Abstract: Abnormal oil pipeline leakage rarely occurs in the production environment. Besides, the abrupt change in the monitoring data curve brought about by the artificial adjustment of pump frequency and instrument calibration, i.e., false anomaly, is hard to be distinguished from the true anomaly, leading to a low recall rate and a high false alarm rate of the traditional abnormal leakage identification method based on machine learning. To solve such problems, an abnormal leakage identification method based on the distinction between true and false anomalies was proposed. Specifically, the normal working mode of oil pipelines were learned with the One-Class Support Vector Machine (OCSVM), and the suspected pipeline anomalies, i.e., true and false anomalies, were screened out with the model. Then, the morphological difference between true and false anomalies curves was increased by superimposing the multi-source data, and in this way, the anomaly patterns of leakage events were found by similarity clustering. Further, the method was verified by applying it to the oil transportation environment. The results show that the recall rate of abnormal leakage identification method is 100%, and the exclusion rate of false anomaly is 83.49%. Generally, this method achieves the real-time and efficient monitoring of abnormal pipeline leakage events in complex operation environments and provides a practical idea for the application of machine learning methods in production environments.