一种数据驱动的控制器性能在线监控策略  

An on-line data-driven strategy for controller performance monitoring

作　　者：王建国[1] 王娟娟[1] 肖前平 

机构地区：[1]上海市电站自动化技术重点实验室,上海大学机电工程与自动化学院,上海200072

出　　处：《计算机与应用化学》2014年第5期521-525,共5页Computers and Applied Chemistry

基　　金：国家自然科学基金资助项目(61171145)

摘　　要：运用基于数据协方差基准的控制性能评估方法,并结合协方差矩阵的递推更新,本文提出了一种控制器性能的在线监控策略。首先给出一种协方差矩阵的递推更新算法,通过实时输出监控阶段数据相对于基准阶段数据的广义特征值及其构成的系统总体性能指标,在线监控系统性能的变化趋势。进一步使用统计推断方法得出每个时刻的特征值所对应的置信区间,得到优/劣子空间的性能指标实时变化趋势,从而判定控制器的性能变化。最后,将本方法应用在重油分馏塔预测控制的实时性能监控中,验证了本方法的有效性。With the combination of data-driven covariance benchmark based control performance monitoring method and recursive update ofcovariance matrix, an on-line controller performance monitoring strategy was proposed in this paper. Firstly, a recursive algorithm forcovariance matrix was presented, and the performance changing trend of the system was monitored instantly by outputting real-timegeneralized eigenvalues of monitored period data relative to benchmark period data and overall performance index which was composed bygeneralized eigenvalues. Then further analysis was made to achieve the confidence interval corresponding to each eigenvalue by usingstatistical inference method, and get the changing trend of better/worse subspace performance indices, based which we determine whether thecontroller performance varied. Finally, this approach was applied to the real-time performance monitoring for model predictive control of aheavy oil fractionators and the utility of the proposed approach was illustrated.

关 键 词：数据驱动 协方差矩阵 实时性能监控 模型失配 递推 

分 类 号：TQ015.9[化学工程]