基于GBDT+LR分类器的工业过程区域控制性能评价

Industrial Process Area Control Performance Assessment Based on GBDT+LR Classifier

由于工业过程控制回路之间往往存在着比较强的相关性,面向工业过程区域的控制性能评价更加具有实际意义。提出了一种将梯度增强决策树与逻辑回归算法(GBDT+LR)相结合的分类器,将其用于工业过程区域控制性能评价。该方法分为线下和线上阶段,在线下阶段,首先对工业过程区域的相关历史数据进行处理,综合偏差平方指标和协方差指标对过程区域控制性能的影响,由操作工程师给出“优”、“良”、“差”的区域控制性能评价标签,并训练GBDT+LR分类器;在线上阶段,实时计算区域控制性能指标送入分类器,给出控制性能评价结果。以精馏塔过程实验对所提方法进行了仿真验证,表明了其有效性。

Due to strong correlations among industrial process control loops,the evaluation of industrial process area control performance is of more practical significance.A classifier based gradient-enhanced decision tree and logistic regression(GBDT+LR)is established to evaluate the industrial process area control performance.The approach consists of the off-line and the on-line phases.In the off-line phase,firstly,the relevant historical data of the industrial process area is processed,and the influence of the deviation square index and the covariance index on the performance of the process area control are synthesized.Then,the area control performance evaluation labels of“excellent”,“good”and“poor”are given by operating engineers and the GBDT+LR classifier is trained.In the on-line stage,real-time calculations of the area control performance index is delivered into the classifier,and evaluation result of the control performance is given.The effectiveness of the proposed method is verified with simulation by distillation column process experiments.