Prediction of Dissolved Gas Concentration in Transformer Oil Based on VMD and ISABO-KELM

In order to solve the problems of significant time series volatility and insufficient prediction accuracy in the prediction of dissolved gas concentration in transformer oil, a new prediction method is proposed, which incorporates the Variational Mode Decomposition(VMD), the improved Subtraction-Average-Based Optimizer(ISABO), and the Kernel Based Extreme Learning Machine(KELM). Firstly, the original gas concentration sequence is decomposed using the VMD method to eliminate the non-stationarity and modal aliasing in the data; then, ISABO incorporating the Bernoulli chaotic mapping, Piecewise chaotic mapping, and the golden sine method is designed to enhance the optimization efficiency and the effect is verified by six test functions; and finally, the The core parameters of the KELM algorithm are optimized using ISABO to achieve accurate prediction of each component of the dissolved gas concentration in oil, and the components of the prediction results are superimposed to arrive at the final predicted value of dissolved gas concentration. The experimental results show that compared with the existing SVM and KELM models, the MAE value of this method on the test set is reduced by 10.32% and 7.95%, the RMSE value is reduced by 12.01 % and 8.98%, and the R2 value is improved by 20.04 % and 17.36%, which indicates that this method has higher accuracy in the prediction of transformer gas concentration.