Deep Learning for Short-Term Load Forecasting—Industrial Consumer Case Study
In the current trend of consumption, electricity consumption will become a very high cost for the end-users. Consumers acquire energy from suppliers who use short, medium, and long-term forecasts to place bids in the power market. This study offers a detailed analysis of relevant literature and prop...
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
MDPI AG
2021-10-01
|
| Series: | Applied Sciences |
| Subjects: | |
| Online Access: | https://www.mdpi.com/2076-3417/11/21/10126 |
| _version_ | 1797512774509658112 |
|---|---|
| author | Stefan Ungureanu Vasile Topa Andrei Cristinel Cziker |
| author_facet | Stefan Ungureanu Vasile Topa Andrei Cristinel Cziker |
| author_sort | Stefan Ungureanu |
| collection | DOAJ |
| description | In the current trend of consumption, electricity consumption will become a very high cost for the end-users. Consumers acquire energy from suppliers who use short, medium, and long-term forecasts to place bids in the power market. This study offers a detailed analysis of relevant literature and proposes a deep learning methodology for forecasting industrial electric usage for the next 24 h. The hourly load curves forecasted are from a large furniture factory. The hourly data for one year is split into training (80%) and testing (20%). The algorithms use the previous two weeks of hourly consumption and exogenous variables as input in the deep neural networks. The best results prove that deep recurrent neural networks can retain long-term dependencies in high volatility time series. Gated recurrent units (GRU) obtained the lowest mean absolute percentage error of <inline-formula><math xmlns="http://www.w3.org/1998/Math/MathML" display="inline"><semantics><mrow><mn>4.82</mn><mo>%</mo></mrow></semantics></math></inline-formula> for the testing period. The GRU improves the forecast by <inline-formula><math xmlns="http://www.w3.org/1998/Math/MathML" display="inline"><semantics><mrow><mn>6.23</mn><mo>%</mo></mrow></semantics></math></inline-formula> compared to the second-best algorithm implemented, a combination of GRU and Long short-term memory (LSTM). From a practical perspective, deep learning methods can automate the forecasting processes and optimize the operation of power systems. |
| first_indexed | 2024-03-10T06:06:26Z |
| format | Article |
| id | doaj.art-4ff471bcf9b84a75918e0375669cd943 |
| institution | Directory Open Access Journal |
| issn | 2076-3417 |
| language | English |
| last_indexed | 2024-03-10T06:06:26Z |
| publishDate | 2021-10-01 |
| publisher | MDPI AG |
| record_format | Article |
| series | Applied Sciences |
| spelling | doaj.art-4ff471bcf9b84a75918e0375669cd9432023-11-22T20:28:11ZengMDPI AGApplied Sciences2076-34172021-10-0111211012610.3390/app112110126Deep Learning for Short-Term Load Forecasting—Industrial Consumer Case StudyStefan Ungureanu0Vasile Topa1Andrei Cristinel Cziker2Department of Electric Power Systems and Management, Technical University of Cluj-Napoca, 400027 Cluj-Napoca, RomaniaDepartment of Electrotechnics and Measurements, Technical University of Cluj-Napoca, 400027 Cluj-Napoca, RomaniaDepartment of Electric Power Systems and Management, Technical University of Cluj-Napoca, 400027 Cluj-Napoca, RomaniaIn the current trend of consumption, electricity consumption will become a very high cost for the end-users. Consumers acquire energy from suppliers who use short, medium, and long-term forecasts to place bids in the power market. This study offers a detailed analysis of relevant literature and proposes a deep learning methodology for forecasting industrial electric usage for the next 24 h. The hourly load curves forecasted are from a large furniture factory. The hourly data for one year is split into training (80%) and testing (20%). The algorithms use the previous two weeks of hourly consumption and exogenous variables as input in the deep neural networks. The best results prove that deep recurrent neural networks can retain long-term dependencies in high volatility time series. Gated recurrent units (GRU) obtained the lowest mean absolute percentage error of <inline-formula><math xmlns="http://www.w3.org/1998/Math/MathML" display="inline"><semantics><mrow><mn>4.82</mn><mo>%</mo></mrow></semantics></math></inline-formula> for the testing period. The GRU improves the forecast by <inline-formula><math xmlns="http://www.w3.org/1998/Math/MathML" display="inline"><semantics><mrow><mn>6.23</mn><mo>%</mo></mrow></semantics></math></inline-formula> compared to the second-best algorithm implemented, a combination of GRU and Long short-term memory (LSTM). From a practical perspective, deep learning methods can automate the forecasting processes and optimize the operation of power systems.https://www.mdpi.com/2076-3417/11/21/10126machine learningdeep learningshort-term forecastingindustrial electricity load |
| spellingShingle | Stefan Ungureanu Vasile Topa Andrei Cristinel Cziker Deep Learning for Short-Term Load Forecasting—Industrial Consumer Case Study Applied Sciences machine learning deep learning short-term forecasting industrial electricity load |
| title | Deep Learning for Short-Term Load Forecasting—Industrial Consumer Case Study |
| title_full | Deep Learning for Short-Term Load Forecasting—Industrial Consumer Case Study |
| title_fullStr | Deep Learning for Short-Term Load Forecasting—Industrial Consumer Case Study |
| title_full_unstemmed | Deep Learning for Short-Term Load Forecasting—Industrial Consumer Case Study |
| title_short | Deep Learning for Short-Term Load Forecasting—Industrial Consumer Case Study |
| title_sort | deep learning for short term load forecasting industrial consumer case study |
| topic | machine learning deep learning short-term forecasting industrial electricity load |
| url | https://www.mdpi.com/2076-3417/11/21/10126 |
| work_keys_str_mv | AT stefanungureanu deeplearningforshorttermloadforecastingindustrialconsumercasestudy AT vasiletopa deeplearningforshorttermloadforecastingindustrialconsumercasestudy AT andreicristinelcziker deeplearningforshorttermloadforecastingindustrialconsumercasestudy |