Amplitude versus offset analysis is a fundamental tool for determining the physical properties of reservoirs but generally hampered by the blurred common image gathers(CIGs).The blurring can be optimally corrected usi...Amplitude versus offset analysis is a fundamental tool for determining the physical properties of reservoirs but generally hampered by the blurred common image gathers(CIGs).The blurring can be optimally corrected using the blockwise least-squares prestack time migration(BLS-PSTM),where common-offset migrated sections are divided into a series of blocks related to the explicit offsetdependent Hessian matrix and the following inverse filtering is iteratively applied to invert the corresponding reflectivity.However,calculating the Hessian matrix is slow.We present a fast BLS-PSTM via accelerating Hessian calculation with dip-angle Fresnel zone(DFZ).DFZ is closely related to optimal migration aperture,which significantly attenuates migration swings and reduces the computational cost of PSTM.Specifically,our fast BLS-PSTM is implemented as a two-stage process.First,we limit the aperture for any imaging point with an approximated the projected Fresnel zone before calculating the Hessian matrix.Then,we determine whether a seismic trace contributes to the imaging point via DFZ during calculating the Hessian matrix.Numerical tests on synthetic and field data validate the distinct speedup with higher-quality CIGs compared to BLS-PSTM.展开更多
Most State-Of-The-Art(SOTA) Neural Machine Translation(NMT) systems today achieve outstanding results based only on large parallel corpora.The large-scale parallel corpora for high-resource languages is easily obtaina...Most State-Of-The-Art(SOTA) Neural Machine Translation(NMT) systems today achieve outstanding results based only on large parallel corpora.The large-scale parallel corpora for high-resource languages is easily obtainable.However,the translation quality of NMT for morphologically rich languages is still unsatisfactory,mainly because of the data sparsity problem encountered in Low-Resource Languages(LRLs).In the low-resource NMT paradigm,Transfer Learning(TL) has been developed into one of the most efficient methods.It is difficult to train the model on high-resource languages to include the information in both parent and child models,as well as the initially trained model that only contains the lexicon features and word embeddings of the parent model instead of the child languages feature.In this work,we aim to address this issue by proposing the language-independent Hybrid Transfer Learning(HTL) method for LRLs by sharing lexicon embedding between parent and child languages without leveraging back translation or manually injecting noises.First,we train the High-Resource Languages(HRLs) as the parent model with its vocabularies.Then,we combine the parent and child language pairs using the oversampling method to train the hybrid model initialized by the previously parent model.Finally,we fine-tune the morphologically rich child model using a hybrid model.Besides,we explore some exciting discoveries on the original TL approach.Experimental results show that our model consistently outperforms five SOTA methods in two languages Azerbaijani(Az) and Uzbek(Uz).Meanwhile,our approach is practical and significantly better,achieving improvements of up to 4:94 and 4:84 BLEU points for low-resource child languages Az ! Zh and Uz ! Zh,respectively.展开更多
Deep learning has achieved great success in a variety of research fields and industrial applications.However,when applied to seismic inversion,the shortage of labeled data severely influences the performance of deep l...Deep learning has achieved great success in a variety of research fields and industrial applications.However,when applied to seismic inversion,the shortage of labeled data severely influences the performance of deep learning-based methods.In order to tackle this problem,we propose a novel seismic impedance inversion method based on a cycle-consistent generative adversarial network(Cycle-GAN).The proposed Cycle-GAN model includes two generative subnets and two discriminative subnets.Three kinds of loss,including cycle-consistent loss,adversarial loss,and estimation loss,are adopted to guide the training process.Benefit from the proposed structure,the information contained in unlabeled data can be extracted,and adversarial learning further guarantees that the prediction results share similar distributions with the real data.Moreover,a neural network visualization method is adopted to show that the proposed CNN model can learn more distinguishable features than the conventional CNN model.The robustness experiments on synthetic data sets show that the proposed method can achieve better performances than other methods in most cases.And the blind-well experiments on real seismic profiles show that the predicted impedance curve of the proposed method maintains a better correlation with the true impedance curve.展开更多
Picking velocities from semblances manually is laborious and necessitates experience. Although various methods for automatic velocity picking have been developed, there remains a challenge in efficiently incorporating...Picking velocities from semblances manually is laborious and necessitates experience. Although various methods for automatic velocity picking have been developed, there remains a challenge in efficiently incorporating information from nearby gathers to ensure picked velocity aligns with seismic horizons while also improving picking accuracy. The conventional method of velocity picking from a semblance volume is computationally demanding, highlighting a need for a more efficient strategy. In this study, we introduce a novel method for automatic velocity picking based on multi-object tracking. This dynamic tracking process across different semblance panels can integrate information from nearby gathers effectively while maintaining computational efficiency. First, we employ accelerated density clustering on the velocity spectrum to discern cluster centers without the requirement for prior knowledge regarding the number of clusters. These cluster centers embody the maximum likelihood velocities of the main subsurface structures. Second, our proposed method tracks key points within the semblance volume. Kalman filter is adopted to adjust the tracking process, followed by interpolation on these tracked points to construct the final velocity model. Our synthetic data example demonstrates that our proposed algorithm can effectively rectify the picking errors of the clustering algorithm. We further compare the performances of the clustering method(CM), the proposed tracking method(TM), and the variational method(VM) on a field dataset from the Gulf of Mexico. The results attest that our method offers superior accuracy than CM, achieves comparable accuracy with VM, and benefits from a reduced computational cost.展开更多
基金supported by the National Key Research and Development Program of China under Grant 2018YFA0702501NSFC under Grant 41974126,Grant 41674116,and Grant 42004101the Project funded by the China Postdoctoral Science Foundation under Grant 2020M680516
文摘Amplitude versus offset analysis is a fundamental tool for determining the physical properties of reservoirs but generally hampered by the blurred common image gathers(CIGs).The blurring can be optimally corrected using the blockwise least-squares prestack time migration(BLS-PSTM),where common-offset migrated sections are divided into a series of blocks related to the explicit offsetdependent Hessian matrix and the following inverse filtering is iteratively applied to invert the corresponding reflectivity.However,calculating the Hessian matrix is slow.We present a fast BLS-PSTM via accelerating Hessian calculation with dip-angle Fresnel zone(DFZ).DFZ is closely related to optimal migration aperture,which significantly attenuates migration swings and reduces the computational cost of PSTM.Specifically,our fast BLS-PSTM is implemented as a two-stage process.First,we limit the aperture for any imaging point with an approximated the projected Fresnel zone before calculating the Hessian matrix.Then,we determine whether a seismic trace contributes to the imaging point via DFZ during calculating the Hessian matrix.Numerical tests on synthetic and field data validate the distinct speedup with higher-quality CIGs compared to BLS-PSTM.
