运用自组织竞争网络进行气体定性分析的研究

Research of Gas Qualitative Analysis Using Self-organizing Competitive Network

太惠玲 谢光忠 蒋亚东

优选了分析H2，CO气体的半导体气体传感器组成阵列，建立了实时数据采集系统，并与自组织竞争网络模式识别技术相结合，以进行气体定性分析的研究；同时为了消除气体浓度变化对传感器阵列输出的影响。提高自组织网络的识别效果，运用三种不同的数据归一化算法对传感器阵列的输出响应进行了预处理，并对各自对应的网络识别结果进行了分析与讨论。实验结果表明，采用相对算法可实现H2，CO气体的准确识别。[著者文摘]

The semiconductor gas sensors sensitive to hydrogen and carbon monoxide are chosen to compose the gas sensor array, and an on-line data acquisition system is constructed, which is combined with the pattern recognition techniques of self organizing competitive network for the research of gas qualitative analysis. In order to eliminate the effect produced by the gas concentration, three kinds of normalization methods are developed to pre-process the sensors responses toward hydrogen and carbon monoxide. The research results show that the accurate identification of hydrogen and carbon monoxide in the finite range could be accomplished by relative method.[著者文摘]

基于主成分分析的自组织竞争神经网络在多光谱遥感影像分类中的应用

Self-Organizing Competition Neural Network Based on Principle Component Analysis in Multi-Spectrum Remote-Sensing Images Classification

周峰[1] 李杏梅[1] 刘福江[1] 孙华山[2]

多光谱遥感影像具有波段多、信息量大的特点，传统的分类方法难以达到提高精度的要求。利用主成分分析对多波段遥感图像进行降维，再采用竞争型自组织神经网络对图像进行非监督分类。这种方法的分类精度为87．5％，Kappa系数为0．86，明显高于最大似然法，最小距离法和基于像元的自组织竞争神经网络法。实验结果表明该方法在多光谱遥感影像分类中具有较好的适用性。[著者文摘]

Due to the characteristics of many wavebands and large information quantity, multi-spectrum remote-sensing images are difficult to be classified with high accuracy by traditional methods. In this paper, we reduce the dimensions of multi-spectrum remote-sensing images with principle component analysis at first, and then perform unsupervised classification with self-organizing competition neural network. The classification accuracy of this method is 87. 5% and the Kappa coefficient is 0. 86. They are obviously higher than that of conventional maximum likelihood method, minimum distance method and selforganizing competition neural network method based on pixels. The results indicate this method can be well applied in multi-spectrum remote-sensing images classification.[著者文摘]

网格环境下信息主体的自组织竞争机制探讨

Self- Organization Competitive Mechanism of Information Agent in Grid Environment

李保珍[1] 朱庆华[2] 周献中[1]

在网格环境下，信息主体间的关联性更加突出，由于每个信息主体都是利益主体，同时又都具有不同程度的信息存储与加工利用的功能，所以在竞争同一信息资源，或在合作提供同一信息服务过程中，不同信息主体间会存在着动态的竞争与协作，这使得信息资源的开发利用具有了新的特征，如具有非线性、非局域性、非定常性等。为深入把握信息主体之间的关联机制，文中将信息主体视为相对独立的神经元，并利用人工神经网络中比较成熟的自组织竞争网络（SOCN）理论和自组织特征映射（SOFM）理论，来研究其竞争和协同的自组织机制，进而为在网格环境下对信息资源主体的协调管理提供借鉴。[著者文摘]

Under the grid environment, the connection between different information agents is more prominent. Each information agent is a benefit entity, simultaneously has the functions of information storage and utilization, therefore different information agents will have the dynamic competition and the cooperation when they compete in the same information resource, or provide cooperatively the same information service. It makes the development and utili- zation of information resources have new characteristics, such as non - linearity, non - location, time - dependent, and so on. To find deeply out the connection mechanism of information agent, this paper looks on the information a- gent as a relatively independent neuron, and uses the mature Self - organization Competition Network （SOCN） and Self- organization Feature Maps （SOFM） of artificial nerve network theory to study its self-organization competitive and harmonious mechanism, and to provide references for managing information agents in the grid environment.[著者文摘]

