Relevance. Reliable and environmentally safe operation of power transformers is an essential requirement for the functioning of modern power systems. Transformer oil degradation and abnormal operating conditions of electrical equipment are key factors leading to emergency situations. The composition of transformer oil serves as an indicator of the technical condition of a transformer and enables assessment of the lifespan of its insulating materials and internal components. Timely replacement of oil contributes to extending the operational lifetime of power transformers, reducing the risk of sudden failures, and enhancing overall reliability of the power system. Forecasting the technical condition of a power transformer, integrating various parameters – such as dissolved gas concentrations and electrical characteristics of the oil – is accepted as a crucial indicator for identifying early signs of wear and potential malfunctions, and allows prediction of the transformer operational lifespan. One of the approaches to addressing challenges in determining the technical condition of power transformers involves the application of artificial intelligence methods. In this context, the development of model-based decision-making systems that integrate predictions from classical machine learning algorithms and models generated using automated machine learning techniques is highly relevant. Such systems combine the advantages of expert-driven algorithm selection with the capabilities of automated searches for optimal model architectures and hyperparameters. This hybrid approach enhances the accuracy of assessing a power transformer technical condition and, consequently, improves the determination of its expected service life based on the evaluation. Aim. To improve the reliability of power transformers while minimizing maintenance costs through the application of artificial intelligence methods. Methods. Statistical analysis of chromatographic data of transformer oil; data preprocessing (elimination of anomalous and duplicate records, z‑transformation); classical machine learning methods (linear regression, Random Forest, Extra Trees, Hist Gradient Boosting), model validation using an 8:2 data split; development of a model structure based on AutoML with the specialized FEDOT software platform; calculation and analysis of model performance metrics (R², MAE, MSE, RMSE); ensemble methods Averaging, Weighted Averaging, Stacking, Blending and XGBoost. Results. An ensemble model was developed for the comprehensive assessment of the technical condition of power transformers based on transformer oil chromatography analysis and operational data, using machine learning methods. This approach eliminates labor-intensive calculations of the effect of individual parameters and reduces human factor impact during expert evaluations. Implementation of the proposed model allows objective estimation of the remaining lifespan of power transformers and justifies the transition to risk-oriented maintenance, thereby reducing operational costs and minimizing the risk of electrical equipment failure.