黄酒发酵过程的建模与参数估计

As a traditional Chinese wine, the annual production value of yellow wine is nearly 20 billion Yuan with extensive consumer group. Currently, the manual manufacture of yellow wine causes a lot of problems (e.g. poor reliability of the quality and different batches). Therefore, there is a great challenge in the production mode of yellow wine, and the creative research on automation in producing yellow wine is desirable for changing such a conventional industry.Benefiting from the computer control system of the rice wine fermentation that has already been developed by this project, the basic idea of this project is that from perspective of statistical learning, we shall utilize the iterative learning algorithm upon the collected data to gradually track the optimized process route, in light of the characteristics of the intermittent control of the rice wine fermentation. Then, the fast track of the designed route can be achieved via the iteration learning rate on the direction of batch of the control quantity. The intelligent optimization methods, mainly including support vector machine and principal component analysis, will be applied with soft-sensing technique, in order to establish the "black box" model for studying the fermentation process on the basis of the existing data. Afterwards, the modeling effect and the actual control results will be compared, and then the relevant parameters will be estimated online accordingly. By studying the mechanism of rice wine fermentation and then setting up the corresponding fermentation dynamic model, it is more accurate with quantized terms to understand both complex biochemical reactions and the mechanism of cell metabolism in the process of rice wine fermentation, achieving the further optimization control. As such, both the quality and production efficiency will be improved. In a word, the creative research of this project on the optimization control of the rice wine fermentation will have great significance to the advancement of rice wine industry.

黄酒是中国传统的酒种，有近200亿的年产值和众多的消费人群。目前，以人工为主的黄酒生产方式,导致质量及批次稳定性差等众多问题，黄酒行业面临巨大的挑战,亟需基于自动化的创新技术来改造这一传统产业。本项目利用已研发的黄酒发酵计算机自动控制系统，针对采集的大量数据样本，用统计学的思路，以支持向量机、主元分析为主的智能优化方法，结合软测量技术，研究基于数据的发酵过程"黑箱"模型，将建模效果与实际控制结果进行比较，对相关的参数在线估计和寻优。结合黄酒发酵间歇控制的特点，研究采用迭代学习控制算法,建立批次方向上的控制量迭代学习率，逐渐跟踪优化后设定的工艺轨线。通过对黄酒发酵机理的研究及建立发酵动力学模型，更加准确和定量的认识其过程中的复杂生化反应，微生物的平衡生长机理，实现更进一步的优化控制，提高黄酒的品质和生产率。本项目的黄酒发酵优化控制创新性理论研究，将对黄酒行业的技术进步有着非常重要的推动作用。

项目背景：黄酒是我国特有的传统酒种，其发酵过程目前仍主要靠人工控制来实现，造成批次质量的稳定性较差。随着产能及质量需求的提升，迫切需要具有高鲁棒性的控制系统来实现其过程自动化，而构建准确描述发酵过程特性的数学模型是建立自动控制系统的基础，黄酒发酵过程是边糖化边发酵过程，其发酵系统极其复杂，且缺乏足够的发酵动力学数据，目前鲜有研究涉及黄酒发酵过程的建模。.主要研究内容与结果和数据：（1）本研究以黄酒酿造过程为研究对象，通过实验模拟黄酒发酵过程，基于高效液相色谱技术(HPLC)测定了不同酶和曲在不同温度梯度下前、后酵结束两个关键点酒醪中主体成分含量，利用SAS分析了产物浓度的差异性，确定前酵阶段为黄酒主体成分生成阶段。另外通过实验获得了不同温度下黄酒前酵过程数据；（2）针对曲中酶的种类及比例不确定，基于所得的数据以低阶为目的建立了适合单酶、双酶和黄酒前酵糖化过程的模型结构；（3）针对黄酒醪液中酵母难以分析的问题，以葡萄糖为底物，系统研究了不同温度和不同起始底物浓度下酵母Su-25的动力学变化。并分别利用典型的Hinshelwood模型和Monod模型对该过程进行了仿真验证，结果表明Monod模型更适合描述Su-25发酵的动力学过程，从而建立了黄酒发酵中酵母的软测量模型；（4）基于前期建立的糖化模型和发酵模型，构建了以寡糖、乙醇和溶氧等为状态变量的黄酒前酵双边发酵过程 (SSF) 模型，并分别利用所采集的实验室水平和工厂实际生产过程数据进行了模型辨识与验证；（5）针对传统反馈控制或在线控制技术难以直接应用于包括黄酒在内的间歇发酵过程，本文提出基于广义预测控制的思想，通过分别优化黄酒发酵初始条件和温度条件来实现黄酒发酵过程优化的策略。.科学意义：本文所获得的实验数据对黄酒工业生产具有实际的指导意义和参考价值，所构建的模型对实现黄酒发酵过程自动化、提高批次稳定性具有重要的理论意义和应用价值。