In order to make up for the shortcomings of the existing transformer fault diagnosis methods in the Dissolved Gas Analysis (DGA) quantity feature selection and diagnostic model,this paper proposes to use three gas ratios in the IEC three-ratio method as the characteristic quantity of transformer fault diagnosis. Three groups of optimal DGA characteristic gases with the highest accuracy are selected from 254 gas combinations containing any three or more dissolved gases in eight kinds of oils, which is taken as the characteristic quantity of the control group. Then, the Empire Competing Algorithm (ICA) optimized support vector machine transformer fault diagnosis model (ICA-SVM) is compared with standard Support Vector Machine (SVM) method, Particle Swarm Optimization Vector Machine (PSO-SVM) and IEC three-ratio method. The results show that the accuracy of fault identification is improved by about 10% compared with the optimal DGA gas combination of three groups; the accuracy of ICA-SVM fault diagnosis model increased by 7% to 35% compared with the standard SVM method, PSO-SVM and IEC method; the accuracy rate of the integrated three-ratio feature and the ICA-SVM fault diagnosis model is 89.3%, which is 7% ~ 35% higher than that of the other methods. Therefore, the validity and accuracy of the proposed method are verified. This work is supported by National Natural Science Foundation of China (No. 51867003), Guangxi Natural Science Foundation (No. 2015GXNSFBA139235), Guangxi Science and Technology Department Project (No. AE020069), and Science and Technology Project of State Grid Henan Electric Power Company (No. 52170215000V).