【作者】 陈仕学；

【导师】 阮晓钢；

【作者基本信息】 北京工业大学 ， 控制理论与控制工程， 2004， 硕士

【摘要】 摘要本文主要研究青霉素补料分批发酵过程的建模与控制。青霉素发酵过程是一个具有复杂非线性的生化反应系统。发酵过程中状态的调节控制决定了青霉素的产物效率，因此对青霉素发酵过程进行建模与控制研究具有现实意义。神经网络是解决非线性系统问题的有效途径，所以将神经网络用于青霉素补料分批发酵是一个值得研究的方向。本文采用径向基神经网络（RBF 神经网络）描述青霉素发酵反应过程，并对其进行控制，取得主要研究成果如下：首先，本文利用青霉素发酵过程的机理模型产生实验用数据,根据得到的数据训练神经网络,建立了基于径向基神经网络的青霉素发酵过程模型 。该模型可用于发酵过程中状态变量的估算和预测，而且在已知初始条件与控制点的情况下，可以藉此模型进行仿真以估计底物、产物与菌体浓度的变化趋势，特别是该模型可用于估计最大产物浓度出现的时间，这对实际工作很有指导意义。然后，本文根据 RBF 神经网络具有参数线性化的结构特点设计了基于 RBF神经网络的非线性内模控制器。利用该方法对青霉素补料分批发酵过程进行研究，当以添加基质浓度作为操作变量以控制菌体生长时，通过仿真评价了所构成的系统的品质指标。仿真实验结果表明，基于 RBF 神经网络的内模控制器可以用于青霉素发酵过程的控制,并具有良好的鲁棒性。本文部分研究成果发表在《第二十二届中国控制会议论文集》和《中南工业大学学报（自然科学版）》上，其中“基于 RBF 神经网络的青霉素发酵过程的模型辨识”一文被 EI 收录。本课题得到了国家自然科学基金的资助。更多还原

【Abstract】 In this thesis, the modeling and control of the penicillin fed-batchfermentation process were studied. Penicillin fermentation process is a complex biochemical and nonlinearsystem. The efficiency of penicillin fermentation is decided by the states of thefermentation process, so adjusting and controlling the states is very important.Thus the study about the modeling and control has practical effect for penicillinfermentation process. Neural networks are effective in solving nonlinear problems.Therefore, it’s promising to apply neural networks to the modeling and control ofthe penicillin fermentation process. This thesis uses Radial Basis Function neuralnetwork (RBF neural network) to describe and control the penicillin fermentationprocess. The main contributions of this thesis are as follows: Firstly, an identification model for penicillin fermentation process isdesigned based on Radial Basis Function network. The training and test data getfrom mechanism model. Using this model, we can know in advance the tendencyof the states change of the fermentation process under given conditions. This isvery useful in practice. Secondly, based on RBF neural network’s advantage of linearity in parameter,an RBF network based nonlinear internal model controller is designed.Simulations results of applying this method to the controlling of penicillinfed-batch fermentation process shows that the proposed controller is efficient. Parts of research results have been published on “Proceeding of the 22ndChinese Control Conference” and “Journal of Central South University ofTechnology (nature science)”. The “Model building in the penicillin fermentationprocess based on RBF neural networks for identification” paper published on“Journal of Central South University of Technology (nature science)” is indexedby EI. The Nature Science Foundation of China under grant 60274060 supportsthis project.更多还原