Andreas Kamvissis, presenterar sitt examensarbete  

Scenario-based stochastic optimization of the Nord Pool bid curves-
 modelling the electricity spot price


Every day the hourly electricity spot price for the next days 24 hours
is set at the Nordic Power Exchange, Nord Pool. In the procedure
setting these prices every power producer must submit a set of so
called bid curves, one for each of the next days 24 hours. A bid curve
consists of several price-volume pairs describing what volumes the
power producer is willing to produce at which spot prices. Simplified,
the process ends here by the matching of these aggregated bid curves
with the electricity demand in this way setting the hourly electricity
spot price for the next days 24 hours. In this thesis we will evaluate
BidOpt, a planning tool for the short term optimization of the power
production in hydropower systems. BidOpt uses scenarios of the hourly
electricity spot price in order to perform a stochastic optimization
of the bid curves to be submitted to Nord Pool. Once the price is set,
BidOpt is used for optimizing the profit from the power production in
the hydropower system, incorporating the dynamics of the river
generating an optimal production plan. In order to use BidOpt, a
technique for modelling the hourly electricity spot price must be
developed, using this model for generation of the necessary spot price
scenarios. In this thesis we have tried three different approaches in
our modelling. We started with a model based on the daily returns of
the hourly spot price, estimated with linear ARMA- and
ARMAX-models. The latter used the daily returns of the electricity
demand as an external signal. The next model was a Vector AR-model,
trying to capture the interdepencies between the days and the hours of
the spot price. Our final model, which also became our best model,
directly used the fact that it is the electricity demand that sets the
hourly spot price and that this demand could be predicted very well on
a short term basis. The model was recursive, only using the most
recent values estimating the dependency between the electricity demand
and the electricity spot price. The scenarios were generated from this
model by simulating an ARMA-model estimated on the residuals of the
recursive model. In addition a mean-value scenario was generated in
order to compare a multiple-scenario based approach with a
single-scenario based dito when optimizing the bid curves in
BidOpt. BidOpt was run on the period of late February to middle March
in 2002. The bid curves from BidOpt showed better characteristica
using the multiple scenarios, bidding more delicately on more
different spot prices. The optimal production planning demonstrated
the hydropower systems dynamics in a fine way, laying out the
aggregated production on the periods with higher spot price. Although
the production was higher than optimal throughout the whole period,
which was explained by a too low valuing of the water in the water
reservoirs. This illustrated the importance of using BidOpt together
with a planning tool with a longer horizon, specifying the volumes to
be produced as input to the BidOpt-model. Even under this constraint
though we would still be optimizing the production, in this way
maximizing the profit of the hydropower system.