The safety of industrial control network is becoming more prominent. Electric power is an important national infrastructure, so the safety protection of smart grid industrial control system is extremely important. In smart grid industrial control system, according to the status quo of the low internal protection level of the control network and the lack of internal network of anomaly traffic detection, this paper analyzes the composition of the industrial control system, the network security demand, and the threats faced by the smart grid industrial control system. It proposes to apply traffic anomaly detection technology to the security protection of smart grid industrial control system, which forms the two-level security protection. Then, the classification and characteristics of traffic anomaly detection methods and the characteristics of network traffic of smart grid industrial control system are studied. And it proposes a dynamic semi-supervised K-means algorithm based on entropy and OCSVM to improve the semi-supervised K-means algorithm for improving the internal protection level of the smart grid industrial control system.