A Method for Predicting Transformer Oil-Dissolved Gas Concentration Based on ICEEMDAN and ISO-KELM

To address the challenges of considerable volatility in time series data and inadequate precision in forecasting the concentration of dissolved gases within transformer oil, a novel prediction technique is introduced. This technique integrates the advanced Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (ICEEMDAN), the enhanced Snake Optimizer (ISO), and the Kernel Extreme Learning Machine (KELM). Initially, the ICEEMDAN technique is applied to decompose the raw gas concentration data, removing any non-stationarity and mode mixing present. Subsequently, the ISO, which is a fusion of Bernoulli chaotic mapping, inverse learning strategy, and subtractive clustering, is developed to amplify optimization efficiency. Ultimately, the ISO is utilized to fine-tune the key parameters of the KELM algorithm, ensuring precise prediction of each individual gas concentration component in the oil. These predicted components are then combined to yield the ultimate forecast for the dissolved gas concentration. Experimental outcomes reveal that the proposed method outperforms the conventional BP and KELM models by reducing the Mean Absolute Error (MAE) by 9.22% and 6.95%, Root Mean Square Error (RMSE) by 12.15% and 9.78%, and improving the R-squared (R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>) by 18.24% and 15.96%, signifying its higher accuracy in predicting transformer gas concentrations.