CAJ | 학술논문

针对复杂发酵过程逆软测量模型难以建立和估计精度不高的问题，提出一种基于偏最小二乘和变量投影重要性指标的PLS-VIP神经网络逆软测量方法。该方法在“内含传感器”逆软测量模型的基础上，通过计算变量投影重要性指标，确定逆模型中各辅助变量对关键生化量的贡献率，建立了辅助变量少、估计精度高的逆软测量模型。与传统逆软测量方法相比，克服了依赖于数学模型和复杂算法推导的问题。将其应用到青霉素发酵过程中生化量估计中，利用Pensim仿真平台进行实验，仿真实验结果表明，该方法的逆模型比传统逆模型构建简单、估计精度高。
침대복잡발효과정역연측량모형난이건립화고계정도불고적문제，제출일충기우편최소이승화변량투영중요성지표적PLS-VIP신경망락역연측량방법。해방법재“내함전감기”역연측량모형적기출상，통과계산변량투영중요성지표，학정역모형중각보조변량대관건생화량적공헌솔，건립료보조변량소、고계정도고적역연측량모형。여전통역연측량방법상비，극복료의뢰우수학모형화복잡산법추도적문제。장기응용도청매소발효과정중생화량고계중，이용Pensim방진평태진행실험，방진실험결과표명，해방법적역모형비전통역모형구건간단、고계정도고。
To overcome the difficulties in establishing an inverse soft-sensing model and its lack of accuracy in the complex fermentation process, an inverse soft-sensing method based on partial least squares and variable importance in projection methods called PLS-VIP is proposed. Based on the “assumed inherent sensor” inverse soft-sensing model, an inverse soft-sensing model with fewer variables and high accuracy is obtained through calculating VIP values. Compared with traditional inverse soft-sensing method, it is independent of mathematical model and complex derivation. It is applied to a soft sensor modeling for penicillin fermentation process using Pensim simulation platform. Simulation results show that the proposed method has higher precision and better performance than the original inverse soft-sensor method.