The figure shows observed inflow series from a reservoir, shown for week 1 to week 52.
Models for production planning help power producers create a strategy that considers changes in electricity prices and inflows, i.e., how much water flows into the reservoir. They must then have good methods for calculating how much water enters the water reservoir. These models are used to create different scenarios for how much water is available in the reservoir for power production.
One of the most used inflow models uses data from previous years to calculate how much water will enter the reservoir in the future. This model has a habit of occasionally showing negative trends. That means, the model believes that large amounts of water evaporate out of the reservoir compared to the amount that flows in.
This has a bad effect on the production strategy for the hydropower plant. Another disadvantage of the model is that it does not capture long-term trends well enough, making it so that particularly dry or particularly wet years do not appear clearly in the possible inflow scenarios.
Increased flexibility without negative inflow
In HydroCen, Sintef researcher Siri Mathisen has looked at a new inflow model that removes these negative influences, and therefore provides more precise and therefore better inflow scenarios. It can be used in planning models that require a linear inflow model. The methodology used will also be relevant for other models, such as for example the generation of synthetic wind data series.
By removing negative inflows in hydropower planning, it is possible for the power companies to create strategies where they can plan to produce electricity even if there is little inflow in the reservoirs, which will increase the flexibility of the hydropower plant. It will thus be possible to deliver more power when it is needed the most by producing during hours of high demand rather than hours of low demand.
Guarantees positive inflow
The mathematical model that creates synthetic inflow series can be divided into a part that depends on the previous week's inflow, and noise. The noise component means that randomness plays a role in it. To solve the problem of the negative inflow, one can model the noise from the inflow model using a 3-parameter lognormal probability distribution. By doing this, you get an inflow scenario that is guaranteed not to be negative, but still meets the mathematical requirements of the production planning model.