基于混合模型及其校正策略的发酵过程实时优化研究

In order to improve production levels and economic benefits of the fermentation process, it is necessary to deeply research its optimal control problems. Current methods for optimization of the fermentation process are mostly off-line, which are based on mechanistic models. However, the models are often not well-matched and the off-line optimization results are difficult to implement, because the fermentation process is extremely complex and has slow time-varying characteristics. In this project, based on hybrid model and its correction strategy, real-time optimization techniques for the fermentation process are proposed: according to specific characteristics of the fermentation process, its multi-phase mechanistic models are established and ensemble modeling methods are used to build the empirical model so as to obtain the hybrid model, while abnormal data detection and processing problems during modeling are deeply investigated; hybrid model correction is carried out on the basis of model prediction reliability and parameter sensitivity analysis; with model prediction reliability constraints being added into the fermentation dynamic real-time optimization model as inequality constraints, high-performance optimization algorithms are developed for the optimization problem, then the implementation strategy of the optimization results is proposed, combined with the ideas of feasibility analysis and iterative learning; based on the fermentation experimental system, the real-time optimal control platform for bio-fermentation processes is built for validation and improvement of the proposed methods and for their application to typical industrial fermentation processes.

为了提高发酵过程的生产水平和经济效益，有必要对其优化控制问题进行深入研究。现有的发酵过程优化方法多为基于机理模型的离线优化，由于发酵过程极其复杂且具有慢时变特性，经常出现模型失配及优化结果难以实施等问题。为此，本项目提出开展基于混合模型及其校正策略的发酵过程实时优化研究：根据发酵过程特性，提出采用分时段思想建立其基本机理模型、并利用集成建模方法建立经验模型的混合建模思路，同时对建模过程中的异常数据检测与处理问题进行深入探讨；基于模型预测可信度和参数灵敏度分析的思想，进行混合模型的校正；将模型预测可信度约束作为不等式约束加入发酵过程的动态实时优化模型中，开发高性能的优化算法求解优化问题，并结合可行性分析和迭代学习的思想制定优化结果的实施策略；在发酵实验系统的基础上开发生物发酵过程实时优化控制平台，对所提出的方法进行验证与完善，并面向典型的发酵工业生产过程进行推广应用。

本项目在充分考虑发酵生产过程具体特性的基础上，深入研究了基于混合模型及其校正策略的发酵优化方法，提出了进行发酵过程优化控制的具体方案，解决了发酵过程优化控制中需要重点关注的问题。1.发酵过程动态混合模型的建立。从发酵动力学和物料平衡的角度出发，建立了补料分批发酵过程的机理模型，利用指示变量对发酵过程进行分段，并将分段机理模型与基于神经网络集成的经验模型相结合进行混合建模，从而获得了能够准确描述发酵过程特性的混合模型，为优化控制奠定了模型基础；2.发酵过程混合模型的校正。针对过程模型随时间推移所出现的“老化”现象给优化控制带来的不利影响，结合对模型预测可信度和参数综合灵敏度的分析进行混合模型的校正。当模型预测可信度开始下降时，首先利用最新积累的过程数据对经验模型部分进行校正。当模型预测可信度进一步下降、单靠更新经验模型部分无法进行有效校正时，基于参数综合灵敏度分析方法选择机理模型中的关键参数进行调整；3.发酵过程优化。鉴于发酵过程优化通常为多目标优化问题的实际情况，对多目标优化方法进行了研究，提出了自适应多目标混沌差分进化算法。建立了以产物量最大和基质消耗量最小为目标的优化问题，并确定了过程的决策变量和约束条件，在约束条件中特别考虑了对模型预测可信度指标的约束，利用所提优化算法对发酵过程优化模型进行了求解，并对优化结果进行了分析； 4.发酵过程优化相关算法的验证。在对发酵过程建模与优化方法研究的基础上，设计了发酵过程优化控制系统，验证了所提方法的有效性和实用性。