The invention discloses an IMABC optimized support vector machine-based transformer fault diagnosis method. The method comprises the steps of 1, dividing a collected sample set S={(x1,x2),(x2,y2)...(xn,yn)}, with class tags, of an oil-immersed transformer into training samples and test samples, wherein xi represents sample attributes including five attributes of hydrogen, methane, ethane, ethyleneand acetylene, yi represents the class tags, and 1, 2, 3, 4, 5 and 6 correspond to a normal state, middle temperature overheat, high temperature overheat, local discharge, spark discharge and arc discharge respectively; 2, proposing an improved artificial bee colony algorithm, fusing population classification and gene mutation in the artificial bee colony algorithm, and optimizing parameters of asupport vector machine; and 3, taking Ci and sigma i as the optimized parameters of the support vector machine, building a multilevel support vector machine fault diagnosis model, and performing transformer fault diagnosis by utilizing data in the step 1. According to the transformer fault diagnosis method, the parameters of the support vector machine can be effectively optimized, so that the accuracy of binary classification is improved.