Automated well-to-seismic tie using deep neural networks

One of the first steps toward integration of seismic with the reservoir model is the well-to-seismic tie. This operation can be subdivided into two parts: depth-to-time conversion and wavelet extraction. Here we focus on the former, which is by far the most time-consuming operation using current industry workflows. In this paper, we present a methodology using deep neural networks that can learn the depth-to-time function derived from Vertical Seismic Profiling (VSP) data. Using both synthetic and field data, we demonstrate how a tuned recurrent neural network (RNN) can predict traces in the two-way-time (TWT) domain with acceptable accuracy. Finally, we present an alternative approach using a temporal convolutional network (TCN), which can help overcome some of the limitations of RNNs while producing similar prediction accuracy. Both methods are proven to automate well tie, reducing the needed time to a fraction. Presentation Date: Tuesday, October 13, 2020 Session Start Time: 1:50 PM Presentation Time: 2:15 PM Location: 360D Presentation Type: Oral