Official PyTorch implementation
Ensemble models for CO₂ leakage detection
This artifact is developed based on the Zeng, S, Huseyin Tuna Erdinc, Ziyi Yin, Abhinav Prakash Gahlot, Felix J. Herrmann. “Enhancing Performance with Uncertainty Estimation in Geological Carbon Storage Leakage Detection from Time-Lapse Seismic Data.
Below is the workflow diagram illustrating the detection process:
Python libraries: See requirements.yml for library dependencies. The conda environment can be set up using these commands:
conda env create -f requirements.yml
conda activate leakage_detection
Put the data under the data directory, and train the dataset by running the Python script:
python scripts/training_loop.py --dataset_path "data/dataset_jrm_1971_seismic_images?dl=0" --data_length 1971 --model_name "vgg16"More details can be found under the notebook training_demo.ipynb.
The multi-criteria decision-making (MCDM)-based Ensemble schema and uncertainty analysis details can be found in artifacts_demo.ipynb
Below is the uncertainty analysis process:
