Energy security is related to national security, and power security is the core of energy security. With the process of intelligent transformation of power, the production network gradually moves from being closed to interconnection. Power production and operation are highly dependent on the power monitoring system and dispatching data network. Once an external attack breaks through The safety protection system will directly threaten the safe and stable operation of the power system, so higher requirements are put forward for the detection of abnormal flow in the power system. This paper designs an intrusion detection algorithm based on the normal flow threshold model based on the deep machine learning algorithm, and conducts a comparison test through the flow characteristic value, and finally verifies the accuracy and reliability of the abnormal flow detection algorithm proposed in this paper for modern power networks in different test environments.