【摘要】 研究了测量向量的特征。提出了利用Voronoi多胞体选择特征样本的新方法。该方法由 3部分组成 :置信判决 ,首正归一及边界样本选择。利用该法生成的训练集不仅可以大大减少样本数从而提高训练速度 ,而且通过调节故障类别分割面位置 ,还可提高故障诊断准确率。更多还原

【Abstract】 In this paper, the character of measurement-vector is studied. Based on voronoi cells, a new approach to conform the training sets of a neural network is proposed. The approach is composed of three parts: confidence determination, normalization, near-bound selection. The approach can be used not only to drastically reduce the numbers of the training samples and increase the training speed, but also to provide a probable way to improve the accuracy of fault diagnosis by adjusting the position and shape of the splitting planes.更多还原