Prior Geological Information Research Articles

Analysis on stable imaging and inverse algorithm for artificial source EM data

Abstract The inversion of artificial source electromagnetic (EM) method data fundamentally involves constructing a mathematical relationship between observable data and geological structures. The aim of imaging and inversion is to construct a geophysical model that matches the observable results, thereby realizing the identification of subsurface targets. The results of EM data inversion, due to the simplicity of geophysical models, limit inversion computing efficiency. Moreover, complexity of actual geological structures, and lack of onsite observable data, are often hindered by non-uniqueness. The challenge in the interpretation of artificial source EM data is in enhancing both the precision and expeditiousness of the inversion process. It can be classified into three main types for EM data inversion: direct imaging inversion, deterministic inversion, and stochastic inversion. To enhance computational efficiency and reduce non-uniqueness in the results, effective inversion methods, prior geological information, geophysical data, and comprehensive analysis can help mitigate the issue of non-uniqueness in EM data inversion, thereby leading to more rational geophysical interpretation results. With the progress of technology such as computing centers and the development of artificial intelligence methods, future inversion techniques will become faster, more efficient, and more intelligent, and will be applied to the interpretation of artificial source EM data.

Deep generative networks for multivariate fullstack seismic data inversion using inverse autoregressive flows

The simultaneous prediction of the subsurface distribution of facies and acoustic impedance (IP) from fullstack seismic data requires solving an inverse problem and is fundamental in natural resources exploration, carbon capture and storage, and environmental risk management. In recent years, deep generative models (DGM), such as variational autoencoders (VAE) and generative adversarial networks (GAN), were proposed to reproduce complex facies patterns honoring prior geological information. Variational Bayesian inference using inverse autoregressive flows (IAF) can be performed to infer the solution to a geophysical inverse problem from the encoded latent space of such pre-trained DGM. Successful applications of such approach on crosshole ground-penetrating radar synthetic data inversion demonstrated that the technique's accuracy is comparable to that of Markov chain Monte Carlo (MCMC) inference methods, while significantly reducing the computational cost. Nonetheless, these application examples did not account for the spatial uncertainty affecting the facies-dependent continuous physical property, from which the geophysical data are calculated. This uncertainty can significantly affect the inversion accuracy and its applicability to real data. In this work, specific VAE and GAN architectures are proposed to simultaneously predict facies and co-located IP, while accounting for their spatial uncertainties. The two types of generative networks are used in Bayesian inversion with IAF for the inversion of seismic data. The results are found to reproduce the statistics of the training images and solve the seismic inversion problem accurately, comparably to MCMC inversion. Furthermore, advantages and limitations of the two DGMs are evaluated by comparing the results obtained.

Basin‐scale prediction of S‐wave Sonic Logs using Machine Learning techniques from conventional logs

AbstractS‐wave velocity plays a crucial role in various applications but often remains unavailable in vintage wells. To address this practical challenge, we propose a machine learning framework utilizing an enhanced bidirectional long short‐term memory algorithm for estimating S‐wave sonic logs from conventional logs, including P‐wave sonic, gamma ray, total porosity, and bulk density. These input logs are selected based on traditional rock physics models, integrating geological and geophysical relations existing in the data. Our study, encompassing 34 wells across diverse formations in the Delaware Basin, Texas, demonstrates the superiority of machine learning models over traditional methods like Greenberg–Castagna equations, without prior geological and geophysical information. Among these machine learning models, the enhanced bidirectional long short‐term memory model with self‐attention yields the highest performance, achieving an R‐squared value of 0.81. Blind tests on five wells without prior geologic information validate the reliability of our approach. The estimated S‐wave velocity values enable the creation of a basin‐scale S‐wave velocity model through interpolation and extrapolation of these prediction models. Additionally, the bidirectional long short‐term memory model excels not only in predicting S‐wave velocity but also in estimating S‐wave reflectivity for seismic amplitude variation with offset applications in exploration seismology. In conclusion, these S‐wave velocity estimates facilitate the prediction of further elastic properties, aiding in the comprehension of petrophysical and geomechanical property variations within the basin and enhancing earthquake hypocentral depth estimation.

