| Abstract Submission No. | ABS-4.1-0151 |
| Title of Abstract | Quantum Inspired AI for Time Lapse Inversion of Seismic Data for simulation of CO2 Injection Scenarios |
| Authors | Karismita Baishya* |
| Organisation | IIT(ISM) Dhanbad |
| Address | Hengrabari, Ujjal Nagar, Guwahati-36, Assam Guwahati, ASSAM, India Pincode: 781036 E-mail: 20je0469@iitism.ac.in |
| Country | India |
| Presentation | Oral |
| Abstract | Seismic full-waveform inversion (FWI) effectively reconstructs subsurface velocity models, but its high computational cost limits practical use.Here a new strategy such as a quantum-inspired artificial intelligence framework is presented for time-lapse seismic full-waveform inversion (FWI), aimed at reconstructing subsurface velocity models that characterize changes due to CO₂ injection in earth¿s subsurface. The Kimberlina-CO₂ synthetic dataset, based on a potential CO₂ reservoir at the Kimberlina site in the southern San Joaquin Basin, serves as the benchmark. This dataset captures the spatial and temporal evolution of a supercritical CO₂ plume over a 200-year period, providing realistic subsurface scenarios for monitoring and detection of CO₂ leakage in deep subsurface reservoirs is vital for safe carbon storage and effective climate management. The proposed hybrid inversion model integrates a convolutional neural network encoder with a PennyLane-based quantum layer wrapped as a TorchLayer, bridging classical and quantum-inspired machine learning to efficiently map seismic inputs to velocity reconstructions. This architecture improves computational efficiency and reduces memory requirements, while maintaining high fidelity in detecting subtle velocity changes induced by the migrating CO₂ plume. Standard error and similarity evaluation metrics¿such as mean absolute error , root mean squared error and structural similarity index ¿are used to quantitatively benchmark the model¿s performance against traditional inversion methods. Results demonstrate that the quantum-inspired hybrid model produces realistic and accurate subsurface velocity images under challenging conditions involving weak seismic signals and limited observations, underscoring its potential to accelerate seismic inversion workflows and enhance reliability in CO₂ storage monitoring. This work makes a substantive contribution to the emerging field of quantum-inspired geophysical inversion and advances the tools available for environmental management and climate change mitigation. |
| Keywords | CO₂ storage monitoring, time-lapse seismic data, full-waveform inversion, quantum neural network |