Introduction: What is Hydropower Production Planning?
"Produksjonsplanlegging er en oppgave som er meget utfordrende på grunn av at det er et stort og komplekst fysisk produksjonssystem..." — Fosso, Kompendium 1, NTNU 2018
The core problem
A hydropower system converts stored water into electricity. The operator must decide, at every point in time: how much water to release through the turbines?
This sounds simple. It is not — for several interlocking reasons:
1. Water is both fuel and storage
Unlike a gas plant that simply burns fuel as needed, a hydropower plant is also a battery. Water not used now is water available later. The decision to generate today is a decision to not generate tomorrow — and tomorrow the price might be higher or a drought might be coming.
This creates a fundamental trade-off:
Release water now → revenue today, less flexibility later
Hold water back → no revenue now, but option value for futureGetting this trade-off right over a season or a year is the central challenge.
2. The resource is stochastic
Natural inflow (tilsig) varies enormously and unpredictably:
- In Norway, annual inflow varies between 95 and 140 TWh (±25% around the mean)
- Within a year: spring flood, summer drought, autumn rain, winter freeze
- Year-to-year: wet years vs. dry years driven by climate patterns
You cannot plan a hydropower system without explicitly accounting for this uncertainty. This is why stochastic methods dominate the planning literature.
3. The system is physically complex
A real hydropower system is not one reservoir and one turbine. It is a cascade:
- 10–100+ reservoirs at different elevations, connected by tunnels and rivers
- Multiple plants along each river reach
- Water released from one reservoir becomes inflow to the next
- Shared transmission lines with capacity limits
- Environmental regulations on minimum flow
Decisions at one reservoir affect all downstream reservoirs — the system must be optimised jointly, not reservoir by reservoir.
4. The time scales interact
Decisions span four orders of magnitude in time:
| Horizon | Granularity | Question being answered |
|---|---|---|
| 5 years | Weekly | How much water to carry into next winter? |
| 1–2 years | Weekly | What should reservoir levels be this autumn? |
| 1–2 weeks | Hourly | When during the week to generate? |
| Day-ahead | 15-minute | Bid into the spot market? |
These levels are coupled: the short-term schedule must respect the seasonal targets; the seasonal targets must be consistent with the long-term water value.
Why open-source tooling matters
Historically, Norwegian hydropower planning has relied on commercial tools: Vansimtap, PowEl, EMPS — developed by SINTEF/NTNU over decades. They are excellent but:
- Closed source — no visibility into algorithms
- Licensed per user/company — high cost for researchers
- Hard to extend or integrate with modern data pipelines
The goal of this project is to explore whether a modern, open, composable tool suite can cover the same ground — starting from first principles, with clean code and documented algorithms.
The Nordic electricity market context
Norway and the Nordic countries operate a liberalised electricity market (Nord Pool). This adds a market dimension to the physical planning problem:
- Spot price (Elspot) is set hourly by supply/demand clearing
- In a hydro-dominated system like Norway, the hydropower producers are the supply curve — their decisions influence the price they receive
- Large price swings between wet years (low prices) and dry years (high prices) are the norm
The planning problem is therefore not just operational but strategic: when to generate in order to maximise revenue over a year or more, under uncertainty about future prices and future inflow.
What we are building
A suite of Julia packages that together cover the full planning hierarchy:
HydroCore — physical data model (reservoirs, plants, cascades, curves)
WaterValues — stochastic DP: compute the value of stored water
ShortTermPlanning — weekly LP/MIP scheduling via JuMP
HydroSimulation — Monte Carlo simulation using water value tablesWe start at the bottom with HydroCore — the data model that all other packages depend on — and work upwards.
Further reading
- Fosso, O.B. (2018). Produksjonsplanlegging i vannkraftbaserte systemer, Del 1. EVU-kurs ET6003, NTNU. (The primary compendium for this project.)
- Wolfgang, O. et al. (2009). Hydro reservoir handling in Norway before and after deregulation. Energy 34(10).
- Pereira, M.V.F. & Pinto, L.M.V.G. (1991). Multi-stage stochastic optimization applied to energy planning. Mathematical Programming 52.