Full Text Available
Note: Clicking the button above will open the full text document at the original institutional repository in a new window.
Forecasting energy consumption on mines using machine learning techniques
Thesis (MEng)--Stellenbosch University, 2023.
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Thesis |
| Language: | en_ZA |
| Published: |
Stellenbosch : Stellenbosch University
2023
|
| Subjects: | |
| Tags: |
No Tags, Be the first to tag this record!
|
| _version_ | 1867614105198657536 |
|---|---|
| access_status_str | Open Access |
| author | Esterhuizen, Lohani |
| author2 | Schutte, C. S. L. |
| author_browse | Esterhuizen, Lohani Schutte, C. S. L. |
| author_facet | Schutte, C. S. L. Esterhuizen, Lohani |
| author_sort | Esterhuizen, Lohani |
| collection | Thesis |
| dc_rights_str_mv | Stellenbosch University |
| description | Thesis (MEng)--Stellenbosch University, 2023. |
| format | Thesis |
| id | oai:scholar.sun.ac.za:10019.1/128708 |
| institution | Stellenbosch University (South Africa) |
| language | en_ZA |
| last_indexed | 2026-06-10T12:46:44.579Z |
| license_str | Other — see source repository |
| provenance_str_mv | Harvested via OAI-PMH from SUNScholar — Stellenbosch University Repository |
| publishDate | 2023 |
| publishDateRange | 2023 |
| publishDateSort | 2023 |
| publisher | Stellenbosch : Stellenbosch University |
| publisherStr | Stellenbosch : Stellenbosch University |
| record_format | dspace |
| source_str | SUNScholar — Stellenbosch University Repository |
| spelling | oai:scholar.sun.ac.za:10019.1/128708 Forecasting energy consumption on mines using machine learning techniques Esterhuizen, Lohani Schutte, C. S. L. Gous, A. G. S. Stellenbosch University. Faculty of Engineering. Dept. of Industrial Engineering. Energy consumption -- Forecasting -- South Africa Machine learning -- South Africa Power resources -- Management -- South Africa Mining engineering -- South Africa Thesis (MEng)--Stellenbosch University, 2023. ENGLISH ABSTRACT: The governing parties of several countries, including South Africa, have made great efforts in managing energy resources. Not only to limit the expulsion of harmful gases into the atmosphere, but also to prevent the depletion of non-renewable resources. The mining sector is one of the major energy consumers in South Africa. In recent years the efforts to adhere to strict energy management protocols have continued to increase. Efficient energy management in mining facilities not only limits the contribution to global warming and resource depletion but greatly reduces operational costs in a mine. Part of energy management includes energy planning and forecasts. Although much research has gone into forecasting energy consumption, few have done so for the total energy consumption on deep-level mines. Due to the complex nature of mining facilities, there is no distinct relationship between energy consumption and other operational variables in a mine, making current forecasting methods used in the industry irrelevant. Machine Learning (ML) techniques have shown promising results in forecasting energy consumption in systems of complex nature. This study applies four differentMLtechniques to four case-study mines in South Africa to forecast energy consumption at different frequencies. The ML models include Seasonal Autoregressive IntegratedMoving Average (SARIMA), the Holt-Winters seasonalmethod (Holt-Winters), the Feed Forward Neural Network (FFNN) model, and the Long-Short TermMemory (LSTM)model. To analyze the accuracy of the ML models, the %Root Mean Square Error (RMSE) and % Mean Absolute Error (MAE) were determined for each model on each case study mine. The results obtained from each model have been validated by comparing the most accurate ML model for each data frequency to the current forecasting method used in the industry. Results show that the data frequency and the mine analyzed impact the accuracy of the ML models. For data at daily and monthly frequency, the FFNN model yielded the most accurate results on most of the case study mines. The Holt-Winters and LSTM models yield more accurate energy consumption forecasts for weekly and monthly data on the fourmines. The most accurate model