基于潜变量迁移模型的复杂工业新过程实时优化方法的研究

项目名称： 基于潜变量迁移模型的复杂工业新过程实时优化方法的研究

项目编号： No.61503384

项目类型： 青年科学基金项目

立项/批准年度： 2016

项目学科： 自动化技术、计算机技术

项目作者： 褚菲

作者单位： 中国矿业大学

项目金额： 21万元

中文摘要： 基于数据驱动模型的复杂工业过程实时优化技术以其建模速度快和易于实现等优势成为当前研究的热点。然而对于刚投入生产的新过程或生物发酵等小批量、多品种的实际工业过程来说，往往因为缺乏充足的过程数据，而导致该技术难以实施。本项目基于潜变量迁移建模思想，提出利用已有相似旧过程的丰富过程数据辅助和加快复杂工业新过程运行优化控制的实现。通过研究离群点对迁移模型的影响，提出集成鲁棒非线性建模技术的潜变量迁移建模方法；在此基础上，结合滑动窗口、数据相似性分析、模型预测误差和置信域的评估，指导迁移模型的在线更新和完全替换；研究潜变量迁移模型预测有效性约束条件的构造方法，并利用梯度信息对优化模型的目标函数和约束条件进行在线补偿，以解决模型不确定性带来的过程-模型最优性条件不匹配问题；最后利用序列二次规划求解优化问题。将所提出的方法在已有的仿真实验平台上进行验证和完善，并逐步推广应用到1-2个典型的复杂工业过程。

中文关键词： 实时优化；优化模型；性能指标；运行优化

英文摘要： Data driven model based real-time optimization of complex industrial process has become the focus of current research due to the superiority of fast modeling and easy to implementing. However, for the new process just went into operation or small quantities multi-species practical industrial processes such as biological fermentation process, this technology is difficult to carry out for the lack of sufficient process data. This project is based on the idea of transfer of process models with latent variable technology, and proposes to employ the sufficient process data of the existing old similar process to aid and accelerate the optimization of the new process. By studying the effect of outlier on the transfer model, the latent variable transfer modeling method integrated robust nonlinear modeling technology is proposed; On the basis of the aforementioned study and combining with moving window and analysis on the similarity of data, the online update and complete replacement of the transfer model is guided by evaluation of the model prediction error and confidence region; Research on the method for constructing the constraints ensuring the validation of latent variable transfer model predictions and use gradient information to compensate the objective function and constraints of the optimization model, solving the problem of the mismatch of the necessary conditions of optimality of the plant and model which is brought by the model uncertainty; Finally, sequential quadratic programming is used to solve the optimization problem. The proposed methods are validated and improved on the existing simulation/experimental platforms, and applied to 1 ~ 2 typical complex industrial processes by degrees.

英文关键词： Real-time optimization;Optimization model;Performance index;Operations optimization