Ocean Alkalinity Enhancement (OAE) is a highly promising marine Carbon Dioxide Removal (mCDR) technique, and the dense shipping network in the Guangdong-Hong Kong-Macao Greater Bay Area can serve as a low-cost mobile platform for nearshore OAE implementation. To address the low carbon sequestration efficiency of conventional uniform alkalinity dosing caused by significant spatiotemporal variability in coastal hydrographic conditions, this study constructs an annual-scale dynamic dosing optimization model based on the TD3 deep reinforcement learning algorithm, using commuter vessels in the Pearl River Estuary as the deployment platform, and derives an adaptive intelligent strategy for alkalinity resource allocation to improve carbon sequestration performance. To verify the engineering feasibility of the strategy, robustness analysis is simultaneously conducted, and realistic validation is performed with the ROMS three-dimensional hydrodynamic-biogeochemical model. Comparative analysis shows that the intelligent dosing strategy outperforms the traditional uniform dosing mode in overall carbon sequestration performance and maintains stable carbon sink benefits even under extreme conditions with significant environmental forecast biases. Numerical simulation results confirm that the strategy can accurately identify favorable periods with intense water mixing and achieve differentiated, precise alkalinity dosing accordingly. With strong environmental adaptability and operational robustness, this model provides preliminary technical references for the iteration of coastal mCDR schemes and the large-scale deployment of OAE in the Greater Bay Area.