Machine-learning, a sector of artificial intelligence (AI), uses methods like clustering, classification and regression for analysing data, find invisible patterns and build predictions from complex and huge sets of data [1]. ML methods are widely divided into unsupervised and supervised techniques, supervised learning utilizes designated data (input and output pairs) to exercise models, in the same time unsupervised learning recognizes invisible patterns in data without predetermined outputs. In petroleum industry, ML became influential in labelling Geo-scientific problems through exploration, development and production steps. Wire-line logging is a key of data source regularly used by Geo-scientists [2]. 

With ML progress, neural networks, particularly convolutional neural networks (CNNs), have been widely adopted for lithofacies prediction using image and log data. Several studies demonstrate diverse ML applications, Chawshin et al. [3] developed a CNN to predict lithofacies from 2D core CT scans image slices, attaining high accuracy in lithofacies prediction, Alzubaidi et al. [4] proposed a CNN-based method for lithology prediction, though with limited accuracy in differentiating rock types. Al-Mudhafar et al. [5] applied boosting algorithms for lithofacies classification in the Majnoon oil field, Iraq, achieving high accuracy validated against core and poro-perm data. Moghanloo et al. [6] used pre-stack seismic inversion to derive key reservoir properties, improving facies identification in Iranian fields. In 2023, Moghanloo and Riahi [7] integrated high-resolution SEM images, porosity logs and Bayesian classifiers to assess the Burgan Formation in SW Iran, enhancing drilling efficiency, while Tian et al. [8] used multi-resolution graph-based clustering (MRGC) for unsupervised facies detection. Chai et al. proposed an automated method for classifying reef-shoal sedimentary facies. Further innovations include, using well logs instead of core samples to reduce data labelling efforts, utilizing multiple log types (e.g., GR, DT, NPHI, RHOB, SP, LLD) to predict lithofacies [9].       
 

Applying Support Vector Machines (SVMs) and Gradient Boosting Decision Trees (GBDT) for lithology classification, implementing PU-learning for partially labelled datasets, improving classification from traditional logs [10]. However, most methods depend on manual interpretations, which introduces subjectivity and increases workload. Thus, automatic facies identification is gaining attention.

This study explores how supervised ML classification can enhance lithofacies recognition within a complex geological data set. These sets of data show an advancing sequence established throughout a great tectonic event that obviously affected facies dispensation and presented volcaniclastic remains, eventually lowering reservoir quality. precise facies prediction is therefore important for perfecting productivity in formations like the Lower Goru of the Kadanwari gas field in southern Pakistan, that is well known for its plentiful reserves of gas. In this research, new logging tools and synthetic gamma ray logs were founded using retrogression approaches and artificial data were created to fill absent values. The sets of data were then trained using four different classification processes, with their functioning estimated to recognize the most effective facies classifier. For more enhancing the accuracy of prediction, an ensemble learning method was implemented, joining together several approaches to accomplish more dependable outcomes [11].

Machine Learning

Machine Learning in Lithofacies Classification: An Integrated Machine learning method, a sector of artificial intelligence (AI), became an essential implement in Geo-sciences to analyse huge and complicated sets of data [12]. Unlike conventional approaches that rely on predefined orders, ML algorithms learn straight from data to make predictions and reveal underlying layouts, enhancing their accuracy with time. This made ML exceptionally valuable in the petroleum field, where large quantities of data from seismic surveys, well logs and core samples demand effectual and dependable explanation [13]. ML approaches mostly divided into three main sections: supervised, unsupervised and reinforcement learning. Supervised learning utilizes labelled data to train predictive approaches; in the same time unsupervised learning represents unseen patterns within unlabelled data. Reinforcement learning, nevertheless less common in Geo-science implementations, focuses on clarifying decision making by trial and error. Because labelled data are generally limited in reservoir researches, unsupervised learning provides a definite advantage, it can independently bring out important patterns with no need for human explanation [14]. In late years, ML has been widely applied to lithofacies categorization, a decisive step in reservoir characterization. Lithofacies show units of rock that share similar sedimentation and petro-physical characteristics, making their recognition important for estimating reservoir quality. Many studies have revealed the success of ML methods in this field. For instant, convolutional neural networks (CNNs) have been used to arrange lithofacies from core photograph and CT scans [4], meanwhile ensemble methods like Support Vector Machines (SVM) and Gradient Boosting Decision Trees (GBDT) have been applied to well log data. Wire-line logging appreciates to their high vertical consistency and resolution serve as a primary input for many ML algorithms. regular parameters such as sonic (DT), neutron porosity (NPHI), gamma ray (GR), resistivity (LLD) and density (RHOB) aid to distinguish lithofacies based on their typical logging reactions [9]. several researches have presented that machine learning models practiced on these logging datasets can precisely recognize facies borders and rock types, even with the lack of core data. One of the distinguished illustrations come from the Zamzama gas field discovered in the southern territory of Pakistan, where investigators implemented unsupervised learning to designate lithofacies by using well logs statistics. researchers performed a Self-Organizing Map (SOM) method an algorithm well accommodated for cases that have insufficient expert input or geological labels [15]. The SOM algorithm advantageously classified data in to facies assemblages and keeping topological compatibilities within the properties. Before implementing SOM, Principal Component Analysis (PCA) was used to bring down the magnitude of seismic characteristics, keeping only the most important information and clarifying the clustering operation [16]. in addition, multi-variate cluster analysis was implemented to a set of well log data into log-facies or electro-facies sets of sedimentary components with indistinguishable logging signatures [17]. This data driven model did not require any pre-fragmentation, making it highly scalable and objective for huge sets of data [18]. The out-coming approach was authenticated with actual lithofacies data, verifying its validity through both graded and ungraded gathering procedures.

