Prediction of Dissolved Gas Content in Power Transformer Oil based on EMMD-CS-SVR Model

Early detection of potential transformer failures is critical to ensure the stability of an electrical power system. Dissolved gas analysis (DGA) is a widely used technology for diagnosing insulation faults in transformers. However, predicting dissolved gas content is challenging due to various factors such as extreme weather, maloperation, and cascading outages. This study uses a decomposition-prediction-reconstruction model based on support vector regression (SVR) to predict dissolved gas content. A 1-dimensional original dissolved gas content is decomposed into a 9-dimensional time sequence by ensemble empirical mode decomposition (EEMD), and the multidimensional gas content time sequences are directly predicted by SVR. To further improve the prediction accuracy, the cuckoo search algorithm (CS) is employed to optimize the parameters of SVR. The gas content predictions are obtained by summing multidimensional prediction values. Experimental findings show that the EEMD-CS-SVR model outperforms the SVR and EEMD-SVR models in predicting the dissolved gas content aspect when applied with identical experimental conditions.