Abstract: The worm shaft pump, a type of vacuum sewage pump, is widely used in marine vacuum sewage systems. Its high-efficiency negative pressure propulsion significantly enhances the system's cleaning and odor removal capabilities. The operational status of the worm shaft pump directly affects the stability of the system. Therefore, accurately and rapidly predicting its performance has become a critical issue that needs urgent resolution. To address this challenge, digital twin technology, with its high precision and real-time prediction capabilities, has emerged as an ideal tool for evaluating pump performance. This paper proposes a digital twin prediction system based on Proper Orthogonal Decomposition (POD) and the Radial Basis Function (RBF) response surface method for rapidly assessing the performance of the worm shaft pump under various operating conditions. A matrix of operating condition parameters is generated through Latin hypercube sampling, and combined with the Twin Builder flow field reconstruction method, a reduced-order model database of the flow field is established. The TB-RBF model is constructed using the radial basis function interpolation method to predict flow field conditions under unknown operating conditions. Compared with the results of control experiments, this method achieves a prediction accuracy exceeding 96.5% under 27 verification conditions, with a single calculation time of only 0.7 seconds, representing a 6676-fold reduction in computation time. This study demonstrates that the digital twin system for the worm shaft pump based on the reduced-order model exhibits significant advantages in real-time performance prediction. The performance indicators include the head of the worm shaft pump, as well as the pressure and velocity distribution of the flow field, which can be utilized for real-time monitoring and optimal design of marine vacuum sewage systems.