Advances in technology has promoted the development of power big data and the research of data mining for the power industry has become a hot topic at present. In view of such issue of traditional support vector machine（SVM）as long consumption time and weak recognition effect in processing big data

the sub-gradient descent algorithm is introduced to train the support vector machine with the aim of reducing the influence of training sample size on training time of the support vector machine. At the same time

with the help of advantages of processing big data by Hadoop platform

the dimension of the original data is reduced by constructing power load characteristic index. The multiple classifiers are randomly constructed under the MapReduce framework and the sample type is classified on the basis of voting mechanism so to construct the classification index for result evaluation. The influence of size of the training sample on the classification effect and the performance of the four machine learning algorithms in the processing of power big data are compared by calculation respectively. The results show that the method used in this paper has a good classification and recognition effect in the field of power big data.