The invention discloses a power grid fault tracking method based on data mining technology. The method includes taking various types of data acquisition systems of an intelligent substation as an original data source; selecting a decision tree to classify the information in the station, and providing decision support for the fault tracking program on a dispatching side; classifying the data filtered by the decision tree according to the supervision model into three classes: protection rejection signals, signal channel fault signals, and breaker rejection signals; and after the classification algorithm of the decision tree on the training samples is completed, predicting the data set in the unknown class, and corresponding to the power grid fault tracking process, namely, analyzing the various kinds of data of the transformer station layer to give the protection or breaker rejection reasons. The power grid fault tracking method can monitor the other data of the lower transformer station layer at the dispatching center, for example, the general level and forecasting level alarm data, can determine the condition of protection and breaker rejection according to the data, timely know the condition of the power system, and saves time for the power grid fault self-healing.