Beyond classification, more researchers have employed Machine Learning to influence insufficient sets of data. Synthetic logs of gamma ray, for example, were created using regression approaches, meanwhile artificial data were generated to fill up absent values enlarging the training datasets and enhancing model dependability. Ensemble learning approaches were also acquired to join the firmness of numerous algorithms, improving the whole prediction reliability. Categorizing rock classifications using deriving log characteristics remarkably lowering explanation time and reduces human prejudice, which is specifically valuable in geologically complicated settings. This is exceptionally important where volcanic fragments deposition and tectonic action have altered reservoir performance. In short, combining innovative ML algorithms basically unsupervised algorithms like assemble analysis and SOM contributed a pivotal, systematic structure for lithofacies categorization and rock type identification. These algorithms not just reducing manual investigation but also transport adjustable, dependable results proper to real field procedures. The extending change toward categorization and data driven calculation continues to adapt the future of underground description in petroleum production [19]. Image Based Lithofacies and Core Data categorization in underground geology, the core data that received from drilling supply the most direct and dependable information for anatomize stratigraphic compositions and facies models. Such details are essential for understanding reservoir framework and development decisions and guiding investigation. with explanation core images, geologists can better recognize sedimentary structures and enhance reservoir prediction [20]. A primary case comes from the Mackay River oil sands in Canada, where thousands of core pictures have been gathered, presenting deep insights into underground formations. conventionally, facies recognition from core pictures depends widely on manual explanation a labour exhaustive process subject to human prejudice, particularly in unclear sectors. This has generated a growing need for mechanistic, objective alternatives and effective. With development in artificial intelligence, computing device can now operate complicated and tedious optical tasks once reserved for human specialists. In Geo-science, image identification has been satisfactorily applied to core lithofacies categorization [21]. One of the early models utilized a Support Vector Machine (SVM) to anatomize texture and colour features in core design, acquiring an accuracy of up to 86% [22]. alternatively, this model was best appropriated for binary categorization and struggled with huge, multi class sets of data. To get better of these restrictions, Song et al. [23] suggested a model using Generative Adversarial Networks (GANs). This method exercised antagonistic training on core slices to create synthetic pictures, which extended the set of data. The trained differentiator’s specifications were then transported to a lithology recognition approach. This combination “WGAN + parameter migration” architecture accomplished an incredible 94.93% identification accuracy. Otherwise, the dependence on synthetic rather than actual core pictures established potential restrictions in real world implementations. Traditional CNNs used for core image anatomization and it also faces some challenges in collecting image characteristics, usually demanding extra deep networks that struggle from gradient linked issues, like vanishing or exploding gradients. Shallow CNNs, otherwise, the absence of ability to extract several features effectively [24]. for more detailed provocations, a deep learning algorithm depended on the ResNet50 framework applied to a wide set of data of core images from the Mackay River field. ResNet50’s use of shortcut link among non-adjacent layers reducing gradient issues, allowing the algorithm to learn complicated features with additional efficiency [25]. When tested on field data, the ResNet50-based algorithm exercised better than regular CNNs, incorporating excellent accuracy and powerful potential for feasible facies recognition in reservoir researches. Data Anatomization and Method Visualization The prevalence of any machine learning workflow relies on thorough data anatomization and clear arithmetical representation of the specimens, that involves conceptualizing the relation between input and output variables. Figure 1 demonstrates the strategy for creating gamma ray (GR) logs from anticipated rock facies and constructing an inclusive facies approach [11].

Figure 1: Flowchart exhibiting the strategy used to create artificial gamma ray (GR) logs from anticipated rock facies and to construct an inclusive facies approach

Lithofacies Identification

The recognition of Lithofacies plays an elementary role in depositional correlation and sedimentary facies investigation. As a key component of sedimentary facies, lithofacies not only indicate differences in rock configuration and appearance but also reveal inequalities in fluid content inside indistinguishable lithologies. Accurate projection of lithofacies is crucial to define the boundaries of reservoir and estimating reservoir characteristics, which in turn braces effective oil and gas exploration, development of the field and long-term production arrangement [26].

