Data Mining Based Approaches To Industrial Process Control Performance Evaluations With Applications

Control performance evaluations are very important for ensuring efficient,stable and safe operations of industrial processes.Most of traditional control performance evaluations adopt model-based minimum variance methods which reveal certain difficulties for practical applications.Consider that human operators can generally evaluate and adjust the control performance taking advantage of operational experiences,and the related knowledge is stored in industrial process monitoring data.Motivated by these observations,this thesis extensively investigates industrial process control performance evaluation methods based on data mining,creating a gradient enhanced decision tree+logistic regression(GBDT+LR)algorithm based conventional control performance evaluation method,and an association rules based control loop correlation performance evaluation method.The main research contents and achievements are presented as follows.1.Conventional control performance indices based on data mining are established to evaluate control loops in terms of deterministic,random performances and effective utilization rates.Control loop operating data are delivered to the GBDT+LR algorithmic learning,so as to establish the control performance classifier.Consequently,the classifier is employed to evaluate the conventional control performance.2.In response to the correlation analysis among control loops,an association rules based performance evaluation method is explicitly proposed.The historical label data are used to mine the association rules among control loops before the correlation performance is evaluated according to the association rules.3.The proposed methods are applied to an industrial DMF recovery process.In addition,an application software system of control performance evaluations is developed.Satisfactory results are achieved,demonstrating the effectiveness of the contribution.