Abstract: The residual moveout correction method based on dynamic time warping is faced with two problems:the dynamic time warping algorithm is sensitive to noise, and it is difficult to calculate the warping path accurately; the algorithm adopts a point-by-point moving method, which corrects residual moveout of seismic trace directly, and may cause seismic waveform distortion. Aiming at these problems, a residual moveout correction method of prestack gather was proposed based on sparse Bayesian learning (SBL) and dynamic time warping (DTW). The implementation steps are as follows:sparse representation of seismic gathers was realized via sparse Bayesian learning, and then the residual moveout correction of the sparse representation results was conducted by dynamic time warping, and the seismic records were reconstructed after processing. This method utilizes sparse Bayesian learning with good noise robustness and few local minimums. The global optimal solution is also the sparsest one. After sparse decomposition, the unit impulse response of subsurface was obtained, and the wavelet effect was eliminated, and then the distortion of waveform caused by using dynamic time directly can be avoided. The highlight of this method is that high fidelity residual moveout correction and random noise suppression are simultaneously achieved. Numerical simulation and actual data processing results show that the proposed method has a good application effect.

Key words: prestack gather, residual moveout, sparse representation, sparse Bayesian learning, dynamic time warping