Seismic data supplies precious information linked to fluid and lithology distribution, improving both predictive accuracy and the lateral resolution among the wells. nevertheless, the relation between lithofacies and seismic responses is intrinsically complicated and nonlinear, effected by many interacting physical and geological aspects [27]. conventional models depend on Geo-statistical methods and rock physical models to delineate and predict reservoir characteristics [28], but these models could be restricted in their capability to hold huge, heterogeneous sets of data. In late years, ML has emerged as a strong system in Geo-sciences, presenting modern capabilities for holding and explaining multi-dimensional and complicated data [29]. Among numerous machine learning approaches, artificial neural networks (ANNs) have been mostly implemented in geophysical researches because of their powerful capability to capture nonlinear relationships. Nonetheless, their efficiency relies deeply on network framework, parameter initialization and rate of learning. These vulnerabilities frequently cause issues like entrapment or slow convergence in limited least, causing less dependable predictions [30]. in contrast Support Vector Machines (SVM) proved a highly effectual in small sample, high geometrical and nonlinear categorization tasks [31]. By increasing the gap between separating hyperplane and the support vectors, SVM accomplishes well-built conception with no to assume any specific data dispensation [32]. In lithofacies categorization, the SVM procedure typically implies two phases: tutoring the model with well-known lithofacies and their correlated features, then predicting unrevealed lithofacies using the tutored decision function [33]. numerous researches have acknowledged the potential of SVM for reservoir depiction and lithofacies categorization using seismic data [34]. To illustrate, Zhao et al. [35] indicated that the Proximal SVM (PSVM) presents both effectiveness and exactness for binary lithofacies categorization in the Barnett Shale. In spite of their firmness, traditional SVMs were primarily progressed for binary classification issues, in the same time lithofacies recognition frequently needs handling with multi class, high dimensional and nonlinear data. To deal this intricacy, kernel functions are presented to calculate input data into higher dimensional attribute spaces where linear detachment becomes practicable [36]. In contrast, depending on a single core function could not represent the full variety of lithofacies features, which can minimize classification precision. To control this restriction, the Multi-Kernel Learning (MKL) model integrates various kernel functions, each emphasizing different details of the data [37]. MKL improves both the rigidity and accuracy of SVM-based lithofacies stratification. still, this development comes at a high price, the synchronous use of diverse kernels raising arithmetical need and memory usage [38], that can block its enforcement in big scale reservoir modelling. To remark these arithmetical provocations, GöNen and Alpaydin [39] presented the Local Multi-Kernel Learning (LMKL) approach. LMKL by adjustable way can select kernels based on domestic data distributions, enhancing the effectiveness and lowering spatial intricacy. And yet, this model could lose general intelligibility and kernel compatibility, resulting in parameter prolixity. Jose et al. [40] more developed this concept via the Local Deep Multi-Kernel Learning (LDMKL) framework, which operates a tree-structured rendering to train both universal and local aspects in high dimensional, sparse sets of data. LDMKL jointly improves kernel functions and the SVM decision parameter, significantly optimizing efficiency and accuracy. nevertheless, its primary formulation was restricted to binary stratification and depended on a one universal kernel. constructing these improvements, an adjusted LDMKL-SVM model has been introduced for multi class lithofacies recognition. This developed method integrates both high and low dimensional cores to better differentiate among lithofacies, joining both local variability and universal structural information. It also contributes automatic parameter improvement, which optimizes prediction accuracy and lowering reliance on manual setting. The graded, tree-based kernel layout improves more arithmetical effectiveness and scalability. investigations utilizing both real world and synthetic sets of data confirm that this method transfers high accuracy of lithofacies categorization and dependable reservoir property predictions. The enhanced LDMKL-SVM structure presents an effective and practical solution to the restrictions of conventional SVM-based models. With remarkable advance in marine shale petroleum evolution in North America, unusual resource investigation became a universal focus. The maximizing attention toward formations of shale drove a major technical and theoretical progress in shale oil research [41]. In China, the Shahejie Formation inside the Bozhong Sag is considered as a standard shale reservoir, identified by several hydrocarbon sources, inter-stratified reservoir and source rocks and disseminated “sweet spots” [42]. otherwise, it’s complicated lithology, thin inter-stratified mud-stone sandstone layers and high diversity present extra challenges for precise lithofacies prediction [43].

