In recent years, the Internet industry has entered a period of rapid development, and smart life has brought people a lot of experiences they have never had before. Rich application scenarios not only bring convenience to people, but also expose more and more network security problems. Network traffic types are more diversified, network security monitoring is becoming more and more difficult, and network communication quality and user host security are constantly facing the threat of network intrusion. This paper studies the current network traffic anomaly detection methods. Aiming at the problems of low accuracy and difficult real-time monitoring caused by the limitations of data scale and processing capacity in the previous methods, combined with machine learning and data simulation technology, a multi model fusion streaming parallel anomaly detection method is proposed, which enables distributed processing of massive streaming data on the basis of ensuring algorithm accuracy, At the same time, a visualization system based on network traffic anomaly detection is developed. The system can monitor the flow, and the reliability experiment can also meet the daily needs.