Design And Implementation Of Real-time Knowledge Discovery For Process Objects

In the process industry,the DCS distributed control system is generally used in the production process.It can monitor the production status of the process industry in real time,form status data.The production process of the process industry emphasizes real-time and highefficiency,and through online data mining of industrial data streams collected in real time,it can mine valuable knowledge of industrial actual production more effectively,do Response quickly and help to decision-making better in production.In this thesis,a complete coal-fired boiler production process is taken as the research object.Aiming at the characteristics of increasing industrial data stream,real-time object updates,and value decay over time,a calculation model for online real-time knowledge discovery of process objects is proposed.The model mainly includes data preprocessing,time series,data stream clustering based on sliding window,concept migration judgment,association rule chain mining,and model prediction update.And use Spark framework to carry on the parallel design to each part of the model,thus improving the efficiency of knowledge discovery.Finally,an online realtime knowledge discovery system for process objects is actually installed and applied in a thermoelectric company.In boiler operation,the industrial data stream is generated by the integration of various links.Data collected process have noise inevitably.First,we must clean the data.Then we need to dig out mine the sequence order of all links and to get the correct order.The algorithm based on differential extreme value used in this thesis is for time series discovery,so as to obtain the time series,and pave the way for the subsequent selection of association rules.For data streams,the issue of data value decay needs to be considered.The workers are more concerned about the changes of the boiler status in the recent period of time in the production,so this thesis uses a sliding window-based data stream clustering algorithm to retain the data changes in the near future and reduce the degree of data dispersion by clustering each link individually.And compare whether the clustering results before and after the arrival of data stream have changed significantly.If there is a significant change,you can determine that the concept migration has occurred,and the production status may be huge at this time.The change requires discover new knowledge,so continue to perform subsequent knowledge discovery algorithms.If the clustering result does not change significantly,continue to wait for the next arrival of data stream.In the stage of association rule chain mining,this thesis uses the FP-Growth based interdimensional association rule algorithm.The relationship between the two links(relationship)is calculated by calculating the frequency ratio of the rules,and then the association degree table between the links is formed;finally,by the time series and association degree tables,the association rule chain is generated.The association rule chain is a kind of chain-type association rule.It can show the influence relationship between the states of multiple links intuitively.The recent boiler production status data in the sliding window and the associated links of each link make sufficient data preparation for the model prediction.This thesis uses a flexible neural tree algorithm and obtain the change trend expressions of key parameters.Finally,the incremental update of the model was implemented,making the model more suitable for changing boiler states.Finally,the model is applied to the online real-time knowledge discovery system of process objects.It realizes the numerical prediction of the link parameters in the production process,realizes the real-time status monitoring of the key parameters,the visualized trend prediction comparison image,and fault detection more effective.Workers can provide auxiliary decision-making to adjust boiler parameters.