Research On Inter-turn Short Circuit Fault Detection Method Of Dry-type Air-core Reactor

Dry-type air-core reactors play an important role in reactive power compensation and harmonic elimination in the power system.However,various faults often occur in the equipment during operation,and even burn and catch fire in severe cases.The inter-turn short circuit fault is the cause of equipment ignition.main cause of destruction.There are various problems in the traditional detection methods.Among them,the electrical quantity detection method pays less attention to the types of fault characteristics,and when combined with machine learning,it often ignores the limitations brought about by the amount of fault sample data.In response to these problems,this thesis studies the characteristics of various electrical faults,and proposes to distinguish the operating stages of equipment according to the amount of fault sample data,and select algorithms in stages to realize the diagnosis of fault locations and the analysis of the importance of features.Firstly,starting from the dry-type air-core reactor structure and fault causes,the equivalent circuit model of the equipment in normal operation and inter-turn short circuit fault is analyzed through physical model derivation.By deriving the key parameters of the equivalent circuit and analyzing the fault process,seven kinds of fault characteristics of electrical quantities that are highly related to faults are given.In view of the fact that the dry-type air-core reactor is put into operation and the amount of fault sample data is small,the PSO-SVM algorithm is used to realize fault diagnosis and feature importance discrimination.Experiments show that the particle swarm optimization algorithm can stably optimize the parameters of the support vector machine;the PSO-SVM fault diagnosis accuracy rate is 0.85;the diagnosis algorithm presents a diagnostic characteristic that is sensitive to equipment edge faults;and can preliminarily judge the importance of features.Then the XGBoost algorithm is selected for the situation that the dry-type air-core reactor is in the middle of operation and the data volume of fault samples has been expanded.Experiments show that the XGBoost algorithm has certain requirements on the number of samples,and its performance is not as good as that of the PSO-SVM algorithm in the case of small samples;however,the accuracy rate can reach 0.92 in the case of expanded samples;it can achieve more accurate feature importance discrimination,which is comparable to PSO-SVM The discrimination results of the algorithms can be mutually verified,and the equivalent resistance is of high importance in the judgment results of each algorithm;it also presents a diagnostic characteristic that is sensitive to the edge of the device,which is explained by analyzing the equivalent resistance from the perspective of data.In the later stage of equipment operation,high-importance features and a large number of fault sample data have been obtained.At this time,because XGBoost has the defect of relying on parameter adjustment,it chooses to quickly build the MLP algorithm,which achieves an accuracy rate of 0.97.Finally,according to the performance of each algorithm in different periods,the optimal fault diagnosis strategy is given.Compared with the traditional electric quantity detection method,the dry-type air-core reactor inter-turn short-circuit fault diagnosis method proposed in this thesis can detect the fault location under different sample numbers in each operation stage of the equipment,and analyze the importance of fault characteristics,which has research value.