竞争网络在中医舌诊中的应用及实现

Competition Nerve Network Applied in Tongue Inspection of Tradition Chinese Medicineand Its Implementation

周玉 汪仁煌 韦玉科 范江涛

自组织竞争神经网络通过对网络权值的调整,使所有矢量都在输入矢量空间相互隔离,形成了各自代表输入空间的一类模式,从而实现聚类功能.在中医舌诊推理的研究过程中,我们建立了基于竞争神经网络的辨识模型,利用该系统对病位和病机进行推断和识别.在分析了Visual C＋＋和Matlab各自特点的基础上,提出了利用Visual C＋＋与Matlab混合编程来实现竞争神经网络的应用这一方法.该程序的界面是由VC＋＋实现的,而神经网络的所有功能是通过调用Matlab引擎的方法实现的,这样做可充分利用Matalb和Visual C＋＋各自的优点,实现更多的功能.实验结果表明该方法是可行和有效的.[著者文摘]

Self-organizing competitive neural network can make the weights separate mutually through the adjustment of the input weight, and the weights can represent different models in different space, so the input can be clustered by the self-organizing competitive neutral net. In the research process of tongue inspection of tradition Chinese medicine, a discriminate mode of neural network with competitive unit was built. We use this mode carry on the inference and discrimination to the disease location and the pathogenesis. On the bases of characteristics of Visual C＋＋ and Matlab, we proposed mixed programming between VC＋＋ and Matlab. Then put forward using this method to realize the competition neural networks processing and application. The interface of the program is implemented by VC＋＋, and the neural network functions are implemented by Matlab engine. The method of mixed programming is most suitable, and we can make use of Matlab and Visual C＋＋ respective merits, realizes more functions. The result is testiffed to be feasible and effective.[著者文摘]

自组织理论视角下企业创新网络研究

陆园园 张阳

企业创新网络是一个自组织系统，自组织理论为其研究提供了新的视角和方法论指导。文章在分析总结国内外有关企业创新网络问题研究现状的基础上，基于自组织理论的崭新视角探讨了企业创新网络的自组织特性，并提出了基于自组织理论的企业创新网络模型，指出了自组织理论在创新网络方面进一步研究的方向。[著者文摘]

自组织网络中UPMA协议的群间仿真

Intercluster simulation of UPMA protocol for self-organizing networks

丁立军 刘凯 李汉涛 张军

基于有效竞争预约接入、无冲突轮询传输的思想，结合分层分布式网络结构为自组织网络提出了依据用户妥善安排的多址接入（UPMA）协议，UPMA协议可以支持节点移动性和多跳网络拓扑，并使用网络仿真工具OPNET仿真评估了它的群间通信性能．该协议利用分群算法将多跳网络拓扑形成轮询所需要的两跳分群结构，包括预约接入和无冲突的轮询服务阶段．有分组发送的节点在每帧的竞争接入时隙中竞争接入．如果成功，则进入轮询服务过程；否则，在本帧重新开始的接入阶段中进行冲突避免和分解的预约接入过程．仿真结果表明，UPMA协议显著提高了多跳群间的业务传输效率，可以提供较高的端到端信道利用率、较低的端到端平均消息时延和较小的平均消息丢弃率．[著者文摘]

Based on the concept of contention reservation, polling transmission and combining distributed clustering network architecture, an user-dependent perfect-scheduling multiple access （UPMA） protocol for supporting node mobility and multihop architecture in wireless self-organizing networks was described, and its performance of intercluster communication was simulated by OPNET. By clustering algorithms, two-hop cluster architecture for polling mechanism was formed to support multihop network topology. The protocol includes reservation access phase and polling phase in channel period. In the protocol, the nodes with packets to send contend to reserve channel resources during access slots of every frame. If successful, they were polled to transmit their packets in subsequent frames.If not, collision avoidance and resolution process was used for efficient access. The simulation resuhs show that the proposed protocol can greatly improve muhihop transmission efficiency, and achieve higher end-to-end channel utilization, lower average end-to-end message delay and less average message dropping rate.[著者文摘]

