Artificial Intelligence in Geosciences最新文献

Compositional reservoir simulation is an important tool to model fluid flow in oil and gas reservoirs. Important investment decisions regarding oil recovery methods are based on simulation results, where hundred or even thousand of different runs are performed. In this work, a new methodology using artificial intelligence to learn the thermodynamic equilibrium is proposed. This algorithm is used to replace the classical equilibrium workflow in reservoir simulation. The new method avoids the stability test for single-phase cells in most cases and provides an accurate two-phase flash initial estimate. The classical and the new workflow are compared for a gas-oil mixing case, showing a simulation time speed-up of approximately 50%. The new method can be used in compositional reservoir simulations.

A direct hydrocarbon detection is performed by using multi-attributes based quantum neural networks with gas fields. The proposed multi-attributes based quantum neural networks for hydrocarbon detection use data clustering and local wave decomposition based seismic attenuation characteristics, relative wave impedance features of prestack seismic data as the selected multiple attributes for one tight sandstone gas reservoir and further employ principal component analysis combined with quantum neural networks for giving the distinguishing results of the weak responses of the gas reservoir, which is hard to detect by using the conventional technologies. For the seismic data from a tight sandstone gas reservoir in the Sichuan basin, China, we found that multi-attributes based quantum neural networks can effectively capture the weak seismic responses features associated with gas saturation in the gas reservoir. This study is hoped to be useful as an aid for hydrocarbon detections for the gas reservoir with the characteristics of the weak seismic responses by the complement of the multi-attributes based quantum neural networks.

The Sulige tight gas field is presently the largest gas field in China. Owing to the ultralow permeability and strong heterogeneity of the reservoirs in Sulige, the number of production wells has exceeded 3,000, keeping the stable gas supply in the decade. Thus, the daily production prediction of gas wells is significant for monitoring production and for implementing and evaluating stimulation measures. Therefore, on the basis of the three data-driven time series approaches, the daily production of 1692 wells over 10 years was mining for the daily production prediction of wells in Sulige. The jointed deep long short-term memory and fully connected neural network (DLSTM-FNN) model was proposed by introducing the recurrent neural network's sequential expression ability and was compared with random forest (RF) and support vector regression (SVR). After the daily production predictions of thousands of wells in Sulige, the proposed DLSTM-FNN model significantly improved the time series prediction accuracy and efficiency in the short training samples and had strong availability and practicability in the Sulige tight gas field.

The spatial prediction of a continuous response variable when spatially exhaustive predictor variables are available within the region under study has become ubiquitous in many geoscience fields. The response variable is often subject to detection limits due to limitations of the measuring instrument or the sampling protocol used. Consequently, the response variable's observations are censored (left-censored, right-censored, or interval-censored). Machine learning methods dedicated to the spatial prediction of uncensored response variables can not explicitly account for the response variable's censored observations. In such cases, they are routinely applied through ad hoc approaches such as ignoring the response variable's censored observations or replacing them with arbitrary values. Therefore, the response variable's spatial prediction may be inaccurate and sensitive to the assumptions and approximations involved in those arbitrary choices. This paper introduces a random forest-based machine learning method for spatially predicting a censored response variable, in which the response variable's censored observations are explicitly taken into account. The basic idea consists of building an ensemble of regression tree predictors by training the classical regression random forest on the subset of data containing only the response variable's uncensored observations. Then, the principal component analysis applied to this ensemble allows translating the response variable's observations (uncensored and censored) into a linear equalities and inequalities system. This system of linear equalities and inequalities is solved through randomized quadratic programming, which allows obtaining an ensemble of reconstructed regression tree predictors that exactly honor the response variable's observations (uncensored and censored). The response variable's spatial prediction is then obtained by averaging this latter ensemble. The effectiveness of the proposed machine learning method is illustrated on simulated data for which ground truth is available and showcased on real-world data, including geochemical data. The results suggest that the proposed machine learning technique allows greater utilization of the response variable's censored observations than ad hoc methods.

