Machine-Learned Inference of Fracture Flowrate From Temperature Logs

Abstract

Reliable identification of a fracture flowrate is essential to successful reservoir exploration and optimization. We developed a machine learning approach to identify fracture flowrate from spatiotemporal wellbore temperature measurements. A long short-term memory fully convolutional network was employed to jointly detect fractures intersecting the wellbore and quantify their contribution to the overall flow during fluid injection. Training data for the algorithm were generated by a wellbore and fractured-reservoir thermal model. The machine learning algorithm trained on single injectionstage temperature shows high fracture flowrate estimation accuracy on synthetic validation cases. It outperforms an optimization-based particle swarm optimization method on a real field case. For the machine learning method, the fusion of various-stage temperature as the input feature set improves the robustness of fracture detection and flowrate estimation to noise interference.