The embodiment of the invention provides a prediction method and device, relates to the technical field of electric power, and solves the problem of how to predict small probability events of an electric power system. The method comprises the steps of determining a data set to which each first positive sample belongs and a neighbor sample of the first positive sample in operation data according toa K neighbor algorithm; according to preset conditions, first positive samples meeting the preset conditions are screened in the data set; determining a second positive sample according to a difference value between the characteristic variables corresponding to any two first positive samples meeting a preset condition; determining a decision tree classification model of each second positive sample according to the classification prediction model of the random forest; determining the number of votes of the decision tree classification model of each second positive sample according to a simplemajority voting method; determining a classification result of the data set according to the number of tickets of the decision tree classification model of each second positive sample; and according to the classification result of each data set, determining the classification result with the highest occurrence frequency in the classification results as the prediction result of the operation data.