A New Machine Learning Procedure to Generate Highly Accurate Synthetic Shear Sonic Logs in Unconventional Reservoirs

Abstract

Abstract Shear sonic travel time (DTS), along with compressional sonic travel time and bulk density are required in order to estimate rock mechanical properties which play an important role in fracture propagation and the success of hydraulic fracture treatments in horizontal wells. DTS logs are often missing from the log suite due to their costs and time to process. The following study presents a machine learning procedure capable of generating highly accurate synthetic DTS curves. A hybrid convolutional-recurrent neural network (c-RNN) was chosen in the development of this procedure as it can learn sequential data which a traditional neural network (ANN) cannot. The accuracy of the c-RNN was superior when compared to that of the ANN, simple baselines and empirical correlations. This procedure is a cost effective and fast alternative to running DTS logs and with further development, has the potential to be used for predicting production performance from unconventional reservoirs.