The hollow cathode, as the core component of an ion thruster system, plays a critical role in ignition and sustaining discharge for the thruster. However, its functional structural components gradually degrade during prolonged operation, ultimately leading to system failure. To effectively enhance the reliability of ion propulsion systems and mitigate operational risks, this study proposes a digital twin-assisted residual temporal convolutional network–bidirectional gated recurrent unit–attention (ResTCN-BiGRU-Attention) model framework for predicting the remaining useful life (RUL) of lanthanum hexaboride (LaB?) hollow cathodes. First, the working mechanism of hollow cathodes is analyzed, and a dynamic simulation model is constructed using digital twin technology to obtain high-fidelity simulated data. Then, a subdomain adaptation mechanism is introduced to align the conditional distributions of fine-grained features by minimizing the discrepancy between simulated and real-world data. Finally, the ResTCN-BIGRU-Atten model is applied to predict the RUL of hollow cathodes. Experimental results show that the proposed model achieves superior performance in four evaluation metrics: mean squared error, root mean squared error, mean absolute error, and coefficient of determination reaching 0.055, 0.2347, 0.1514, and 0.98738, respectively. The prediction accuracy is significantly better than that of other methods.