Research On Power Transformer Fault Diagnosis Method Based On Hybrid Feature Selection And PSO-KELM

The power transformer is the hub of the power system.It bears the tasks of voltage conversion,power distribution and transmission.Therefore,its reliability directly affects the safety and stability of the power system.Any internal fault of the transformer may cause a certain range of power outages and produce immeasurable economic losses.Therefore,it is important to find and accurately judge the latent fault of the transformer in time.Dissolved gas analysis(DGA)in transformer oil is one of the most extensive and effective methods for online fault diagnosis of transformers.However,traditional DGA analysis methods often have shortcomings such as lack of coding,excessive coding,and low diagnostic performance.In addition,due to the lack of widely recognized fault feature sets,DGA analysis based on intelligent algorithms often ignores the completeness and compactness of fault features,resulting in unsatisfactory diagnostic performance.In view of the above shortcomings,this paper proposes a transformer fault diagnosis method based on hybrid feature selection and Particle Swarm Optimization(PSO)algorithm for kernel based extreme learning machine(KELM),which provides the optimal feature subset while providing a more compact and efficient transformer fault diagnosis model to improve the accuracy of the final transformer fault diagnosis.This thesis mainly carried out the following aspects of research:First,for the current DGA analysis method,there is no widely recognized feature set.By searching a large number of cases and literature research,and combining the "Guidelines for Dissolved Gas Analysis in Transformer Oil",a comprehensive fault feature subset is established to provide feature optimization space for subsequent methods.Secondly,the extreme learning machine(ELM)is applied to transformer fault diagnosis,and the ELM is optimized by using the kernel function to build a transformer fault diagnosis model based on the KELM.In addition,the fault characteristics of transformers are sorted and filtered by combining F-score and Information Gain to achieve dimensionality reduction of fault features,and the importance of feature selection is analyzed.The experimental results show that the subsets after feature selection can effectively improve the fault diagnosis accuracy of the model,and KELM has better diagnostic performance than other commonly used fault diagnosis algorithms and ELM.Finally,Aiming at the shortcomings of Filter method applied to transformer fault feature selection,and in order to reduce the difficulty of tuning,a transformer fault diagnosis model based on hybrid feature selection and KELM was established.Firstly,the filter model is used to sort the fault features and the preliminary screening of redundant features.Then,The PSO algorithm is combined with the KELM model to further feature selection and parameter optimization.Finally,the influence of the initialization strategy on the PSO optimization results is analyzed.The experimental results show that the method achieves more accurate fault feature selection and provides a more compact and effective fault diagnosis model,which can effectively improve the fault diagnosis accuracy.