【作者】 刘军；

【导师】 王博；

【作者基本信息】 江苏大学 ， 控制工程（专业学位）， 2024， 硕士

【摘要】 毕赤酵母表达系统拥有目前调控机制最为严格的启动子,具有高表达、高分泌和高稳定性的特点,非常适合于外源基因的表达和调控,被认为是最成功的外源蛋白表达系统之一,在基因工程、食品工程等领域得到了广泛应用。至今,已有数千种外源蛋白成功地在毕赤酵母表达系统中得到了表达,并实现了规模化生产,如胰岛素、人血清白蛋白。该表达系统不仅在经济上具有巨大的价值,在当今社会发展中也具有重要的意义。为了使得毕赤酵母表达系统实现高质量的表达,需要对毕赤酵母发酵过程进行实时优化和动态控制,将毕赤酵母准确地控制在最佳工艺条件下进行生产。然而,毕赤酵母发酵过程机理复杂,具有高度的非线性、多变量、多阶段以及大时变特性,并且影响外源蛋白高质量表达的各个参量之间呈现出强耦合性,此外,不同的发酵批次之间呈现多工况的特性,常规以动力学模型为基础的控制方法已经难以获得良好的动态性能,优化控制更难以实现,难以满足实际发酵工业控制需求,同时,毕赤酵母发酵过程中,某些直接影响光合细菌发酵过程品质的关键参量如菌体浓度、产物浓度难以直接在线测量(新型的生物传感器在测量稳定性、操作维护条件、价格等方面还存在着严重问题,其在毕赤酵母发酵过程中的实际应用还有待时日),目前这些关键参量仍是采用离线采样、实验室分析等手段,但离线检测滞后时间长、测量误差大、易引入人为污染,且测量精度受菌种死亡、测量误差等因素影响,不能及时反映毕赤酵母发酵过程中发酵过程当前状态,难以满足毕赤酵母发酵过程中实时动态调控的要求。针对这些问题,本文提出了一种基于BDA-IPSOLSSVM的新型毕赤酵母软测量建模方法,解决了由于不同批次毕赤酵母发酵过程中操作条件的不同(发酵数据分布不同)而导致软测量模型失效的问题,实现对关键参量的精确实时预测,进一步将软测量预测模型与模型预测控制策略相结合,提出基于BDA-IPSOLSSVM的毕赤酵母发酵过程非线性补料预测控制方法,实现补料流加速率的实时动态调控。本文主要研究内容如下:首先,基于最小二乘支持向量机(LSSVM)构建毕赤酵母发酵过程软测量模型,并利用个体进化差异机制改后的粒子群优化算法对软测量模型参数进行优化。其次,针对毕赤酵母发酵时不同工况之间的数据分布差异较大导致软测量模型失效的问题,利用模糊集概念对迁移学习中的平衡分布自适应(BDA)方法进行改进,将分类问题转化为发酵过程的回归预测问题,引入改进的BDA算法对建立的毕赤酵母发酵软测量模型进行改进。将BDA-IPSO-LSSVM软测量模型应用于毕赤酵母发酵过程中关键参量的预测中。仿真结果表明,本文所提出的软测量方法优于现有的方法,在预测关键参量时表现出更高的预测精度,能够准确预测不同操作条件下毕赤酵母的发酵过程的关键参量。其次,针对传统控制算法难以对毕赤酵母发酵流加补料过程进行实时动态控制的问题,提出利用模型预测控制策略对毕赤酵母流加补料过程进行控制,并利用已构建的软测量模型建立基于BDA-IPSO-LSSVM的毕赤酵母发酵过程非线性模型预测控制模型,并利用提出的IPSO算法进行滚动优化,构建毕赤酵母发酵过程闭环补料控制系统,以实现发酵过程流加补料的精确实时控制以及毕赤酵母表达系统的高效表达。仿真结果表明利用IPSO算法进行滚动优化的预测控制算法能够有效对毕赤酵母补料的流加速率进行稳定精确控制,对于参考轨迹具有良好的跟踪效果。最后,为实时监控毕赤酵母发酵过程的动态信息,实现毕赤酵母发酵过程的可视化以及智能化控制,设计了毕赤酵母发酵过程实时监控系统。该系统将软测量模块和预测控制模块嵌入到监控系统中实现发酵过程的动态调控,首先通过传感器模块实时采集发酵过程的环境数据,然后通过数据处理模块以及通信模块将数据传入到上位机系统中,最后利用内嵌的软测量模块实现对关键参量的实时预测,利用内嵌的模型预测控制模块自动做出补料控制决策,从而实现发酵过程的实时在线预测和远程调控。更多还原

