基于Walsh变换的过程神经网络建模及应用

Modeling and Application of Process Neural Network Based on Walsh Conversion

着重研究了基于离散数据的过程神经网络建模问题。考虑到来自现场的过程变量数据基本都是离散的采样数据,并且其中存在伪数据的情况,故先对离散采样数据进行预处理,然后采用离散Walsh变换法对数据进行转换,即将网络输入函数和权函数在Walsh基下映射为一组新的时变向量,将积分聚合运算简化为向量内积运算,实现离散采样数据对连续网络的直接输入。应用所建立的过程神经网络模型对发酵过程菌体浓度进行了预测,取得了较好的效果。

The modeling problem of the process neural network based on the discrete data is studied. Considering the data of process variables with the prosperities of being discrete and including some pseudo ones, the data pretreatment is given. And then a method based on discrete Walsh conversion is used to convert the sampled dada to be the direct inputs to network, The input of the network is mapped as a set of the new variable vectors. The model of the process neural network with two hidden-layers based on Walsh conversion is used to forecast the cell concentration of the glutamate fermentation process, and the good results are obtained.