Shared Autonomous Vehicles (SAVs) are gradually developing, and their parking demand exhibits a short-term and frequent nature that traditional parking facilities cannot meet. To address this issue, this paper proposes a two-stage temporary parking facility location model based on mixed-integer programming for SAV parking. The first stage of the model focuses on the location selection of parking facilities, while the second stage addresses the dispatching of SAVs, with the objective of minimizing the combined costs of facility construction and vehicle dispatching. To enhance solution efficiency, this study introduces a Callback mechanism to improve the classical Benders decomposition algorithm, thus avoiding the computational overhead of repeatedly constructing the master problem. The feasibility of the model and the effectiveness of the algorithm were validated through case studies on both the Sioux Falls road network and the large-scale road network of Langfang City. The results demonstrate that the Callback-enhanced Benders decomposition algorithm not only significantly improves solution speed but also performs effectively on large-scale urban road networks, offering valuable insights and methods for large-scale SAV parking location optimization.