Transformer Top Oil Temperature Prediction Using Deep Learning Time Series Model

作者：Xuemin Huang, Xiaoliang Zhuang, Fangyuan Tian, Zheng Niu, Yujie Chen, Zhou Qian, Chao Yuan · 年份：2024 · DOI：10.1109/isneet64164.2024.10956109 · 被引用次数：2 · 研究领域：Advanced Data Processing Techniques、Fault Detection and Control Systems、Energy Load and Power Forecasting

Transformers playa vital role in voltage regulation, and their oil temperature serves as an effective indicator of operational conditions, enabling early fault detection through accurate temperature prediction. However, transformer top oil temperature data often exhibit a combination of linear and nonlinear characteristics, challenging traditional single-model approaches. This paper presents a hybrid ARIMA-LSTM model for high-accuracy prediction of transformer top oil temperature. The ARIMA model is first employed to capture the linear trends in the data, while the residuals—the differences between predicted and actual values—are processed by an LSTM network to learn the nonlinear features. The final prediction integrates both the linear trend and the nonlinear residuals. The ARIMA-LSTM model is validated using real transformer operation data and compared with a standalone LSTM model. Results demonstrate that the ARIMA-LSTM model achieves higher prediction accuracy and lower error rates, providing an effective solution for forecasting complex time series data.