for each data frequency on Mine A was compared to the linear regression model with production as an independent variable. The daily, weekly, and monthly forecasts by the ML models were respectively 3.8%, 3.5%, and 6.2% more accurate when comparing the %RMSE of the linear regression model. AFRIKAANS OPSOMMING: Die regerende partye van veelvuldige lande, insluitend Suid-Afrika, het al groot pogings aangewend om energiebronne te bestuur. Nie net om die vrylating van skadelike gasse in die atmosfeer te beperk nie, maar ook omdie uitputting van nie-hernubare hulpbronne te voorkom. Die mynbousektor is een van die grootste energieverbruikers in Suid-Afrika. In onlangse jare het die pogings om aan streng energiebestuurprotokolle te voldoen toegeneem. Doeltreffende energiebestuur in mynboufasiliteite beperk nie net die bydrae tot aardverwarming en hulpbronuitputting nie, maar kan ook operasionele koste in ’n myn aansienlik verminder. Suksesvolle energiebestuur sluit energiebeplanning en vooruitskattings in. Alhoewel baie navorsing reeds gedoen is om energieverbruik vooruit te skat, het min dit gedoen vir die totale energieverbruik op diepvlakmyne. As gevolg van die komplekse aard van mynfasiliteite, is daar geen duidelike verband tussen energieverbruik en ander bedryfsveranderlikes in ’n myn nie, wat huidige vooruitskattingsmetodes wat in industrie gebruik word irrelevant maak. Masjienleer (ML) tegnieke toon belowende resultate in die vooruitskatting van energieverbruik in stelsels van komplekse aard. Hierdie studie pas vier verskillende ML-tegnieke toe op vier gevallestudiemyne in Suid-Afrika om energieverbruik by verskillende frekwensies vooruit te skat. Die ML-modelle sluit in SARIMA, die Holt-Winters, die FFNN-model en die LSTM-model. Om die akkuraatheid van die ML-modelle te ontleed, is die %RMSE en %MAE vir elke model op elke gevallestudiemyn bepaal. Die resultate wat van elke model verkry is, is bekragtig deur die mees akkurateML-model vir elke datafrekwensie te vergelyk met die huidige voorspellingsmetode wat in industrie gebruik word. Resultate toon dat die datafrekwensie en die myn self ’n impak het op die akkuraatheid van die ML-modelle. Vir data teen daaglikse en maandelikse frekwensie, het die FFNN-model die mees akkurate resultate op meeste van die gevallestudiemyne opgelewer. Die Holt-Winters- en LSTMmodelle lewer meer akkurate energieverbruik vooruitskattings vir weeklikse en maandelikse data oor die vier myne. Die mees akkurate model vir elke datafrekwensie op Myn A is vergelyk met die lineere regressiemodel met produksie as ’n onafhanklike veranderlike. Die daaglikse, weeklikse en maandelikse voorspellings deur die ML-modelle was onderskeidelik 3.8%, 3.5% en 6.2% meer akkuraat wanneer die %RMSE vergelyk wordmet die lineere regressiemodel. Masters 2023-02-13T11:39:07Z 2023-11-16T09:08:35Z 2023-02-13T11:39:07Z 2023-11-16T09:08:35Z 2023-03 Thesis https://scholar.sun.ac.za/handle/10019.1/128708 en_ZA Stellenbosch University xiii, 110 pages : illustrations application/pdf Stellenbosch : Stellenbosch University |
| spellingShingle | Energy consumption -- Forecasting -- South Africa Machine learning -- South Africa Power resources -- Management -- South Africa Mining engineering -- South Africa Esterhuizen, Lohani Forecasting energy consumption on mines using machine learning techniques |
| title | Forecasting energy consumption on mines using machine learning techniques |
| title_full | Forecasting energy consumption on mines using machine learning techniques |
| title_fullStr | Forecasting energy consumption on mines using machine learning techniques |
| title_full_unstemmed | Forecasting energy consumption on mines using machine learning techniques |
| title_short | Forecasting energy consumption on mines using machine learning techniques |
| title_sort | forecasting energy consumption on mines using machine learning techniques |
| topic | Energy consumption -- Forecasting -- South Africa Machine learning -- South Africa Power resources -- Management -- South Africa Mining engineering -- South Africa |
| url | https://scholar.sun.ac.za/handle/10019.1/128708 |
| work_keys_str_mv | AT esterhuizenlohani forecastingenergyconsumptiononminesusingmachinelearningtechniques |