一种新的过程神经网络训练算法及其应用(英文)

New Training Algorithm for the Process Neural Network and Its Application

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,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。

A new training algorithm for the process neural network is presented when it is used to model an industrial process. Considering the process variables data with the prosperities of being discrete and including some pseudo ones, the data pretreatment has to be required, and then a new algorithm based on discrete Walsh conversion was used to convert the sampled data to be the direct inputs, then it can shorten the network training time and improve the network mapping precision. The model of the process neural network with the new training algorithm and two hidden-layers structure was applied to forecast the mycelium density of the glutamate fermentation process, and the simulation results were excellent.