【Abstract】 The Pichia pastoris expression system features a promoter with the most stringent regulatory mechanisms currently available.This system is characterized by high expression levels,efficient secretion,and remarkable stability,making it highly suitable for the expression and regulation of foreign genes.Widely recognized as one of the most successful platforms for foreign protein expression,it has found extensive applications in genetic engineering and food engineering.To date,thousands of foreign proteins have been successfully expressed using the Pichia pastoris system,including insulin and human serum albumin,demonstrating its capability for large-scale production.Beyond its considerable economic value,this expression system holds significant importance in advancing various sectors of contemporary society.To achieve highquality expression in the Pichia pastoris expression system,real-time optimization and dynamic control of the fermentation process are essential.Accurate control of Pichia pastoris production under optimal conditions is crucial for achieving desired protein expression levels.However,the fermentation process of Pichia pastoris is inherently complex,characterized by nonlinearity in each parameter and strong coupling between various parameters influencing the expression of foreign proteins.Conventional control methods based on kinetic models have struggled to deliver satisfactory dynamic performance due to the complexity of the system.Moreover,achieving optimized control remains challenging,failing to meet the actual control requirements of the fermentation industry.Additionally,certain factors during the fermentation process directly impact photosynthesis,and key parameters such as cell concentration and product concentration are challenging to measure directly online.Although new biosensors are under development,they face issues related to measurement stability,operational maintenance,cost,and practical application in the Pichia pastoris fermentation process.Currently,key parameters continue to be measured through offline sampling and laboratory analysis,which are associated with significant lag time,measurement errors,and potential human contamination.These methods also suffer from reduced accuracy due to bacterial strain death and measurement errors influenced by various factors,making them inadequate for real-time dynamic control during the fermentation process.In response to these challenges,this study introduces a novel soft measurement modeling approach for Pichia pastoris based on BDA-IPSO-LSSVM.This method addresses the variability in operating conditions across different fermentation batches and overcomes the limitations of traditional soft sensor models.By achieving accurate real-time prediction of key parameters,the proposed approach integrates the soft sensor model with a model predictive control strategy.This leads to the development of a nonlinear multi-step fedfeed predictive control method based on BDA-IPSO-LSSVM for Pichia pastoris fermentation,enabling real-time dynamic control of the feeding flow acceleration rate.The main research contributions are as follows:Firstly,to address the challenge of significant variations in data distribution across different operational conditions during Pichia pastoris fermentation,leading to the failure of soft measurement models,the concept of fuzzy sets is employed to enhance the Balanced Distribution Adaptation(BDA)method in transfer learning.This transformation reclassifies the problem into a regression prediction issue for the fermentation process.The enhanced BDA algorithm is then introduced to refine the established soft measurement model for Pichia pastoris fermentation.Secondly,a soft measurement model for the Pichia pastoris fermentation process is developed based on the Least Squares Support Vector Machine(LSSVM).The Particle Swarm Optimization(PSO)algorithm,modified by the Individual Evolutionary Difference Mechanism,is utilized to optimize the parameters of the soft measurement model.The resulting BDA-IPSOLSSVM soft measurement model is applied to predict key parameters in the Pichia pastoris fermentation process.Simulation results demonstrate that the proposed soft sensor method outperforms existing approaches,exhibiting higher prediction accuracy when forecasting key parameters.Furthermore,it effectively predicts the fermentation process of Pichia pastoris under varying operational conditions.Secondly,a new soft sensing model for the Pichia pastoris fermentation process is proposed based on the Least Squares Support Vector Machine(LSSVM)algorithm.The model parameters are optimized using an individual evolutionary difference mechanism that enhances the traditional particle swarm optimization algorithm.The proposed BDA-IPSO-LSSVM soft sensing model is used to predict cell concentration and product concentration during the Pichia pastoris fermentation process.Fermentation experiments were conducted to establish a fermentation database and perform soft sensing simulation experiments.The results indicate that the proposed soft sensing method outperforms existing methods in predicting key parameters with higher accuracy,accurately predicting the fermentation process under different operational conditions.Thirdly,to address the real-time precise control challenges of the Pichia pastoris fermentation flow and feeding process that conventional control algorithms struggle with,model predictive control algorithms are proposed.The established BDA-IPSO-LSSVM soft sensing model serves as the predictive model in the control algorithm to achieve real-time control of the fermentation flow and efficient expression of the Pichia pastoris expression system.Simulation results demonstrate that the predictive control algorithm effectively stabilizes the flow rate of Pichia pastoris feeding and exhibits excellent tracking performance for reference trajectories.Lastly,to monitor the Pichia pastoris fermentation process in real-time and achieve process visualization and intelligence,a real-time monitoring system for the Pichia pastoris fermentation process based on the soft sensing model is designed.The system embeds the soft sensing model into the monitoring system for precise control.Environmental data from the fermentation process are collected in real-time through sensor modules,processed through data processing and communication modules,and transmitted to the host computer system.The constructed soft sensing model enables real-time prediction of key parameters,and the model predictive control algorithm automatically makes feeding control decisions,realizing remote monitoring and realtime accurate control of the fermentation process,achieving intelligence and visualization of the fermentation process.更多还原