The invention discloses a power user feature recognition and classification method based on a random forest. The method adopts a random forest algorithm to carry out feature recognition and classification on power users. Firstly, feature data are extracted as many as possible through analyzing a load curve, and learning samples and test samples of the random forest are obtained; the sample data are then learnt through the random forest algorithm, a final classification decision making model is acquired, and the test samples are used to check the accuracy of the model; and finally, the model isused to analyze to-be-classified power user feature data, and a power user classification condition is acquired. Besides, the adopted random forest is a statistical theory and is a combined classifier algorithm, and the algorithm has a high accuracy rate of prediction, good tolerance to abnormal values and noise and is not easy to be over fitting, the user feature recognition and classification problems can be effectively solved, recognition and classification demands on loads by a power grid can be met, and a foundation is laid for realizing management on different load clusters at a demandside by the power grid.