To explore the actual operation mechanism of ports under the background of diversified ship be-haviors, ship behaviors in port waters are identified by superposing and matching ship trajectory data and geographic information data. Navigation behaviors of inbound and outbound ships passing through the same waters are aggregated to form ship behavior patterns. Navigable waters with inbound and outbound ship activities are recognized and set as virtual nodes. Port areas and an-chorages are taken as actual geographical nodes. Routes of ship movement and transfer among port areas, anchorages and other navigable waters with ship activities are defined as network edges. A construction method of multi-port network considering ship behavioral characteristics is pro-posed.It is indicated that, under random attacks and descending betweenness centrality attacks, the average slope values of the variation curves of the maximum connected component size in the multi-port network with ship behavior patterns are reduced from -0.016406 and -0.012138 of the traditional network to -0.031445 and -0.02152 respectively. The earliest collapse points are delayed from 4 and 4 of the traditional network to 10 and 7. Meanwhile, the constructed network presents a stronger retention capacity of network efficiency compared with the traditional one.The constructed multi-port network possesses better robustness when coping with emergencies such as ship collisions and severe weather, which are regarded as attacks on network nodes. This study can provide the-oretical reference and technical support for the research on ship behaviors and the formulation of scheduling strategies in port waters.