Data Analysis And Application Research On Text Defects Of High Voltage Relay Protection In A Regional Large Power Grid

Recently,the scale of the power grid has become increasingly large,and the number of relay protection devices has also experienced a leap-forward growth.The completion of the defect management of the relay protection device directly affects whether the relay protection system can maintain the safe and stable operation of the power system.How to solve the contradiction of "more equipment and less people" under the current personnel allocation,to meet the timely and efficient operation and maintenance work,so as to ensure the security of power grid,has become a difficult problem for major power enterprises.Thanks to the construction of intelligent relay protection information platform,the data generated in the daily operation of relay protection devices can be quickly stored.Among these data,the text defect data can directly reflect the status and reliability level of relay protection devices.Therefore,mining and analyzing such data will be of great significance to improve the operation and management level of relay protection.However,traditional statistical methods are difficult to meet the requirements of feature extraction and effective information mining of text defect data,and it is urgent to find new data mining methods.With the integration of artificial intelligence,natural language,other emerging technologies,the physical laws and professional knowledge of power system,"electric artificial intelligence" arises at the historic moment,which creates favorable conditions for the in-depth mining and application of defective text data.This paper analyzes and studies the text data of high-voltage relay protection defects in a certain area.The main contents of this paper include the following four points:(1)The statistical characteristics of relay protection standard defect data are analyzed.Combined with the overall normative defect data of a regional power grid,statistical analysis is made from the three perspectives of defect distribution,defect cause and defect location,and the concentrated distribution of defects in relay protection device,the main factors of defects occurrence and the location of frequent defects are obtained.Furthermore,the data of normative defects of different manufacturers are analyzed,and the difference of defect rate of different manufacturers is compared.(2)Based on Apriori algorithm,association rules among relay protection canonical data are mined.Combined with the whole defect data of a regional power network,the common defect characteristics of relay protection devices are analyzed.Combined with the defect data of main manufacturers,the family defect characteristics of relay protection devices are analyzed.Furthermore,FP-Growth algorithm is applied to improve the efficiency of rule mining.Finally,according to the defect characteristics,the index suitable for the reliability evaluation of relay protection device is put forward.(3)Based on the non-standard relay protection data(defect log)of a regional power network,a short text preprocessing model is constructed and a dictionary is generated.Specifically,the regular expression is used to establish the stop word list,and the stuttering word is used to divide the defect log,and the domain dictionary of relay protection device is obtained.At the same time,this article analyzes the similarities and differences between the language characteristics of relay and general text corpus(4)Relay protection defect grading models based on KNN algorithm and Naive Bayes algorithm are built respectively.Combined with the idea of textual vectorization,a word bag model is constructed for the dictionary of relay protection device domain.Through a practical example,the applicability of the two algorithms in the classification of relay protection defects is compared and analyzed.