Supervised machine learning for predicting shear sonic log (DTS) and volumes of petrophysical and elastic attributes, Kadanwari Gas Field, Pakistan

Abstract

Shear sonic log (DTS) availability is vital for litho-fluid discrimination within reservoirs, which is critical for field development and production. For certain reasons, most of the wells in the Lower Indus Basin (LIB) lack DTS logs, which are modeled using conventional techniques based on empirical relations and rock physics modeling. However, in their extensive computation, these approaches need assumptions and multiple prerequisites, which can compromise the true reservoir characteristics. Machine learning (ML) has recently emerged as a robust and optimized technique for predicting precise DTS with fewer input data sets. To predict the best DTS log that adheres to the geology, a comparison was made between three supervised machine learning (SML) algorithms: random forest (RF), decision tree regression (DTR), and support vector regression (SVR). Based on qualitative statistical measures, the RF stands out as the best algorithm, with maximum determination of correlation (R2) values of 0.68, 0.86, 0.56, and 0.71 and lower mean absolute percentage error (MAPE) values of 4.5, 2.01, 4.79, and 4.65 between the modeled and measured DTS logs in Kadanwari-01, -03, -10, and -11 wells, respectively. For detailed reservoir characterization, the RF algorithm is further employed to generate elastic attributes such as P-impedance (Zp), S-impedance (Zs), lambda-rho (λρ), mu-rho (μρ), as well as petrophysical attributes such as effective porosity (PHIE) and clay volumetric (Vcl) utilizing seismic and well data. The resultant attributes helped to establish a petro-elastic relationship delineated at the reservoir level. Possible gas zones were determined by zones with high PHIE (8%–10%) and low values of other attributes like Vcl (30%–40%), Zp (10,400–10,800 gm/cc*m/s), and Zs (6,300–6,600 gm/cc*m/s). The potential bodies are also validated by low λρ (27–30 GPa*g/cc) cross ponding to higher μρ (38–44 GPa*g/cc).

Harnessing Advanced Machine-Learning Algorithms for Optimized Data Conditioning and Petrophysical Analysis of Heterogeneous, Thin Reservoirs

“…In the Central Indus Basin, Pakistan, few studies evaluated ML methods. Ali et al used the Random Forest Regressor (RFR) to forecast facies with an accuracy of 83.85%, and Ahmed et al . used a stacking method to combine the outputs of numerous models, including the Extra Tree Regressor (ETR) reservoir, with an accuracy of 87.23%.…”

“…In the Central Indus Basin, Pakistan, few studies evaluated ML methods. Ali et al 57 used the Random Forest Regressor (RFR) to forecast facies with an accuracy of 83.85%, and Ahmed et al 37 used a stacking method to combine the outputs of numerous models, including the Extra Tree Regressor (ETR) reservoir, with an accuracy of 87.23%. In a nutshell, this research highlights the effectiveness of ML models in removing outliers from raw well logs and modeling the missing and bad well logs in optimized and systematic ways.…”

“…Similarly, missing sonic logs in the splice zone and density logs due to bad holes cause misleading results. , Rock physics modeling is usually used to predict and optimize bad logs. RPM requires a detailed set of reservoir parameters along with a complex set of equations and models to reproduce reservoir conditions from the mineral level to formation and dry conditions to the saturation level. , Therefore, it requires a very expert-level approach and extra care; otherwise, it may produce erroneous results that later result in failure or misleading information. Therefore, a more robust approach is needed that not only detects outliers but also helps address data limitations with great accuracy and authenticity.…”

ResNet and CWT Fusion: A New Paradigm for Optimized Heterogeneous Thin Reservoir Evaluation

“…DTP and DTS play a vital role in building rock physics templates for the classification of lithofacies. , Previously, to address the issues regarding DTP and DTS, linear relationships were taken among the seismic velocities and relevant rock properties, such as pores, fluids, shale volumetrics, etc., for assessing poor or missing zones. − Contrarily, ML acts as an efficacious tool having the capability of building nonlinear relationships amid logs based on prominent features for unrecorded log estimation especially DTS that comprehensively appraise the reservoir characteristics. Due to the intricate reservoir properties of the B-interval along with the limitations in data, the ML techniques proved critical and acted as an enhanced tool for approximating the DTS . The accuracy of the estimated DTS is valued by statistical measures, including R2 and MAPE .…”

“…Additionally, meticulous approaches like deterministic and stochastic seismic inversions have been employed in various nearby producing fields to tackle these challenges. − However, their effectiveness in such complex geological settings often falls short, primarily due to their inability to handle missing data effectively, and the increased number of procedural steps may cause uncertainty. In recent years, the comparable use of advanced ML techniques has gained significant interest in subsurface imaging and reservoir characterization due to their enhanced result, robustness, and efficiency . A comprehensive solution has been developed, focused on an integrated strategy that combines different types of data along with various approaches (petrophysics, rock physics, and seismic inversion) and employs modern ML algorithms to successfully handle reservoir challenges.…”