Artificial intelligence models for real-time synthetic gamma-ray log generation using surface drilling data in Middle East Oil Field

Abstract

Gamma-ray logging (GR) is one of the most crucial elements in petroleum engineering that would assist evaluate oil and gas reservoirs and identify the formation lithology. GR can be measure for a retried core or by running the logging while drilling (LWD) tools. Both methods require a long time and high cost besides the complexity of applying them. In this study, three artificial intelligence (AI) techniques were assessed for their ability to produce accurate models for predicting a synthetic GR log using surface drilling parameters. These techniques were artificial neural networks (ANN), adaptive network-based fuzzy logic (ANFIS), and functional networks (FN). A total of 4609 data entries from three wells in the Middle East were used to train and test the models and later validate them. The data from wells 1 and 2 were used to develop the AI models, while well 3 was used to validate these models. The obtained results showed the accuracy of predicting the GR using ANN and ANFIS. The correlation coefficient (CC) values of the prediction and measured GR were 0.96, 0.96, and 0.83 for ANN, ANFIS, and FN, respectively. Also, the average absolute percentage error (AAPE) values for the three models in the same order were 3.69%, 3.72%, and 6.7% which shows slightly less accuracy of the FN model. For the validation data set, the ANN and ANFIS models were able to capture all changes in the GR log trend with depth and accurately predict the GR values. A new empirical correlation for GR prediction was developed based on the weight and biases of the optimized ANN model. The new correlation can be used to predict the GR log with high accuracy as a function of the drilling surface data with no need to run AI methods.