A constrained 3D gravity inversion for complex density distributions: Application to Brazil rifted continental margin

Gravity inversion is a highly effective method for investigating regional geological structures, and this paper proposes an optimization scheme for constrained three-dimensional (3D) gravity inversion to obtain a 3D density model, utilizing prior geological and geophysics information. Specifically, the proposed method enhances deep structural imaging resolution and minimizes false structures by progressively inverting deep and shallow-density structures using long and short-wavelength signals of gravity anomaly with prior information. The scheme is applied in the southeast passive continental margin of Brazil, and the results show that the density model is consistent with the previous reflection and refraction seismic data. Moreover, the 3D density model reveals several insights: (i) The Abimael Ridge (AR) and the São Paulo Plateau (SPP) exhibit thin crustal thickness (∼5–7 km) indicative of proto-oceanic crust. The SPP and AR area crustal thinning may be related to an aborted opposing rift propagator pair. (ii) The rifting modes of Santos and Campos Basins differ significantly. Campos Basin exhibits a depth-dependent lithospheric stretching model with a relatively intact upper crust. In contrast, Santos Basin shows a highly brittle upper crust that is partially thinned and, in some regions, even absent under the far-field effect of spreading failed rifts, while the lower crust remains relatively intact. Moreover, the upper crustal stretching factor is about five to ten higher than the lower crustal. Thus, the constrained 3D gravity inversion scheme provides a new avenue for continental rifted margin geological structure studies.

A knowledge-embedded close-looped deep-learning framework for intelligent inversion of multisolution problems

Deep learning is prevalent in many fields and attempts have been made to use it in nonbidirectional mapping problems, such as seismic inversion. These nonbidirectional mapping problems have two special issues, that is, insufficient labels and uncertainty of solution. Therefore, current deep-learning structures are not suitable for handling this kind of problem. A distinctive knowledge-embedded close-looped (KECL) deep-learning framework is developed, tuned to the characteristics of the seismic inverse problem. The KECL deep-learning framework is composed of a reservoir parameter generator (RPG) and a reservoir parameter updater (RPU). The former half-loop is RPG, which takes the seismic data as input to generate the initial reservoir parameters. The latter loop is RPU, which takes the initial parameters as input to output synthetic seismic data. Through the training by well data, the difference between field seismic data and synthetic seismic data modeled by the RPU is used to optimize the RPG and RPU. In this deep-learning framework, knowledge of the Robinson convolutional model is embedded to address the problem of insufficient labels. Furthermore, semisupervised learning is used as prior information to reduce the uncertainty of solution. After the training, with the help of prior geologic information data, the RPU is used to update the initial reservoir parameters generated by RPG for final reservoir parameter inversion. Numerical models and field data are used to test the feasibility of our deep-learning framework. We find that intelligent inversion results using data from one well to train the KECL network are consistent with results using multiple well data. Experiments demonstrate that it is adaptable to situations in which insufficient well data are available and is able to achieve reliable intelligent inversion.

An attempt to use convolutional neural network to recover layered-earth structure from electrical resistivity tomography survey

The Electrical Resistivity Tomography (ERT) method is a geophysical prospecting technique for determining subsurface structure in terms of electrical resistivity distribution. Direct current is injected into the ground through a pair of electrodes, and the potential difference is measured using another pair of electrodes. The estimated resistivity model of the subsurface is then obtained using a mathematical optimization algorithm called the inversion algorithm, which incorporates multiple measured potential differences from different electrode configurations as the input data. The inverted resistivity model serves as a resource for the interpretation process of the survey. Specialists employ the model and the prior geological information in the area to extract the subsurface geological structure under the data measurement profile. The inverted model can be treated as a two-dimensional image in which each element represents resistivity. The image can be visualized as a blurred or distorted version of the actual resistivity image of the subsurface. We, therefore, developed a post-inversion image enhancement code using a neural network aiming to recover the true subsurface model with predetermined features. We employed the convolutional neural network (CNN) that uses the inverted models as input and returns an enhanced model as output. We tested our developed neural network with a case study that utilized the proposed method to recover the layered earth structure which is often found in real surveys. The electrode spacing was 5 meters in the Schlumberger array configuration. The true subsurface resistivity models of training data and validating data for evaluating the neural network were randomly generated based on the predetermined feature of the layered structure. We then generated their respective inverted models from those data through simulated inversion routine. As a result, 94.75% of the enhanced models in the validation dataset exhibited a lower loss metric compared to the original inverted models in the same validation dataset. We also applied the proposed method to practical field data to recover the subsurface layers and compared them with stratigraphic information from a borehole in each area. The result showed that the proposed method gives well-resolved structures in the enhancement process after the typical inversion.

