Abstract: The accuracy of reservoir prediction has directly restricted the economy and effectiveness of oil and gas development, and the compensation of low-frequency information is one of the favorable ways to improve the inversion effect of the reservoir. However, the conventional modeling method of well interpolation low-pass filter is difficult to represent the low-frequency models in complex geological background, which seriously restricts the effect of seismic inversion. In order to reduce the impact of the presence of special geological bodies on reservoir inversion in the study area, the authors have adopted the iteration method to build low-frequency model in this paper. The factors such as compaction and special lithology have been considered step by step during modeling, and the low-frequency model was iteratively updated through the inversion effect to accurately establish the inversion low-frequency model for reflecting different lithologies (including volcanic rocks). The results show that this method can reasonably guarantee the reliability of reservoir inversion results and fully provide the data foundation for reservoir detailed description. The favorable reservoir (P1) of the main layer of W gas field in Xihu Sag was also re-characterized based on the inversion results, and the horizontal section of the development well was optimized. Besides, the good reservoirs were confirmed by the implementation results. Therefore, the iteration method provides a more effective modeling method for reservoir prediction in complex geological background.