Near-surface Q Value Inversion And Machine Learning Modeling

Seismic exploration method is an important means of oil and gas exploration.The relatively more complex near-surface underground stratigraphic structures can strongly attenuate the effective seismic signals and affect the quality of seismic data.The compensation for attenuation of energy can effectively enhance the resolution of seismic records.The key problem of attenuation compensation is how to accurately extract the attenuation coefficient from data,that is,the extraction of quality factor Q value.Among the commonly used quality factor estimation methods,the logarithmic spectral ratio method has better effect,but the direct method has poor noise resistance and poor stability in practice.In this paper,in order to complete the near-surface 3D Q value model based on micro-log data,we use a new Q value extraction method based on matching pursuit algorithm.The change of the first break energy received from the microlog data directly reflect the result of seismic wave attenuation.Therefore,we use the first break information to complete near-surface attenuation analysis.In this paper,firstly we study the matching pursuit which is a signal decomposition and reconstruction algorithm,and use the method to extract the first break of actual data to obtain its high-resolution spectrum.Then based on the conventional spectral ratio method,we use the inversion method of shaping regularization,according to regularization the spectral ratio calculation in the spectral ratio method will be more stable,so as to obtain a more accurate Q value estimation result.The processing of the model data and the actual data respectively indicates the accuracy of estimation.Finally,the artificial intelligence method of machine learning is introduced to construct a multivariate nonlinear regression model by deep neural network.According to the actual data,the Q value obtained by the matching pursuit is used to construct the training set to complete the training of the network model.We use the model to complete the near-surface 3D Q value modeling of the actual data in the entire work area.