To tackle these provocations, scientists and researchers used multiple multivariate and statistical analysis approaches like cluster, discriminant, principal and factor component analyses to investigate relations between Geo-chemical indexes and shale lithofacies [44]. Lithofacies delineate the compositional and physical features of rocks, including characteristics like texture, colour and sedimentary structure and are important to understand sedimentary conditions and reservoir performance [45]. In lacustrine settings, shale lithofacies are made by various depositional circumstances. Rock mineral configuration brings precious insights into sedimentation procedures and lithological variance [46], still its utilization is restricted by poor structural continuity and high costs. Well logging data, in comparison, supplies dependable, constant and cost-effective information about underground formations [47]. Hence, more researches have sought to construct relations between lithofacies and logging data using arithmetical and machine learning approaches like cluster analysis, discriminant analysis, principal component analysis and SVM. nevertheless, experimental methods frequently compete with underlying and unevenly scattered data, decreasing categorization precision. Artificial neural networks (ANNs) deliver a strong alternative capability of modelling complicated nonlinear relations by simulating human psychological procedures. In supervised learning, ANNs are learned by utilizing marked data, regulating inner weights via forward and backward propagation tile projection faults are reduced [48,49]. Their adaptability makes them exceptionally acceptable for lithofacies categorization based on mineralogical and logging data. researches have revealed favourable results: Myles et al. [50] perfectly implemented neural networks for lithology categorization from logging data, in the same time Yu-Jiang et al. [51] revealed their advantage in predicting lithology when adequate learning samples are accessible. correspondingly, Li et al. [52] utilized Kohonen networks for effectual lithofacies identification. In spite of these progresses, conventional neural networks are still facing restrictions in convergence and accuracy rate. To remark this, the ASO-BP neural network joining the Atomic Search Optimization (ASO) approach with the conventional back propagation (BP) method has been improved. This hybrid algorithm enhances local accuracy, generalization and convergence, making it extremely acceptable for lithofacies identification missions.

In this research, shale lithofacies were recognized via desegregated analysis of core specimen, thin section petro-graphy and whole rock structure from key wells. These results were utilized to learn an ASO-BP neural network, setting up a powerful connection among mineral composition, lithofacies types and logging responses. By exploiting the substantial availability of logging data, this method authorizes more systematic and precise lithofacies prediction, minimizing dependence on restricted core data. The improved substructure allows a scalable and comprehensive solution for lithofacies categorization in the Bozhong Sag, improving our understanding of shale reservoir diversity and reinforcing enhanced investigation and production techniques [53].

Rock Facies Classification

Well logs reveal one of the most dominant sources of underground geological details, supplying precious understandings into structural features, mineral layering and key reservoir characteristics like permeability, porosity and water saturation. By distinguishing and anatomizing several well logs, it becomes feasible to differentiate geological formations that present similar log indication. This is described in Figure 2, which explains the lithological chain of the Buzurgan oilfield [54].

Figure 2: Lithological column of Buzurgan oilfield

Stratigraphic units demonstrated by wire-line log data are currently indicated to as log-facies or electro-facies. via the explanation of dia-genetic facies and their proportional log responses, different lithological facies can be recognized using multiple logging instruments, such as density, sonic, neutron, resistivity and gamma-ray logs. Each facies represent a typical association of log responses, meanwhile core data presents significant estimation for permeability and porosity values [55].

Petro-physical approaches improved autonomously from facies models are established using permeability and porosity evaluations alongside with arithmetical analyses of their allocation. These approaches frequently represent patterns of high and low values that nearly correspond to variances detected in the facies model, authenticating a powerful relation among petro-physical properties, facies type and grain-size distribution [56]. Log-facies analysis takes a vital part in petroleum engineering, subsurface characterization and sediment-ology, especially because well logs frequently bring the most continuous and dependable data available. Such analyses could be dedicated manually by expert explanation or automatically utilizing statistical and arithmetical methods [57]. Among these, diversified cluster analysis has shown to be one of the most precise and effective methods for clustering log data, especially in both carbonate and clastic reservoirs [58]. In reality, rock types within a reservoir are typically recognized by a conjunction of machine learning methods like neural network-based clustering, petro-graphic observations and petro-physical measurements. The correlation between electro-facies and geological features proposes that both sedimentary and dia-genetic operations play key roles in managing reservoir quality. Core reservoir variables permeability, porosity and water saturation are used to estimate the condition of each electro-facies [59]. The ultimate objective in reservoir geology is to evolve methods and analytical models that remark geological heterogeneity and develop a full understanding of reservoir behaviour. In this research, porosity, gamma-ray and water-saturation logs were categorized into definite log-facies groups to greater capture lithological difference inside the reservoir. Log-facies categorization was carried out employing cluster analysis to collect log data with indistinguishable responses, indicating differences in lithology and differentiating among data collections based on arithmetical similarity. The categorization procedure was applied using Interactive Petro-physics software, which brought an efficient platform for data visualization and cluster explanation.

