基于粒子群算法优化LVQ神经网络的应用研究

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PSO—LVQ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。

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

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.