基于粒子群算法优化LVQ神经网络的应用研究  

Research and Application of LVQ Neural Network Based on Particle Swarm Optimization Algorithm

作　　者：张超[1] 魏三强[1] 胡秀建[1] 梁西陈[1] 

机构地区：[1]宿州职业技术学院计算机信息系,安徽宿州234101

出　　处：《贵州大学学报（自然科学版）》2013年第5期95-99,共5页Journal of Guizhou University:Natural Sciences

基　　金：安徽省高校省级自然科学基金项目(KJ2012Z405);安徽省高等学校省级优秀青年人才基金资助项目(2012SQRL263)

摘　　要：为了解决LVQ神经网络在应用时对初始权值敏感的问题,基于粒子群算法提出PSOLVQ算法。PSO-LVQ算法利用PSO为LVQ神经网络寻找最适应的初始权值。算法的适应度函数定义为初始权值和输入样本集的平均聚集距离与最大聚集距离的变化率。该定义将输入样本集的数据分布特征作为PSO优化LVQ初始权值的依据。利用PSO-LVQ算法对乳腺癌进行诊断实验,并与其它相关算法进行比较。研究结果表明:PSO-LVQ神经网络算法在收敛性和分类准确率上都有改善和提升,乳腺癌诊断平均准确率可达95.94203%,最高可达100%,适用于乳腺癌的辅助诊断。In order to solve the problem of LVQ being sensitive to the initialization, PSO-LVQ algorithm was proposed based on particle swarm algorithm. PSO-LVQ made use of PSO algorithm to find the best initial weights for LVQ neural network. Algorithm fitness function is defined as the rate of the clustering distance changing, the average and maximum clustering distances between the initial weights and the set of input samples. The PSO algo-rithm was used to optimize the initial weight of LVQ neural network on the basis of sample set of data distribution characteristics. Based on PSO-LVQ algorithm, a breast cancer diagnostic experiment was conducted and the results were compared with other algorithms. The results show that, using this algorithm, the average accuracy rates are 95. 94203% and the best accuracy rates are 100%, and the classification accuracy is higher and appli- cable for breast cancer diagnosis.

关 键 词：LVQ PSO SOM 适应度函数 神经网络 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]