Reservoir Characterization

To define the characterization of a Reservoir is an extensive procedure that combines multiple sets of data to describe the physical characteristic distribution, geometry and behaviour of fluid flow inside a reservoir. It contributes as a keystone to understand reservoir diversity and recruits a great domain of algorithms for geological complexity evaluation. the efficiency of reservoir categorization contributes in improving oil recovery, leading to production enhancement, renovate mature fields, boosts reservoir performance forecast and minimizing unimportant functional expenses. This period links the space between the early discovery of a reservoir and the application of effective reservoir management techniques. Key components affecting reservoir performance consist the natural diversity of the formation, spatial differences in permeability and porosity, fluid dispensation and the aspects of the porous medium itself. Because of these features span several research-based spots, reservoir categorization is essentially integrated, relying on competence from geophysics, geology, petro-physics, reservoir engineering, economics and data science. Among the most vital instruments for reservoir categorization in the petroleum industry are well logs, which present a comprehensive underground detail from boreholes drilled for gas, oil, groundwater, minerals and geothermal resources [60]. Well logging is significantly used to recognize producing region, estimate formation characteristics and evaluate hydrocarbon capacities. Wireline log data allows the determining of indispensable reservoir variables for example water saturation (Sw), shale volume (Vsh), permeability (k) and porosity (φ). To exactly identify a geological formation, formation estimation incorporates down-hole measurements, formation evaluation and lab survey to evaluate both fluid and rock characteristics. Petro-physical inspection merges these datasets to establish a comprehensive consideration of reservoir properties, furthermore leading conclusions to ideal recovery and producing tactics. The dominating object of reservoir description is to represent and explain the lithology diversity that take control of fluid influx in the reservoir. This includes building a structural and geological model that combines the differences in petro-physical aspects like porosity, permeability, capillary pressure, relative permeability and fluid saturation all of which are important to estimating reservoir performance. The pore-size distribution is considered a key link among these characteristics, implementing a great influence on reservoir performance and fluid flow potential [61].

To have a summary on petro-physical aspects including porosity, permeability, capillarity, relative permeability and water saturation are interconnected via the pore-size distribution of the rock. Within the context of these properties, porosity is one of the most crucial aspects. The structure of a rock includes mineral grains of different sizes and shapes, creating a complicated network of pores. The geometry and capacity of these pores define two significant features: interconnected pores permit fluid motion; meanwhile segregated pores originally serve as storage. From a reservoir engineering viewpoint, porosity quantifies the ratio of a rock’s bulk volume that is obtainable for hydrocarbon storage. It is computationally revealed as the ratio of pore volume to total rock volume. throughout sediment accumulation and succeeding diagenesis, some pore spaces became sealed by cementation process, keeping just a part of the interconnected pores. These differences indicate two kinds of porosity:

• Absolute porosity presents the whole pore spaces related to the rock’s bulk volume. Despite the fact that a rock could have a high absolute porosity, it may reveal poor fluid conductivity if the pores aren't connected

• Effective porosity demonstrates interconnected pore space that plays a part in fluid flow

This exact parameter is extremely essential in reservoir engineering, as it indicates to the storage capacity of retrievable hydrocarbons that takes an important role in estimating redeemable reserves. Effective porosity mostly governs the transportation of fluids in the reservoir. otherwise, this exact measurement in compacted formations is usually considered a challenge because of complicated pore geometries, fluctuating mineral compositions and lithofacies diversity.

This article reveals a distinctive algorithm to make a description of lithofacies and calculation of petro-physical aspect utilizing an artificial neural network which consists of hard and soft statistics. This method is distinguished with regular based strategies, referring to a marked improvement in lithology recognition precision. Error anatomization represents the lithologies that are sensitive to mis-classification, bringing a precious comprehension for purposed manual accentuation after automated clarification. decreasing singularity and minimizing unwanted human mistakes, the proposed algorithm enhances the efficiency and preserving time, labour and resources. the enhanced dependability of lithology recognition presents a rigid framework for the subsequent geological investigations, embracing conviction to upcoming analyses and permits extra precise evaluation of oil reservoirs. Moreover, this article represents a powerful prospective for boundary implementations in other oil sands reservoirs projects, contributing in an empirical and expandable reference for effectual reservoir enhancement.

1. Ehsan, M. and H. Gu. “An integrated approach for the identification of lithofacies and clay mineralogy through Neuro Fuzzy, cross plot and statistical analyses, from well log data.” Journal of Earth System Science, vol. 129, 2020, pp. 1–13.

2. Anees, A. et al. “Sand-ratio distribution in an unconventional tight sandstone reservoir of Hangjinqi area, Ordos Basin: Acoustic impedance inversion-based reservoir quality prediction.” Frontiers in Earth Science, vol. 10, 2022, article 1018105.

3. Chawshin, K. et al. “Classifying lithofacies from textural features in whole core CT-scan images.” SPE Reservoir Evaluation and Engineering, vol. 24, no. 2, 2021, pp. 341–357. https://doi.org/10.2118/205354-PA.

