Intra-Week Dispatch
The dispatch module provides greedy allocation of weekly plant discharge across intra-week price segments to maximise revenue.
Greedy Dispatch Algorithm
Given a plant's total weekly discharge (from the SDDP LP), the greedy dispatch concentrates production in the highest-price segments:
- Sort segments by price (descending)
- For each segment (highest price first):
- Allocate as much discharge as possible (limited by plant capacity per segment and remaining weekly total)
- Compute production via PQ curve at that flow rate
- Accumulate revenue = price × production × hours
For a concave PQ curve, this greedy allocation is optimal — there is no benefit to spreading discharge across segments when prices differ.
Capacity Factor Effect
- Full capacity (e.g., TORPA at 41 m³/s): all 42 segments are used at max power regardless of price ordering → zero dispatch uplift.
- Partial capacity (e.g., 50% load): plant runs in only the top 21 highest-price segments → 13-24% revenue uplift over mean-price dispatch.
API
# Single plant dispatch
(revenue, production_mwh) = greedy_dispatch(
weekly_discharge_mm3, # Mm³/week from the LP
segment_prices, # [(hours, price), ...] for this week
plant, # Plant with PQ curve
conv_weekly, # m³/s → Mm³/week conversion
)
# All plants in a cascade
result = cascade_greedy_dispatch(
plant_discharges, # Dict{plant_id => Mm³/week}
segment_prices,
plant_list,
conv_weekly,
)
# result: Dict{plant_id => (revenue, production_mwh)}Nested Decomposition
The dispatch module enables the nested decomposition strategy:
- SDDP (outer): simple weekly LP with scalar price → fast cuts
- Dispatch (inner): greedy allocation of weekly discharge → actual revenue from intra-week price variation
The dispatch callback is integrated into the SDDP forward pass via the dispatch_fn parameter to solve_sddp. This ensures the upper bound reflects actual dispatch-achievable revenue.
See the SDDP Theory page for details on the multi-cut formulation.