Research On Weak Seismic Signal Denoising By Using Time-frequency Algorithm

With the increasing demand for oil and gas and the exploration techniques become mature gradually, the target for global oil and gas exploration has turned to the research on the complex oil-gas reservoirs and unconventional oil-gas exploitation in recent years. Then it comes along the adverse collection environment and the seismic data which is collected with poor quality. The effective information in the collected seismic data is hard to be identified due to a large number of noise interference. Thus, it’s necessary for us to eliminate random noise effectively and extract the weak effective signals from the seismic data with low signal-to-noise ratio(SNR). The research in this dissertation is focused on random noise suppression for the land seismic exploration records and microseismic records with low SNR. For the two different records above, the research is conducted by using two time-frequency denoising algorithms, which includes time-frequency peak filtering(TFPF) and shearlet transform, respectively. This dissertation constructs a series of improved schemes according to the shortcomings of the two algorithms in random noise attenuation, respectively. The theoretical analysis and the processing results of sythetic and real data have proved these improved schemes are effective to improve the SNR of the records.Time-frequency peak filtering is a relatively new time-frequency denoising algorithm. This algorithm can realize the recovery of the non-stationary signal effectively without any assumptions. Therefore, TFPF algorithm has been applied to seismic signal processing field and it is used to suppress the random noise in the land seismic exploration records. This algorithm has achieved some progress in seismic random noise attenuation. For the problem on random noise suppression in the land seismic exploration records with low SNR, this dissertation presents some improved schemes after analysising the shortcomings of the conventional TFPF algorithm in seismic random noise attenuation and then applies them to the denoising process for the land seismic exploration records. First, for the problem on divergence of time-frequency energy which is caused by the window function, we construct a time-frequency distribution with a high time-frequency concentration, which makes the TFPF algorithm based on the time-frequency distribution we construct has a good performance in signal preservation and random noise attenuation. Second, we analysis the influence induced from the window length in the TFPF algorithm. For the contradiction between signal preservation and noise suppression by using a fixed window length, this dissertation separates noisy signals into signal-dominant sections and noise-dominant sections according to the different smoothness in time domain and the different distribution ranges in frequency domain, and then the adaptive window length adjustment scheme is proposed in the TFPF algorithm. A short window length is adopted for signal-dominant sections to preserve the amplitudes of effective signals and a long window length is adopted for noise-dominant sections to eliminate more noise. Thus, the adaptive window length adjustment scheme improves the filtering precision of TFPF algorithm for non-linear seismic signals processing under low SNR. At last, in order to take the spatial correlation of effective signals in the data to be processed into account and make signals to be filtered better satisfy the unbiased estimation condition of the TFPF algorithm, we construct multi-directional temporal-spatial traces TFPF model. This algorithm conducts a linear segmentation to the events of effective signals and makes the segmented events are characterized by linear type. Then constructing the parallel radial traces according to the direction of the segmented event, finally the TFPF is done along different directions of traces, and thus we can do TFPF along directions of effective signals though the SNR of the seismic data is low. However, when the TFPF is done along directions of noise, the noise with a high correlation in that direction is easy to be regarded as effective signals and will be reserved, which brings the interference for effective signals identification. For this problem, this dissertation studies direction characteristics of random noise from land seismic exploration record in temporal-spatial domain. Then it gives the concept of directional random noise and proposes the identification and suppression method for the directional random noise. This reasearch makes the temporal-spatial traces TFPF algorithm effective in noise suppression.The unconventional oil and gas resources have been a new highlight in the global oil and gas exploration and development. There are rich reserves of unconventional oil and gas resources in our country and thus the unconventional oil and gas resources have a huge exploration prospect. Microseismic technique is the important mean in unconventional oil and gas reservoir exploration. However, the energy of effective signals in microseismic data is extremely weak and the signals are contaminated by strong interference, the conventional filtering methods can not make an effective processing for microseismic data which limits the application of microseismic data in the subsequent processing. Therefore, this dissertation carries out a study on microseismic noise suppression technique based on shearlet transform and constructs a filtering scheme and makes it suitable for microseismic signals multidirectional detection with variable window and irregular noise attenuation. Shearlet transform is a multiscale and multidirectional time-frequency analysis technique, it can realize the multiscale and multidirectional decomposion for signals. The information of effective signals can be reconstructed by a few nonzero coefficients combination, which provides a new idea for noise reduction. In this dissertation, first we make the most of directional characteristics of shearlet transform and set an adaptive threshold scheme according to the distinguishing characteristics between coefficients associated with effective signals and random noise at each scale and direction. The threshold we set will be low adaptively in the directions which are mapped by effective signals, and thus the effective signals can be better preserved in the noise elimination. Second, for the problem of identifying coefficients associated with effective signals under very low SNR, we compute the gradient of shearlet coefficients and extract the local structural information of coefficients associated with effective signals through the modulus value and direction of gradient. Then the diffusion process is done to the coefficients dominated by noise to attenuate their amplitudes, and thus the coefficients associated with effective signals can be identified. Finally, the effective signals can be recovered by reconstructing coefficients associated with effective signals in shearlet domain even under very low SNR.This dissertation constructs a series of optimized filtering schemes based on TFPF and shearlet transform algorithms, which overcome the shortcomings of two algorithms in deal with the data with low SNR and improve the denoising performance of conventional algorithms. Finally, these optimized filtering algorithms realize the better recovery of effective signals under low SNR, which provide reliable support for oil and gas basin fine structure reconstruction and raise development effectiveness of oil and gas reservoirs.