This paper proposes a method for evaluating the condition of large power transformers based on a health index. It constructs a four-layer deep architecture transformer health assessment system. The state indicators' deterioration is evaluated using extension cloud theory. Indicators are weighted at the indicator level by combining the Analytic Hierarchy Process and the Entropy Weight Method. The enhanced Dezert-Smarandache Theory (DSmT) is applied to successfully merge evaluation results from multiple layers, addressing any inconsistencies and disputes among conclusions. The study results demonstrate that the proposed approach can precisely and efficiently evaluate the health condition of transformers and their components, offering crucial insights for equipment management and maintenance planning.