基于小波融合的ASTER数据自组织特征映射神经网络分类研究

哈斯巴干 马建文 李启青 韩秀珍 刘志丽

重点介绍了有自组织功能的两层结构的神经网络Kohonen自组织特征映射，两层之间各神经元实现全连接并且在竞争层各神经元之间还存在侧连接，实现了有效的竞争和抑制，提高了自适应的学习能力，因此成为国际遥感数据分类的研究热点，ASTER卫星数据是新型遥感数据，包括3个15m分辨率波段和3个30m分辨率的短波红外波段，选择天津大港ASTER数据作为方法实验数据，首先对数据进行了小波融合，然后进行了土地覆盖类型的神经网络分类研究，研究结果与相同训练点的最大似然监督分类比较，总体提高分类精度7％，特别对城镇分类精度提高近一倍。

基于泛化竞争和局部渗透机制的自组织网TSP问题求解方法

Self Organizing Map with Generalized and Localized Parallel Competitions for the TSP

张军英 周斌

旅行商问题（TSP）是组合优化中最典型的NP完全问题之一，具有很强的工程背景和应用价值．文章在分析了标准SOM（Self-Organizing Map）算法在求解TSP问题的不足和在寻求总体最优解的潜力的基础上，引入泛化竞争和局部渗透这两个新的学习机制，提出了一种新的SOM算法——渗透的SOM（Infiltrative SOM，ISOM）算法．通过泛化竞争和局部渗透策略的协同作用：总体竞争和局部渗透并举、先倾向总体竞争后倾向局部渗透、在总体竞争基础上的局部渗透，实现了在总体路径寻优指导下的局部路径优化，从而使所得路径尽可能接近最优解．通过对TSPLIB中14组TSP实例的测试结果及与KNIES、SETSP、Budinich和ESOM等类SOM算法的比较，表明该算法既简单又能使解的质量得到很大提高，同时还保持了解的良好的稳健特性。[著者文摘]

As one of the most typical NP-complete combinational optimization problems, TSP （Traveling Salesman Problem）, which has a diversity of applications in real world, has attracted extensive research lem has been paid interest. Recently, Self-Organizing Map（SOM） based approaches to this prob great attention for its simplicity and novelty. By analyzing drawbacks of stand ard SOM algorithm for solving TSP problem, it was found that the standard SOM has a great potential for finding overall optimal solution rather than globally optimal solution for a TSP problem. Based on this, the paper proposes a new SOM algorithm for solving TSP problem, the infiltrative SOM （ISOM）, by introducing two new learning schemes, competition generalization and local infiltration. By the collaboration of the two learning schemes in that both the schemes work together in the whole learning process and initial learning focuses more on overall optimization, which is conducted by the competition generalization, while the afterward learning focuses more on local optimization, which is conducted by the local infiltration, the near-optimal solution is much more easy to be found. Experiments on public TSPLIB data show that not only the quality of the solutions is higher, but also the solutions are more pared with those by several typical SOM-based methods robust, by the proposed method comsuch as the KNIES algorithms, the SETSP, the SOM developed by Budinich, and the ESOM.[著者文摘]

雷达网数据融合的自组织模糊神经网络模型

张尤赛[1] 刘维亭[2]

对雷达网的数据融合提出了一种自组织模糊神经网络模型，分析了该模型的自组织特征及神经元感应场内模糊隶属度函数最大的WTA竞争机制，通过二维舰船目标的识别和航迹融合仿真实验，证明了该方法的有效性。

新型模糊神经网络控制器的设计及应用研究

程启明

讨论自组织竞争网络优化模糊神经网络的设计及应用研究。该设计方法运用自组织竞争神经网络（SCNN）来优化模糊神经网络结构，并采用遗传逄法来训练模糊神经网络的连接权参数，获得同时具有最佳结构和模糊神经网络（FNN），船舶操纵的仿真结果表明此法明显优于传统的PID控制器。