Summary P/S wave mode separation is an essential tool for single-mode analysis from multi-component seismic data. Wave separation methods in recorded data require expert knowledge to choose parameters in different shots of data. To make this process automatic, we propose a machine learning-based method to separate P/S waves. This method employs a multi-task neural network that extracts P-and S-potential simultaneously from multi-component VSP data. Targeting at a specific testing dataset, we derive an efficient building strategy to construct training datasets. Synthetic data experiment shows NN trained on our training dataset performs well both in training and testing datasets. We also make further verifications from the view of acoustic reverse time migration. The separated waves by using NN trained on our training datasets have a considerable high-resolution PP and PS imaging.
Paper
Full text
Multi-task learning based P/S wave separation and reverse time migration for VSP
Semantic Scholar · Engineering · 2020
Abstract
Summary P/S wave mode separation is an essential tool for single-mode analysis from multi-component seismic data. Wave separation methods in recorded data require expert knowledge to choose parameters in different shots of data. To make this process automatic, we propose a machine learning-based method to separate P/S waves. This method employs a multi-task neural network that extracts P-and S-potential simultaneously from multi-component VSP data. Targeting at a specific testing dataset, we derive an efficient building strategy to construct training datasets. Synthetic data experiment shows NN trained on our training dataset performs well both in training and testing datasets. We also make further verifications from the view of acoustic reverse time migration. The separated waves by using NN trained on our training datasets have a considerable high-resolution PP and PS imaging.