基于平行坐标轴的BP网络可视化技术探析

Research on visualization of BP neural network based on parallel coordinates

基于平行坐标轴的BP网络可视化技术能够将BP网络的多维训练样本以合理、易读的方式展现出来,并揭示其中的潜在规律。根据BP网络中的输入节点,输出节点信息构造相应的平行坐标轴,结合聚类方法和颜色、密度、大小等图形属性,得到清晰、易于理解的视图。实验结果表明:该方法简单有效,可以帮助挖掘数据背后隐藏的关系、规律及模式。

In order to illustrate the training samples of BP neural network and reveal the potential law,a parallel coordinates-based BP network visualization technique is proposed.In this approach,the parallel coordinates are constructed according to the model of BP neural network.In particular,the clusters＇colors,size,density and other features are used to improve the parallel coordinates to create clear intelligible visualization.The results show that the parallel coordinates-based BP network visualization is simple,effective and suitable to reveal the hidden relationships,regularity and mode in training samples.