Scholarlink

<p>Prediction of shear wave velocities from well logs is vital for understanding rock and fluid properties, improving seismic interpretations, and supporting reservoir management strategies. The uncertainty of predicting the shear waves (Vs) has a significant impact on resource exploration and production efficiency. However, the traditional methods for predicting the shear wave velocity face a major challenge due to the high cost and limitations in old wells due to the heterogeneity and complexity of the subsurface reservoir, such as in the case of the Sirt Basin. To address this deficiency, a technique has been enhanced to reduce the uncertainty of shear wave prediction. This technique is based on an ensemble model that combines several machine learning (ML) techniques, such as the Multi-Layer Perceptron (MLP) optimization by Particle Swarm Optimization (PSO), Genetic Algorithm (GA), Random Search (RS), and Bayesian optimization (BO). The main objective of combining these optimized algorithms with the MLP is to explore various scenarios and leverage the strengths of each method. Wireline logging data has been obtained from two vertical wells in the Sirt Basin, which is the most prolific embayment in North Africa, while the other wells in the basin lack advanced tools for measuring the shear wave. The data underwent preprocessing and dimensionality reduction using the principal component analysis (PCA). Next, the data has been divided into 50% for training, 20% for testing, and 30% for validation. Each model was then developed and evaluated, integrating MLP with Particle Swarm Optimization (PSO) to achieve a relatively high coefficient of determination value (<i>R</i><sup>2</sup> = 0.947), i.e., of high reliability. Next, an ensemble model was developed by stacking all four MLP models (PSO, GA, RS, and BO) using a linear combiner. The results demonstrated the superiority of the proposed integrated ensemble model in improving the reliability and precision of predictions compared to individual models. The findings provided a high coefficient of determination (<i>R</i><sup>2</sup> = 0.996) and a relatively low Mean Squared Error (MSE = 17.7). This proposed technique can be applied in similar cases worldwide to accurately predict shear waves to enhance exploration and production activities by leveraging the complementary strengths of different optimization algorithms.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

New approach to reduce the uncertainty level of the shear wave velocity prediction by utilizing ensemble model: A case study from Sirt Basin, Libya

  • 期刊论文

  • 发表日期：

    2025年7月3日

查看全文
  • Mohammed A. Amir,
  • Hamzah S. Amir,
  • Mohamed I. Abdel-Fattah,
  • Mohamed O. Elsteil,
  • Saad Mogren,
  • Elkhedr Ibrahim,
  • Bassem S. Nabawy,
  • Mohamed Ramah,
  • Mohamed Reda

摘要

Prediction of shear wave velocities from well logs is vital for understanding rock and fluid properties, improving seismic interpretations, and supporting reservoir management strategies. The uncertainty of predicting the shear waves (Vs) has a significant impact on resource exploration and production efficiency. However, the traditional methods for predicting the shear wave velocity face a major challenge due to the high cost and limitations in old wells due to the heterogeneity and complexity of the subsurface reservoir, such as in the case of the Sirt Basin. To address this deficiency, a technique has been enhanced to reduce the uncertainty of shear wave prediction. This technique is based on an ensemble model that combines several machine learning (ML) techniques, such as the Multi-Layer Perceptron (MLP) optimization by Particle Swarm Optimization (PSO), Genetic Algorithm (GA), Random Search (RS), and Bayesian optimization (BO). The main objective of combining these optimized algorithms with the MLP is to explore various scenarios and leverage the strengths of each method. Wireline logging data has been obtained from two vertical wells in the Sirt Basin, which is the most prolific embayment in North Africa, while the other wells in the basin lack advanced tools for measuring the shear wave. The data underwent preprocessing and dimensionality reduction using the principal component analysis (PCA). Next, the data has been divided into 50% for training, 20% for testing, and 30% for validation. Each model was then developed and evaluated, integrating MLP with Particle Swarm Optimization (PSO) to achieve a relatively high coefficient of determination value (R2 = 0.947), i.e., of high reliability. Next, an ensemble model was developed by stacking all four MLP models (PSO, GA, RS, and BO) using a linear combiner. The results demonstrated the superiority of the proposed integrated ensemble model in improving the reliability and precision of predictions compared to individual models. The findings provided a high coefficient of determination (R2 = 0.996) and a relatively low Mean Squared Error (MSE = 17.7). This proposed technique can be applied in similar cases worldwide to accurately predict shear waves to enhance exploration and production activities by leveraging the complementary strengths of different optimization algorithms.