The invention discloses a state evaluation method for a power transformer, and the method comprises the steps of obtaining the state parameters of the power transformer, carrying out the splitting andsample synthesis of a conventional data set, and forming a plurality of complete balanced data sets, so as to serve as the training data of a component learner; training a plurality of component learners by using the training data, and forming an integrated learner by the plurality of trained component learners; obtaining the state parameter of the power transformer to be identified, inputting aconventional data set in the state parameter into the integrated learner, and using the integrated learner to output a first state evaluation result of the power transformer, so that the problems of training data loss and data classification imbalance are solved, the cost sensitivity requirement of power transformer state evaluation is met, and the state of the whole network transformer can be accurately evaluated.