As the processor's power consumption continually increases, power has become the most critical problems during the design and implementation of high performance computer (HPC) system. Nowadays, the heterogeneous system has been one important trend for HPC system. Compared with the traditional homogeneous system, the heterogeneous system has a higher theoretical performance and energy efficiency. However, explointing the potential advantage under the performance constraint is still a challenging problem. This work first establishs the energy optimization model for heterogeneous parallel system via abstracting common applications into the general program model which consists of sequential section and parallel section. Through theoretical analysis, the study conclude the relationship among heterogeneous processors for the minimum energy consumption during the single parallel section and full application (including multiple sections) and provide the corresponding algorithms to decide the operating frequencies under the given performance constraint. Finally, the study evaluates of the proposed model with eight typical applications on a CPU-GPU heterogeneous system.