Abstract: To address the engineering requirements for real-time and accurate prediction of extreme mooring line tension in deepwater floating production storage and offloading (FPSO) single-point mooring systems, where tension sensors are difficult to install, existing neural-network-based methods suffer from insufficient generalization capability, low training efficiency, and strong dependence on sensor. This paper proposes a real-time prediction method for FPSO mooring line tension based on mooring-chain motions in a local coordinate system and a NARX neural network. The time-series data of the mooring chain top-end motion in the local coordinate system are obtained from the OrcaFlex numerical model and used as input features. A double-hidden-layer NARX neural network is constructed for direct mooring tension estimation. Tests are carried out on FPSOs with different mooring configurations and structural forms, and the proposed method is compared with the global motion model and the Bi-LSTM model. The results show that the proposed method achieves high prediction accuracy, strong generalization capability, and high computational efficiency. The prediction error for the peak tension is only 1.6%, and the method he method enables real-time prediction at the millisecond level while smoothly reproducing high-frequency oscillations observed in numerical simulations. High prediction accuracy is maintained across FPSOs with different structures, displacements and mooring systems. The proposed mooring digital twin framework provides a feasible technical approach for real-ship mooring condition monitoring and safety early warning.