How to solve various faults and abnormal behavior problems in the current power supply process is a challenge to the stability and security of power supply. In order to solve the problems of power small sample detection, this study uses the method based on machine learning algorithm, and takes the Support Vector Machines (SVM) algorithm as the main classification and detection means. The SVM algorithm uses the manner of constructing an optimal decision boundary to classify samples and detect abnormal points. The proposed power small sample detection system based on machine learning algorithm is evaluated by conducting experiments on a real power data set to verify the feasibility and effectiveness of the system. The experimental results suggest that the accuracy of this method is 89% to 97% when dealing with small sample data, which can effectively address the problems in the detection of power small sample, and detecting the classification of small sample data accurately. Compared with the traditional methods, the experimental results have demonstrated that this method can greatly improve the detection of power small sample.