Research And Application Of Alarm Information Recognition Based On Text Representation

The power information system is the data center for the normal production activities of the power grid,and it undertakes the data transmission,message distribution and personnel scheduling of the system.In order to ensure the normal operation of the power information system,it is necessary to deal with the fault alarms in the system in time.At present,the identification of alarm information relies on manual analysis and judgment of alarm information,and then a specific disposal plan is obtained.Facing the increasing business demands of the power information system,more business subsystems have been incorporated into the information system,resulting in the collection of a large amount of alarm data by the system monitoring platform.Complex system network relationships are also more likely to induce alarm flooding.The previous method of relying on manual identification of alarm information has become a bottleneck in the development of power information systems.Aiming at identifying the warning information,this thesis has carried out a series of related research work.Firstly,based on the large-scale pre-trained Roberta text representation model,the method of fusing the alarm domain knowledge is studied,and then the alarm text representation result of this method is used to construct the alarm classification model.And studied the multiple factors that affect the judgment of the alarm level,and put forward the classification method of the alarm information through the weight distribution of each influencing factor.The main research content of this thesis is as follows:(1)Research on the text representation method of alarm information.In this thesis,the Roberta text representation model is used to realize the mapping of alarm text information to vector space.At the same time,in order to strengthen the text representation ability of the Roberta model in the warning field,the construction of the warning field dictionary based on the solidification degree and the degree of freedom is studied,and it is used in the pre-training task of the Roberta model,and the FD-Roberta text suitable for the current field is obtained.representation model.It is verified by experiments that the FD-Roberta model has better performance,which provides better data support for the downstream research on alarm information identification methods.(2)Research on the identification method of alarm information.The FD-Roberta model and TextRCNN are integrated to build a classification model of alarm information,which solves the problem of directly using the FD-Roberta model to classify alarm information.At the same time,considering multiple factor indicators that affect the judgment of the alarm level,a hierarchical alarm model is constructed.Using the constructed alarm classification and grading model,the identification of real-time alarm data can be realized,and the human interference in the previous alarm handling process can be reduced.(3)Construction of the alarm identification subsystem.Based on the above two research results and software engineering specifications,the text designs and develops the alarm identification subsystem.The system is based on the B/S architecture,the front end adopts the Vue framework,the back end adopts the SpringBoot framework,and uses the RabbitMQ message middleware to implement the algorithm model call.The system mainly includes an alarm identification module,a data visualization module and a model update module.