The invention discloses a probability graph-based transformer state association rule mining method. The method comprises the steps of constructing a transformer state variable association rule mining data set; finding out all frequent 2-item sets in the transformer state variable data set; screening the frequent 2-item sets and calculating condition probability distribution among all the screened frequent 2-item sets; constructing a probability graph by taking the frequent 2-item sets and the condition probability distribution among different frequent 2-item sets as edges; finding out all association rules and generating an association rule set; determining an association degree among transformer state variables according to a support degree and a confidence degree obtained by calculation; and introducing the association rules among state variable data into transformer state prediction to correct a prediction result. The method has the beneficial effects that the association rules among the data obtained by data mining are introduced in the state prediction to correct the prediction result, so that the prediction precision can be remarkably improved, and the average prediction error is reduced to 10% from 20%.