Management of Dissipative and Uncertain Trust in a Digital Twin Based Control System for the MASS

With the rapid development of autonomous ships and digital twin technology, the control of autonomous ships is gradually shifting from traditional onboard control to a hybrid model that integrates shore-based control with onboard monitoring, ultimately enabling fully autonomous navigation under the supervision of Shore Control Centers (SCCs) and advancing Mari time Autonomous Surface Ships (MASS). However, this new model introduces a critical trust challenge: when a ship initially begins to interact with SCCs or other unfamiliar ships, the absence of prior interaction history causes a “cold-start” trust hindrance. As MASS increasingly depends on dynamic collaboration among ships and SCCs, it has become safety-critical to determine the appropriate control levels for delegation to SCCs, as well as to evaluate the trustworthiness of information from unfamiliar ships, particularly during remote control failures. To address this issue, this article constructs a directed trust network based on ships' his torical interactive behaviors and proposes a trust model, GAG STrust, which combines GraphSAGE with Graph Attention Network (GAT). To overcome the intrinsic limitations of GraphSAGE and GAT, an Adaptive Neighborhood Sampling (ANS) mechanism is introduced to enable dynamic neighborhood selection with respect to node degree. The proposed model further enhances node representations by incorporating both in-degree and out-degree trust into comprehensive attention weights, and this help to enrich multi-dimensional ship features. Finally, it utilizes the prior ship control levels to tackle with the challenges outlined above. Experimental results demonstrate that GAGSTrust can accurately classify the control levels for SCCs' handover of MASS ships with complex and cascaded trust. Ablation studies confirm that the ANS mechanism improves the balanced accuracy of control levels prediction by 3.62%, which effectively mitigates the trust “cold start” problem in the MASS system.