基于软测量的纺织工业生产过程鲁棒运行优化问题研究

Taking the production process of textile industry as the background, this project is focused on a class of robust operational optimization problem of production processes in textile industry, in which it is very difficult to achieve the on-line measurement of control objectives and there exist many disturbances in the control variables. The research task of this project is to explore and propose the soft sensor modeling methods based on data analytics and the optimization theory, modeling methods, and solution methods for the robust operational optimization problem of production processes in textile industry. For the difficulty of achieving the on-line mesurement of control objectives, the project will research the AdaBoost learning algorithm based on dynamic threshold and the abnormal data detection method combining the modeling process so as to propose a soft sensor modeling method with high precision. For the disturbances in control variables, the project will introduce the robust otpimization into the operational optimization problem of production processes in textile industry and research the modeling and solution methods of their integrated optimization problem. A multi-objective robust optimzation algorithm based on evolutionary computation will be proposed, in which a population diversity maintainance method based on estimation of distribution method and a local search based on optimization methods will be also incorporated. Finally, the proposed theory and methods will be applied to the sizing production process and a software system for the robust operational optimization will be developed so as to test the practical efficiency of the propoased theory and methods. The reseach results of this project can provide a set of theories and optimization methods for the soft sensor and the robust operational optimization of production process in textile industry, and consequently can help the textile enterprises to improve product quality, reduce production cost and environmental pollution. Therefore, the research of this project has important theoretical significance and application value.

针对从纺织工业生产过程中提炼出的一类控制目标难以在线检测且控制变量存在扰动的生产过程鲁棒运行优化问题，探索基于数据解析的软测量建模技术，以及生产过程鲁棒运行优化理论、建模方法和求解算法。针对控制目标难以在线检测问题，研究新型AdaBoost动态阈值回归学习算法，以及与建模过程相结合的异常数据检测方法，提出高精度的软测量建模方法；针对控制变量存在动态扰动问题，将鲁棒优化引入进来，研究纺织工业生产过程鲁棒运行优化问题的建模与求解方法，提出基于进化计算的多目标鲁棒智能优化算法，以及基于分布估计的种群多样性保持方法和基于最优化的局部搜索算法；最后，将所提出的理论与方法应用到纺织工业的上浆生产过程中，开发鲁棒运行优化软件系统，进行应用验证。本项目的研究将为纺织工业提供一套软测量和生产过程鲁棒运行优化的理论与方法，帮助纺织工业企业提高产品质量、降低生产成本、减少环境污染，具有重要的理论意义和应用价值。

本项目按照任务书的要求，对纺织工业生产过程中控制目标难以在线检测且控制变量存在扰动的生产过程鲁棒运行优化问题进行了理论研究和应用验证。1. 在工业软测量建模问题上，首先，针对工业大数据不可避免地存在异常点的问题，从对数据本身进行分析入手，提出了基于AFEMD的自相关去噪方法；另一方面将将数据的异常点检测与软测量建模过程相结合，提出了基于KMDB的异常点检测方法。其次，针对预测精度难以提高以的问题，建立了集成软测量模型，并结合工业应用中对模型在线更新的需求，将增量学习思想加入到集成建模算法中，提出了增量AdaBoost.RS集成建模算法和基于动态数据的在线增量算法，实现了模型的动态更新。2. 针对不确定情况下的纺织工业生产过程鲁棒运行优化问题，在上述软测量基础上，考虑工艺约束和运行优化要求，建立了纺织生产过程鲁棒运行优化模型；并针对模型特点设计了鲁棒智能优化算法的混合框架，提出了基于HPSO-GA算法的鲁棒优化算法和改进HB算法，并获得较好的鲁棒最优解。3. 以纺织工业中浆纱生产过程为背景，对所提出的软测量建模及生产过程鲁棒运行优化理论和算法进行了应用验证。. 在该项目的支持下，共发表和录用论文20篇(标注资助)，其中SCI收录期刊论文11篇，EI收录期刊论文3篇，国际会议论文3篇(EI收录2篇)；授权发明专利1项，申请发明专利3项；培养硕士生5名。