基于自联想神经网络的谷氨酸发酵故障诊断

Fault diagnosis for glutamic acid fermentation by an auto-associative neural network

研究了用自联想神经网络对谷氨酸发酵进行故障诊断。自联想神经网络采用一种带有瓶颈层的特殊结构,且具有单位总增益。在经过大量样本的训练之后,各变量之间能够建立起内在联系。输入信息通过瓶颈层前的压缩及瓶颈层后的解压缩过程,信息中的精华将被提取。应用自联想神经网络对发酵过程变量进行预处理,可以准确及时的进行谷氨酸发酵故障诊断。

An auto-associative neural network （ AANN ） was used to detect faults from glutamic acid fermentation. The AANN adopted a special structure with a bottle-neck layer and had a unit overall gain. The internal relation between the variables could be built after the training of a large amount of samples. The quintessence of the information could be obtained through the compression before the bottle-neck layer and the decompression after that. The simulation results showed that the AANN could ac- curately and immediately detect the faults from glutamic acid fermentation.