Abstract: The common reflection surface (CRS) stack imaging technique enhances the signal-to-noise ratio of seismic data by utilizing information from adjacent common midpoint gathers, making it a key technology for processing seismic data with low signal-to-noise ratios and low fold coverage. Whether using classical stepwise search or synchronous global search methods for CRS three-parameter optimization, a local optimization algorithm is ultimately required to enhance accuracy. The traditional Nelder-Mead simplex algorithm, with its simple search approach, suffers from relatively low computational efficiency. This paper adopts a derivative-free optimization algorithm called NEWUOA (new unconstrained optimization algorithm) as the precision enhancement method for CRS stacking. This algorithm dynamically constructs a quadratic interpolation model, employs a trust region iterative strategy, and utilizes curvature information of the objective function to improve computational efficiency. The algorithm was tested using a layered undulating model and the Marmousi2 model. The results show that for the simple layered undulating model and the complex Marmousi2 model, the NEWUOA generally outperformed or equaled the Nelder-Mead simplex algorithm in improving coherence values, with a computational efficiency increase of approximately 48% and 45%, respectively, compared to the Nelder-Mead simplex algorithm. This demonstrates that the NEWUOA can significantly enhance computational efficiency while maintaining the accuracy of CRS stack optimization, making it well-suited as a computational tool for precision refinement.

Key words: CRS stack imaging technique, three-parameter optimization, NEWUOA, derivative-free optimization, Nelder-Mead simplex algorithm