Learning from prior geological information for geotechnical soil stratification with tree-based methods

Geotechnical subsurface stratification based on sparse measurements presents a significant challenge. Learning from prior geological information, such as learning soil layer distribution patterns from stratification results (2D images) of adjacent data-rich projects, is an emerging approach to reducing uncertainties caused by data scarcity. Existing methods rely on pixel-based techniques that require training images to have identical soil types as the testing image. Additionally, pixel-based methods are prone to error accumulation. To address these issues, this study introduces a new framework that focuses solely on learning boundary information from training image rather than soil types. This eliminates the need for matching soil types between training and testing images. Tree-based model is proposed to learn boundary information from training images. Contrary to conventional tree-based models that use coordinates as input, this study employs a set of designed distance fields (boundary dictionary) to represent complex boundary patterns. A selection process is introduced to identify the most important distance fields from the training images. Using these selected distance fields for model input results in soil stratification that aligns well with both borehole data and training image boundaries. The proposed method's efficacy is validated through multiple simulated and real-world cases. The proposed method outperforms pixel-based methods in multiple cases, achieving up to a 25% improvement in accuracy. This method is robust and it yields consistent results for a variety of training images considered. Additionally, the proposed method also provides quantification of interpolation uncertainty through the Gini impurity method.

Introducing conceptual geological information into Bayesian tomographic imaging

AbstractGeological process models typically simulate a range of dynamic processes to evolve a base topography into a final two‐dimensional cross section or three‐dimensional geological scenario. In principle, process parameters may be updated to better align with observed geophysical or geological data. However, it is hard to find any process model realisations that fit all observations if data sets are complex and sparse in space or time because the simulations typically depend highly non‐linearly on base topography and dynamic parameters. As an alternative, geophysical probabilistic tomographic methods may be used to estimate the family of models of a target subsurface structure that are consistent both with information obtained from previous experiments and with new data (the Bayesian posterior probability distribution). However, this family seldom embodies geologically reasonable images. Here we show that the posterior distribution of tomographic images obtained from travel time data can be fully geological by injecting geological prior information into Bayesian inference and that we can do this near‐instantaneously by using trained mixture density networks (MDNs). We invoke two geological concepts as prior information about the possible depositional environment of an imaged target structure: a braided river system and a set of marine parasequences. Each concept is parameterised by the latent parameters of a generative adversarial network. Data from a target structure can then be used to infer the family of compatible latent parameter values using either geological concept using MDNs. Our near‐instantaneous MDN solutions closely resemble those found using relatively expensive Monte Carlo methods. We show that while the use of incorrect geological conceptual models provides significantly less accurate results, a classifier neural network can infer which geological conceptual model is most consistent with the data. It is thus demonstrated that even apparently barely related geophysical data may contain information about abstract geological concepts, and that geological conceptual models are key to creating reasonable images from geophysical data.