4. Alzubaidi, F. et al. “Automated lithology classification from drill core images using convolutional neural networks.” Journal of Petroleum Science and Engineering, vol. 197, 2021, article 107933. https://doi.org/10.1016/j.petrol.2020.107933.

5. Al-Mudhafar, W. et al. “Performance evaluation of boosting machine learning algorithms for lithofacies classification in heterogeneous carbonate reservoirs.” Marine and Petroleum Geology, vol. 145, 2022, article 105886. https://doi.org/10.1016/j.marpetgeo.2022.105886.

6. Moghanloo, H. et al. “Application of simultaneous prestack inversion in reservoir facies identification.” Journal of Geophysics and Engineering, vol. 15, no. 4, 2018, pp. 1376–1388. https://doi.org/10.1088/1742-2140/aab249.

7. Ghanbarnejad Moghanloo, H. and M. Riahi. “Integrating watershed segmentation algorithm and supervised Bayesian classification for the assessment of petrophysical parameters, pore properties and lithofacies: a case study from Abadan Plain, SW Iran.” Earth Science Informatics, vol. 16, no. 4, 2023, pp. 3913–3930. https://doi.org/10.1007/s12145-023-01129-x.

8. Tian, Y. “Multi-resolution graph-based clustering analysis for lithofacies identification from well log data: case study of intraplatform bank gas fields, Amu Darya Basin.” Applied Geophysics, vol. 13, no. 4, 2016, pp. 598–607. https://doi.org/10.1007/s11770-016-0588-3.

9. Wu, D. “Investigation and prediction of diagenetic facies using well logs in tight gas reservoirs: evidences from the Xu-2 Member in the Xinchang structural belt of the western Sichuan Basin, western China.” Journal of Petroleum Science and Engineering, vol. 192, 2020, p. 107326. https://doi.org/10.1016/j.petrol.2020.107326.

10. Lan, X. et al. “Log facies identification in carbonate reservoirs using multiclass semi-supervised learning strategy.” Fuel, vol. 302, 2021, pp. 121145. https://doi.org/10.1016/j.fuel.2021.121145.

11. Ali, M. et al. “Data-driven lithofacies prediction in complex tight sandstone reservoirs: A supervised workflow integrating clustering and classification models.” Geomechanics and Geophysics for Geo-Energy and Geo-Resources, 2024. https://doi.org/10.1007/s40948-024-00787-5.

12. Ashraf, U. et al. “Application of unconventional seismic attributes and unsupervised machine learning for the identification of fault and fracture network.” Applied Sciences, vol. 10, 2020, article 3864.

13. Safaei-Farouji, M. et al. “Application of robust intelligent schemes for accurate modelling interfacial tension of CO₂–brine systems: implications for structural CO₂ trapping.” Fuel, vol. 319, 2022, p. 123821.

14. Unglert, K. et al. “Principal component analysis vs. self-organizing maps combined with hierarchical clustering for pattern recognition in volcano seismic spectra.” Journal of Volcanology and Geothermal Research, vol. 320, 2016, pp. 58–74.

15. Al-Anazi, A. and I.D. Gates. “A support vector machine algorithm to classify lithofacies and model permeability in heterogeneous reservoirs.” Engineering Geology, vol. 114, 2010, pp. 267–277.

16. Imamverdiyev, Y. et al. “Lithological facies classification using deep convolutional neural network.” Journal of Petroleum Science and Engineering, vol. 174, 2019, pp. 216–228.

17. Abbas, A. et al. “Integrated seismic interpretation, 2D modeling along with petrophysical and seismic attribute analysis to decipher the hydrocarbon potential of Missakeswal area, Pakistan.” Journal of Geology and Geophysics, vol. 7, 2019, pp. 1–12.

18. Wang, G. et al. “Identifying organic-rich Marcellus Shale lithofacies by support vector machine classifier in the Appalachian Basin.” Computers & Geosciences, vol. 64, 2014, pp. 52–60.

19. Hussain, M. et al. “Application of machine learning for lithofacies prediction and cluster analysis approach to identify rock type.” Energies, vol. 15, 2022, pp. 4501. https://doi.org/10.3390/en15124501.

20. Zhu, Y. et al. “Implementation overview of Chinese continental scientific drilling (CCSD) project and technical systems of core boring.” Acta Geologica Sinica, vol. 92, 2018, pp. 1971–1984.

21. Al-Farisi, O. et al. “Machine learning for 3D image recognition to determine porosity and lithology of heterogeneous carbonate rock.” Proceedings of the SPE Reservoir Characterisation and Simulation Conference and Exhibition, Abu Dhabi, United Arab Emirates, September 2019.

22. Wang, H. et al. “Research on core recognition algorithm based on support vector machine.” Journal of Hubei Engineering University, vol. 41, 2021, pp. 61–66.

