Abstract

This study evaluates the effectiveness of neural network testing on well-log data in the study area. The Artificial Neural Networks (ANNs) and Convolutional neural networks (CNNs) models are developed to predict the missing part of the data or verify the values due to errors in the measurement process. In addition, neural networks are also used to create virtual logs at any location in the reservoir based on log data from existing wells to get a better view of the geological characteristics in the subsurface without any new drilling wells. The dataset used in this study includes qualified logs in twenty wells located in Cuu Long Basin. Data sets for neural networks are designed based on the characteristics of the log data, including the direction of the target well, the angle of the goal well, the position, the depth, and the log values of the nearest wells. Min-max normalization is used to scale the well length before training the dataset. The database is divided into three different sets: training data set, test set validation data set, and test data set. The reliability and accuracy of the methods are expressed through the loss function or the correlation coefficient R2. The accuracy of these logs was tested for newly drilled wells at the time the system was developed and trained. Log values generated by CNNs have higher correlation coefficients than those of ANNs with R2 equal to 0.7994, while R2 of ANNs is only 0.6701. Results showed that predicting using CNNs was better than ANNs. Therefore, the use of CNNs will increase decision-making efficiency by avoiding time-consuming procedures and processes.