Efficient Gas Concentration Prediction for Power Transformers Using Data-Fused Co-Trained AE-MLP

The early detection of gas concentration in power transformers plays a vital role in preventive maintenance and ensures the reliability of power supply systems. In this paper, we propose a novel approach for predicting gas concentration in power transformers using a data fusion technique and deep learning method called co-trained Autoencoder-Multilayer Perceptron (CoAMLP). The proposed model integrates temperature, voltage, current, and power data to accurately forecast the concentration of gases, which enables timely maintenance and improves the reliability of power transformers. To improve the accuracy and generalization, the model is constructed with autoencoder and three distinct MLPs with varied seeds. The model was trained and evaluated using real-world data, demonstrating its efficiency and effectiveness in gas concentration prediction.