23. Song, W. et al. “Intelligent lithology identification method of core slices based on generative adversarial network.” Journal of Yangtze University, Natural Science Edition, vol. 19, 2022, pp. 39–46.

24. Miao, Y. “Review of image classification algorithms based on convolutional neural networks.” Remote Sensing, vol. 13, no. 23, 2021, p. 4712.

25. Zhao, M. et al. “Deep residual networks with dynamically weighted wavelet coefficients for fault diagnosis of planetary gearboxes.” IEEE Transactions on Industrial Electronics, vol. 65, no. 5, 2018, pp. 4290–4300.

26. Xiong, W. et al. “Semi-automatic determination of the number of seismic facies in waveform classification.” Proceedings of the 72nd EAGE Conference and Exhibition Incorporating SPE EUROPEC, 2010. https://doi.org/10.15258/sst.2008.36.3.10.

27. Liu, X.Y. et al. “Bayesian discriminant analysis of lithofacies integrate the Fisher transformation and the kernel function estimation.” Interpretation, vol. 5, no. 2, 2017, pp. SE1–SE10. https://doi.org/10.1109/cvpr.2007.383084.

28. Jalalalhosseini, S.M. et al. “Predicting porosity by using seismic multi-attributes and well data and combining these available data by geostatistical methods in a South Iranian oil field.” Petroleum Science and Technology, vol. 32, no. 1, 2014, pp. 29–37. https://doi.org/10.1080/10916466.2011.584102.

29. Huang, W.L. et al. “Signal extraction using randomized-order multichannel singular spectrum analysis.” Geophysics, vol. 82, no. 2, 2016, pp. V69–V84. https://doi.org/10.1190/geo2015-0708.1.

30. Caruana, R. and A. Niculescu-Mizil. “An empirical comparison of supervised learning algorithms.” Proceedings of the 23rd International Conference on Machine Learning, 2006, pp. 161–168. https://doi.org/10.1145/1143844.1143865.

31. Li, J. et al. “Reservoir prediction via SVM pattern recognition.” 74th SEG Technical Program Expanded Abstracts, 2004, pp. 425–428. https://doi.org/10.1190/1.1845255.

32. Suykens, J. and J. Vandewalle. “Least squares support vector machine classifiers.” Neural Processing Letters, vol. 9, no. 3, 1999, pp. 293–300. https://doi.org/10.1023/A:1018628609742.

33. Abedi, M. et al. “Support vector machine for multi-classification of mineral prospectivity areas.” Computers and Geosciences, vol. 46, 2012, pp. 272–283. https://doi.org/10.1016/j.cageo.2011.12.014.

34. Chai, H. “Automatic discrimination of sedimentary facies and lithologies in reef-bank reservoirs using borehole image logs.” Applied Geophysics, vol. 6, no. 1, 2009, pp. 17–29. https://doi.org/10.1007/s11770-009-0011-4.

35. Zhao, T. et al. “Lithofacies classification in Barnett Shale using proximal support vector machines.” SEG Technical Program Expanded Abstracts, vol. 84, 2014, pp. 1491–1495. https://doi.org/10.1190/segam2014-1210.1.

36. Qu, S. et al. “Automatic high-resolution microseismic event detection via supervised machine learning.” Geophysical Journal International, vol. 218, no. 3, 2019, pp. 2106–2121. https://doi.org/10.1093/gji/ggz273.

37. Crammer, K. and Y. Singer. “On the algorithmic implementation of multiclass kernel-based vector machines.” Journal of Machine Learning Research, vol. 2, no. 12, 2001, pp. 265–292. https://doi.org/10.1162/15324430260185628.

38. Lin, Y.Y. et al. “Local ensemble kernel learning for object category recognition.” 2007 IEEE Conference on Computer Vision and Pattern Recognition, 2007, pp. 1–8. https://doi.org/10.1109/cvpr.2007.383084.

39. Gönen, M. and E. Alpaydin. “Localized multiple kernel learning.” Proceedings of the 25th International Conference on Machine Learning, 2008, pp. 352–359. https://doi.org/10.1145/1390156.1390201.

40. Jose, C. et al. “Local deep kernel learning for efficient non-linear SVM prediction.” International Conference on Machine Learning, 2013, pp. 486–494. https://doi.org/10.1145/2623330.2623759.

41. Hao, F. et al. “Lacustrine source rock deposition in response to co-evolution of environments and organisms controlled by tectonic subsidence and climate, Bohai Bay Basin, China.” Organic Geochemistry, vol. 42, 2011, pp. 323–339.

42. Huang, C. et al. “Sedimentology and lithofacies of lacustrine shale: a case study from the Dongpu sag, Bohai Bay Basin, Eastern China.” Journal of Natural Gas Science and Engineering, vol. 60, 2018, pp. 174–189.