3D pseudo-lithologic modeling via iterative weighted k-means++ algorithm from Tengger Desert cover area, China

The bedrock beneath the Tengger Desert is covered by Quaternary deposits, making it difficult to directly observe the underlying geological information using traditional geological methods. In areas with limited prior geological information, employing geophysical methods to obtain deep-seated information, constructing a multi-source geophysical dataset, and performing three-dimensional modeling can significantly enhance our understanding of the underground geological structures. Cluster analysis is a fundamental unsupervised machine learning technique employed in data mining to investigate the data structure within the feature space. This paper proposes an iterative weighted distance-based extension to the k-means clustering algorithm, referred to as the Iterative Weighted Distance K-means (IW k-means++) algorithm. It incorporates the farthest distance method to select the initial centroid, performs iterative centroid updates based on weighted distance, and dynamically adjusts feature weights during training. The Davies-Bouldin index shows that the performance of IW k-means ++ clustering algorithm is better than the traditional K-Meme ++ clustering algorithm in 3D pseudo-lithology modeling.

An Object‐Oriented Bayesian Gravity Inversion Scheme for Inferring Density Anomalies in Planetary Interiors

AbstractGravity inversions have contributed greatly to our knowledge of the interior of planetary bodies and the processes that shaped them. However, previous global gravity inversion methods neglect the inference of mantle density anomalies when using techniques to decrease the non‐uniqueness of the inversion. In this work, we present a novel global gravity inversion algorithm, named THeBOOGIe, suited to inferring global‐scale density anomalies within the crust and mantle of planetary bodies. The algorithm embraces the nonuniqueness inherent in gravity inversions by not prescribing at the outset a density interface or depth range of interest. Instead, the method combines a Bayesian approach with a flexible incorporation of prior geological or geophysical information to infer density anomalies at any depth. A validation test using synthetic lunar‐like gravity data shows that THeBOOGIe can constrain the lateral location of crustal density anomalies but tends to overestimate their thicknesses. Importantly, THeBOOGIe can detect deep mantle density anomalies and quantify the level of confidence in the inferred density models. Our results show that THeBOOGIe can provide complementary information to one‐dimensional seismic models of the interior of the terrestrial planets and the Moon by constraining density anomalies that are not spherically symmetric. Additionally, THeBOOGIe is specially suited to constraining the interior of partially differentiated bodies where these large‐scale density anomalies are more likely to exist. Finally, thanks to the flexible use of priors, THeBOOGIe is an essential tool to understand the interior of planetary bodies lacking additional constraints.

Near-Surface Structure Investigation Using Ambient Noise in the Water Environment Recorded by Fiber-Optic Distributed Acoustic Sensing

Near-surface structure investigation plays an important role in studying shallow active faults and has various engineering applications. Therefore, we developed a near-surface structure investigation method using ambient noise in a water environment. This newly developed seismic acquisition technology, fiber-optic distributed acoustic sensing (DAS), was used to acquire ambient noise from the Yangtze River. The recorded data were processed to reconstruct surface waves based on the theory of seismic interferometry. The fundamental-mode dispersion curves were extracted and inverted to obtain a shear-wave velocity model below the DAS line. We compared the inverted velocity model with the subsurface geological information from near the study area. The results from the inverted model were consistent with the prior geological information. Therefore, ambient noise in the water environment can be combined with DAS technology to effectively investigate near-surface structures.

Sensing prior constraints in deep neural networks for solving exploration geophysical problems

One of the key objectives in geophysics is to characterize the subsurface through the process of analyzing and interpreting geophysical field data that are typically acquired at the surface. Data-driven deep learning methods have enormous potential for accelerating and simplifying the process but also face many challenges, including poor generalizability, weak interpretability, and physical inconsistency. We present three strategies for imposing domain knowledge constraints on deep neural networks (DNNs) to help address these challenges. The first strategy is to integrate constraints into data by generating synthetic training datasets through geological and geophysical forward modeling and properly encoding prior knowledge as part of the input fed into the DNNs. The second strategy is to design nontrainable custom layers of physical operators and preconditioners in the DNN architecture to modify or shape feature maps calculated within the network to make them consistent with the prior knowledge. The final strategy is to implement prior geological information and geophysical laws as regularization terms in loss functions for training the DNNs. We discuss the implementation of these strategies in detail and demonstrate their effectiveness by applying them to geophysical data processing, imaging, interpretation, and subsurface model building.

