Encyclopedia
Please note this is an old version of this entry, which may differ significantly from the current revision.
Subjects:
Engineering, Petroleum
The accurate prediction of underground formation lithology class and tops is a critical challenge in the oil industry. This research presents a machine-learning (ML) approach to predict lithology from drilling data, offering real-time litho-facies identification.
- lithology prediction
- machine learning
- drilling data
1. Introduction
Since the early days of the oil industry, human ingenuity has been at work to overcome various challenges. Among these challenges, real-time and precise determination of the tops of the formation and its lithology whereas drilling is of utmost importance to guarantee efficient and safe drilling operations [1]. Accurate information about the formation tops is essential when designing a casing program, as it is necessary to select the right depths for placing the casing to ensure efficient zonal separation, and to correctly design the correct mud weight in order to keep the wellbore conditions in check [2,3,4].
Drilling engineers in oil fields use four distinct methods to identify different reservoir zones or formation tops: rate of penetration (ROP) charts, gamma-ray (GR) logs, formation cuttings, and mud logging [5,6,7]. These techniques are helpful for drillers to delineate the formation tops, but each one has drawbacks such as high costs, lower accuracy, and substantial labor. Moreover, the majority of these measurements encounter temporal or depth-related delays, which restrict the ability to instantly estimate the formation tops. These limitations pose challenges to the feasibility and efficacy of current techniques employed for formation top determination.
Although lithology significantly impacts the ROP, other drilling parameters also exert considerable influence on ROP fluctuations [8]. Consequently, relying solely on ROP for estimating lithology changes or formation tops is inadequate, particularly when other drilling parameters experience significant variability. Furthermore, as the wellbore depth increases, there is a noticeable time lag in obtaining geophysical logs or drilled cuttings, which delays the prediction of the currently drilled formation [6]. Employing techniques such as GR, measurement, or logging while drilling, or mud logging in each section is also economically impractical and does not offer prompt and essential information.
On the other hand, to address these challenges, researchers from the realm of the oil and gas industry have harnessed the power of ML to transform key aspects of this industry. In drilling optimization, Berrehal et al. (2022) [9] proposed an ML-based approach for a real-time mechanical earth model, enhancing wellbore stability in the Volve field. Similarly, Al-Sudani et al. (2017) [10] introduced a control engineering system for real-time monitoring of drilling mechanical energy and bit wear, optimizing drilling performance. Meanwhile, for fracturing, Erofeev et al. (2021) [11] predicted post-hydraulic fracturing oil and liquid production with 80% accuracy, enabling real-time HF candidate selection. In the domain of oil recovery, Ouadi et al. (2023) [12] introduced high-accuracy models for predicting gas well productivity using Fishbone drilling, demonstrating its potential to enhance hydrocarbon recovery and reduce environmental impact. Although Ahmed et al. (2017) [13] showcased the potential of AI techniques in estimating oil recovery factors in early-time reservoirs, surpassing existing correlations. Additionally, Hamadi et al. (2023) [14] presented a robust machine-learning framework to predict key parameters in CO2 -enhanced oil recovery, delivering superior accuracy and insights for efficient CO2-EOR design. Furthermore, Mouedden et al. (2022) [15] proposed FIScreT, a decision-making tool based on fuzzy logic for automating candidate well selection in stimulation processes.
2. Lithology Prediction
Lithology prediction using machine learning has rapidly evolved from early foundational models to an array of sophisticated techniques, integrating real-time data and achieving high accuracy. The initial explorations into the field of lithology prediction were laid by Rogers et al. (1992) [16], Benaouda et al. (1999) [17], and Wang and Zhang (2008) [18]. Utilizing well-log data, these pioneers developed predictive models; however, they faced challenges in predicting thin formations and dealing with missing density logs. As the field progressed, Qi and Carr (2006) [19] and Al-Anazi and Gates (2010) [20] contributed to the development of machine-learning models for lithofacies and permeability prediction, using refined artificial neural network (ANN) and support vector machine (SVM) techniques. Moazzeni and Haffar (2015) [21] highlighted the impact of external factors on drilling parameters, revealing the need for further refinement of these machine-learning techniques. Addressing this need, Raeesi et al. (2012) [22], Wang and Carr (2012) [23], and Al-Mudhafar (2017) [24] introduced the use of Artificial Neural Networks (ANNs) and comprehensive integrated workflows, which significantly improved lithofacies classification.
