The horizontal ground heat exchanger (HGHE) possesses a complicated heat transfer mechanism as its extensive spatial scale, long operational duration, and vulnerability to meteorological conditions. Consequently, the long-period simulations of HGHE usually involve significant computational costs, posing challenges for its dynamic optimization implementation. Therefore, this study initially established and validated a conventional numerical (full-order) model for HGHE as the reference model. Subsequently, a hybrid model was developed using the proposed adaptive proper orthogonal decomposition (POD) method. By analyzing the influential characteristics, the study identified the solution strategy and the key parameter values for adaptive POD, followed by the generality tests. The hybrid model proved to successfully mitigate the issue of error accumulation commonly associated with native POD extrapolation. Finally, employing a long-running engineering case study, the accuracy and the solution efficiency of the hybrid model were compared against those of the conventional (full-order) model. The results demonstrated that the hybrid model maintained computational accuracy at a comparable level while exhibiting a computational efficiency 326 % higher than that of the conventional (full-order) model, without requiring additional computational resources. This study can provide efficient modeling support for the dynamic optimization design of HGHE.