Gravity inversion requires much computation,and inversion results are often non-unique.The first problem is often due to the large number of grid cells.Edge detection method,i.e.,tilt angle method of analytical signal...Gravity inversion requires much computation,and inversion results are often non-unique.The first problem is often due to the large number of grid cells.Edge detection method,i.e.,tilt angle method of analytical signal amplitude(TAS),helps to identify the boundaries of underground geological anomalies at different depths,which can be used to optimize the grid and reduce the number of grid cells.The requirement of smooth inversion is that the boundaries of the meshing area should be continuous rather than jagged.In this paper,the optimized meshing strategy is improved,and the optimized meshing region obtained by the TAS is changed to a regular region to facilitate the smooth inversion.For the second problem,certain constraints can be used to improve the accuracy of inversion.The results of analytic signal amplitude(ASA)are used to delineate the central distribution of geological bodies.We propose a new method using the results of ASA to perform local constraints to reduce the non-uniqueness of inversion.The guided fuzzy c-means(FCM)clustering algorithm combined with priori petrophysical information is also used to reduce the non-uniqueness of gravity inversion.The Open Acc technology is carried out to speed up the computation for parallelizing the serial program on GPU.In general,the TAS is used to reduce the number of grid cells.The local weighting and priori petrophysical constraint are used in conjunction with the FCM algorithm during the inversion,which improves the accuracy of inversion.The inversion is accelerated by the Open Acc technology on GPU.The proposed method is validated using synthetic data,and the results show that the efficiency and accuracy of gravity inversion are greatly improved by using the proposed method.展开更多
The S-wave velocity is a critical petrophysical parameter in reservoir description,prestack seismic inversion,and geomechanical analysis.However,obtaining the S-wave velocity from field measurements is difficult.When ...The S-wave velocity is a critical petrophysical parameter in reservoir description,prestack seismic inversion,and geomechanical analysis.However,obtaining the S-wave velocity from field measurements is difficult.When no measured Swave data are available,petrophysical modelling provides the most accurate S-wave velocity prediction.However,because of the complexity of underground geological structures and diversity of rock minerals,the prediction results of petrophysical modelling are easily affected by factors such as the cognition and experience of the modeller.Therefore,the development of novel robust and simple S-wave velocity inversion and prediction methods independent of the modeller is critical.Inspired by ensemble learning and based on the geologic sedimentation law of reservoirs and their characteristics in logging response,an Swave velocity inversion and prediction method based on deep hybrid neural network was developed by combining the classical convolution neural network(CNN)with the long short-term memory(LSTM)network.Considering the conventional logging data such as acoustic and density as the input in the proposed method,the CNN was used to establish the nonlinear mapping relationship between the input data and S-wave velocity,and the LSTM network was used to integrate the vertical variation trend of the stratum.Thus,intelligent data-driven inversion and prediction of the S-wave velocity were realised.The experimental results revealed that the proposed method exhibited a strong generalisation ability and could obtain prediction results comparable to those of petrophysical modelling with a single-well data set for training.Thus,a novel methodology for robust and convenient S-wave velocity prediction was devised.The proposed method has considerable academic and application implications.展开更多
基金supported by the National Key Research and Development Program of China Project(Grant No.2018YFC0603502)
文摘Gravity inversion requires much computation,and inversion results are often non-unique.The first problem is often due to the large number of grid cells.Edge detection method,i.e.,tilt angle method of analytical signal amplitude(TAS),helps to identify the boundaries of underground geological anomalies at different depths,which can be used to optimize the grid and reduce the number of grid cells.The requirement of smooth inversion is that the boundaries of the meshing area should be continuous rather than jagged.In this paper,the optimized meshing strategy is improved,and the optimized meshing region obtained by the TAS is changed to a regular region to facilitate the smooth inversion.For the second problem,certain constraints can be used to improve the accuracy of inversion.The results of analytic signal amplitude(ASA)are used to delineate the central distribution of geological bodies.We propose a new method using the results of ASA to perform local constraints to reduce the non-uniqueness of inversion.The guided fuzzy c-means(FCM)clustering algorithm combined with priori petrophysical information is also used to reduce the non-uniqueness of gravity inversion.The Open Acc technology is carried out to speed up the computation for parallelizing the serial program on GPU.In general,the TAS is used to reduce the number of grid cells.The local weighting and priori petrophysical constraint are used in conjunction with the FCM algorithm during the inversion,which improves the accuracy of inversion.The inversion is accelerated by the Open Acc technology on GPU.The proposed method is validated using synthetic data,and the results show that the efficiency and accuracy of gravity inversion are greatly improved by using the proposed method.
基金supported by the National Natural Science Foundation of China(Grant Nos.42030812,42042046,41974160)the project of SINOPEC Science and Technology Department(Grant No.P20055-6)。
文摘The S-wave velocity is a critical petrophysical parameter in reservoir description,prestack seismic inversion,and geomechanical analysis.However,obtaining the S-wave velocity from field measurements is difficult.When no measured Swave data are available,petrophysical modelling provides the most accurate S-wave velocity prediction.However,because of the complexity of underground geological structures and diversity of rock minerals,the prediction results of petrophysical modelling are easily affected by factors such as the cognition and experience of the modeller.Therefore,the development of novel robust and simple S-wave velocity inversion and prediction methods independent of the modeller is critical.Inspired by ensemble learning and based on the geologic sedimentation law of reservoirs and their characteristics in logging response,an Swave velocity inversion and prediction method based on deep hybrid neural network was developed by combining the classical convolution neural network(CNN)with the long short-term memory(LSTM)network.Considering the conventional logging data such as acoustic and density as the input in the proposed method,the CNN was used to establish the nonlinear mapping relationship between the input data and S-wave velocity,and the LSTM network was used to integrate the vertical variation trend of the stratum.Thus,intelligent data-driven inversion and prediction of the S-wave velocity were realised.The experimental results revealed that the proposed method exhibited a strong generalisation ability and could obtain prediction results comparable to those of petrophysical modelling with a single-well data set for training.Thus,a novel methodology for robust and convenient S-wave velocity prediction was devised.The proposed method has considerable academic and application implications.
