Hanyang University Reservoir Imaging with Seismic & EM technology Laboratory’s student Lee Jae-woo won 2nd place by developing an "interpolation module for elastic wave exploration data using LSTM" in the Undergraduate AI Contest held by Mathworks. Student Lee solved the previous machine learning techniques' problem of data quality had by applying deep learning technologies in the high-quality physical prospecting methods. It can also be applied to oil exploration, carbon collection, and storage monitoring.

Student Lee introduced the "elastic wave exploration data interpolation module using LSTM(Long Short-Term Memory)."

Student Lee said that “I used the traces-to-trace approach in the module.” The traces-to-trace is a method of inferring the missing parts of a specific area of information using data already collected. Here, interpolation functions are trained to predict the missing trace signal to complete the picture.

He explained that “The method went through four steps.” Firstly, using Mathworks’ SeisLab add-on and hyperparameter, variables for trainNetwork and prediction functions were created. Next, hyperparameters were selected for learning. The weights were also calculated on the graph using Matlab’s Deep Learning Toolbox. Lastly, the prediction target trace and saving settings processes were conducted.

Student Lee emphasized that “I derived the missing seismology data predictions from seismic survey information gathered from the coast of Vincent oilfield in Western Australia and the Gulf of Mexico.” He added, “At that time, I was able to confirm a high accuracy of the interference algorithm based on the traces-to-trace approach” and that "It recorded a lower error rate compared to existing research techniques.”

Currently, Hanyang University's Reservoir Imaging with Seismic & EM technology Laboratory is applying for domestic and international patents related to the traces-to-race approach to elastic wave exploration.