43. He, J. et al. “Logging identification and characteristic analysis of the lacustrine organic-rich shale lithofacies: a case study from the Es 3 L shale in the Jiyang Depression, Bohai Bay Basin, Eastern China.” Journal of Petroleum Science and Engineering, vol. 145, 2016, pp. 238–255.

44. Kim, J. “Lithofacies classification integrating conventional approaches and machine learning technique.” Journal of Natural Gas Science and Engineering, vol. 100, 2022.

45. Shi, J. et al. “Lithofacies classification and origin of the Eocene lacustrine fine-grained sedimentary rocks in the Jiyang Depression, Bohai Bay Basin, eastern China.” Journal of Asian Earth Sciences, vol. 194, 2020.

46. Wu, J. et al. “The significance of organic matter–mineral associations in different lithofacies in the Wufeng and Longmaxi shale-gas reservoirs in the Sichuan Basin.” Marine and Petroleum Geology, vol. 126, 2021.

47. Wang, G. et al. “Applications of molecular fossils in lacustrine stratigraphy.” Chinese Journal of Geochemistry, vol. 29, 2010, pp. 15–20.

48. Chang, H.C. et al. “Lithofacies identification using multiple adaptive resonance theory neural networks and group decision expert system.” Computers and Geosciences, vol. 26, 2000, pp. 591–601.

49. Ameur-Zaimeche, O. et al. “Lithofacies prediction in non-cored wells from the Sif Fatima oil field (Berkine Basin, southern Algeria): A comparative study of multilayer perceptron neural network and cluster analysis-based approaches.” Journal of African Earth Sciences, vol. 166, 2020.

50. Myles, P.S. and J. Cui. “Using the Bland–Altman method to measure agreement with repeated measures.” Statistics in Medicine, vol. 26, no. 1, 2007, pp. 309–311.

51. Yu-Jiang, J. et al. “Fine lithofacies of China’s marine shale and its logging prediction: a case study of the Lower Silurian Longmaxi Marine Shale in the Weiyuan area, southern Sichuan Basin.” Fine Lithofacies of China’s Marine Shale, vol. 23, 2016, p. 107.

52. Li, B. et al. “Lithofacies and pore characterization in an argillaceous-siliceous-calcareous shale system: a case study of the Shahejie Formation in Nanpu sag, Bohai Bay Basin, China.” Journal of Petroleum Science and Engineering, vol. 173, 2019, pp. 804–819.

53. Cui, Q. et al. “Identification of lithofacies and prediction of mineral composition in shales: A case study of the Shahejie Formation in the Bozhong Sag.” Unconventional Resources, vol. 2, 2022, pp. 72–84. https://doi.org/10.1016/j.uncres.2022.09.002.

54. AlHusseini, A. and S. Hamd-Allah. “Rock facies classification and its effect on the estimation of original oil in place based on petrophysical properties data.” International Journal of Nonlinear Analysis and Applications, vol. 14, no. 1, 2023, pp. 527–540. https://doi.org/10.22075/ijnaa.2022.6855.

55. Lai, J. et al. “Prediction of diagenetic facies using well logs: Evidences from Upper Triassic Yanchang Formation Chang 8 Sandstones in Jiyuan Region, Ordos Basin, China.” Oil & Gas Science and Technology – Revue d’IFP Energies Nouvelles, vol. 71, no. 3, 2016, pp. 34.

56. Veloso, F.M.L. et al. “Facies and petrophysical modelling of a thick Lower Cretaceous tsunami deposit in eastern Spain: up-scaling from samples to outcrop scales.” Sedimentary Geology, vol. 343, 2016, pp. 38–55.

57. Al-Baldawi, B.A. “Applying the cluster analysis technique in logfacies determination for Mishrif Formation, Amara oil field, southeastern Iraq.” Arabian Journal of Geosciences, vol. 8, no. 6, 2015, pp. 3767–3776.

58. Gill, D. et al. “Numerical zonation of log suites and logfacies recognition by multivariate clustering.” AAPG Bulletin, vol. 77, no. 10, 1993, pp. 1781–1791.

59. Mirzaee, R. et al. “Formation evaluation and rock type classification of Asmari Formation based on petrophysical–petrographic data: a case study in one of super fields in Iran southwest.” Petroleum, 2022, article S2405656122000153.

60. Ganguli, S. et al. Reservoir Characterization: State-of-the-art, Key Challenges and Ways Forward. Reservoir Characterization, Modeling and Quantitative Interpretation, 2024, https://doi.org/10.1016/B978-0-323-99593-1.00015-X. Accessed December 08, 2025.

61. Esiri, A. et al. Enhancing reservoir characterization with integrated petrophysical analysis and geostatistical methods. OARJ Open Access Research Journal of Multidisciplinary Studies, 2024, https://doi.org/10.53022/oarjms.2024.7.2.0038. Accessed December 10, 2025.