Sunday 16 March 2025
A team of researchers has developed a novel approach to predicting the top- oil temperature in power transformers, which could lead to significant improvements in their maintenance and lifespan.
Power transformers are critical components of electrical grids, responsible for transmitting and distributing electricity across vast distances. However, they can be prone to failures due to overheating, which can have devastating consequences for both the environment and human safety. To mitigate this risk, it’s essential to monitor the top-oil temperature within these devices, as high temperatures can lead to degradation of the insulation system.
Traditionally, power transformers are monitored using standard regression models, such as those outlined in the IEC 60076-7 standard. These models, however, have limitations and often rely on simplifying assumptions that don’t accurately reflect real-world conditions. As a result, they may not provide reliable predictions of top-oil temperature, which can lead to incorrect maintenance decisions.
In an effort to overcome these challenges, the researchers developed a machine learning-based approach that combines advanced time-series forecasting techniques with data-driven models. The team used a range of architectures, including artificial neural networks (ANNs), Time- series Dense Encoder (TiDE), and Temporal Convolutional Networks (TCN), to develop predictive models that can accurately estimate top-oil temperature.
The researchers’ approach is based on the idea that historical measurements of top-oil temperature, ambient temperature, and load current can be used to train machine learning models that can generalize well to new, unseen data. The team’s experiments demonstrated that their proposed models outperformed traditional regression-based approaches, achieving lower mean absolute errors (MAEs) and mean squared errors (MSEs).
One of the key findings of this study is that the best-performing model was an ANN with 8 layers, 128 neurons per layer, and a look-back window of 4 hours. This configuration achieved an MAE of just 1.49°C for top-oil temperature predictions.
The researchers also explored the use of quantile regression to estimate the uncertainty of temperature predictions. By constructing prediction intervals using conditional quantiles, they were able to provide a range of possible values for the expected top-oil temperature at a given point in time. This approach has significant implications for power transformer maintenance, as it allows operators to better understand the potential risks and uncertainties associated with different operating conditions.
Overall, this study demonstrates the potential benefits of machine learning-based approaches for predicting top-oil temperature in power transformers.
Cite this article: “Machine Learning-Based Approach Improves Prediction of Top-Oil Temperature in Power Transformers”, The Science Archive, 2025.
Power Transformers, Machine Learning, Top-Oil Temperature, Time-Series Forecasting, Data-Driven Models, Artificial Neural Networks, Tide, Tcn, Regression-Based Approaches, Uncertainty Estimation
