Prediction research of transformer fault based on regular extreme learning machine

To predict the concentration of dissolved gas in transformer oil, and then realize the transformer latent fault prediction, can effectively prevent unnecessary loss caused by the transformer faults. In order to improve the transformer fault prediction ability, this paper proposes a new transformer fault prediction model-Regular Extreme Learning Machine (RELM) prediction model. RELM algorithm introduce structure risk minimization principle on the basis of traditional ELM, using the balance factor to weigh the empirical risk and the risk of structure size, further enhance the generalization performance of ELM. Verified by examples, the proposed prediction model based on the RELM in this paper achieve better generalization performance and prediction accuracy in the forecast of gases concentration dissolved in transformer oil.