ISSA-SVM Transformer Fault Diagnosis Method Based On RF Feature Selection

Transformers play a vital role in the transmission,transformation and distribution tasks of the power system and are the key equipment of the power system.Once a transformer fails,it can bring about a short circuit or other faults in electrical equipment,ranging from the stalling of machines to the extinguishing of lighting equipment,and even leading to fire and serious personal injury,resulting in huge economic losses.Therefore,a more scientific method for condition monitoring and fault diagnosis of transformers is needed to achieve more accurate fault prediction and maintenance and to improve the reliability and stability of transformers.In this thesis,a transformer fault diagnosis method based on Random Forest(RF)feature selection with Improved Sparrow Algorithm(ISSA)optimized Support Vector Machine(SVM)is proposed.First,to address the problem that the existence of irrelevant,redundant or noisy features in the original data affects the accuracy of transformer fault diagnosis,an SVM transformer fault diagnosis model based on RF feature selection is established,and the feature selection results are obtained and analyzed for comparison.The diagnostic accuracy of the SVM transformer fault diagnosis model with feature selection results as input is not only higher than that of the SVM model with data without feature selection as input,but also higher than that of the SVM model with other conventional methods of inputting features as input.At the same time,the superiority of SVM fault diagnosis model compared with other fault diagnosis models is verified by using the feature selection results as the input of the model.Second,to address the problems of slow convergence and easy to fall into local optimum of the sparrow search algorithm,an improved sparrow search algorithm is proposed: first,the Tent chaotic mapping initialization strategy is used to increase the diversity of the algorithm and improve the search efficiency,and second,the Levy flight strategy is introduced into the sparrow optimal position of the vigilantes’ position update formula in SSA to reduce the risk of the sparrow falling into local optimum.Then the effectiveness of the Improved Sparrow Search Algorithm(ISSA)is demonstrated based on 12 test functions and Wilcoxon rank sum test,compared and analyzed with Particle Swarm Algorithm(PSO),Gray Wolf Algorithm(GWO),Whale Algorithm(WOA)and Sparrow Search Algorithm(SSA),and the effectiveness of each improvement is shown by ablation experiments.The results show that the ISSA algorithm has some improvement in convergence speed and number of iterations as well as the ability to jump out of local optimum.Finally,the improved sparrow search algorithm is selected to find the optimization of the relevant parameters of the SVM model and establish the transformer fault diagnosis model based on RF feature selection and ISSA-SVM.The results of RF feature selection are used as input for transformer fault diagnosis and compared with SVM,PSO-SVM and SSA-SVM transformer fault diagnosis models to verify the superiority of the transformer fault diagnosis model based on RF feature selection and ISSA-SVM in this paper.