Research On Intelligent Modeling Method Of Transformer Condition Assessment Based On PCA Feature Reconstruction

As energy transmission and conversion equipment,transformers plays an important role in power systems.Real-time diagnosis and evaluation of transformer health status and timely detection of transformer potential faults are the key ensure the safe and stable operation of power system.Aiming at the shortcomings of transformer condition assessment,the transformer condition assessment method by combining dissolved gas in transformer oil with continuous hidden semi Markov model(CHSMM)and adaptive deep learning model are studied in this dissertation.The main contents of this dissertation are as follows:The generation mechanism of dissolved gas in transformer oil is analyzed.According to the corresponding relationship between gas components in transformer oil and internal faults,the evaluation index and diagnosis type of power transformer state are established,Principal component analysis(PCA)is used to reconstruct the characteristics of dissolved gas data in transformer oil,and min-max normalization is performed for subsequent diagnosis of transformer intelligent modeling of state assessment;Combined with the theoretical basis of hidden Markov model(HSMM),aiming at the shortage of initial parameter setting and training information loss of original HSMM,the algorithm is re-deduced and evolved into CHSMM,The method of transformer condition assessment based on CHSMM is deeply discussed.By adding the state parameters of transformer,the dynamic modeling of time series is carried out,and the multiple group state observation values of transformer are continuously processed;In order to improve the accuracy of transformer condition assessment,the dissolved gas in oil is adopted as the characteristic quantity of condition assessment.According to the theory of deep learning,a transformer condition assessment method based on adaptive deep learning model is proposed,This method adjusts the depth learning rate adaptively by the change characteristics of the iterative process,which improves the accuracy and training rate of the depth learning model.Finally,through the comparative analysis of the performance of adaptive deep learning model and CHSMM model,it is concluded that the accuracy of adaptive deep learning state assessment model is better than CHSMM,and the convergence speed of CHSMM model is better than that of adaptive deep learning model.The research results of this dissertation have certain theoretical significance and engineering application value.