The challenge of real-time lithology prediction during drilling operations was addressed by Mohamed et al. (2019) [25] and Nanjo and Tanaka (2019, 2020) [26], utilizing machine-learning models and image analysis methods. Elkatatny et al. (2019) [27] took a significant step towards real-time prediction, using ANN models to determine formation tops based on drilling parameters. Gupta et al. (2020) [28] designed a real-time machine-learning workflow for lithology prediction during drilling, marking a milestone in the field.
In their research, Zhang and Baines (2021) [29] explored machine-learning models such as ANN, SVM, and CNN, which yielded promising results, especially the 2D CNN combined with PCA feature extraction. Similarly, Wei Zhoucheng et al. (2019) [30] proposed a multi-well lithology identification method that involved feature engineering, machine-learning model training, and optimal model selection. Additionally, Aniyom et al. (2022) [31] demonstrated the potential of ensemble methods to improve lithology prediction performance through the development of a voting classifier machine-learning model.
Several studies have integrated additional features or methods into their models to improve classification performance. Xi Chen et al. (2020) [32] combined the Reducing Error Correcting Output Code algorithm with the Kernel Fisher Discriminant Analysis, outperforming conventional methods. Jiang et al. (2021) [33] introduced a stratigraphic unit as an additional feature, significantly improving lithology classification. Zerui Li et al. (2020) [34] proposed a semi-supervised lithology identification workflow using a Laplacian support vector machine, enhancing classification performance by utilizing feature and depth similarities.
Mou et al. (2016) [35] employed support vector machine models to estimate volcanic lithology in the Liaohe Basin, China, achieving high accuracy. In another study, Moazzeni et al. (2015) [21] accurately predicted formation and lithology in the South Pars gas field, Iran, using artificial neural networks. Similarly, Wang De-ping et al. (2007) [36] attained a 96% correctness rate in predicting lithology in the Bayantala oil field by utilizing an SVM model. These studies focused on specific geological contexts and demonstrated impressive accuracy levels.
Flexible and adaptive models have been explored for lithology prediction. Jia et al. (2012) [37] demonstrated the efficacy of an adaptive neuro-fuzzy inference system for lithology identification from well-log data. Cheng et al. (2010) [38] combined a particle swarm optimization (PSO) algorithm with the least squares support vector machines (LSSVM) for higher precision lithology identification.
Sebtosheikh and Salehi (2015) [39] employed support vector machines (SVMs) with various kernel functions for accurate lithology prediction in a multilayered carbonate reservoir in Iran. Building on this, Avanzini et al. (2016) [40] presented a workflow using cluster analysis to identify productive sweet spots in unconventional reservoirs, focusing on the Barnett Shale Formation. In a different approach, Gu et al. (2019) [41] achieved high (>75%) lithology prediction accuracies by integrating CRBMs and PSO into PNNs. Additionally, Imamverdiyev and Sukhostat (2019) [42] introduced a deep learning 1D CNN model that outperformed other methods in geological facies classification.
Moazzeni et al. (2019) [43] made significant advancements through their research about real-time prediction models by developing an ANN model optimized with a genetic algorithm and Taguchi experimental design for lithology and formation prediction. In a related study, Zhang and Baines (2021) [29] demonstrated the potential of real-time models with their 2D CNN model, achieving over 90% accuracy in identifying four lithology classes. These notable developments have contributed to the progress of real-time prediction techniques.
Finally, recent studies have demonstrated the feasibility of rapid, automated lithology prediction. Popescu et al. (2020) [44] created a supervised machine-learning pipeline that enabled rapid, scalable, and confident lithology prediction. Ao et al. (2019) [45] combined mean–shift feature extraction and random forest classification to improve prediction accuracy. Zhang et al. (2017) [46] used a convolutional neural network for accurate lithology identification from borehole images with a success rate of about 95%.
This entry is adapted from the peer-reviewed paper 10.3390/eng4030139
This entry is offline, you can click here to edit this entry!
