Research On Early Warning Of Capacitive Equipment Based On Data Fusion

With the development of economy,the demand of electricity is increasing,the stable operation of power system is very important.In substations,capacitance equipment accounts for 40%-50%,far more than other types of equipment.Therefore,it is of great significance for the stable operation of the entire power system to monitor the operation status of capacitor equipment,distinguish normal equipment from defective equipment,and give early warning.During the operation of the power system,a large number of operation data and maintenance records are generated,which are accumulated as historical data.These data cover all kinds of information of capacitive devices and have many characteristics of attributes.How to effectively integrate these data information to serve the operation and management of power system is always an important issue.Based on the data fusion theory and method,this thesis studies and analyzes the data of China southern power grid corporation’s capacitor equipment ledger and maintenance record data,and puts forward the corresponding algorithm.This thesis firstly preprocesses the data set.There are two problems in the data of capacitor equipment ledger and maintenance record of China southern power grid corporation: the data type is not uniform,and the data set is not balanced in positive and negative sample size.To solve the first problem,this thesis adopts two encoding methods of numerical coding and onehot encoding for type conversion,and discards some attributes that are not convenient for coding conversion.Then the model is tested for the two coding methods to determine their advantages and disadvantages.To solve the second problem,this thesis adopts up-sampling and down-sampling respectively to increase rare samples and reduce large samples.Then,using the data fusion method,this thesis constructs the equipment defect prediction model respectively by the neural network,the support vector machine,the random forest method.In order to determine the parameters of the model,this thesis USES the network search method to adjust,so that the prediction accuracy of the model after parameter adjustment can reach or approach the highest.At the same time,the cross-validation method is adopted to ensure that the model does not overlearn the data set.Then,the accuracy,precision,recall rate,F1_score and other indicators were used to evaluate the accuracy of the model,and the experimental results were compared and analyzed.