Modeling Electricity Prices in the German Energy market - with Applications to Renewables
(2022) DABN01 20221Department of Economics
- Abstract
- Recent focus on the negative effects of climate change has amplified the importance of renewable sources of energy for electricity generation. The contribution of renewables to the energy mix is growing steadily with profound effects on the price of electricity and implications for market participants. In this paper, we employed a similar model to that developed in (Green, 2015) to model the very important shaping vector of the Hourly Price Forward Curve (HPFC) in the German electricity market that is dependent on solar and wind sources of renewable energy. We trained our model using Artificial Neural Networks. However, instead of using price weights as our response variable, we used deviations from the mean to model the shape of the HPFC.... (More)
- Recent focus on the negative effects of climate change has amplified the importance of renewable sources of energy for electricity generation. The contribution of renewables to the energy mix is growing steadily with profound effects on the price of electricity and implications for market participants. In this paper, we employed a similar model to that developed in (Green, 2015) to model the very important shaping vector of the Hourly Price Forward Curve (HPFC) in the German electricity market that is dependent on solar and wind sources of renewable energy. We trained our model using Artificial Neural Networks. However, instead of using price weights as our response variable, we used deviations from the mean to model the shape of the HPFC. We also included calendar information as a variable in the model. We tested the effects of renewables on the shape vector with scenarios of a 15% increase and a 15% decrease in wind and solar generation. Our model indicates that a 15% increase in renewable generation reduces the average price of electricity while a 15% decrease leads to an increase in price. This finding is consistent with the literature and in line with our intuition and proves the existence of a merit order effect. Additionally, we trained a model for short-term price forecasting using a combination of Light Gradient Boosting Machine (LightGBM) and Artificial Neural Networks (ANN) in a two stage forecasting scheme. We used the LightGBM to identify the spike prices and then train both spike prices and normal prices separately using ANN. The predicted spike prices are added to the predicted normal prices. This approach performed better than just training an ANN on the original dataset without separately training the spike prices. We observed that a variable selection using LassoNet did not include both solar and wind generation as important variables for short term normal price prediction, but did include both variables for spike price forecasting. (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/9084692
- author
- Jin, Kaisen LU and Azuka, Chibuzo LU
- supervisor
- organization
- course
- DABN01 20221
- year
- 2022
- type
- H1 - Master's Degree (One Year)
- subject
- keywords
- Hourly Price Forward Curves, Power Market, Renewables, Electricity spot prices, Day-ahead market
- language
- English
- id
- 9084692
- date added to LUP
- 2022-06-08 12:50:58
- date last changed
- 2022-06-08 12:50:58
@misc{9084692,
abstract = {{Recent focus on the negative effects of climate change has amplified the importance of renewable sources of energy for electricity generation. The contribution of renewables to the energy mix is growing steadily with profound effects on the price of electricity and implications for market participants. In this paper, we employed a similar model to that developed in (Green, 2015) to model the very important shaping vector of the Hourly Price Forward Curve (HPFC) in the German electricity market that is dependent on solar and wind sources of renewable energy. We trained our model using Artificial Neural Networks. However, instead of using price weights as our response variable, we used deviations from the mean to model the shape of the HPFC. We also included calendar information as a variable in the model. We tested the effects of renewables on the shape vector with scenarios of a 15% increase and a 15% decrease in wind and solar generation. Our model indicates that a 15% increase in renewable generation reduces the average price of electricity while a 15% decrease leads to an increase in price. This finding is consistent with the literature and in line with our intuition and proves the existence of a merit order effect. Additionally, we trained a model for short-term price forecasting using a combination of Light Gradient Boosting Machine (LightGBM) and Artificial Neural Networks (ANN) in a two stage forecasting scheme. We used the LightGBM to identify the spike prices and then train both spike prices and normal prices separately using ANN. The predicted spike prices are added to the predicted normal prices. This approach performed better than just training an ANN on the original dataset without separately training the spike prices. We observed that a variable selection using LassoNet did not include both solar and wind generation as important variables for short term normal price prediction, but did include both variables for spike price forecasting.}},
author = {{Jin, Kaisen and Azuka, Chibuzo}},
language = {{eng}},
note = {{Student Paper}},
title = {{Modeling Electricity Prices in the German Energy market - with Applications to Renewables}},
year = {{2022}},
}