Characterization Of Local Attributes For Non-stationary Seismic Data And Its Applications

The seismic attributes are defined as the characterization of geometry,kinematics,dynamics,and statistics for seismic data.As the measurement of the seismic data,seismic attributes are able to reflect the information of subsurface media.Therefore,seismic attributes play an important role in the seismic data processing and interpretation.At present,there are lots of categories of seismic attributes.Considering the global and local characterization of seismic data,we can classify the seismic attributes as global attributes and local attributes.Global attributes reflect the characterization for the whole seismic data.Local attributes reflect the characterization for the local seismic data which in general refers to the single data point.Local attributes use the information of every data point and its neighbors.It's able to reflect the local information of seismic data accurately,which is beneficial to process and interpret the seismic data accurately and visually.Local attributes have lots of classifications.Common ones are local time-frequency attribute,local similarity,local signal-and-noise orthogonalization attribute,and so on.We studied two local attributes which are local time-frequency attribute and local signal-to-noise ratio(SNR).Local time-frequency attribute applies the mathematical theory of local attributes to the time-frequency transformation and calculate the time-varying Fourier series in the least-squares sense.Local time-frequency spectrum is able to reflect the time-frequency characterization of seismic data accurately,which could be applied to the seismic data processing and interpretation based on time-frequency analysis.We studied the compensation method of seismic wave attenuation on the basis of local time-frequency attributes.The attenuation was compensated by using Kolsky model and the spectral ratio in the local time-frequency spectrum.The other local attribute we studied is local signal-to-noise ratio.To solve the problem that the existing SNR can't reflect the local quality of seismic data,we proposed the concept of local SNR and defined the SNR function varying with time and space.The adaptivity turns the calculation of SNR into ill-posed problem.Therefore,we utilize the information of every data point and its neighbors to obtain the least-squares solution of local SNR by using the regularized conjugate gradient algorithm.The space-time varying SNR can reflect the quality of every data point,which provides an accurate and visual evaluation criteria for the subsequent processing and interpretation of seismic data.When solving the local SNR of field data,we also need to know the signal and noise in the noisy field data.We proposed the cascading signal estimation method to estimate the signal based on over-filtering and local signal-and-noise orthogonalization attribute.The proposed method uses the correlation between signal and noise to estimate the signal for calculating the local SNR.Local SNR estimation measures signal characteristics not instantaneously at each signal point but locally in the neighborhood of each point.Meanwhile,it can reduce the influence of global noise to local SNR.Local SNR estimation produces a visual and accurate SNR image.Moreover,it can also evaluate the denoising result.The mathematical theory of local attributes is using the regularized conjugate gradient algorithm to solve the inverse problem.The iterative manner has a high computational cost which limits the application of local attributes to the seismic processing and interpretation.We studied the streaming algorithm of local attributes.This algorithm updates the coefficients on a fly as each new data point arrives.The continuous updated calculation is the reason why this algorithm is called “streaming”.Streaming algorithm uses the updated coefficients to replace the adaptive coefficients,uses the algebraic similarity to replace the regularization,and uses the simple algebraic operation to replace the complicated iteration.Therefore,the streaming manner reduces the computational cost and computer memory greatly.Combining the theory of streaming local attributes and streaming prediction filter,we can obtain the streaming orthogonal prediction filtering method to attenuate the random noise in seismic data.It is able to protect the signal well as attenuate the random noise effectively.The streaming orthogonal prediction filtering method is the application of streaming local attributes.