|
| 1 | +# Requirements: Grid Guardian |
| 2 | + |
| 3 | +**Defined:** 2026-01-27 |
| 4 | +**Core Value:** Detect anomalies in energy distribution networks without labeled data, using physics constraints and self-play learned patterns on graph-structured grid data |
| 5 | + |
| 6 | +## v1 Requirements |
| 7 | + |
| 8 | +Requirements for this milestone. Each maps to roadmap phases. |
| 9 | + |
| 10 | +### GNN Verifier Architecture |
| 11 | + |
| 12 | +- [ ] **GNN-01**: Build PyTorch Geometric graph from SSEN feeder/substation topology |
| 13 | +- [ ] **GNN-02**: Implement spatial-temporal GNN architecture (GAT/GraphSAGE + GRU) |
| 14 | +- [ ] **GNN-03**: Replace MLP Verifier with GNN Verifier in self-play training loop |
| 15 | + |
| 16 | +### Self-Play Enhancement |
| 17 | + |
| 18 | +- [ ] **SELF-01**: Implement graph-aware Proposer generating topology-respecting scenarios |
| 19 | +- [ ] **SELF-02**: Scenarios cascade through neighbors (COLD_SNAP, OUTAGE propagation) |
| 20 | + |
| 21 | +### Three-Layer Ensemble |
| 22 | + |
| 23 | +- [ ] **ENS-01**: Integrate GNN learned detector as Layer 2 with physics constraints |
| 24 | +- [ ] **ENS-02**: Implement cascade early exit logic (physics violations skip GNN) |
| 25 | + |
| 26 | +### Evaluation |
| 27 | + |
| 28 | +- [ ] **EVAL-01**: Implement precision/recall/F1 evaluation against baselines |
| 29 | +- [ ] **EVAL-02**: Compare GNN Verifier vs IsolationForest, Autoencoder, DecompositionAnomalyDetector |
| 30 | +- [ ] **EVAL-03**: Measure physics compliance rate of detected anomalies |
| 31 | + |
| 32 | +## v2 Requirements |
| 33 | + |
| 34 | +Deferred to future milestone. Tracked but not in current roadmap. |
| 35 | + |
| 36 | +### Uncertainty Quantification |
| 37 | + |
| 38 | +- **UQ-01**: Monte Carlo Dropout for epistemic uncertainty |
| 39 | +- **UQ-02**: Conformal prediction for calibrated prediction intervals |
| 40 | + |
| 41 | +### External Data Integration |
| 42 | + |
| 43 | +- **DATA-01**: Elexon BMRS API integration for real-time UK grid data |
| 44 | +- **DATA-02**: Live demonstration capability for FYP presentation |
| 45 | + |
| 46 | +### Self-Play Enhancement |
| 47 | + |
| 48 | +- **SELF-03**: Verifier weakness targeting (Proposer conditions on struggles) |
| 49 | + |
| 50 | +### Academic Deliverables |
| 51 | + |
| 52 | +- **ACAD-01**: FYP report following Aston University format |
| 53 | +- **ACAD-02**: Presentation/viva preparation |
| 54 | + |
| 55 | +## Out of Scope |
| 56 | + |
| 57 | +Explicitly excluded. Documented to prevent scope creep. |
| 58 | + |
| 59 | +| Feature | Reason | |
| 60 | +|---------|--------| |
| 61 | +| Real-time production deployment | Academic demonstration sufficient per PROJECT.md | |
| 62 | +| Multi-country grid support | UK-only scope with SSEN data | |
| 63 | +| Ground-truth anomaly labels | Defeats self-play purpose; use physics validation | |
| 64 | +| Forecasting-based detection | Proven to fail in v1-v3 experiments | |
| 65 | +| Mobile/web frontend | CLI and notebooks sufficient for FYP | |
| 66 | +| Proprietary data sources | Public APIs and datasets only | |
| 67 | + |
| 68 | +## Traceability |
| 69 | + |
| 70 | +Which phases cover which requirements. Updated during roadmap creation. |
| 71 | + |
| 72 | +| Requirement | Phase | Status | |
| 73 | +|-------------|-------|--------| |
| 74 | +| GNN-01 | TBD | Pending | |
| 75 | +| GNN-02 | TBD | Pending | |
| 76 | +| GNN-03 | TBD | Pending | |
| 77 | +| SELF-01 | TBD | Pending | |
| 78 | +| SELF-02 | TBD | Pending | |
| 79 | +| ENS-01 | TBD | Pending | |
| 80 | +| ENS-02 | TBD | Pending | |
| 81 | +| EVAL-01 | TBD | Pending | |
| 82 | +| EVAL-02 | TBD | Pending | |
| 83 | +| EVAL-03 | TBD | Pending | |
| 84 | + |
| 85 | +**Coverage:** |
| 86 | +- v1 requirements: 10 total |
| 87 | +- Mapped to phases: 0 |
| 88 | +- Unmapped: 10 |
| 89 | + |
| 90 | +--- |
| 91 | +*Requirements defined: 2026-01-27* |
| 92 | +*Last updated: 2026-01-27 after initial definition* |
0 commit comments