Prediction of Oil Temperature for Transformers Using Gated Recurrent Unit

Yuwen Liu, Yihong Yang, Yuqing Wang

Abstract: The service life of transformer is determined by the insulating materials’ aging degree. The factors causing insulation aging are temperature, oxidation and moisture in insulating materials. Among them, temperature is the decisive factor. Therefore, managers need to predict the oil temperature changes in time, which has great significance on maintaining the transformers’ lives and ensuring the normal operation of the power system. However, prediction methods are always based on linear regression or artificial neural network. These methods hardly consider the interaction between historical oil temperatures. However, the historical oil temperature is precisely an important factor affecting the future changes of oil temperature. Therefore, we propose an oil temperature prediction model (GRU-OTP) that can simultaneously pay attention to the long-term and short-term effects between historical oil temperatures. In order to pay attention to the short-term effects of historical oil temperature, we add a 5-h time sliding window to preprocess the data. Then, We use the GRU model to explore the long-term effects between historical oil temperatures. Experiments on different transformers show that GRU-OTP model has higher prediction accuracy and applicability.

External IDs:doi:10.1007/978-981-16-7476-1_36