<p>Intrusion detection, as an active network security protection technology, has become an important part of the network security field. However, with the increasing complexity and diversification of network attacks, it is a challenging topic to design an intrusion detection scheme with high detection accuracy and strong ability to detect unknown attack types. Therefore, this paper proposes a graph convolutional neural network (GCN)-based multi-classification network intrusion detection algorithm, named as GCNID algorithm. Firstly, in order to better represent the similarity and correlation between detection data, the preprocessed detection data is transformed into an undirected graph structure by using the <i>k</i> nearest neighbor method. Subsequently, a multi-classification network based on three-layer GCN is constructed to enhance the detection performance. Specifically, information from adjacent nodes is integrated into the central node through the graph convolutional layers of GCN, so that effective identification features can be extracted from the intrusion data, and the intrusion data can be identified and classified according to the obtained features. Simulation results indicate that the proposed algorithm not only enhances the precision of detection but also effectively categorizes unknown attack varieties.</p>