【作者】 潘丰；

【导师】 陈坚； 伦世仪；

【作者基本信息】 江南大学 ， 发酵工程， 2001， 博士

【摘要】 对于具有高度非线性、时变性和复杂相关性的生化过程,由于一些重要的过程变量不能在线测量,使得经典控制方法在生化过程优化控制的应用很难取得好的效果。人工智能技术为生化过程的优化控制提供了全新的方法,它能够绕过生化过程控制中建立数学模型的难点,也能够很好地利用工程师和熟练工人的一些经验,而这些经验是传统数学方法所无法利用的。本论文研究了模糊数学、人工神经网络、专家系统等人工智能方法在生化过程控制各个方面的应用。涉及生化过程变量测量和控制、数据处理、机理状态(生化时期)识别、故障诊断、状态变量的在线估计和优化控制等内容。模糊控制方法的本质优势在于能够处理过程的不确定性信息。生化过程的非线性、时变性使得技术人员无法精确确定生化过程的状态,更不能对其加以精确控制。本论文研究了生化过程的罐温和pH 的模糊控制,其应用效果良好。专家系统能够有效地利用非数值形式的信息,能够方便地引进规则库系统。基于模糊集方法和神经网络方法的知识表达大大促进了它的应用。本研究成功地应用专家系统于生化过程故障诊断、溶氧变区域控制。人工神经网络方法的一个优势在于能够通过样本学习,准确模拟各种非线性特性。这使得神经网络在生化过程建模、状态变量估计等方面具有突出的优势。准确的状态预估使得人工神经网络又能有效地应用于补料分批发酵过程优化控制问题的求解。如何将各种人工智能控制方法和经典控制方法有机结合起来,是本研究努力探讨的内容。基本观点是:(1)各类方法做各自擅长的事,并没有必要去勉强做一些不擅长的工作;(2)充分利用经典方法。因为经典方法成熟、可靠、可分析性强。本研究提出的生化过程智能控制各种方法已在工业现场应用,研究和应用结果如下: 1) 针对我国发酵企业规模相对较小而控制要求较高的情况,开发了适用于生化过程优化控制的低成本、开放型、高性价比的集散控制系统。硬件采用分层阶递的分布式结构,软件采用面向对象的模块式编程方法。自主研制、开发的FPC2000 生化过程集散控制系统具有方便、灵活、易用、简单、可靠、高性能特点,总体技术方面达到先进水平,已在十多家单位成功地推广应用; 2) 创建了智能控制系统的新型结构体系,将生化过程智能控制子系统与更多还原

【Abstract】 The control of biotechnological process is a technical challenge to the traditional control techniques, not only because that process is a strong nonlinear system with time varying parameters, but also because some variables of that process are difficult to measure. The artificial intelligence (AI) is used to solve some problems of the optimization of the biotechnological processes. An obvious merit of the proposed method is to incorporate the rich experiences of experts and engineers into the optimization without modeling the system or the process in a mathematical view. The application of the artificial intelligent technology, such as fuzzy mathematics, artificial neural network and expert system, to the optimization control of the biotechnological processes is considered in this dissertation. Some important engineering problems, which are the measurement and control of the biotechnological processes, the data processing, the identification of functional states (bioprocess phases), the fault diagnosis, the on-line estimation of state variables, and the optimization control, are discussed. The uncertain information of biotechnological process can be handled by using the fuzzy control method. In general, engineers could not easily measure and control the state variables of the process because of the time varying parameters and high nonlinearity of biotechnological processes. Thus, fuzzy control is introduced to control the temperature and the pH value in biotechnological processes. The good results suggest the effectiveness of the application of fuzzy control to the process. The information with non-value form can be used to construct the rule base system. Based on the idea of fuzzy and artificial neural network method, the express system can promote the applications of the expert system to some extent. The fault diagnosis, and the varying areas control of dissolved oxygen concentration in the biotechnological processes are achieved by using the expert system. An outstanding advantage of the artificial neural network method is that it can pass the sample learning, and then accurately emulate every kind of non-linearity. This implies the superiority of the artificial neural network method in the application of modeling and state estimation of the biotechnological processes. The optimization problem of fed-batch ferment could be solved by the accurate state prediction via the method of neural network. How to combine the artificial intelligent control with the traditional control is the researching emphasis in this dissertation. The basic viewpoint is that: (1) the suitable applications of various methods, and (2) the sufficient utilizations of the traditional methods. The proposed methods have achieved control in the industrial engineering of biotechnological processes. And the control results are listed as follows: 1) The industrial scale of the ferment factories in our country is small, and the requirement of the control performances is high, so the distribute control system, with low cost and high capability, is developed in the optimization control of biotechnology process. The layer construction is adopted in the hardware design, and the modularization programming method with the object orientation is utilized in the software design. The FPC2000 distribute control system of biotechnological processes is independently developed and manufactured with the convenient, flexible, easy, simple, and credible characteristics. It has achieved the advanced level in total technical viewpoint, and has already been successfully applied to the biotechnological process in more than ten factories. 2) The new construction of intelligent control is made. The biotechnological processes intelligent control system and the distribute control system are combined to possess advanced control functions. 3) In order to handle the strong nonlinear biotechnological processes with time varying parameters, the intelligent control technology is incorporated into the traditional distribute control system. The temperature fuzzy-PI control system, pH parameter self-tuning fuzzy control system, and expert control system of dissolved oxygen concentration varying areas were designed by using fuzzy control, expert system and tradition control technique in a inter-combining way. The intelligent control method was introduced to control the environment variables, and the control accuracy increases by 50% in comparison with the accuracy via normal control method. 4) The micro expert system of exceptional status diagnosis examination is constructed to diagnose the fault in the ferment process. The estimation technique is adopted to predict draw-off timing for batch fermentation process, in order to provide effective helps for workers to handle complicated working status. 5) The neural network model is established to estimate biomass concentration, substrate concentration and produce concentration, so as to overcome the strong non-linearity in the biotechnological processes with time varying parameters. The process is modeled by using the improved BP learning algorithm. In ploymyxin production, the neural network model trained by the production datum has high estimate accuracy. The estimate error is less than 5%. This indicates that the type of neural network model, the category of import variants, the method of network learning have been suitably selected. 6) Although lacking the accurate mathematics model of the biotechnological processes, the neural network non-line prediction control method that absorbs the genetic algorithm can search for the optimal trajectories of substrate fed rate, pH and dissolved oxygen concentration in ferment process. The optimal trajectories can be trailed by on-line adjustments. The optimization results can be revised by the technologic engineers, and then be applied in the production. By the optimization, the ferment time shortens 5%, and the ploymyxin production increases 3%. The economic benefits of these applications are obvious by increasing the production and reducing the energy consume. By incorporating the intelligent control technique into the tradition distribute control system, the new construction of biotechnological processes intelligent control is made. The full intelligent treatments of control, fault diagnosis, on-line estimation of state variants, and optimization of fed-batch strategy in ferment process are carried out elementarily, and the goal of optimizing ferment production is achieved. The research on the application of artificial intelligent method in the biotechnological processes is still developing. And it needs urgent improvement. Therefore, in my opinion, researcher could consider the research in the following aspects: to perfect theoretical results, to further study of the biotechnological processes function, to combine different control systems, to apply the mixing intelligent system in the biotechnological processes.更多还原