Author(s)

Jianpeng Hu, Lichen Zhang

Corresponding Author

Jianpeng Hu

ABSTRACT

The power transformer is a key piece of equipment in power plants and substations. However, abnormal oil temperature in power transformers accelerates insulation aging, shortening their lifespan and leading to accidents. Therefore, predicting and monitoring oil temperature is crucial. To overcome the reliance on experience and rules in traditional prediction methods, an oil temperature forecasting method based on a multi-model combination under the Stacking framework was proposed. Taking into account the differences in data observation and training principles among various algorithms, fully leveraging the strengths of each model, we construct a Stacked ensemble learning oil temperature prediction model embedded with multiple machine learning algorithms. The base learners of the model include Light Gradient Boosting Machine (LightGBM) and Category Boosting (CatBoost).The experimental results indicate that the oil temperature prediction method based on the Stacking ensemble learning approach with multiple model fusion achieves a high level of prediction accuracy.

KEYWORDS

Oil temperature forecasting, Power transformer, Stacking, LightGBM, CatBoost