Electricity price forecasting for the day-ahead market in Croatia
Date Issued
2022
Author(s)
Kosteski, Kristijan
Dedinec, Aleksandra
Abstract
This paper presents a Machine Learning model,
specifically a Neural Network, to forecast the prices of the
Croatian day ahead power exchange. The price forecasting is of
great importance for building and maintaining the smart energy
grids that are a part of Smart Cities around the globe. Knowing
the price of electricity can make a switch to renewable energy
sources easier to predict and maintain so that it is available to
the citizens of the city.
The day ahead market requests bids for the next day for all
of the 24 hours in that day. From all of the participants’ bids, a
price is calculated where supply meets the demand and this is
the price that the models try to predict without using the bids
from the participants. Predicting the price of electricity for
markets that produce a big part of its electricity from renewable
sources can be unpredictable and cause a lot of problems for
researchers. Croatia is trying to increase production of
electricity from renewable sources in the following years, which
will make the prices more unpredictable.
The data used is collected for the period from 11.02.2016 to
24.08.2021. Multiple Neural Network structures were explored,
but the best one was a Neural Network with 2 hidden layers with
11 neurons each. The testing period was the last 20 % of the data
as the training and testing data was split 80:20 and some years
were dropped for better results.
The results for the model were evaluated using multiple
evaluation metrics, most of which being error rates.
specifically a Neural Network, to forecast the prices of the
Croatian day ahead power exchange. The price forecasting is of
great importance for building and maintaining the smart energy
grids that are a part of Smart Cities around the globe. Knowing
the price of electricity can make a switch to renewable energy
sources easier to predict and maintain so that it is available to
the citizens of the city.
The day ahead market requests bids for the next day for all
of the 24 hours in that day. From all of the participants’ bids, a
price is calculated where supply meets the demand and this is
the price that the models try to predict without using the bids
from the participants. Predicting the price of electricity for
markets that produce a big part of its electricity from renewable
sources can be unpredictable and cause a lot of problems for
researchers. Croatia is trying to increase production of
electricity from renewable sources in the following years, which
will make the prices more unpredictable.
The data used is collected for the period from 11.02.2016 to
24.08.2021. Multiple Neural Network structures were explored,
but the best one was a Neural Network with 2 hidden layers with
11 neurons each. The testing period was the last 20 % of the data
as the training and testing data was split 80:20 and some years
were dropped for better results.
The results for the model were evaluated using multiple
evaluation metrics, most of which being error rates.
Subjects
File(s)![Thumbnail Image]()
Loading...
Name
CIIT_2022_11.pdf
Size
709.91 KB
Format
Adobe PDF
Checksum
(MD5):fa4c9ee5227ae0ae7d168ab0404aa611