Prediction of concentration for dissolved gas in oil based on CEEMDAN and TCN

Abstract

Accurately predicting the concentration trend of dissolved gas in oil has a positive effect on the evaluation of transformer status and life assessment. In order to improve the accuracy of dissolved gas in oil prediction, a dissolved gas in oil prediction method based on complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) and time convolution network (TCN) is proposed in this paper. Firstly, the CEEMDAN method is used to decompose the original sequence of dissolved gas in oil into multiple intrinsic mode functions(IMFs), separating the stable and unstable IMFs. Secondly, TCN are established for IMFs, then predictions based on these trained TCNs are made. Finally, the prediction results of IMFs are overlaid to reconstruct the prediction results of the original sequence. Analysis in this paper shows that the root mean square error, mean absolute error and maximum error of the prediction method are 1.01 μL/L, 1.53 μL/L, 5.54 μL/L respectively, which are reduced by 53.47%, 41.18%, 13.36% compared to the case without using the CEEMDAN. When using CEEMDAN, the three errors are the smallest compared to commonly used recurrent neural networks. The proposed dissolved gas in oil prediction method has higher prediction accuracy and can provide more effective support for condition based maintenance strategy.

Keywords

• dissolved gas in oil
• transformer
• complete ensemble empirical mode decomposition with adaptive noise(ceemdan)
• time convolution network (tcn)
• time series forecasting
• condition based maintenance