改进的单层前向神经网络用于核电百万千瓦级汽轮发电机的故障诊断

IMPROVED SINGLE-LAYER FEEDFORWARD NEURAL NETWARKS APPLIED TO FAULT DIAGNOSIS FO THE MILLION-KILOWATT-CLASS NUCLEAR POWER TURBOGENERATOR

为对核电百万千瓦级汽轮发电机进行故障诊断 ,本文作者提出了一种“改进的单层前向人工神经网络 ,”这种神经网络克服了常规的单层前向神经网络仅是线性可分及学习过程中需重复提交一系列要记忆的输入模式来调整权值之不足 ,从而大大减少了分类时的迭代次数。本文在独特设计的多功能“核电百万千瓦级汽轮发电机”轴系振动模拟试验台上验证了这种方法的有效性。

The Improved Single_layer Feedforward Neural Networks are presened in the paper for fault diagnosis of the million_kilowatt_class nuclear power turbogenerator.This kind of neurat networks avoids the disadvantages of the normal singlelayer feedforward neural networks which can only be linearly disbranched and has to be repeatedly fed a series of input models for adjusting the weight values in the learning process.The effectiveness of the Improved Single_layer Feedforward Neural Networks is confirmed on a specially desigced multifunctional simulating test_bed for the million_kilowatt_class nuclear power turbogenerator.