Multi-Constrained Seismic Multi-Parameter Full Waveform Inversion Based on Projected Quasi-Newton Algorithm

The multi-parameter full waveform inversion (FWI) that integrates velocity and density can make full use of the kinematic and dynamic information of the measured data to reconstruct the underground model. However, it faces problems of crosstalk between multiple parameters and strong nonlinearity. This research proposes a multi-constrained, multi-parameter FWI framework based on the projected quasi-Newton algorithm. This framework can introduce multiple types of prior geological information, which can effectively improve the problem of multi-parameter inversion. Additionally, the quasi-Newton method can eliminate the crosstalk phenomenon to further improve the inversion convergence speed. Taking the 1994BP model as an example, the results show that the projected quasi-Newton method has a faster convergence speed than the spectral projected gradient method, and reduces the crosstalk between parameters; multiple constraint sets are uniquely projected onto the intersection to ensure that the estimated values of model parameters meet multiple constraints. We also experiment with the overthrust model, which shows that the framework we proposed can improve the inversion accuracy and has good adaptability. The proposed multi-parameter inversion framework can be compatible with more prior information to obtain an inversion model that conforms to geological understanding and shows great potential in seismic exploration.

Characterizing hydraulic heterogeneity of Bayin River Basin using river stage tomography

The drastic spatial and temporal variations of water resources are one of the causes of the fragile ecological environment in the Bayin River Basin. While the temporal variation is less predictable, a detailed spatial distribution of hydrological parameters of the basin is invaluable for developing guidance for allocating and utilizing water resources in the Bayin River Basin. This study estimates the two-dimensional spatial distribution of hydraulic diffusivity (D) of the Bayin River Basin through the relationship between river flow and groundwater using limited available river stages data along the river. It first conducted a wavelet analysis to illustrate strong correlations between river flow and groundwater during the flood season in the basin. Subsequently, it used a simple linear model to estimate temporal and spatial variations of the river stage along the river due to flood. Then, the study applied a recently developed river stage tomography (RST) to the alluvial fan to estimate the spatial distribution of the D values. The RST treats the flood river stage as a signal pressure transmitter, and the groundwater fluctuation recorded in the observation wells as the images of the aquifer heterogeneity between the river and observation wells. The RST then synthesizes these images to characterize the heterogeneous distribution of D in the alluvial fan in the Bayin River Basin. The results show that the northeast area of the alluvial fan has a relatively high D, while the D is relatively low in the tail and the southwest area of the fan. These findings are consistent with the general concept of alluvial fan evolution.Moreover, the predicted groundwater levels during the flood event in 2019 based on the RST-estimated D is more accurate than the preliminary stratigraphic survey, confirming that capturing groundwater responses at various parts of a basin caused by some disturbances could reveal hydraulic connectivity, which the prior geologic information cannot. This result is essential to advancing aquifer characterization technologies and affirming the contribution of RST and the view toward the future subsurface sciences advocated by Yeh et al. (2008).

Shear wave velocity structure at the Fukushima forearc region based on H/V analysis of ambient noise recordings by ocean bottom seismometers

SUMMARYThis study presents the shear wave velocity (VS) structures of sedimentary sequences and a section of the upper crustal layer in the Fukushima forearc region of the Japan Trench subduction zone, which were obtained by analysing the horizontal-to-vertical (H/V) spectral ratios of ambient vibration records. The H/V curves were derived using 31 d of continuous seismic data from 3 broad-band and 16 short-period ocean bottom seismometer (OBS) stations. Using the broad-band data, H/V ratios from 0.01 to 10 Hz were derived, but the ratios below 0.1 Hz frequencies were unusually large and temporally unstable. Characterization of seismic noise energy from ∼1 yr of seismic data of three broad-band OBSs revealed variable and elevated energy conditions below 0.1 Hz due to typical long-period oceanic noise; we link these observations with the unstable H/V ratios below this frequency. Therefore, H/V analysis was performed in the frequency range of 0.1–10 Hz for both broad-band and short-period OBSs to obtain subsurface VS profiles. For the forward calculation of the H/V ratios in the inversion process, we used the recently developed ‘hvgeneralized’ method, which is based on the diffuse field assumption, and accounts for the water layer on top of stratified media. Moreover, available prior geological and geophysical information was utilized during the inversion of the H/V curves. We found that subsurface VS ranged from approximately 30 m s−1 at the seabed to approximately 4900 m s−1 at 7000 m below the sea floor (mbsf). Starting with the best model candidate at each OBS location, the effect of the water layer on the H/V curve in the deep ocean was investigated by comparing synthetic H/V curves with and without the water layer. The synthetic H/V analysis revealed that the water layer had a significant effect on H/V amplitudes at higher frequencies (>1 Hz), whereas comparatively little effect was observed at lower frequencies (<1 Hz). This study provides an empirical basis for H/V analysis using OBS data to determine VS down to several kilometres of sedimentary sequences to the upper crust with high-resolution.

