Natural Resources Research最新文献

Predicting flyrock is key to safety and efficiency in open pit mining. In this study, we developed and tested four hybrid models utilizing an adaptive neuro–fuzzy inference system (ANFIS) integrated with metaheuristic optimization techniques: Lévy-enhanced Jaya (ANFIS–LJ), bat algorithm (ANFIS–BA), firefly algorithm (ANFIS–FA) and social spider optimization (ANFIS–SSO). Remarkably, the Lévy technique was applied to enhance the JA algorithm and improve the performance of the ANFIS model for predicting flyrock distance. The models were trained and tested using a dataset from Ta Phoi copper mine with 204 blast events and flyrock distance as the target variable. A drone was used to measure flyrock distance in this study with high resolution to capture the entire flyrock phenomenon of each blast. The k-fold cross-validation technique (with 5 folds) was applied to ensure that AI-based models are not only accurate but also generalize well to new data. It helps in evaluating model performance, tuning hyperparameters, reducing overfitting, and providing a more reliable estimate of how the model will perform in predicting blast-induced flyrock. The models were evaluated using MAE (mean absolute error), RMSE (root mean-squared error) and R². The result showed that ANFIS–LJ outperformed the other models with MAE of 1.423, RMSE of 1.895 and R² of 0.981 on the testing dataset. It was also validated through 13 blasts in practice and achieved a high R² of 0.988, indicating excellent agreement between predicted and observed flyrock distances. Besides, the low MAE (1.322) and RMSE (1.825) values confirmed the model's precision and reliability in predicting flyrock distances. These results confirmed its potential as a valuable tool for optimizing blast designs, enhancing safety, and reducing environmental impacts in real-world engineering applications. This study showed that combining ANFIS with metaheuristic algorithms, especially Lévy-enhanced Jaya algorithm, can produce accurate flyrock prediction. The result can be used to improve the predictive model in open pit mining and decision making. Future study can focus on refining the models and applying them in different mining environments to improve the accuracy.

The overmature Lower Cambrian shale in southern China typically exhibits underdeveloped organic matter (OM) pores, with low porosity, and it is commonly believed that both of them have a causal linkage. However, there remains a lack of in-depth study on the characteristics of OM pores and their controlling factors for this shale. In this study, a suite of Lower Cambrian shale samples was taken from a well in the Upper Yangtze Platform, and their isolated kerogen was used to represent their OM. The shale samples were subjected to the analysis of TOC (total organic carbon) contents, mineral composition and maturity, and the kerogen samples were measured by low pressure gas absorption and X-ray photoelectron spectroscopy to characterize the OM nanopore structure and heterogeneity, and the degree of graphitization, respectively. These data were jointly used to investigate the influencing factors of OM nanopores. The results show that the shale samples were rich in quartz and clay minerals, mainly belonging to siliceous shale in lithofacies, their OM was overmature, with average equivalent vitrinite reflectance (EqVRo) values of 3.48–3.49% and graphitization degrees of 12.77–18.56%. The development of their OM nanopores varied widely, with total pore volumes of 0.321–0.786 cm³/g and total specific surface areas of 142.27–206.02 m²/g (the data were normalized by the elemental carbon content of kerogen samples). The variable graphitization degree and pore structure parameters of OM in the shale samples are primarily attributable to the differential compaction caused by their differences in TOC content and mineral composition. The shale samples with higher TOC contents tended to have lower ratios of quartz to TOC, with the disadvantage to the formation of an effective rigid framework, which increases the compaction of OM particles in shale, and enhances their graphitization degree as well as the collapse of their larger nanopores (such as mesopores and macropores) to form smaller nanopores (typically micropores). However, these processes are weakened to some extent by the pressure-shielding effect of OM-clay aggregations. In contrast, as the graphitization degree increases, the orderly arrangement of carbon atoms is enhanced, leading to the OM particles are easier to be deformed. Combined with the influence of compaction, the graphitization can promotes the transformation of OM mesopores and macropores into micropores, which also complicates the pore structure to enhance the heterogeneity. Therefore, the OM nanopore characteristics and heterogeneity of the studied overmature shale samples were directly affected by their compositions, and the primary mechanism was the synergistic effect of compaction and graphitization.

