基于深度极限学习机的蔗汁澄清过程协同优化控制方法研究

Sugarcane juice clarification is a key process in sugar production, which is very complicated and affected by many factors. Therefore, most of the domestic sugar-refinery adopt manual control mode or local link control mode, which is low clarification effect, low production efficiency and high exhaust emissions. In view of this situation, this project starts from the analysis of material flow, energy flow, information flow and entropy in the clarification process. The whole clarification system is regarded as three subsystems based on material flow, energy flow and information flow, and then the project analyze the clarification system by analyzing the mechanism of mutual coupling influence, transformation and entropy distribution of each subsystem. In order to achieve the best results of clarification, the highest efficiency and the lowest total emissions, the system of material flow, energy flow and information flow dynamic collaborative optimization modeling method based on minimum entropy production is studied to explore a new way that can dynamically adjust the optimal process parameters with the change of production conditions. Applying the extreme learning machine as the core, the deep learning machine as the main line and the optimal control technology index of the collaborative optimization as the control target, this project combine the theory of deep extreme learning machine and research a new control method for optimal control of the clarification process based on deep extreme learning machine. The goal of this research is to coordinate various interactions in the production process by optimizing the synergistic control of the entire process. Establishing such a multi-factor, strong coupling, large nonlinearity and uncertainty process of collaborative optimization control method in complex production process, not only can significantly improve the juice clarification process performance and efficiency, but also reduce the difficulty of process control and decrease pollution emissions.

蔗汁澄清是制糖生产的关键环节，过程异常复杂。目前国内糖厂大都采用人工或局部环节参数的控制模式，清净效果差、效率低、残硫量和废气排放高。针对此现状，项目拟从澄清过程物质流、能量流和信息流的熵产分析入手，将澄清系统看作三者组成的系统，分析各子系统相互耦合作用和有效协调的机制；以澄清过程效果最好、效率最高和排放最低为总目标，研究基于熵产最小的澄清系统物质流、能量流和信息流动态协同优化的方法，探寻一种能跟随生产工况变化动态调整最优工艺指标的新方式。有机结合深度极限学习机，以极限学习机为核心，以深度学习机为主线，以系统全局协同优化的最优工艺指标为控制目标，研究一种基于深度极限学习机的澄清过程运行优化控制的新方法。目标是通过整个澄清过程的全局协同优化控制，协调生产过程各种相互作用力，提升澄清效果和效率, 降低排放，并形成此类具有多因素、强耦合、大非线性和不确定性复杂生产过程的协同优化控制方法。

国内外蔗汁澄清的工艺方法主要包括亚硫酸法、碳酸法和“两步法”。目前国内企业甘蔗混合汁澄清大都采用亚硫酸澄清方法，生产过程基本采用人工操作或局部环节参数自动控制模式，存在人为因素影响大、生产能耗高、效率低、清净效果差、残硫量和废气排放高等问题。. 项目有机结合协同原理、多目标协同优化和深度极限学习机理论，基于熵产分析研究了澄清过程各工作单元之间的基本特性及其相互影响的机制关系，将理论探索、数学推演、计算机模拟仿真以及实验研究相互呼应和有机结合，揭示了澄清过程物质流、能量流和信息流作用机理和有效协同的机制。以澄清过程效果最好、效率最高和排放最低为总目标，基于深度极限学习机建立澄清系统动态多目标协同优化模型，利用小生境多目标粒子群优化算法进行优化求解，突破了澄清生产过程自适应动态优化控制技术关键，实现了生产过程各种相互作用力的有效协调和整个澄清过程的协同优化控制。鉴于蔗糖生产过程糖浆在线锤度测量的难题，研究一种基于深度学习算法和微波雷达脉冲反射信号特征实现糖浆浓度信息检测和量化的方法，突破制炼生产过程糖浆锤度稳定、准确及可靠检测的技术关键。为了验证本项目所提方法的有效性，将所研究的方法及系统在南宁糖业股份有限公司多家公司分别进行实施应用，大大改善澄清质量和生产效率，优化效果十分明显，为企业带来显著经济效益。研究成果为蔗汁澄清生产过程创新设计与优化控制提供理论指导，对同类复杂机械设计与优化硏究同样具有借鉴参考价值。. 项目取得重要理论创新研究成果，公开发表SCI论文共10篇，其中中科院一区和二区期刊5篇，获授权发明专利5项、软件著作权3件，培养研究生7人，其中1人获广西大学优秀论文，并在自治区研究生论文抽查中获全优评价。