Three-dimensional hydraulic tomography analyses to investigate commingling issues of reproducibility, data density, and geological prior models

Over the past two decades, geostatistics-based hydraulic tomography (HT) has been proven to be a robust method for subsurface heterogeneity characterization. However, the reproducibility of estimated parameter fields, namely the hydraulic conductivity (K) and specific storage (Ss), obtained over different data collection periods and under different flow scenarios have been rarely studied. In this study, we investigate the reproducibility and predictivity of HT estimates using data collected in 2019 and 2020 at a field site of varying boundary conditions underlain by sedimentary units in Japan, with different data density and prior geological information. Model performance of each case is evaluated qualitatively and quantitatively by examining results from model calibration, self-validation and cross-validation, in terms of parameter fields comparison, drawdown matches and cross-correlation distributions. Results of this study reveal that: 1) overall, the parameter fields for 2019 and 2020 HT estimates are generally reproducible, while fine-scale discrepancy exists; 2) they are also found to be reproducible in terms of consistency of estimated geostatistical parameters; 3) higher levels of reproducibility and predictivity associated with estimated K and Ss fields are obtained when higher data density is included for geostatistical inverse modeling with more accurate prior geological information; and 4) the different spatial distributions of cross-correlation between head and parameter fields induced through the significant variation of the groundwater flow field from 2019 to 2020 strengthen the notion of reproducibility of HT results. These results collectively suggest that HT has the potential for mapping changes to K and Ss heterogeneity over time broadening its future applications to various practical problems.

3D Geophysical Predictive Modeling by Spectral Feature Subset Selection in Mineral Exploration

Several technical challenges are related to data collection, inverse modeling, model fusion, and integrated interpretations in the exploration of geophysics. A fundamental problem in integrated geophysical interpretation is the proper geological understanding of multiple inverted physical property images. Tackling this problem requires high-dimensional techniques for extracting geological information from modeled physical property images. In this study, we developed a 3D statistical tool to extract geological features from inverted physical property models based on a synergy between independent component analysis and continuous wavelet transform. An automated interpretation of multiple 3D geophysical images is also presented through a hybrid spectral feature subset selection (SFSS) algorithm based on a generalized supervised neural network algorithm to rebuild limited geological targets from 3D geophysical images. Our self-proposed algorithm is tested on an Au/Ag epithermal system in British Columbia (Canada), where layered volcano-sedimentary sequences, particularly felsic volcanic rocks, are associated with mineralization. Geophysical images of the epithermal system were obtained from 3D cooperative inversion of aeromagnetic, direct current resistivity, and induced polarization data sets. The recovered cooperative susceptibilities allowed locating a magnetite destructive zone associated with porphyritic intrusions and felsic volcanoes (Au host rocks). The practical implementation of the SFSS algorithm in the study area shows that the proposed spectral learning scheme can efficiently learn the lithotypes and Au grade patterns and makes predictions based on 3D physical property inputs. The SFSS also minimizes the number of extracted spectral features and tries to pick the best representative features for each target learning case. This approach allows interpreters to understand the relevant and irrelevant spectral features in addition to the 3D predictive models. Compared to conventional 3D interpolation methods, the 3D lithology and Au grade models recovered with SFSS add predictive value to the geological understanding of the deposit in places without access to prior geological and borehole information.