Residual magnetic error remains after standard levelling process. The weak non-geological effect, manifesting itself as streaky noise along flight lines, creates a challenge for airborne geophysical data processing and interpretation. Microleveling is the process to eliminate this residual noise and is now a standard areogeophysical data processing step. In this paper, we propose a two-step procedure for single aerogeophysical data microleveling: a deep convolutional network is first adopted as approximator to map the original data into a low-level part with nature geological structures and a corrugated residual which still contains high-level detail geological structures; second, the mixture of Gaussian robust principal component analysis (MoG-RPCA) is then used to separate the weak energy fine structures from the residual. The final microleveling result is the addition of low-level structures from deep convolutional network and fine structures from MoG-RPCA. The deep convolutional network does not need dataset for training and the handcrafted network serves as prior (deep image prior) to capture the low-level nature geological structures in the areogeophysical data. Experiments on synthetic data and field data demonstrate that the combination of deep convolutional network and MoG-RPCA is an effective framework for single areogeophysical data microleveling.

Most of the existing machine learning studies in logs interpretation do not consider the data distribution discrepancy issue, so the trained model cannot well generalize to the unseen data without calibrating the logs. In this paper, we formulated the geophysical logs calibration problem and give its statistical explanation, and then exhibited an interpretable machine learning method, i.e., Unilateral Alignment, which could align the logs from one well to another without losing the physical meanings. The involved UA method is an unsupervised feature domain adaptation method, so it does not rely on any labels from cores. The experiments in 3 wells and 6 tasks showed the effectiveness and interpretability from multiple views.

Machine learning methods are increasingly used for spatially predicting a categorical target variable when spatially exhaustive predictor variables are available within the study region. Even though these methods exhibit competitive spatial prediction performance, they do not exactly honor the categorical target variable's observed values at sampling locations by construction. On the other side, competitor geostatistical methods perfectly match the categorical target variable's observed values at sampling locations by essence. In many geoscience applications, it is often desirable to perfectly match the observed values of the categorical target variable at sampling locations, especially when the categorical target variable's measurements can be reasonably considered error-free. This paper addresses the problem of exact conditioning of machine learning methods for the spatial prediction of categorical variables. It introduces a classification random forest-based approach in which the categorical target variable is exactly conditioned to the data, thus having the exact conditioning property like competitor geostatistical methods. The proposed method extends a previous work dedicated to continuous target variables by using an implicit representation of the categorical target variable. The basic idea consists of transforming the ensemble of classification tree predictors' (categorical) resulting from the traditional classification random forest into an ensemble of signed distances (continuous) associated with each category of the categorical target variable. Then, an orthogonal representation of the ensemble of signed distances is created through the principal component analysis, thus allowing to reformulate the exact conditioning problem as a system of linear inequalities on principal component scores. Then, the sampling of new principal component scores ensuring the data's exact conditioning is performed via randomized quadratic programming. The resulting conditional signed distances are turned out into an ensemble of categorical outputs, which perfectly honor the categorical target variable's observed values at sampling locations. Then, the majority vote is used to aggregate the ensemble of categorical outputs. The effectiveness of the proposed method is illustrated on a simulated dataset for which ground-truth is available and showcased on a real-world dataset, including geochemical data. A comparison with geostatistical and traditional machine learning methods show that the proposed technique can perfectly match the categorical target variable's observed values at sampling locations while maintaining competitive out-of-sample predictive performance.

In this study, we present a machine learning-based method to predict trace element concentrations from major and minor element concentration data using a legacy lithogeochemical database of magmatic rocks from the Karoo large igneous province (Gondwana Supercontinent). Wedemonstrate that a variety of trace elements, including most of the lanthanides, chalcophile, lithophile, and siderophile elements, can be predicted with excellent accuracy. This finding reveals that there are reliable, high-dimensional elemental associations that can be used to predict trace elements in a range of plutonic and volcanic rocks. Since the major and minor elements are used as predictors, prediction performance can be used as a direct proxy for geochemical anomalies. As such, our proposed method is suitable for prospective exploration by identifying anomalous trace element concentrations. Compared to multivariate compositional data analysis methods, the new method does not rely on assumptions of stoichiometric combinations of elements in the data to discover geochemical anomalies. Because we do not use multivariate compositional data analysis techniques (e.g. principal component analysis and combined use of major, minor and trace elements data), we also show that log-ratio transforms do not increase the performance of the proposed approach and are unnecessary for algorithms that are not spatially aware in the feature space. Therefore, we demonstrate that high-dimensional elemental associations can be modelled in an automated manner through a data-driven approach and without assumptions of stoichiometry within the data. The approach proposed in this study can be used as a replacement method to the multivariate compositional data analysis technique that is used for prospectivity mapping, or be used as a pre-processor to reduce the detection of false geochemical anomalies, particularly where the data is of variable quality.

