The invention provides a network intrusion detection method based on a time sequence neural network. The method comprises the following steps: collecting network traffic data in a point to be detected, and storing the network traffic data in a classified manner; Preprocessing the classified and stored network traffic data, wherein the preprocessing comprises symbolic feature numeralization and feature value normalization; Training a GRU network, and detecting the preprocessed network traffic data by using the trained GRU network; If the detection data result is normal flow, allowing the flow to pass through the current detection point; And if the detection data result shows that the attack flow exists, sending an alarm to the user. By adopting the GRU network, the network structure is further simplified on the basis of the LSTM, the defects existing in the RNN can be effectively overcome, and compared with the LSTM, the method is closer to the requirement of network instantaneity.