The stable approach of inverse Q filtering based on wavefield continuation theory can simultaneously perform amplitude compensation and phase correct

The inverse Q filtering method with a stabilization factor solves the noise amplification problem caused by the exponential growth of the amplitude compensation operator in traditional inverse Q filtering to a certain extent, but it is not ideal for reducing noise and amplitude compensation of deep seismic records. The inverse Q filtering method with a variable stabilization factor is superior to the inverse Q filtering method with a constant stable factor in reducing noise, but the amplitude compensation for the deep seismic record is still not ideal, especially when the quality factor Q value is small. In addition, because its compensation function changing with time and frequency is fixed, its amplitude compensation cannot be flexibly carried out according to the characteristic of seismic records. We have developed a novel inverse Q filtering method with a new variable stabilization factor under the constraint of signal-to-noise ratio (S/N). The S/N of seismic data is taken as prior information, and the stabilization factor is a function of the S/N and the quality factor. The amplitude compensation operator changes dynamically with time, frequency, S/N, and the quality factor, which can accurately compensate for the effective amplitude while avoiding enhancing the noise energy. Theoretical synthetic seismic data and real seismic data using the new method find significant improvements in amplitude compensation and suppression of noise energy. The new method has the strongest comprehensive ability of amplitude compensation and noise suppression. The resolution and S/N of seismic data processed by the new method have been improved.

This paper proposes a coal structure prediction technology based on deep learning, which uses logging data to achieve single-well prediction of the coal structure. This paper introduces the genetic algorithm (GA) to optimize the BP neural network, which can speed up its convergence to the global optimal solution, improve its training speed, and avoid the problems of easily producing the local optimal value and requiring a long training time. Taking the main coal seam of the Shizhuang block in the south of the Qinshui Basin as the research object and using the coal core data and logging data of nine parameter wells, the mapping relationship between the logging curve and coal structure is constructed based on the GA-BP neural network structure, and the coal structure is predicted. The prediction results are highly consistent with the coal structure measured from coal core sampling, with only a small error, and the prediction accuracy is 90%. It is shown that the GA-BP neural network structure can be used to effectively identify the coal structure, as well as predict the coal structure of uncored wells. Moreover, the findings of this study will be helpful for efforts to study the distribution law of the coal structure.

The quality factor Q has a wide application range in seismic data processing and interpretation. Compared with the commonly used methods of Q estimation (e.g., the spectral ratio method and the centroid frequency shift method), the logarithmic spectral area difference (LSAD) method exhibits better noise immunity. However, the LSAD method uses the area difference in one frequency band, and the Q values estimated using this method are still unstable under the influence of noise. We have developed a new parabolic fitting (PF) method to estimate Q. This method is an extension of the LSAD method. We derive an analytical relationship between Q and area difference curve of the logarithmic amplitude spectra. Using the logarithmic area difference of multiple frequency bands to fit the parabola, we improve the accuracy and stability of Q estimation. In addition, the PF method does not require special assumptions regarding the source wavelet. We test the PF method in the presence of noise and at varying bandwidths and compare the results with those obtained using the LSAD method. The results of the theoretical examples indicate that the PF method is noise resistant and stable. Applying the PF method to real zero-offset vertical seismic profiling records also indicates that the proposed method can reasonably and stably estimate the Q value of the formation.