Research on Intelligent Diagnosis of Dissolved Gas in Transformer Oil Using XGBoost Algorithm

As core equipment in power transmission and distribution, the stability of transformers is crucial for the safety of the power system. Analysis of dissolved gases in oil is an important method for assessing the condition of transformers, and the XGBoost algorithm, with its efficiency, accuracy, and flexibility, has demonstrated excellence in handling large-scale data. This study identified key characteristic gases such as hydrogen ($\mathrm{H}_{2}$), methane ($\text{CH}_{4}$), ethane ($\mathrm{C}_{2} \mathrm{H}_{6}$), ethylene ($\mathrm{C}_{2} \mathrm{H}_{4}$), acetylene ($\mathrm{C}_{2} \mathrm{H}_{2}$), carbon monoxide (CO), and carbon dioxide ($\text{CO}_{2}$) through feature selection. Variations in the concentrations of these gases can indicate different types of faults within the transformer. The intelligent diagnosis model constructed by XGBoost algorithm can effectively capture the complex patterns in the data and improve the prediction accuracy by iteratively optimizing multiple decision trees. The experimental results show that the model shows high diagnostic accuracy under different parameter combinations, and the accuracy can reach 93.1 % under the optimal parameter combination. During the model training, with the increase of training rounds, the prediction error gradually decreases and tends to be stable, showing a good performance level. Confusion matrix analysis further proves the high accuracy of the model in various fault identification, especially in the identification of overheating fault and arc discharge fault. Compared with traditional methods, XGBoost intelligent diagnosis model has high automation, real-time performance and high accuracy, which provides scientific basis for transformer maintenance strategy and promotes the development of power system to intelligence and efficiency.