Process Mining Based Industrial Process Operating Knowledge Discovery Method And Application

In modern industrial processes,data acquisition(SCADA)systems store a large amount of process operating time series data which record information about the production processes operating and operational behavior.Researches on the establishment of effective process operating knowledge discovery methods can effectively monitor the states of production process operating,particularly providing a basis for detecting and diagnosing early abnormal operating conditions.Towards discovering knowledge contained in data,traditional data mining methods focus on handling numerical data.However,limited information provided by numerical data is often insufficient for deep knowledge discovery for multidimensional time series.Process mining is a class of deep knowledge discovery methods for events with comprehensive information.Currently,process mining methods are rarely used for knowledge discovery of industrial process operating events.Motivated by these observations,this thesis proposes a novel process mining-based knowledge discovery method for industrial process operating events and applies it to the abnormal state detection of practical industrial processes.The main research work and achievements of the thesis are presented as follows.1.For industrial process operating data containing temporal information,the time series feature expression method is explicitly studied,in which,fusion information of data trends and values is employed to describe industrial process operating events.By means of identifying trend features and mean values of industrial process operating data,effective descriptions of industrial process operating events are formulated.The representation of process operating data is enriched by fused information of evolving trends and numerical values,providing support for mining deep process operation knowledge.2.A novel process discovery algorithm is proposed for mining the process knowledge in industrial process operating events.Specific processing methods are used to generate industrial process operating event logs before an inductive discovery algorithm is applied to extract behavioral patterns and sequence rules in the events.Eventually,a visual Petri net process model is constructed by means of the use resultant process operating information.3.The proposed method is applied to an actual coal gasification industrial process.Using discovered knowledge of the established Petri net process operating model,inference rules for industrial process abnormal state detection are created by the sequential relationship among the transition events,which are suggested to assist human operators in early screening of process operating abnormal states from the process operating perspective.The application study has achieved satisfactory results.