项目名称: 融合数据驱动并跟踪模型切换的鲁棒数据协调及显著误差检测算法研究
项目编号: No.61304136
项目类型: 青年科学基金项目
立项/批准年度: 2014
项目学科: 自动化技术、计算机技术
项目作者: 周凌柯
作者单位: 南京理工大学
项目金额: 26万元
中文摘要: 基于生产过程机理模型和数据误差分布模型的数据协调技术是流程工业优化控制的关键。目前已有的研究成果多是基于固定不变的机理模型以及事先假定的误差分布模型进行数据协调,并且不能对非冗余变量进行有效的显著误差检测。本项目基于已知机理模型框架,跟踪模型切换信息,融合基于支持向量机的可变参数数据驱动模型,研究建立结合数据驱动并跟踪模型切换的混合数据协调约束模型,并进一步展开变量分类及模型降维研究。基于广义T分布模型以及仪表可靠度变化模型,研究建立兼顾数据协调估计性能的符合数据误差实际分布,以及对显著误差具有鲁棒抗差特性的数据协调目标函数。根据数据协调结果的残差统计分析以及基于证据决策理论融合其它先验信息,实现对冗余及非冗余变量的显著误差检测。该项目的成功实施将丰富数据协调领域的理论研究成果,同时也将对数据协调技术用于实际工业过程产生积极影响。
中文关键词: 数据协调;显著误差;鲁棒估计;多工况;广义T分布模型
英文摘要: Data reconciliation technology is the key to optimal control of the process industry by integrating process mechanism model and data error distribution model. Most current research results are based on fixed process mechanism model and presupposed data error distribution model. In addition, the gross error of non-redundant variable can't be detected effectively. This project intends to track model switching information, apply support vector machine method to build data-driven parameter estimation model, and develop hybrid data reconciliation model by integrating the known mechanism model with the data-driven parameter estimation model. Furthermore, the variable classification and model dimension reduction algorithm are studied also. Based on generalized T distribution model and instrument reliability model, this project develops data reconciliation objective function which considers both the estimated data accuracy and the gross error robust characteristics. By analyzing the data reconciliation residual results and using evidence theory to combine evidence from prior information, a gross error detection method is proposed to detect gross errors of redundant and non-redundant variable. The successful implementation of this project will have important significance and application prospect in the research field of
英文关键词: data reconciliation;gross error;robust estimator;multiple mode;generalized T distribution