High-precision prediction of vessel traffic flow is crucial for managing traffic during adverse weather conditions and enhancing navigation efficiency. Aiming to eliminate the deficiencies of traditional prediction methods and improve prediction precision and adaptability, this paper proposed a vessel traffic flow prediction method based on an origin-destination (O-D) matrix and a Spatio-Temporal Zero-Inflated Negative Binomial Graph Neural Network (STZINB-GNN) for port cluster navigational networks. To understand the internal coupling relationships within the traffic network from a macro perspective, a port cluster navigational network was constructed based on vessel origin-destination (O-D) data. Traditional prediction methods often overlook the Euclidean spatial characteristics between network nodes and face challenges due to the sparsity of vessel traffic flow. To address these issues, a spatial-temporal graph neural network prediction method incorporating a sparsity parameter \(\pi\) was proposed. A case study with the Qiongzhou Strait port group demonstrated that the STZINB-GNN achieved the highest prediction accuracy across three temporal resolutions compared to four established methods: STNB-GNN, STG-GNN, STGCN, and HA. Specifically, the STZINB-GNN attained superior accuracy when the temporal resolution was extended to one day.