Oil dissolved gas analysis is an important means for transformer fault diagnosis. However, in practical applications, there are problems such as data imbalance and insufficient generalization ability of the diagnostic model. This paper proposes a transformer fault diagnosis method that integrates the improved K-nearest neighbor data augmentation and QPSO-SVM-Attention-Adaboost. Firstly, the improved K-nearest neighbor data augmentation algorithm is used to generate high-quality synthetic samples, balancing the distribution of various data categories and improving the model’s ability to identify minority fault types; Secondly, the quantum particle swarm optimization (QPSO) algorithm is used to optimize the parameters of the support vector machine (SVM), constructing the optimal QPSO-SVM base classifier; Then, the dynamic Attention mechanism is introduced to improve the Adaboost integration framework, achieving adaptive fusion of base classifiers by calculating the dynamic weights of the classifiers, and strengthening the contribution of effective classifiers; Finally, based on the DGA dataset, multiple sets of comparative experiments are designed to verify the effectiveness of the proposed method. Experimental results show that the overall diagnostic accuracy of the proposed method in transformer fault diagnosis reaches 94.73%, which is superior to methods such as ABC-SVM and PSO-SVM, providing a new technical path for precise diagnosis of transformer faults.