Seismic Data De-noising Based On Noisy Independent Component Analysis

Random noises in petroleum exploration seismic data can affect its reliability. Improving the signal-to-noise ratio(SNR) of seismic data is a basic research project. Meanwhile, it’s of great significance to seismic data interpretation. Random noises in seismic data have no uniform rules, so the common denoising methods usually have certain limitations. The paper studies how to overcome the shortcomings of traditional seismic data denoising methods and improve the SNR of seismic data with the method of noisy independent component analysis(NICA)Firstly, the paper analyzes the statistic characteristics of petroleum exploration seismic data. Effective seismic data and random noises are independent statistically, which is in line with the application conditions of independent component analysis(ICA). Random noises consist of non-Gaussian noise and Gaussian noise. If non-Gaussian noise is regarded as an independent signal and Gaussian noise is regarded as additive noise, the problem of seismic data denoising can be transformed into a NICA problem. And the NICA model can match the seismic model better than ICA model.Aiming at the problem of conventional seismic data denoising, an improved noisy ICA algorithm is proposed. It is a new Fast ICA algorithm based on the maximum likelihood estimation and bias removal technique(MLBR-FICA). The algorithm can reduce bias caused by the noise through bias removal technique. Then an unbiased objective function based on the likelihood of observed signals is established. And the mixing matrix is optimized by the Newton iteration method. Simulation results illustrate that the proposed method can estimate the mixing matrix parameters more accurately and have fast convergence speed. Finally, the algorithm is applied to conventional seismic data denoising. Experiment results show that the algorithm can attenuate random noises as well as protecting effective seismic data.Aiming at the problem of strong noise in microseismic data, the paper proposes a noisy independent component analysis algorithm with low signal-to-noise ratio(LSNR-NICA). The sum of negentropy of the separated signals is utilized as the objective function. The gaussian-distribution density model is selected as the nonlinear function to estimate the negentropy. Meanwhile a confirm rule of the model parameters is established. It can suppress the noise effects in low signal-to-noise ratio(LSNR). Finally the objective function is optimized by artificial bee colony algorithm with good global convergence ability. Meanwhile the mixing matrix is obtained. Simulation results illustrate that the algorithm can estimate source signals more accurately. The algorithm is applied to microseismic data denoising. Experiment results show that the algorithm can attenuate random noises and improve quality of seismic sections.