基于关键参数动态估计的复杂工业过程实时优化控制方法

With the characters of the complex industrial process considered, such as broad ranges of operating conditions and various disturbances, this project studies a real-time optimization control method which combines technologies of dynamic modeling, performance assessment and intelligent optimization. A semi-supervised clustering method based on active learning is presented to give reasonable classification for the process data by mining the association between data items. In order to estimate the key process parameters accurately and with real time manners modeling method based on ensemble learning and the confidence coefficient are studied. To make all-around and accurate performance assessments to the accurate industrial process, the comprehensive performance assessment method with multi-indexes and multi-hierarchies based on industrial data is, and multi-indexes are synthesized by the information entropy. A multi-objective robust optimization method based on event driven mechanism is proposed. In this method, the optimization problem is divided into optimization events which correspond to different optimization models; and the solutions to the optimization models which are the parameters of the controller should guarantee good robustness as well as acceptable control performance. This project will build a real-time optimization control structure including dynamic estimation of key parameters, comprehensive performance assessment and real-time optimization, and provide a real-time optimization control method based on dynamic estimation of key parameters for complex industry process to improve the adaptive ability and make the industrial process run with safety , stability and high efficiency.

针对工况变化范围大、干扰因素多的复杂工业过程，综合动态建模、性能评估和智能优化技术，研究一种实时优化控制方法。提出基于主动学习的半监督聚类方法，通过挖掘过程信息的关联特性，实现过程数据的合理分类，研究基于集成学习和置信度的动态建模方法，实现对关键参数的实时准确估计。研究基于过程数据的多性能多层次的综合性能评估方法，通过信息熵的方法对多个性能进行综合，实现过程性能的全面准确评估。研究事件驱动的多目标鲁棒优化方法，基于评估结果将优化问题分解为多个优化事件，在保证系统性能的前提下，通过对不同优化事件下优化模型的求解获得具有较好鲁棒性的控制器参数。通过研究，建立一种包括关键参数动态估计、控制性能综合评估以及控制器参数实时优化的控制系统结构，提出一种基于关键参数动态估计的复杂工业过程实时优化控制方法，提高控制系统的自适应能力，保证复杂工业过程控制系统稳定、安全、高效运行。

本项目建立了一种包括关键参数动态估计、控制性能综合评估以及控制器参数实时优化的控制系统结构，提出一种基于关键参数动态估计的复杂工业过程实时优化控制方法，提高控制系统的自适应能力，保证复杂工业过程控制系统稳定、安全、高效运行。首先，综合利用已知的过程机理知识、实际操作经验和历史生产数据等，研究一种面向复杂工业过程数据的主动半监督聚类方法，在聚类过程中基于主动查询的方式获取过程数据中的关联特性，实现过程数据的合理分类；为了降低数据噪声对模型的影响，建立基于不敏感损失的半监督协同训练模型，实现对关键参数的实时准确估计。研究基于过程数据的多性能多层次的综合性能评估方法，采用直觉乘法层次分析法建立评估模型，实现过程性能的全面准确评估。基于性能评估的结果，研究事件驱动的多目标鲁棒优化方法，基于评估结果将优化问题分解为多个优化事件，在保证系统性能的前提下，通过对不同优化事件下优化模型的求解获得具有较好鲁棒性的控制器参数。以炼焦生产过程为典型生产过程，对所研究的方法采用实际生产数据进行了仿真，验证了方法的有效性。