The invention provides an online collected electric power data classifying method. The online collected electric power data classifying method includes the steps of (1) collecting data and establishing a database; (2) selecting data and samples from an original database; (3) training a linear SVM (supported vector machine) by the aid of organized data, and saving a training result; (4) judging whether an event is a transformer fault event via a trained model; (5) explaining a classifying result and adjusting power quality. The online collected electric power data classifying method has the advantages that an SVM classifier based on a stochastic gradient descent algorithm is applied to classification and recognition of the transformer fault event, so that the online collected data classifying problem of actually measured data of an electric power system can be solved; the stochastic gradient descent algorithm is adopted, and by means of subjecting every sample to iterative refinement for one time, even under the condition of large number of samples, the optimal solution can be iterated probably by only using tens of thousands or thousands of the samples; the online collected electric power data classifying method is more applicable to increasing demands of online collected electric power data processing nowadays.