Aiming at the nonlinear prediction problem of three-degree-of-freedom (3-DOF) maneuvering motion attitudes of unmanned surface vessels (USVs) in waves, this paper introduces a hybrid prediction model that combines Variational Mode Decomposition (VMD), Sparrow Search Algorithm (SSA), and Long Short-Term Memory (LSTM) network. Experimental data from the 35° turning test of the KCS ship model in regular waves are used. VMD is applied to conduct multimodal decomposition on the 3-DOF motion parameters of USVs, extracting dynamic response features, while SSA is utilized to optimize the hyperparameters of the LSTM network, constructing a time-series prediction model. The prediction results suggest that the proposed hybrid model enhances the effectiveness and generalization capability of motion identification modeling for USV maneuvering in waves, providing a reference for subsequent modeling research.