Quantitative Evaluation of Faults by Combined Channel Wave Seismic Transmission-Reflection Detection Method

The quantitative detection of faults using the channel wave seismic method has been a major but challenging area of interest. In this study, we adopted an effective technical process to evaluate fault attribution. First, we use integrated transmission and reflection channel wave information to improve the accuracy of extraction velocity. Then, the location of the fault is determined by the elliptical tangent offset method, and feature extraction and fault location extension determination are achieved through logistic regression and a neural network. This is combined with the prior geological information, the fractional dimension D to the quantitative analysis of the fault throw. Data regarding the 4203 working face of a mine in Shanxi, China, are considered as an example. Two groups of faults were predicted, with the location error in the f30 fault position as 6.7 m. In addition, the f29 fault throw first increased, and then gradually decreased from the return airway to the haulage gateway. These predicted results have been drill-verified and were used to modify the original design. The proposed method has good stability and promising application prospects for fault evaluation.

Cooperative multinetworks semi-supervised pre-stack seismic inversion

SUMMARY The elastic properties of the subsurface, such as density, P-velocity and S-velocity, can be estimated in pre-stack seismic inversion. In recent research, the deep neural network is widely used in pre-stack seismic inversion for its strong non-linear fitting and feature extraction ability. However, the label data is generally inadequate due to high drilling costs and strong data-sharing barriers in the field of exploration geophysics. In order to reduce the dependence of network performance on label data and ensure the accuracy of inversion mostly, semi-supervised learning is adopted. Here, we develop a cooperative multinetworks semi-supervised pre-stack seismic inversion method. In the cooperative multinetworks inversion framework, the inversion network, mapping network, and modification network are adopted to complete the inversion task cooperatively. A forward network is constructed to automatically generate seismic data from density, P-velocity and S-velocity, which can assist the above networks to complete semi-supervised learning. Compared with some published deep learning pre-stack inversion methods, the spatio-temporal correlation of data can be fully mined, the prior geological structure and low-frequency information can be utilized effectively, and reflectivity is adopted as an intermediate output parameter to improve the robustness of the method. The experiments on the Marmousi2 model demonstrate that cooperative multinetworks semi-supervised inversion strategy is superior to conventional semi-supervised inversion methods in both inversion accuracy and antinoise performance. In addition, the susceptibility experiments of the initial model indicate that the proposed method can maintain a high inversion accuracy with little effective information in the initial model. Finally, the proposed method is successfully applied to the field data and obtains a high-resolution inversion result.

Rayleigh Wave Dispersion Curve Inversion with the Artificial Bee Colony Algorithm

Rayleigh wave dispersion curve inversion is a non-linear iterative optimization process with multi-parameter and multi-extrema. It is difficult to carry out inversion and reconstruction of stratigraphic parameters quickly and accurately with a single linear or non-linear inversion for the data processing of Rayleigh waves with complex seismic geological conditions. We proposed a new method that combines artificial bee colony algorithm (ABC) and damped least squares algorithm (DLS) to invert Rayleigh wave dispersion curve. First, food sources are initialized in a large scale of the model based on the prior geological information. Then, after three kinds of bee operators (employed bees, onlooker bees and scout bees) transform each other and perform search optimization with several iterations, the targets are converged near the optimal solution to obtain an initial S-wave velocity model. Finally, the final S-wave velocity model is obtained by local optimization of DLS inversion with fast convergence and strong stability. The correctness of the method has been verified by one high-velocity interlayer model, and it was further applied to a real Rayleigh wave dataset. The results show that our method not only absorbs the advantages of ABC global search optimization and strong adaptability, but also makes full use of the advantages of DLS inversion, such as high accuracy and fast convergence speed. The inversion strategy can effectively suppress the inversion falling into local extrema, get rid of the dependence on an initial model, enhance the inversion stability, further improve the convergence speed and inversion accuracy, while has good anti-noise ability.