The present invention relates to a big data analysis method for warning of heavy-load and overload of an electric power system. The method comprises: preprocessing historical electric power load data; extracting a set of attributes that are strongly related to a load attribute in preprocessed historical electric power load data D, establishing a data set D' that comprises load attribute data and the set of attributes strongly related to the load attribute; dividing the data set D into a normal operation data class, a heavy-load data class and an overload data class by using a clustering method; and constructing an electric power load data prediction model according to the data set D, predicting a heavy-load probability and an overload probability of a transformer. According to the method provided by the present invention, the heavy-load probability and the overload probability of each covered area in a specific time interval of a specific period and region is predicted by using historical load data, weather data, and other external data, so as to optimize allocation of human and material resources for testing, maintenance and the like of the distribution network in the period, improve system operation safety and stability, and reduce the labor intensity the staff of the power grid.