Dynamic data driven sonic well log model for formation evaluation

The lack of acoustic measurements places severe limitations on the application of well log data to analyze rock physics. In such conditions, other petrophysical data can be used to predict the shear and compressional sonic travel time. This study presents a novel data-driven model based on a nonlinear autoregressive neural network with exogenous (NARX) input to estimate the shear and compressional sonic travel time due to its ability to accurately determine nonlinearity in sequential and temporal data. The architecture of the model comprises three-layers and ten hidden neurons with gamma ray log as exogenous input. The proposed NARX methodology is developed using 11 wells, six from the Norwegian continental shelf and five from West Africa. The results show that the wells provide sufficiently accurate predictions of the actual sonic well logs using the NARX model. The predicted sonic logs are used to estimate formation property parameters like sonic ratio, sonic difference, sonic porosity, and Poisson's ratio. This paper proves NARX is an affordable, efficient and accurate means to reproduce sonic well logs for formation evaluation.