Stimulation Initiative for European Neural Applications
Esprit Project 9811
Short-Term Load Forecast for German Power Utility
Company background
Bayernwerke AG, located in Bavaria, Germany, is one of the largest
utility in Germany. Precise Load Forecast solutions using ANN are
required for optimizing their energy management system.
The problem
A main concern in Power Distribution is the prediction of the future
electrical load requirements for a given region in order to plan the
power generation optimally with minimal loss and costs. The load
required by a region is affected by factors including the general trend,
time dependent factors (seasons, holidays) weather, and special events.
This predicted power system load is the quantity for planning sufficient
generation, spinning reserve and standby reserve.
Neural network application
The idea was to inprove the precision of the forecast by applying neural
network technologies and create a modular system. The forecast is a
short-term prediction which makes possible predictions for the next 24
hours on a hourly basis (or more frequently if data is available). The
forecast can be done for each day up to three weeks in the future.
Benefits
When compared to conventional methods , this method based on neural
networks and modular system architecture performes the best. The
prediction error ranges between 1 - 2 % whereas multiple regressions,
for instance, gives around 2 - 3 % error. The difference in terms of
absolute power is in the terms of 10 MW. Furthermore, the time required
for calculating a forecast is less than one second and forecasts for
special days (legal and religuous holidays) are possible.
Generalization
This application proves training to be possible for a neural network to
forecasting tasks. It can be trained to forecast the future electrical
load requirements for certain regions, taking into consideration
changing factors with great significance for the electricity demand,
such as time dependend factors.
Contact person
Mr. Selmayr,
Bayernwerke AG.
Bayernwerkstraße 39,
D-85757 Karlsfeld.
Phone: +49 8131 88 1
Fax: +49 89 1254 330.
References for further reading
Yurtsever, K.: Neural networks, the New Technology Applied to Power
System Control. In: Proceedings EUROSIM 95, Vienna, Austria.