Pretražite po imenu i prezimenu autora, mentora, urednika, prevoditelja

Napredna pretraga

Pregled bibliografske jedinice broj: 1283175

ELECTRICITY PRICES FORECASTING ON THE DAY-AHEAD MARKET – PERFORMANCE COMPARISON OF SELECTED MACHINE LEARNING MODELS


Jurčević, Jura; Vlah Jerić, Silvija; Zoričić, Davor
ELECTRICITY PRICES FORECASTING ON THE DAY-AHEAD MARKET – PERFORMANCE COMPARISON OF SELECTED MACHINE LEARNING MODELS // Proceedings of 12th International Scientific Symposium REGION, ENTREPRENEURSHIP, DEVELOPMENT / Leko Šimić, Mirna (ur.).
Osijek: Ekonomski fakultet Sveučilišta Josipa Jurja Strossmayera u Osijeku, 2023. str. 591-602 (predavanje, međunarodna recenzija, cjeloviti rad (in extenso), znanstveni)


CROSBI ID: 1283175 Za ispravke kontaktirajte CROSBI podršku putem web obrasca

Naslov
ELECTRICITY PRICES FORECASTING ON THE DAY-AHEAD MARKET – PERFORMANCE COMPARISON OF SELECTED MACHINE LEARNING MODELS

Autori
Jurčević, Jura ; Vlah Jerić, Silvija ; Zoričić, Davor

Vrsta, podvrsta i kategorija rada
Radovi u zbornicima skupova, cjeloviti rad (in extenso), znanstveni

Izvornik
Proceedings of 12th International Scientific Symposium REGION, ENTREPRENEURSHIP, DEVELOPMENT / Leko Šimić, Mirna - Osijek : Ekonomski fakultet Sveučilišta Josipa Jurja Strossmayera u Osijeku, 2023, 591-602

Skup
12th International Scientific Symposium REGION, ENTREPRENEURSHIP, DEVELOPMENT

Mjesto i datum
Osijek, Hrvatska, 15.06.2023. - 16.06.2023

Vrsta sudjelovanja
Predavanje

Vrsta recenzije
Međunarodna recenzija

Ključne riječi
forecast accuracy, market volatility, predictor importance

Sažetak
This paper examines performance of selected frequently used machine learning models for the purpose of forecasting the day-ahead prices in the German power market. Data for 2017 is used as the initial training data set in order to forecast the day-ahead prices up to one month ahead. The forecasting process is repeated by rolling the training set one month forward while keeping its length fixed. Thus, a moving window of training data which serves as a base for each of the subsequent forecasts is obtained. By employing this methodological framework forecasts for the 2018 and for the 2021 are made. This allows the comparison of forecasting performance in the period of low market volatility (2018) with the high market volatility observed in the 2021. The following machine learning methods are employed: Random Forest, Lasso, XGBoost and Support Vector Machine. The accuracy of forecasts for each method is assessed in both periods analysed based on the root-mean-squared error (RMSE). Computational complexity in terms of software running time needed to produce the forecasts is also reported. Research findings show that Lasso model seems to produce more accurate forecasts than the other three analysed models in the period of increased volatility (2021) which, combined with the significantly lower computational complexity, adds to its superiority. However, in the less volatile period (2018) accuracy of Lasso model was by far the worst. Support Vector Machine model’s forecasts dominated other models in this period but overall performance is overshadowed by the model’s computational complexity. Random Forest and XGBoost models seem to exhibit more robust performance with forecast accuracy deteriorating slightly under adverse market conditions while also performing similarly well. Importance analysis points out wind forecast and 24-hour lagged day-ahead price as the most important external predictor variables, while the load forecast completes the top three list.

Izvorni jezik
Engleski

Znanstvena područja
Ekonomija, Interdisciplinarne društvene znanosti



POVEZANOST RADA


Projekti:
EK--KK.01.1.1.04.0034 - Umreženi stacionarni baterijski spremnici energije (USBSE) (Pandžić, Hrvoje, EK ) ( CroRIS)

Ustanove:
Ekonomski fakultet, Zagreb

Poveznice na cjeloviti tekst rada:

www.efos.unios.hr

Citiraj ovu publikaciju:

Jurčević, Jura; Vlah Jerić, Silvija; Zoričić, Davor
ELECTRICITY PRICES FORECASTING ON THE DAY-AHEAD MARKET – PERFORMANCE COMPARISON OF SELECTED MACHINE LEARNING MODELS // Proceedings of 12th International Scientific Symposium REGION, ENTREPRENEURSHIP, DEVELOPMENT / Leko Šimić, Mirna (ur.).
Osijek: Ekonomski fakultet Sveučilišta Josipa Jurja Strossmayera u Osijeku, 2023. str. 591-602 (predavanje, međunarodna recenzija, cjeloviti rad (in extenso), znanstveni)
Jurčević, J., Vlah Jerić, S. & Zoričić, D. (2023) ELECTRICITY PRICES FORECASTING ON THE DAY-AHEAD MARKET – PERFORMANCE COMPARISON OF SELECTED MACHINE LEARNING MODELS. U: Leko Šimić, M. (ur.)Proceedings of 12th International Scientific Symposium REGION, ENTREPRENEURSHIP, DEVELOPMENT.
@article{article, author = {Jur\v{c}evi\'{c}, Jura and Vlah Jeri\'{c}, Silvija and Zori\v{c}i\'{c}, Davor}, editor = {Leko \v{S}imi\'{c}, M.}, year = {2023}, pages = {591-602}, keywords = {forecast accuracy, market volatility, predictor importance}, title = {ELECTRICITY PRICES FORECASTING ON THE DAY-AHEAD MARKET – PERFORMANCE COMPARISON OF SELECTED MACHINE LEARNING MODELS}, keyword = {forecast accuracy, market volatility, predictor importance}, publisher = {Ekonomski fakultet Sveu\v{c}ili\v{s}ta Josipa Jurja Strossmayera u Osijeku}, publisherplace = {Osijek, Hrvatska} }
@article{article, author = {Jur\v{c}evi\'{c}, Jura and Vlah Jeri\'{c}, Silvija and Zori\v{c}i\'{c}, Davor}, editor = {Leko \v{S}imi\'{c}, M.}, year = {2023}, pages = {591-602}, keywords = {forecast accuracy, market volatility, predictor importance}, title = {ELECTRICITY PRICES FORECASTING ON THE DAY-AHEAD MARKET – PERFORMANCE COMPARISON OF SELECTED MACHINE LEARNING MODELS}, keyword = {forecast accuracy, market volatility, predictor importance}, publisher = {Ekonomski fakultet Sveu\v{c}ili\v{s}ta Josipa Jurja Strossmayera u Osijeku}, publisherplace = {Osijek, Hrvatska} }




Contrast
Increase Font
Decrease Font
Dyslexic Font