Tailoring TTS voice font, or pruning redundant synthesis instances, usually results in loss of non-uniform. In order to solve this problem, this paper proposes the algorithm named NuClustering-VPA. According to this algorithm, the high level non-uniforms containing same syllables are clustered to several centers, then the centers are projected to low level non-uniforms. Therefore, the centerrs projections can guide the clustering of low level nonuniforms. These series of processes avoid erasing or destroying those key non-uniforms for synthesis. In experiments, the naturalness scored by MOS does not severely degrade when reduction rate is above 39.63%. And this approach has been applied in software products of Ifytek Co. Ltd.[著者文摘]

基于聚类和支持向量机的入侵检测研究

Research of intrusion detection based on clustering and support vector machine

倪霖[1] 郑洪英[2]

提出了一种基于聚类和支持向量机的入侵检测算法,该算法可以有效地减小计算复杂性,提高检测性能。算法首先使用K-MEANS聚类算法对数据做一个初步的划分得到多个类;然后考察每个类中数据的标识,只有当类中的数据标识不止一个时才进行支持向量的查找。最后使用KDDCUP1999进行了计算机仿真实验,实验结果说明了该算法的有效性。[著者文摘]

A new approach based on clustering and Support Vector Machine （SVM） was proposed. It could reduce the computational complexity and get better performance in the intrusion detection. Firstly, clustering was imposed on data set using K-MEANS partition algorithm and clusters were established. Secondly, the support vector was searched only when the labels in a cluster were different. Finally, the experimental results using KDD CUP 1999 demonstrate the efficiency of our approach.[著者文摘]

基于新聚类学习的属性离散化方法

Attribute Discretization Method Based on New Clustering Learning Algorithm

赵晓霞

粗糙集理论中，属性的离散化是预处理中的关键问题。基于新聚类学习算法提出了一种新的属性离散化方法，并将该方法用于车牌字符识别中。首先根据车牌字符的特征建立决策表，给出了基于新聚类学习算法对决策表属性值进行离散化的算法。然后应用粗糙集理论对离散后的决策表属性进行约简，由约简后的属性构造神经网络识别器。字符识别的结果分析表明基于新聚类学习方法的离散化算法对于车牌字符属性较为适用。[著者文摘]

Attribute discretization is an important research issue in rough set. In this article,a strategy for a discretization method of decision attribute is proposed based on a new clustering learning algorithm and applied to the car＇s plate characters recognition. Firstly,the features of the training samples are extracted to build up the decision table. After comparing discretization algorithm of decision attributes,the rough set theory is applied to reduce the decision table. Finally, the reduced decision attribute is used to generate neural network recognizing machine. The results of characters recognition show that a strategy for the discretization algorithm based on SOFM networks is applicable for the attribute of the car＇s plate characters.[著者文摘]

基于PSO的RBF神经网络学习算法及其应用

RBF Neural Network Algorithm Based on PSO Algorithm and Its Application

张顶学 关治洪 刘新芝

提出了一种基于粒子群优化（PSO）算法的径向基函数（RBF）神经网络学习方法，首先利用减聚类算法确定网络径向基层的单元数，再用PSO对基中心和宽度进行优化，并与最小二乘法相结合训练RBF神经网络。将此算法用于混沌时间序列的预测，实例仿真表明此方法是有效的。[著者文摘]

This paper introduces a method of Radial Basis Function（RBF） neural network algorithm based on Particle Swarm Optimization（PSO） algorithm.The first to determine units＇ number in RBF layer using subtractive clustering method, the second to optimize central position and directional width based on PSO algorithm,and the last to train RBF neural network combine with least-square method.The new training algorithm is used to predict chaotic time series,and the simulation results are efficiency.[著者文摘]

快速发现任意形状的聚类

刘必红 符红光

提出FFCAS（Fast Finding the Clusters of Arbitrary Shape)聚类算法，用于快速发现任意形状的聚类，先将每个对象分配到很小的ε－邻域，即原子聚类，然后找出高浓度的核心原子聚类，再消除所有的冗余原子聚类，仅用边界来表示聚类，大大减小了存储空间。因为ε是一个很小的值，原子聚类能自然地描述聚类。该算法最坏时间复杂度为O（n log n)(n为数据库中的对象个数），只需访问一次数据库，实验表明，FFCAS运行时间与数据库中的对象数目成线性关系，能发现任意形状的聚类，对异类的敏感性低，对大型、高维数据库也有效。

聚类分析算法CLIQUE的研究及改进

Research and Impovement of Cluster Analysis CLIQUE Algorithm

陈朝华[1,2] 王伟平[1]

CLIQUE是一种基于密度和基于网格的混合聚类方法。在高维空间中,它能够有效地进行聚类,并且能够发现嵌套在高维数据空间子空间中的聚类。但是,CLIQUE算法存在着很多的局限性,主要有以下两点：首先是子空间的剪枝;其次是CLIQUE算法追求方法简单化。针对CLIQUE算法的局限性,采用基于约束条件的聚类技术、自适应网格技术和边界调整技术来对CLIQUE算法进行改进,提出了基于约束条件和自适应网格的CAG-CLIQUE算法。[著者文摘]

CLIQUE is an omnibus cluster method that is density-based and grid-based. It can effectively cluster on data set with high dimensions and discover clusters in the subspace of high dimensional data. But, there are many limitations in CLIQUE algorithm. The main two limitations are as follows. First , Subspace pruning. Second , CLIQUE algorithm oversimplifies at some aspects and many steps are approximate algorithm. In view of CLIQUE algorithm limitations, CAG-CLIQUE algorithm is proposed that is constraint-based and self-adaptive grid. CLIQUE is modified by constraint-based cluster technique, self-adaptive grid division technique and boundary adjustment technique.[著者文摘]