The process of attack identification of data-driven complex oil and gas production systems is often mixed with fault data, which leads to the failure of timely attack defense or fault relief, resulting in the occurrence of attacks and oil and gas production accidents. Most of the current attack detection focuses on the detection of external attacks, ignoring the impact of the possible fault data of the oil and gas system itself on the attack detection. To distinguish abnormal events such as system faults from information attacks in complex oil and gas production systems and improve the accuracy of information physical attack detection in complex oil and gas production systems, an SVM-based undirected graph joint detection method is proposed. Firstly, the key sensors in the complex system of oil and gas production are topologized to form an undirected graph. Secondly, SVM is used to detect the anomalies of the undirected graph sensor system. Finally, the receiving station low-pressure pump system is taken as an example to verify. The results show that the accuracy rate, precision rate, recall rate, and F1 of the proposed attack detection method are above 99%. Taking the receiving station tank system as an example to verify, the results show that the accuracy rate, precision rate and recall rate of the proposed attack detection method are more than 97%, and F1 is more than 99%. Compared with the K-means method, the proposed method has a good performance in the accuracy and completeness of detection.