Abstract: Lysine fermentation process ,which is characterized by nonlinear、large time-delay、multivariable dynamic coupling, is very difficult to realize crucial biochemical parameters on-line measurement. The soft-sensing modeling method was proposed based on PSO-SVM inversion theory. Firstly, the reversibility of system was analyzed and the inverse extention model was constructed by introducing the feature information and abandon the secondary information, and the inverse extention model was built offline with SVM which has a good fitting capability and adjusted online by PSO based on the error information between actual fermentation process input and model output. Finally, the inverse extention model adjusted was cascaded after the fermentation process to be soft sensing model. Simulations on a fed-batch L-lysine fermentation process shows that the crucial biochemical parameters can be predicted. The results also show that proposed method is more accurate compared with the alternative modeling methods.

摘要： 针对赖氨酸发酵过程非线性、大滞后、多变量动态耦合，关键生化参数难以实时在线测量等问题，提出一种改进的粒子群-支持向量机(PSO-SVM)逆发酵过程软测量建模方法。首先分析逆系统的存在性，并结合赖氨酸发酵过程，引入发酵特征信息和舍弃次要信息构造逆扩展模型；然后利用支持向量机离线辨识初始逆扩展模型，并根据系统输入与模型输出的偏差信号，采用粒子群算法对初始逆扩展模型进行在线校正；最后将校正后的逆扩展模型串联在原发酵过程之后构成复合伪线性系统，实现不直接可测关键生化参数的在线预测。以L-赖氨酸流加发酵过程为例，验证了所提算法能够对发酵过程关键生物量参数进行较准确的在线预测，较普通的SVM逆建模方法具有更高的预测精度。

关键词: 粒子群算法, 支持向量机, 逆扩展模型, L-赖氨酸发酵, PSO algorithm, SVM, Inversion extention model, L-lysine fermentation process