Abstract: The main mechanisms of high-frequency scattering include specular reflection, edge diffraction, and creeping wave diffraction. To overcome limitations of single-algorithm methods for electrically large targets, which suffer from incomplete mechanisms and trade-off between efficiency and accuracy, this paper proposes a collaborative modeling framework with heterogeneous parallel acceleration. The framework integrates multilevel fast physical optics (MLFPO) for specular scattering, incremental length diffraction coefficient (ILDC) for creeping waves, and truncated wedge ILDC(TWILDC) for edge diffraction, effectively computing edge and creeping wave diffraction, and establishes a physical model covering three major scattering mechanisms, improving accuracy. For efficiency, a fine-grained GPU parallelization strategy and an event-triggered heterogeneous scheduling mechanism are designed for efficient CPU-GPU collaborative execution. Numerical examples show that the proposed model achieves good agreement with multilevel fast multipole algorithm(MLFMA), reducing mean absolute error by 1.82 dB versus standalone MLFPO. The proposed scheme delivers a maximum 318.43× speedup over MLFMA, exceeding two orders of magnitude. This method provides solid technical support for high-confidence and efficient radar cross section(RCS) prediction of electrically large targets.