The invention relates to a support vector machine power grid business classification method based on feature weight learning, which utilizes a feature weight learning method to combine with the multi-classification characteristic of the support vector machine to perform processing on the business classification. The support vector machine power grid business classification method comprises steps of dividing collection data into a training set and a testing set, adopting a 1-a-1SVM classification method to decompose a multi-classification problem into a series of SVM secondary classification problems, performing parameter optimization and characteristic subset choosing on each secondary classifier, introducing characteristic learning ideology, and endowing various weight values to various characteristics in order to express the degree of importance. The model which is obtained through choosing the optimal characteristic subset and the optimal parameter training model is a classification model, and the classification is performed on a testing set sample. The support vector machine power grid business classification method disclosed by the invention gives full consideration to the difference between various sub-classifiers through respectively choosing the optimal parameter in each SVM secondary classifier and the characteristic subset retraining SVM classification model, and has better classification precision.