基于可拓神经网络的变压器故障诊断

Transformer Fault Diagnosis Based Extension Neural on Networks

针对BP神经在变压器故障诊断中用于模式识别时,存在训练准则和分类准则不一致而导致的样本识别率降低和网络训练速度缓慢的问题,采用了可拓学的扩缩变换,通过在输出空间中用一个特定的区域(称作教师区域)来代替教师信号,然后将可拓神经网络用于变压器故障诊断中。通过实例证明,可拓神经网络模型的训练速度有了极大提高,模式识别问题得到彻底解决。

There is a problem that the training rule and classifying rule is not consistent when the mostly used BP neural network is applied in the transformer fault diagnosis. It causes recognition rate of samples to reduce and network training slow. By extension transform, using a region in the outputting space replaces apoint, then using the extension neural network in the transformer fault diagnosis.The example proves the training speed is extraordinarily improved and the problem of inconsistent training rule and classifying rule is solved.