Abstract: Fish in nature use the lateral line system to sense surrounding flow fields in real time, enabling agile obstacle avoidance and efficient swimming. Inspired by this mechanism, integrating a similar sensing approach into underwater vehicles holds promise for enhancing maneuvering accuracy and environmental adaptability. A real-time prediction method based on surface pressure sensing and long short-term memory (LSTM) networks is proposed to realize real-time hydrodynamic prediction for underwater vehicles with high efficiency and accuracy, and to overcome the limitations of traditional methods in both speed and precision. The method employs a limited number of pressure sensors to capture local flow information. An LSTM network is then trained to establish a nonlinear mapping between surface pressure data and global hydrodynamic forces, so as to realize the real-time prediction with the local pressure information as the input and the global hydrodynamic force as the output. Using the SUBOFF submarine model as a numerical test case, key hydrodynamic coefficients—including resistance, lift, and moment—are predicted. Results demonstrate that the proposed approach achieves high prediction accuracy while maintaining computational efficiency. This work presents a new solution for real-time hydrodynamic force estimation in autonomous underwater vehicles.