Dokka Watercourse — Case Study
The Dokka watercourse is the primary validation case, used to compare the Julia SDDP solver against the commercial ProdRisk system and a Python SDDP implementation.
Topology
M802 (DOKKFLØYVATNET) ← main reservoir, 251.8 Mm³
│
├── P802 (TORPA) ← 160 MW, head 465 m
│ │
│ ▼
│ M800 (KJØLJUA) ← buffer reservoir, 0.5 Mm³
│ │
│ ├── P800 (DOKKA) ← 48 MW, head 133 m
│ │ │
│ │ ▼
│ │ SINK (discharge)
│ │
│ └── BY800 (bypass, min 4.75 m³/s env flow)
│ │
│ ▼
│ M801 (junction) → P801 (0.5 MW) → SINK
│
└── SP802/BY802 (spillway/bypass) → M800Data
- 37 weather years (1984–2020): weekly inflows for M800 and M802
- 37 price scenarios: weekly NOK/MWh
- 42×4h price segments: intra-week price variation per scenario
- 220 stages (~4.2 years) starting week 41 (October)
- 7 Markov price states fitted from all price data
Key Physical Features
- Buffer reservoir M800: 0.5 Mm³, fixed at min_volume in the LP. Water passes through instantly — TORPA discharge enters M800 and exits via DOKKA in the same time step.
- Cascade energy: water from M802 generates power at TORPA (1139 MWh/Mm³) AND DOKKA (304 MWh/Mm³) = 1443 MWh/Mm³ total.
- Min environmental flow: 4.75 m³/s bypass from M800 (BY800).
- Volume constraint: M802 max 210 Mm³ (vs physical 251.8 Mm³).
Results
| Solver | Revenue | Rev gap | Volume | Vol gap | Runtime |
|---|---|---|---|---|---|
| Julia SDDP (full segment) | 147.6M | 6.0% | 134.2 | 3.8% | 96 min |
| Python SDDP (42×4h) | 148.3M | 5.6% | 127.3 | 1.6% | 280s |
| ProdRisk (8-core) | 157.1M | — | 129.3 | — | 890s |
Running
# Full segment strategy (20 forward scenarios, 12 iterations)
julia --project=HydroRunner HydroRunner/bin/run_case.jl cases/dokka
# With nested decomposition
julia --project=HydroRunner HydroRunner/bin/run_case.jl cases/dokka --nested-decomp
# Charts saved to cases/dokka/output/Config
resolution: weekly
n_stages: 220
n_iterations: 12
n_forward_scenarios: 20
convergence_tol: 0.03
price_model:
type: markov
n_states: 7
initial_price_state: 2
start_week: 41