| # GeoForce v2.0 — Transformation Progress |
|
|
| **Last Updated:** 2026-04-16 |
|
|
| --- |
|
|
| ## Phase Status |
|
|
| | Phase | Name | Status | Started | Completed | |
| |-------|------|--------|---------|-----------| |
| | 1 | Real Data Foundation | Not Started | — | — | |
| | 2 | Real Simulator Setup | Not Started | — | — | |
| | 3 | Model Architecture Upgrade | Not Started | — | — | |
| | 4 | Validation and Benchmarking | Not Started | — | — | |
| | 5 | Platform Integration | Not Started | — | — | |
| | 6 | Publish and Position | Not Started | — | — | |
|
|
| --- |
|
|
| ## Detailed Task Tracking |
|
|
| ### Phase 1: Real Data Foundation |
|
|
| **1A. NREL Brady Hot Springs Dataset** |
| - [ ] Clone NREL/geothermal_osr repository |
| - [ ] Explore data format, understand scenario structure |
| - [ ] Retrain current CNN on Brady data |
| - [ ] Compare results against published USGS benchmark (<3.68% T error) |
| - [ ] Document gaps |
| |
| **1B. Utah FORGE Dataset** |
| - [ ] Download FORGE well log data |
| - [ ] Download 3D geological model |
| - [ ] Map FORGE data structure to GeoForce input requirements |
| |
| **1C. Indonesian Field Parameters** |
| - [ ] Compile parameter table from published papers |
| - [ ] Document permeability ranges for Indonesian volcanic systems |
| - [ ] Document fracture characteristics |
| - [ ] Build scenarios.yaml v2 with realistic Indonesian ranges |
| |
| ### Phase 2: Real Simulator Setup |
| |
| **2A. Simulator Installation** |
| - [ ] Request TOUGH3 license from LBNL |
| - [ ] Install Waiwera (backup, open-source) |
| - [ ] Install PyTOUGH |
| - [ ] Run TOUGH2 standard benchmark problems |
| |
| **2B. Simulation Campaign** |
| - [ ] Write simulation campaign generator script |
| - [ ] Define 3D grid with vertical layering |
| - [ ] Run pilot batch (10 scenarios) |
| - [ ] Estimate total compute time |
| - [ ] Run full campaign (1,000+ scenarios) |
| |
| ### Phase 3: Model Architecture Upgrade |
| |
| - [ ] Implement 3D U-Net architecture |
| - [ ] Design input preprocessing for TOUGH3 outputs |
| - [ ] Design output denormalization pipeline |
| - [ ] Implement upgraded physics loss functions |
| - [ ] Benchmark training time |
| |
| ### Phase 4: Validation and Benchmarking |
| |
| - [ ] Train on Brady OSR data |
| - [ ] Evaluate against USGS published benchmarks |
| - [ ] Build Indonesian validation scenarios |
| - [ ] Run qualitative validation against published field data |
| - [ ] Write v2.0 technical report |
| |
| ### Phase 5: Platform Integration |
| |
| - [ ] Update serve.py for v2.0 model |
| - [ ] Update platform GeoForcePage.tsx |
| - [ ] Add 3D visualization |
| - [ ] Update HuggingFace model card |
| - [ ] Update landing page copy |
| |
| ### Phase 6: Publish and Position |
| |
| - [ ] Submit to Stanford Geothermal Workshop |
| - [ ] Publish v2.0 model on HuggingFace |
| - [ ] LinkedIn announcement |
| - [ ] ITB faculty review outreach |
| - [ ] Pertamina data-sharing proposal |
| |
| --- |
| |
| ## Log |
| |
| ### 2026-04-16 |
| - Created transformation plan |
| - Completed deep research on state-of-the-art |
| - Identified NREL Brady OSR as immediate training data source |
| - Identified TOUGH3/Waiwera as simulator options |
| - Identified 3D U-Net as v2.0 architecture target |
| - Identified GNN (SeqSage) as v3.0 architecture for fracture systems |
| |
