Removal Of Random Noise In Seismic Data By Time-frequency And Spatio-temporal Filtering Algorithms

With the increasing market demand and declining conventional oil and gas production,the complex oil-gas reservoirs and unconventional oil-gas exploitation have become the key contents in the geophysical prospecting recently.Seismic data processing and interpretation receives great difficulties by complex stratigraphic environment and random noise,as well as the seismic data with poor quality.Therefore,it is of great significance to eliminate random noise effectively,to recovery the weak effective signals and to improve resolution ratio.And this dissertation focuses on the research of the random noise suppression for the land seismic exploration records and microseismic downhole records in low noise-signal ratio(SNR)environment.Improved schemes for time-frequency peak filtering(TFPF)algorithm are constructed as to the land seismic exploration denoising,where main emphases are the time-varying-WL process and two-dimensional expansion.Meanwhile,polarization filtering in Shearlet domain is presented for microseismic downhole signal reservation and noise attenuation.The improvements have been proved by the theoretical analysis,together with synthetic model and field data performance.Time–frequency peak filtering is a valid random noise attenuation algorithm,and enables revovering non-stationary desired seismic signal and suppressing random noise under a low low SNR.Traditional TFPF can not get rid of the the window length(WL)restriction and ignore spatial correlation,so the dissertation constructs improved schemes according to these shortcomings.Firstly,a fixed WL leads to a pair of contradiction between signal retention and noise attenuation requirements,so the time-vary-WL processing is added depend on the nonlinearity difference between desired seismic signal and random noise.The new proposed delay vector variance based on the straight line sequence(L-DVV)method is essentially an extension of the DVV method,and enables analyzing the nonlinearity by compareing to the obtained for the linearized version,the so-called surrogate.It has great superiority in the non-stationary signal processing without modeling for the time series;while the straight line sequence defined as the surrogate,so that L-DVV method has high stability and strong sensitivity to the random noise for discriminating the seismic signal masked by strong noise accurately.Since the linearity criterion of L-DVV method meets the unbiased estimation of TFPF algorithm,the nonlinearity cloud determine the location of the wave crest,rest of the desired signal and random noise.The different WLs deals with different sections: a lower WL to recover the reflect events and a higher one to suppress the noise.The time-varying-WL TFPF achieves a satisfactory performance in both signal retention and noise attenuation aspects in the synthetic model and real record.Secondly,traditional TFPF fliters along time domain andmay lead to weak continuity for the reflection events.The proposed delay-compensation TFPF establishes the reconstructed signal,and this time-space expansion filters along the propagation direction according to similarity among the desired signals in time-space domain.The reconstructed signal has a dynamic correction process to overcome the limitation of the direction curve fitting,which relieves waveform distortion problem occurs by the other time-space TFPF algorthms in complex stratum records.The reconstructed signal completely meets the TFPF unbiased estimation,so integrates enhancing reflection events and attenuating noise.As a result,delay-compensation TFPF preserves the desired seismic signals with irregular propagation law,suppresses random noise with low frequency,and improves the feasibility of TFPF.Experiments on both synthetic model and field data indicate that delay-compensation TFPF produces stronger and more continuous signal without distorting reflection events' the shape and location.Microseismic technique has been a new highlight in the unconventional oil and gas resources exploitation.The microseismic downhole desired signal are charactered by high frequency,short duration and low energy.At the same time,the signals are contaminated by strong random noise,which brings difficulties in the data processing and the following data interpretation.Microseismic records have “weak signal,strong interference”characteristics,and the high-frequency signal and extremely complex noise brings difficulty to remove the random noise by conventional filtering methods.This dissertation presents polarization filtering in Shearlet domain to divide the signal from the irregular microseismic noise.Shearlet transform involves the construction of multi-direction and multi-scale feature,and offers a directional multiscale framework to precisely analyze the optimal direction,location,and scale representations.Consequently,it presents more significant difference between the signal and random noise relative to time domain,frequency domain and some other time-space transform.But the high-frequency bands are often shared by the signal and noise,and part of the desired signals is filtered out with the random noise when using pure threshold filtering to restore the weak and complex reflection events.Polarization filtering weakens the noise interference according to the three-dimensional polarization feature of the microseismic downhole signals.This feature is highlighted in the Shearlet transform with each scale and direction.Therefore,polarization filtering in Shearlet transform could retain the high-frequency downhole signal effectively,suppress the strong noise,and avoid the false axis.The performance is explained and discussed thoroughly by the synthetic model and a real microseismic downhole data.The results show that this algorithm significantly suppresses the complex random noise and effectively preserves the desired signal.This dissertation constructs a series of optimized chemes,which overcome the shortcomings and improve the feasibility in the seismic random noise attenuation.L-DVV method maintains the superiority in desired seismic signal extraction by nonlinearity level,and applied in the WL-time-varying processing to overcome the limitation by a fixed WLof traditional TFPF algorithm.Delay-compensation TFPF establishes a reconstructed signal by the similarity of reflection event,so this two-dimensional expansion completely meets the TFPF unbiased estimation and dispenses the direction restraint.Polarization filtering in Shearlet domain provides a multiscale and multi-direction condition to meet the three-dimensional polarization filtering condition for the micrioseismic downhole record.Finally,these optimized filtering algorithms have satisfactory performances in realizing the signal preservation under low SNR and complex environments,which provide reliable support and raise development for unconventional oil-gas exploitation.