Prediction of Reservoir Parameters Based on Deep Learning for CO2 Storage Monitoring
Abstract
The increase in fossil energy consumption has led to the emission of a large amount of CO2 gas into the atmosphere. Carbon capture and storage (CCS) is an effective method to reduce the concentration of CO2 in the air and is of great significance in achieving carbon neutrality. CCS technology is to capture and compress CO2 and inject it into underground for storage. The key to the success of this project is that CO2 does not leak. In order to ensure long-term and safe geologic storage of CO2, a method for CO2 plume detecting based on full convolutional neural network (FCN) that applies time-lapse difference seismic data is proposed. The non-linear relationship between time-lapse difference data and reservoir parameter changes is established through FCN. The training data of background geological model is generated by way of logging curve interpolation and random simulation of folding structures. The changes in reservoir parameters caused by CO2 injection are simulated by the shapes of various animals, birds, and clouds. Based on the time-lapse geological model, the corresponding time-lapse seismic data can be obtained through seismic forward modeling. This provides a large number of data sets for the training of the FCN network. The network was trained and tested based on 3000 sets of geological models and data with actual geological features, of which 2800 sets were used for training and 200 sets were used for testing. Once the training of the network is completed, the prediction process will be very fast and does not need to be trained again. The test results show that 2800 sets of time-lapse difference data and reservoir parameter changes can train a better network. and the prediction accuracy of noise-free and noisy data are all greater than 0.95. Therefore, the CO2 storage monitoring method based on deep learning has the advantages of high efficiency and high precision, and is an effective tool for CO2 geologic storage monitoring.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2021
- Bibcode:
- 2021AGUFM.S31B..03L