Mineral exploration campaigns are financially risky. Several state-of-the-art methods have been developed to mitigate the risk, including predictive modelling of mineral prospectivity using principal component analysis (PCA) and geographic information systems (GIS). The PCA and GIS approach is currently considered acceptable for generating mineral exploration targets. However, some of its limitations are the dependence on sample stoichiometry (e.g., the existence of minerals), the necessity of log-ratio transformations when dealing with compositional data, and manual interpretation and use of principal components to enhance potential geochemical anomalies for prospectivity mapping. In this study, we generalize the fundamental ideas behind the PCA and GIS approach by developing a new data-driven approach using ML. We showcase a new workflow capable of generating either intermediate evidence layers or final prospectivity maps that depict major regional geochemical anomalies using multi-element geochemical data from Southeastern Churchill Province (Québec and Labrador), Canada. The region is known for its REEs endowment and the data were gathered for prospectivity mapping. A comparison with the established multivariate hybrid data- and knowledge-based approach revealed that on a roughly comparable basis of the amount of manual effort, our new data-driven procedure can much more accurately identify geochemical anomalies in both univariate and multivariate applications. The results of our prospectivity mapping corroborate with the ground truth or known geological anomalies in the studied region. These findings have potentially wider implications on exploration target generation, where project risks (financial, environmental, political, etc.) and geochemical anomalies must be quantified using robust and effective data-driven approaches. In addition, our methodology is more replicable and objective, as manual geoscientific interpretation is not required during the detection of geochemical anomalies.

Artificial Intelligence, or AI, is a method of data analysis that learns from data, identify patterns and makes predictions with the minimal human intervention. AI is bringing many benefits to petrophysical evaluation. Using case studies, this paper describes several successful applications. The future of AI has even more potential. However, if used carelessly there are potentially grave consequences.

A complex Middle East Carbonate field needed a bespoke shaly water saturation equation. AI was used to ‘evolve’ an ideal equation, together with field specific saturation and cementation exponents. One UKCS gas field had an ‘oil problem’. Here, AI was used to unlock the hidden fluid information in the NMR T1 and T2 spectra and successfully differentiate oil and gas zones in real time. A North Sea field with 30 wells had shear velocity data (Vs) in only 4 wells. Vs was required for reservoir modelling and well bore stability prediction. AI was used to predict Vs in all 30 wells. Incorporating high vertical resolution data, the Vs predictions were even better than the recorded logs.

As it is not economic to take core data on every well, AI is used to discover the relationships between logs, core, litho-facies and permeability in multi-dimensional data space. As a consequence, all wells in a field were populated with these data to build a robust reservoir model. In addition, the AI predicted data upscaled correctly unlike many conventional techniques. AI gives impressive results when automatically log quality controlling (LQC) and repairing electrical logs for bad hole and sections of missing data.

AI doesn’t require prior knowledge of the petrophysical response equations and is self-calibrating. There are no parameters to pick or cross-plots to make. There is very little user intervention and AI avoids the problem of ‘garbage in, garbage out’ (GIGO), by ignoring noise and outliers. AI programs work with an unlimited number of electrical logs, core and gas chromatography data; and don’t ‘fall-over’ if some of those inputs are missing.

AI programs currently being developed include ones where their machine code evolves using similar rules used by life’s DNA code. These AI programs pose considerable dangers far beyond the oil industry as described in this paper. A ‘risk assessment’ is essential on all AI programs so that all hazards and risk factors, that could cause harm, are identified and mitigated.