A New Artificial Intelligence Based Empirical Correlation to Predict Sonic Travel Time

Abstract

Abstract Sonic logs can be used to estimate rock elastic parameters that are useful in obtaining the in-situ stresses of the rock. In cases where the sonic log is missing, the log values are estimated using empirical correlations. Since conducting sonic logging involves time and cost, the industry relies on empirical models. However, none of the available models are universally accepted by log analysts because each model estimates the sonic log values significantly different from the others. The error in sonic log values affects the elastic parameters, which can result in either underestimation or overestimation of the in-situ stresses. The objective of this paper is to develop simple and accurate model to determine sonic travel time based on gamma ray, bulk density and neutron porosity because these three logs are generally available for every well. Artificial Neural Networks are used for the development of the model. An attempt has also been done to converge the results into one simple empirical correlation using the weights of ANN model in order to make the model equation universal and usable for field applications. Several empirical correlations from literature were selected and subjected to same real field well log data as the AI model to predict sonic travel time and the results showed that ANN model was able to predict sonic logs with 96% accuracy with a correlation coefficient of 0.96 while other correlations did the same but with lesser accuracy. In addition, the developed empirical correlation from the weights of ANN model gives exactly the same results as predicted by the ANN model. The developed model along with the empirical correlation can serve as handy tool to help geo-mechanical engineers detremine the reservoir geo-mechanical parameters and design any operation in cases where sonic logs are not available.