Abstract: The rapid identification of vibration sources near the oil and gas pipelines is an important technical means to ensure the safety of pipelines. As the existing single machine learning algorithms varied greatly in the monitoring performance in different application scenarios, and different learning algorithms were complementary in classification and recognition, a classification algorithm based on BP-SVM fusion was proposed. Based on the ϕ-OTDR system, the vibration signals of four different vibration sources, i.e. manual knocking, vehicle passing, mechanical construction and train passing, were collected, and a four-dimensional feature space was constructed as the feature vector of machine learning. The algorithm was applied in Hangu–Wuqing Product Oil Pipeline of North China Branch of China Oil & Gas Pipeline Network Corporation for recognition of site vibration signals, and the test results show that, the recognition rates of BP neural network classifier, SVM classifier and BP-SVM fusion classifier are 75.2%, 68.5% and 81.4% respectively, indicating that the performance of fusion classifier is obviously better than that of a single classifier.