The petrochemical industry is large in production scale, it uses a large number of toxic and harmful devices in the process, and intensively relies on the safe, effective and long-term operation of these devices. Chemical accidents are often caused by abnormal conditions in the process, so the anomaly detection is an important tool to ensure the safe production of petrochemicals. The traditional anomaly detection methods based on mechanism model and knowledge model are limited by the complexity of the model and the artificial subjective experiences, and it is difficult to quickly apply them to new process systems. The anomaly detection methods based on mathematical statistics have problems such as slow calculation speed and inconsistent with actual parameters. Aiming at the problems of traditional anomaly detection methods, this paper proposes a petrochemical anomaly detection method based on the neural network, this algorithm uses the echo state network (ESN) to construct the feature data model, which can simplify the complexity of feature extraction and combine with the support vector machine (SVM) algorithm for the judgement and detection of abnormal data. At the same time, combining with the process of petrochemical industry, this paper proposes application examples of anomaly detection in petrochemical industry based on the neural network. This anomaly detection method can quickly perform feature data extraction and model construction, and can be applied to different processes, and has good versatility and practicability.