Oil and Gas Reservoir Characterization in Predicting Missing Velocity Logs Using Machine Learning

Interpreting well-logging data is crucial in the oil and gas industry as it helps make decisions on drilling, completion, and production processes. Traditional methods of interpretation are time-consuming and require specialized knowledge and experience. Over the past few years, machine learning techniques have emerged as a viable alternative to conventional interpretation methods. This research proposes a hybrid ML model that combines the strengths of different algorithms for predicting missing velocity log data in the gas and oil industry. The model is trained and tested on a large dataset from several oil fields and evaluated using multiple metrics. The proposed model effectively predicts porosity, permeability, and water saturation from well logs, outperforming traditional methods in terms of accuracy and computational efficiency. After evaluating two different proposals for predicting well-logging data, it was determined that the Ensemble Machines approach was more effective than the neural network approach. However, during testing, the neural network suffered from parameter limitations, which resulted in processing and time consumption issues. Despite these challenges, the neural network still exhibited significant potential and demonstrated excellent generalization capabilities. It highlights the promise of neural networks in predicting well-logging data, while also underscoring the importance of carefully considering processing limitations and parameter optimization in their implementation. During analysis, the ensemble model gives the highest accuracy (i.e., 82.64%) for predicting missing velocity logs.