赖氨酸发酵过程关键参数的模糊神经网络逆软测量研究

Research of soft sensor based on fuzzy neural network inverse system for lysine fermentation process

针对微生物发酵关键生物量参数(基质浓度、菌丝浓度和产物浓度)难以直接测量的问题,以赖氨酸发酵过程为研究对象,采用基于"虚拟子系统"的模糊神经网络逆系统软测量方法对关键生物量参数进行在线估计。假定在发酵过程内部存在一个以不可直接测量参数为输入,直接可测参数为输出的"虚拟子系统",并建立"虚拟子系统"的数学模型。再构造"虚拟子系统"的模糊神经网络逆系统,将逆系统串接在"虚拟子系统"后构成复合伪线性系统,得到动态软测量模型,实现不可直接测量参数的在线估计。实验结果表明:该方法能很好地实时估算赖氨酸发酵过程关键参数,为进行赖氨酸发酵过程补料优化控制打下良好的基础。

In order to overcome the difficulties of crucial biochemical variable on-line measurement in fermentation process, a soft sensor method based on Fuzzy Neural Network （FNN） inversion of the ＂subsystem＂ is proposed to estimate the crucial process variables in lysine fermentation process. It is supposed that there exists a ＂virtual sub- system＂ with direct-immeasurable variables as its inputs and direct-measurable variables as its outputs. Based on this supposition, the mathematics model of the ＂subsystem＂ is established. And the corresponding inversion of the ＂virtual subsystem＂ is constructed using FNN and is connected in series after the ＂virtual subsystem＂, so a pseudo-linear system is established, which can be regarded as the desired dynamic soft sensor model, because its outputs are just the direct-immeasurable variables. Experiment results show that this method can on-line estimate the direct-immeasurable variables and lays a good foundation for the fed-batch optimal control of lysine fermentation process.