Process Variable Sequential Event Recognition Methods For Industrial Process Monitoring-Control

Due to the influence of smooth operations and stochastic perturbations,industrial process variable time series curves often reveal specific frequent spatio-temporal evolution patterns.Most current machine learning and feature mining methods for industrial process monitoring-control are usually carried out directly on process variable temporal data,leading to that a large amount of redundant temporal data largely obscures the changes of temporal curves.In response,this thesis specifies the frequent temporal evolution patterns of time series curves as sequential events,thereby proposing a method for identifying sequential events of industrial process variables.The main contents and results of the research are presented as follows.1.Primitive sequential events of industrial process variables are collected and transformed into two-dimensional images by the gramian angular field(GAF).A gramian angular field-convolutional neural network classifier(GAF-CNN)is built to classify the primitive temporal events of industrial process variables.2.For industrial process variable time series curves,a GAF-CNN classifier-based sequential event segmentation algorithm is constructed.A deep denoising auto-encoder(DDAE)is established to extract features from the process variable sequential timing events.Consequently,the industrial process variable timing event recognition is achieved.3.The proposed method is applied to TE simulation platform and an industrial coal gasification process.Experimentally,the process variable primitive timing events are collected,GAF-CNN classifications are performed to segment process variable sequential events,and the DDAE is established to extract features from the sequential events.Satisfactory results have been achieved,verifying the effectiveness of the contribution.