A hierarchical control framework integrating trajectory optimization and tracking is proposed to address the issues of complex trajectories, tracking difficulties, and high energy consumption in trajectory tracking of differential-driven unmanned surface vehicles. The upper layer employs a Bézier-based trajectory optimization method constrained by safety domains to fit and smooth reference paths, ensuring trajectory feasibility and continuity. In the lower layer, an improved nonlinear model predictive control algorithm incorporates a selective reference strategy: the optimized trajectory is highly discretized, and candidate points beyond the prediction horizon within the kinematically reachable range are selected to form a reference point index set. A rolling optimization process is then applied to solve for the optimal sequence of reference points, thereby reducing trajectory tracking energy consumption. Additionally, a state observer is introduced to compensate for wave disturbances. Simulation results demonstrate that the proposed method effectively improves trajectory fitting accuracy and tracking stability while reducing overall energy consumption without compromising control performance.