基于SPA-WOA-SVR的青霉素发酵过程软测量建模方法  

Soft sensor modeling method for penicillin fermentation process based on SPA-WOA-SVR

作　　者：吴萍 WU Ping(Automatic Control Department,Jingjiang Secondary Specialized School Campus,Jiangsu Union Technical Institute,Jingjiang 214500,China;Automatic Control Department,Jiangsu Jingjiang Secondary Specialized School,Jingjiang 214500,China)

机构地区：[1]江苏联合职业技术学院靖江中专办学点自动控制系,江苏靖江214500 [2]江苏省靖江中等专业学校自动控制系,江苏靖江214500

出　　处：《镇江高专学报》2024年第3期88-93,共6页Journal of Zhenjiang College

基　　金：2022年泰州市科技支撑计划(社会发展项目)(SSF20220133);2023年泰州市科技支撑计划(社会发展项目)(SSF20230166)。

摘　　要：针对青霉素发酵过程中关键参数难以直接在线测量这一工程技术难题提出基于SPA-WOA-SVR的软测量方法。采用连续投影算法(SPA)选取发酵过程辅助变量,以选取的特征辅助变量(CO_(2)浓度η、溶解氧浓度D_(O)、pH值、酸流加速率μ、碱流加速率l)为输入,以发酵过程关键参数(菌体浓度X、基质浓度S和产物浓度P)为输出,基于支持向量机回归(SVR)建立软测量模型。为提高软测量模型的预测精度和稳定性,采用鲸鱼优化算法(WOA)对模型的重要参数(核函数宽度σ、惩罚因子c和不敏感系数ε)进行寻优。将构建的软测量模型应用于青霉素发酵过程关键参数预测,仿真结果表明,与传统SVR软测量建模方法相比,采用SPA-WOA-SVR软测量方法,关键参数X、S、P测试集中的决定系数R^(2)均在0.953 3以上,均方根误差R_(MSE)均小于0.029 3,特别是菌体浓度R^(2)由0.904 4提升到0.987 9,R_(MSE)由0.031 4下降为0.009 8,说明基于SPA-WOA-SVR的软测量建模方法有效提升了模型的预测性能,具有更高的预测精度和稳定性,可推广应用于一般非线性系统的软传感器建模。A soft senor method based on SPA-WOA-SVR is proposed to address the engineering technical challenge of difficult direct online measurement of key parameters during penicillin fermentation process.The Continuous Projections Algorithm(SPA)is used to select auxiliary variables for the fermentation process.The selected feature auxiliary variables(CO_(2) concentration,dissolved oxygen concentration,pH value,acid flow acceleration rate,alkali flow acceleration rate)are used as inputs,and the key parameters of the fermentation process(bacterial concentration,substrate concentration,and product concentration)are used as outputs.A soft senor model is established based on Support Vector Regression(SVR).To improve the prediction accuracy and stability of the soft senor model,Whale Optimization Algorithm(WOA)is used to optimize the important parameters of the model(kernel width,penalty factor,and insensitivity coefficient).The constructed soft senor model was applied to predict key parameters in the penicillin fermentation process.Simulation results showed that,compared with the traditional SVR soft senor modeling method,using the SPA-WOA-SVR soft senor method,the determination coefficients of key parameters and test sets were all above 0.9533,and the root mean square error was less than 0.0293.Especially,the bacterial concentration increased from 0.9044 to 0.9879,and decreased from 0.0314 to 0.0098.This indicates that the SPA-WOA-SVR based soft senor modeling method effectively improves the predictive performance of the model,has higher prediction accuracy and stability,and can be applied to soft senor modeling of general nonlinear systems.

关 键 词：青霉素发酵 连续投影算法(SPA) 鲸鱼优化算法(WOA) 软测量 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]