Prediction of gases content dissolved in power transformer oil based on gene expression programming

In order to predicting the operational status and the latent faults of a power transformer effectively. A new method to forecast the dissolved gases´ concentration in transformer oil based on GEP sliding window model is proposed. According to the change characteristics of the dissolved gases´ concentration in transformer oil, selects a appropriate embedding dimension, terminals, functions and other running parameters of GEP, then evolve each gas´s forecasting models which driven by the fitness function for genetic operation. With a running instance of a power transformer, prediction results for seven major gases and the prediction formula of H2 are given in this paper, then contrasts with the MGM (1,7) model. The comparative results show that GEP model can improve the prediction accuracy effectively.