The invention provides a power industrial control attack classification method and system based on machine learning. The method and the system are characterized by utilizing the historical message data of the electric power industrial control, after completing the default value of the data, extracting the characteristic variable, inputting the stochastic forest model for multi-fold cross-validation, and adjusting the model parameters according to whether the stochastic forest model has occurred fitting and/or under-fitting phenomenon to determine the optimal stochastic forest model to classifythe electric power industrial control attacks. Compared with that prior art, by collecting the history message data of electric power industry control for machine learning, the random forest model isbuilt, and the messages generated by the electric power industrial control system are imported into the random forest model to realize the classification of the electric power industrial control attacks, thereby improving the status quo of the passive defense of the industrial control system, enabling the system to detect and intercept the attacks before being attacked, and improving the safety performance of the electric power industrial control system.