CAJ | 학술논문

针对生物发酵过程中生化变量难以在线检测的问题,提出一种基于变量投影重要性( Variable importance in the project,VIP)方法优化的神经网络逆系统软测量模型. 根据逆系统理论建立发酵过程生化变量的软测量模型,由于发酵系统的复杂性,逆系统软测量模型具有不惟一性,且难以得到精确的表达式. 文中提出采用VIP方法对逆系统软测量模型的辅助变量进行优选,以对主变量贡献率较高的变量作为软测量模型的辅助变量,离线采集发酵过程各变量值,训练神经网络近似逆系统软测量模型,得到优化的神经网络逆系统软测量模型,实现发酵过程中菌体浓度和基质浓度的在线估计. 利用Pensim平台采集数据,对所提方法做了仿真实验,结果表明:经过优化辅助变量的神经网络逆系统软测量方法具有更高的估计精度和泛化能力.
침대생물발효과정중생화변량난이재선검측적문제,제출일충기우변량투영중요성( Variable importance in the project,VIP)방법우화적신경망락역계통연측량모형. 근거역계통이론건립발효과정생화변량적연측량모형,유우발효계통적복잡성,역계통연측량모형구유불유일성,차난이득도정학적표체식. 문중제출채용VIP방법대역계통연측량모형적보조변량진행우선,이대주변량공헌솔교고적변량작위연측량모형적보조변량,리선채집발효과정각변량치,훈련신경망락근사역계통연측량모형,득도우화적신경망락역계통연측량모형,실현발효과정중균체농도화기질농도적재선고계. 이용Pensim평태채집수거,대소제방법주료방진실험,결과표명:경과우화보조변량적신경망락역계통연측량방법구유경고적고계정도화범화능력.
To solve the online measuring of biochemical variables in the fermentation process, a neural network inverse soft sensing method which is optimized using the variable importance in the project( VIP) is proposed. According to the inverse system theory,a soft sensing model of biochemical variables is constructed. Due to the complexity of the fermentation process,the soft sensing model is not unique and not exact. This paper proposes that secondary variables should be optimized using the VIP method. The variables which have greater contribution to key variables are selected as the secondary variables of the soft sensing model. This paper collectes the fermentation process data offline and trains neural network approximating complex soft sensing model. The optimal neural network inverse system soft sensing model is obtained. It can estimate the mycelium concentration and substrate concentration online. Numerical simulations based on the Pensim data platform show that the optimal soft sensing model has higher estimation accuracy and stronger generation ability.