The discovery of new lithium resources is essential because lithium plays a vital role in the manufacturing of green technology. Along with brines and volcano–sedimentary deposits, approximately a one-third share of global lithium resources is associated with lithium-cesium-tantalum (LCT) pegmatites, with Canada hosting numerous examples. This research applied generative adversarial networks, natural language processing, and convolutional neural networks to generate mineral prospectivity models and support exploration targeting for Canadian LCT pegmatites. Geoscientific text data included within public bedrock geology maps and natural language processing were used to convert conceptual targeting criteria into evidence layers that complement more traditional, geophysical and geochronological data used for mineral prospectivity modeling (MPM). A multilayer architecture of convolutional neural networks, including an attention mechanism, was designed for data modeling. This architecture was trained and validated using variable synthetically generated class labels, input image sizes, and hyperparameters, resulting in an ensemble of 1000 models. The uncertainty of the ensemble was analyzed using a risk–return analysis, yielding a bivariate choropleth risk–return plot that facilitates the interpretation of prospectivity models for downstream applications. This was further complemented by employing post hoc interpretability algorithms to translate the black-box nature of neural networks into comprehensible content. The low-risk and high return class of our prospectivity models reduces the search space for discovering LCT pegmatites by 88%, delineating 99% of known LCT pegmatites in Canada. The results of this study suggest that our workflow (i.e., combining synthetic data generation, natural language processing, convolutional neural networks, and uncertainty propagation for MPM) facilitates decision-making for regional-scale lithium exploration and could also be applied to other mineral systems.

The goaf residual coal is more susceptible to oxidation after long-term soaking that raises the spontaneous combustion risk level. This paper investigates the oxidation thermodynamic mechanism and the active functional group proliferation trend in soaked coal. A coal macromolecular model C₂₁₇H₁₇₁O₄₄N₃S₂ is constructed to evaluate the oxygen adsorption effectiveness via molecular dynamics simulation. The results show that the soaking behavior reduces the coal intramolecular hydrogen bonds and expands the coal pore volume. The soaked coal creates 1.18 times average oxygen adsorption loading higher than the raw coal. The soaking decreases the crossing point temperature from 147.40 °C to 144.15 °C and enlarges the CO production rate by 1.087 times, increasing the potential hazard of coal oxidation. The reactive functional groups -CHO, –CH₂, –CH₃, and –OH exhibit an evident increase of 0.73, 0.01, 0.42, and 0.72 after water soaking. The –CH₂/–CH₃ drops from 1.95 to 1.13, indicating that aliphatic chain consists of shorter and more branched chains. The increase in oxygen adsorption and reactive functional group of soaked coal promotes the coal oxidation chain reaction, which boosts oxidation temperature rise and gas release.

Machine learning is becoming a popular and appealing tool in mineral prospectivity mapping (MPM); however, it has always been challenged by some essential limitations, such as scarcity of training samples, overfitting, and uncertainties. Data augmentation has been proven to be effective in addressing these issues and improving the performance of artificial intelligence models, but its mechanism regarding how augmented data influences predictive modeling processes, improves model performance, and alleviates overfitting has yet to be elucidated due to the black-box nature of machine learning modeling. In this study, the synthetic minority oversampling technique (SMOTE), proven to perform best among five commonly used data augmentation methods, was selected and utilized to enhance the training data and improve model performance. The results indicate that the convolutional neural network (CNN) model trained by rational-feature ordering and SMOTE-augmented data achieved better performance, with higher test accuracy (0.9306), recall (0.9167), F1-score (0.9296), and alleviated overfitting (0.0215), compared with the model trained on original data. A set of black-box visualization tools, including filter weight visualization, individual conditional expectation (ICE) plots, derivative ICE (d-ICE) plots, partial dependence plots (PDPs), and Shapley additive explanations (SHAP), were employed to explore the beneficial mechanism of SMOTE when applied to enhance the predictive capabilities of CNN in MPM. The visualization of the weight filters reveals that the optimal model activates favorable excitations of W anomalies, Mn anomalies and proximity to Yanshanian intrusions, which are associated with tungsten mineralization, thus optimizing feature extraction, refining convolutional operation, and improving model performance. The ICE and d-ICE analyses reveal that the SMOTE-augmented model exhibites a more consistent decision trend in key ore-associated features and reduces variability in derivative estimates, particularly beyond decision thresholds, leading to stabler predictions. The PDP results show that SMOTE-augmented data increase the decision boundary difference between positive and negative samples, suggesting a broader decision width that favored more accurate classification. The SHAP analyses indicate that the SMOTE-augmented data boost the recognition ability of the CNN model by clearly separating feature values of key ore-associated factors with contrasting SHAP values and help the model make more convergent decision paths, especially for samples with top probabilities. Our findings provide a straightforward view for explaining how a superior algorithm can benefit model predictions through black-box modeling processes, and contribute to understanding the decision-making mechanism of machine learning in MPM.

