Low fault diagnosis accuracy is caused by the ineffectiveness of traditional shallow machine learning methods un exploring the connection between the oil-immersed transformer fault data. In response, this study suggests a method for diagnosing transformer faults based on multi-class adaBoost algorithms solves this issue. First, the SVM and the adaBoost algorithm are linked. The SVM is improved by the adaBoost approach, and the transformer defect data is thoroughly investigated. The IPSO is then used to optimize the SVM's parameters when the dynamic weight is added to the PSO algorithm. This is accomplished by updating the particle inertia weight in real-time. Lastly, by examining the relationship between the type of fault and the dissolved gas in the transformer oil, the uncoded ratio technique develops a novel gas set collaboration. The feature vector used as the input is produced using the enhanced ratio approach. The diagnosis method suggested in this paper has a significant increase in diagnostic accuracy when compared to conventional methods, according to simulations using 419 collection of transformer fault data and 117 groups of IECTC10 standard data that were gathered in China. Additionally, it has a fast confluence speed and a powerful search capability.