To diagnose a transformer fault accurately, dissolved gas analysis (DGA) is combined with artificial intelligence technology, such that a transformer fault diagnosis model based on LogitBoost-decision tree optimized by whale optimization algorithm (WOA) can be obtained. The model takes the decision tree as the weak learner. The LogitBoost ensemble algorithm is used as an integration framework to integrate multiple decision trees into a strong learner. A model optimization strategy based on the whale optimization algorithm is constructed to optimize the decision tree and the maximum splitting times of the decision tree in the LogitBoost-decision tree model. The experiments show that the WOA-LogitBoost-DT transformer diagnosis model improves the comprehensive diagnosis accuracy by about 4%, 10% and 21% compared with the commonly used decision tree, support vector machine and three ratio diagnosis models, respectively. The proposed model can provide technical support for transformer fault diagnosis.