基金supported by the National Key R&D Program of China (No. 2017YFB0202204)the National Natural Science Foundation of China (Nos. 61925601, 61761166008, and 61772302)+1 种基金Beijing Advanced Innovation Center for Language Resources (No. TYR17002)the NExT ++ project which supported by the National Research Foundation, Prime Ministers Office, Singapore under its IRC@Singapore Funding Initiative。
文摘Most State-Of-The-Art(SOTA) Neural Machine Translation(NMT) systems today achieve outstanding results based only on large parallel corpora.The large-scale parallel corpora for high-resource languages is easily obtainable.However,the translation quality of NMT for morphologically rich languages is still unsatisfactory,mainly because of the data sparsity problem encountered in Low-Resource Languages(LRLs).In the low-resource NMT paradigm,Transfer Learning(TL) has been developed into one of the most efficient methods.It is difficult to train the model on high-resource languages to include the information in both parent and child models,as well as the initially trained model that only contains the lexicon features and word embeddings of the parent model instead of the child languages feature.In this work,we aim to address this issue by proposing the language-independent Hybrid Transfer Learning(HTL) method for LRLs by sharing lexicon embedding between parent and child languages without leveraging back translation or manually injecting noises.First,we train the High-Resource Languages(HRLs) as the parent model with its vocabularies.Then,we combine the parent and child language pairs using the oversampling method to train the hybrid model initialized by the previously parent model.Finally,we fine-tune the morphologically rich child model using a hybrid model.Besides,we explore some exciting discoveries on the original TL approach.Experimental results show that our model consistently outperforms five SOTA methods in two languages Azerbaijani(Az) and Uzbek(Uz).Meanwhile,our approach is practical and significantly better,achieving improvements of up to 4:94 and 4:84 BLEU points for low-resource child languages Az ! Zh and Uz ! Zh,respectively.
基金financially supported by the NSFC(Grant No.41974126 and 41674116)the National Key Research and Development Program of China(Grant No.2018YFA0702501)the 13th 5-Year Basic Research Program of China National Petroleum Corporation(CNPC)(2018A-3306)。
文摘Deep learning has achieved great success in a variety of research fields and industrial applications.However,when applied to seismic inversion,the shortage of labeled data severely influences the performance of deep learning-based methods.In order to tackle this problem,we propose a novel seismic impedance inversion method based on a cycle-consistent generative adversarial network(Cycle-GAN).The proposed Cycle-GAN model includes two generative subnets and two discriminative subnets.Three kinds of loss,including cycle-consistent loss,adversarial loss,and estimation loss,are adopted to guide the training process.Benefit from the proposed structure,the information contained in unlabeled data can be extracted,and adversarial learning further guarantees that the prediction results share similar distributions with the real data.Moreover,a neural network visualization method is adopted to show that the proposed CNN model can learn more distinguishable features than the conventional CNN model.The robustness experiments on synthetic data sets show that the proposed method can achieve better performances than other methods in most cases.And the blind-well experiments on real seismic profiles show that the predicted impedance curve of the proposed method maintains a better correlation with the true impedance curve.
基金supported in part by the National Key Research and Development Program of China under Grant 2018YFA0702501in part by NSFC under Grant 41974126,41674116 and 42004101。
文摘Picking velocities from semblances manually is laborious and necessitates experience. Although various methods for automatic velocity picking have been developed, there remains a challenge in efficiently incorporating information from nearby gathers to ensure picked velocity aligns with seismic horizons while also improving picking accuracy. The conventional method of velocity picking from a semblance volume is computationally demanding, highlighting a need for a more efficient strategy. In this study, we introduce a novel method for automatic velocity picking based on multi-object tracking. This dynamic tracking process across different semblance panels can integrate information from nearby gathers effectively while maintaining computational efficiency. First, we employ accelerated density clustering on the velocity spectrum to discern cluster centers without the requirement for prior knowledge regarding the number of clusters. These cluster centers embody the maximum likelihood velocities of the main subsurface structures. Second, our proposed method tracks key points within the semblance volume. Kalman filter is adopted to adjust the tracking process, followed by interpolation on these tracked points to construct the final velocity model. Our synthetic data example demonstrates that our proposed algorithm can effectively rectify the picking errors of the clustering algorithm. We further compare the performances of the clustering method(CM), the proposed tracking method(TM), and the variational method(VM) on a field dataset from the Gulf of Mexico. The results attest that our method offers superior accuracy than CM, achieves comparable accuracy with VM, and benefits from a reduced computational cost.