Determining the Intensity and Occurrence Location of Faults in Transformers using Frequency Response Analysis (FRA) with Novel Multistage Optimization Algorithm and SVMD Decomposition Technique

Frequency response analysis (FRA) has become a worldwide accepted technique for detecting winding and core deformation in transformers. The main weakness of this technique is its reliance on the level of expertise and experience of personnel and the lack of standards and automatic codes. It is necessary to create reliable FRA interpretation codes for the high-frequency transformer model that can implement the frequency characteristics of real transformers in a wide frequency range. This paper presents an artificial intelligence method to estimate these parameters from the FRA diagram of the transformer. In the proposed method, a three-step optimization algorithm is implemented on the real data of a 33 kV disc winding to find the intensity and occurrence location of faults. At first, the frequency response amplitude signal is decomposed into oscillating modes using successive variational mode decomposition (SVMD), the output of which is much less complicated than the original signal. The frequency response of the modeled circuit decomposition is also obtained in the next stage and in the optimization process, whose decision variables are the RLC values of the detailed (lumped) model of the transformer. Based on the ability to hunt sharks in nature, the new meta-heuristic algorithm of shark smell optimization (SSO) will search for the optimal solution by minimizing the error between the actual and modeled winding frequency response. This process is implemented gradually, with the addition of each oscillatory mode in each stage. The accuracy of the proposed method is evaluated with the data of the tests performed on a 33 kV high voltage disc winding to estimate the parameters of their high frequency electrical equivalent circuit in normal and fault conditions. The results show that the proposed method can estimate the parameters of the equivalent circuit with high accuracy and help to interpret the FRA diagram based on the numerical changes of these parameters.