The Sousse governorate (Tunisian Sahel) is an important socio-economic pole with a strong need for water. Intense drought during the last years has harmed the governorate’s activities. Therefore, groundwater exploitation is increasingly becoming necessary for ensuring sustainable development. It takes in-depth knowledge of aquifers to create an appropriate exploitation plan. In this context, the present study aimed for precise delineation of Mio–Plio–Quaternary aquifers in the central part of the Sousse governorate by thoroughly reconstructing their geometry and understanding their functioning. To reach this goal, a rich database, including 142 water boreholes and 123 vertical electrical soundings (VES), was used. Water borehole data containing lithological columns and their corresponding well logs as well as hydrodynamic and hydrochemical measurements were exploited. Second, VES data were interpreted and geoelectrical cross sections are accordingly established. Two aquifer formations showing variable arrangement were differentiated from well log analysis and correlation: AI and AII, which are Quaternary and Mio–Pliocene in age, respectively. Aquifer AI is absent at the Kalaa Kebira anticline, while on either side of this structure, both formations are present with deepening of aquifer AII. The established piezometric map exhibits groundwater flow toward the north and south of the Kalaa Kebira anticline. In the same directions, water salinity values increase gradually from 1 to 5 g/l. The geoelectrical cross sections highlighted that tectonic deformations control the water reservoirs arrangement and the groundwater circulation. Cross-comparison of the deduced information regarding the aquifers geometry, hydrodynamics, and water quality brings new elements to the hydrogeological scheme in the central part of the Sousse governorate. The Mio–Plio–Quaternary deposits encompass two multilayered aquifers, which are the Balaoum–Sidi Bou Ali aquifer to the north and the Oued Laya aquifer to the south. These aquifers are juxtaposed with a groundwater divide at the Kalaa Kebira anticline. The present study will guide groundwater exploitation in the Sousse governorate and thereby support sustainable development in the Tunisian Sahel. More broadly, it constitutes a model of hydrogeophysical application for better groundwater management in other arid regions.

Determination of reservoir production schemes has always been a difficult problem during the close-loop management of waterflooding reservoir. Different well control results in significant influence on production, water breakthrough time and recovery rate of producing wells, especially in heterogeneous reservoirs. To optimize well controls, a new method using transpose convolution neural network (TCNN) surrogate model and adaptive differential evolution with optional external archive (JADE) algorithm was introduced. In this method, the TCNN surrogate model, which uses image processing, took well controls (i.e., bottom hole pressure and injection rate) and production time as parameters to predict oil saturation and pressure distribution fields at different time periods. It could well replace a numerical simulator, accurately predict the regional production dynamics at different production time steps, and significantly reduce the simulation time during the optimization process. Meanwhile, the JADE algorithm, as an improved differential evolution algorithm, greatly improved the convergence rate while ensuring the search breadth and it was suitable for solving multi-parameter well control optimization problems. Using a comprehensive reservoir optimization problem as an example, the selection and setting of some parameters during the TCNN training and JADE optimization are discussed. Finally, the method was applied to a real 3D reservoir. The computational speed of the TCNN model was about 3600 times and 2300 times faster than that of a numerical simulation model for the synthetic reservoir and L43 block, respectively.

In petroleum and natural gas exploration, lithology identification—analyzing rock types beneath the Earth’s surface—is crucial for assessing hydrocarbon reservoirs and optimizing drilling strategies. Traditionally, this process relies on logging data such as gamma rays and resistivity, which often require manual interpretation, making it labor-intensive and prone to errors. To address these challenges, we propose a novel machine learning framework—contrastive learning-transformer—leveraging self-attention mechanisms to enhance the accuracy of lithology identification. Our method first extracts unlabeled samples from logging data while obtaining labeled core sample data. Through self-supervised contrastive learning and a transformer backbone network, we optimize performance using techniques like batch normalization. After pretraining, the model is fine-tuned with a limited number of labeled samples to improve accuracy and significantly reduce reliance on large labeled datasets, thereby lowering the costs associated with drilling core annotations. Additionally, our research incorporates shapley additive explanations (SHAP) technology to enhance the transparency of the model’s decision-making process, facilitating the analysis of the contribution of each feature to lithology predictions. The model also learns time-reversal invariance by reversing sequential data, ensuring reliable identification even with variations in data sequences. Experimental results demonstrate that our transformer model, combined with semi-supervised contrastive learning, significantly outperforms traditional methods, achieving more precise lithology identification, especially in complex geological environments.

In the realm of mineral prospectivity mapping, a novel hybrid approach for optimizing hyperparameters of the support vector machine (SVM) algorithm is proposed here. The concept of ant colony optimization (ACO) algorithm, inspired by collective intelligence of ant colonies, and grid search (GS) that systematically evaluate all hyperparameter combinations to find the optimal model configuration are leveraged to fine-tune SVM parameters, enhancing its predictive capabilities. A dataset comprising geophysical, geochemical, geological, tectonic, and remote sensing evidence layers from the Sardouyeh region in Kerman province, Iran, is utilized for model development aimed the prediction of areas favorable for porphyry-Cu mineralization. After generating the regular and tuned predictive models, a comparison was carried out using quantitative performance metrics such as confusion matrix and success rate curves. The results demonstrated that the optimized versions of SVM using ACO (ACO–SVM) and GS (GS–SVM) models exhibit superior performance, achieving better accuracy and predictive capability in identifying locations favorable for porphyry-Cu mineralization. The study highlights the potential of incorporating optimization algorithms, especially ACO, into SVM, leading to the development of more effective predictive models for mineral prospectivity mapping.