基于变分模态分解参数优化的地震随机噪声去除方法

为解决变分模态分解在地震数据去噪中依赖人工经验，模态分解和去噪效果具有一定随机性和偶然性的问题，提出基于频域奇异值分解信噪比估计的参数优化方法。该方法在参数范围内以较高的估计信噪比为评价参数对模态分量数目与有效模态进行选取，自适应寻找去噪最有效的参数，从而避免主观选取参数的随机性，改善去噪效果。仿真模型实验表明：估计信噪比与真实信噪比的误差为正相关关系，能够有效反映地震数据中噪声程度，所估计信噪比可以作为去噪效果的评价参数。通过仿真模型和实际地震数据对方法进行验证，结果表明基于估计信噪比参数优化后的变分模态分解方法能够有效压制噪声、凸显同相轴信息。

Seismic random noise removal based on variational mode decomposition with parameter optimization

Due to the influence of artificial, environmental and geological conditions, seismic data is inevitably mixed with random noise, and noise must be effectively suppressed before subsequent processing and interpretation. Variational mode decomposition is one of the effective methods for seismic denoising, with good robustness and decomposition accuracy. However, the number of modal components decomposed and the recognition of effective modalities after decomposition mainly depend on manual experience selection, which leads to modal decomposition and denoising effects with a certain randomness and contingency. In order to solve this problem, this paper proposes a parameter optimization method based on singular value decomposition for signal-to-noise ratio (SNR) estimation. The modal component number and effective mode are selected by using the higher estimated SNR as the evaluation parameter in the parameter range. The method can be adapted to find the most effective parameters for denoising, thus avoiding the randomness of subjective selection parameters and improving the denoising effect. Simulation model experiments show that the estimated SNR is positively correlated with the true SNR error, which can effectively reflect the noise level in the seismic data. The estimated SNR ratio can be used as the evaluation parameter of the denoising effect. The simulation method and actual seismic data are used to verify the method, and the results of wavelet denoising and empirical mode decomposition are compared. The results show that the variational mode decomposition method based on the optimized SNR parameter can effectively suppress the noise and highlight the in-phase axis information.