Hydrodynamic prediction of geothermal water injection using machine learning

Research output: ThesisMaster's Thesis

Standard

Bibtex - Download

@mastersthesis{37a4db00166b4d16a09a877494aed97d,
title = "Hydrodynamic prediction of geothermal water injection using machine learning",
abstract = "To ensure efficient and sustainable reinjection of water into geothermal reservoirs, a proper prediction of the hydrodynamics of the geothermal system is necessary. However, a deep understanding of the reservoir's properties is typically required. This study investigates the use of machine learning to forecast the wellhead pressure of geothermal injection wells using only historical data of wellhead pressure, fluid injection rates, and temperatures. These datasets will be used to construct several machine learning (ML) algorithms, including Multiple Linear Regression (MLR), Random Forest (RF), eXtreme Gradient Boosting (XGBoost), Support Vector Machine (SVM), as well as Artificial Neural Network (ANN) such as Multilayer Perceptron (MLP) and Long Short-Term Memory (LSTM). Water injection data from four injection wells underwent rigorous data cleaning and outlier removal to ensure the fidelity of the results. The performance of the six algorithms was then validated and examined using metrics like the coefficient of determination (R squared), the root mean squared error (RMSE), and the symmetric mean absolute percentage error (sMAPE). Results show that LSTM outperforms other tested algorithms with a sMAPE of 8.02%, followed closely by XGBoost with 8.28% and MLP with 8.55%. RF and SVM achieved a sMAPE of 10.18% and 10.48%, respectively, while the MLR, used in this thesis as a way to better gauge the upper limit of the models, got a sMAPE of 20.56%. It's essential to note that these results are subjective on the specific dataset used in this study, as well as the particular preprocessing, feature engineering, and hyperparameter tuning. Future studies might consider implementing other ML techniques or hybrid learning models, offering valuable insights for geothermal optimization. As ML becomes increasingly important in the geothermal energy sector, such exploration is of significant interest.",
keywords = "Maschinelles Lernen, Geothermie, Wassereinspritzung, Bohrkopfdruck, machine learning, geothermal energy, water injection, well head pressure",
author = "Redha Hamel",
note = "no embargo",
year = "2023",
doi = "10.34901/mul.pub.2023.201",
language = "English",
school = "Montanuniversitaet Leoben (000)",

}

RIS (suitable for import to EndNote) - Download

TY - THES

T1 - Hydrodynamic prediction of geothermal water injection using machine learning

AU - Hamel, Redha

N1 - no embargo

PY - 2023

Y1 - 2023

N2 - To ensure efficient and sustainable reinjection of water into geothermal reservoirs, a proper prediction of the hydrodynamics of the geothermal system is necessary. However, a deep understanding of the reservoir's properties is typically required. This study investigates the use of machine learning to forecast the wellhead pressure of geothermal injection wells using only historical data of wellhead pressure, fluid injection rates, and temperatures. These datasets will be used to construct several machine learning (ML) algorithms, including Multiple Linear Regression (MLR), Random Forest (RF), eXtreme Gradient Boosting (XGBoost), Support Vector Machine (SVM), as well as Artificial Neural Network (ANN) such as Multilayer Perceptron (MLP) and Long Short-Term Memory (LSTM). Water injection data from four injection wells underwent rigorous data cleaning and outlier removal to ensure the fidelity of the results. The performance of the six algorithms was then validated and examined using metrics like the coefficient of determination (R squared), the root mean squared error (RMSE), and the symmetric mean absolute percentage error (sMAPE). Results show that LSTM outperforms other tested algorithms with a sMAPE of 8.02%, followed closely by XGBoost with 8.28% and MLP with 8.55%. RF and SVM achieved a sMAPE of 10.18% and 10.48%, respectively, while the MLR, used in this thesis as a way to better gauge the upper limit of the models, got a sMAPE of 20.56%. It's essential to note that these results are subjective on the specific dataset used in this study, as well as the particular preprocessing, feature engineering, and hyperparameter tuning. Future studies might consider implementing other ML techniques or hybrid learning models, offering valuable insights for geothermal optimization. As ML becomes increasingly important in the geothermal energy sector, such exploration is of significant interest.

AB - To ensure efficient and sustainable reinjection of water into geothermal reservoirs, a proper prediction of the hydrodynamics of the geothermal system is necessary. However, a deep understanding of the reservoir's properties is typically required. This study investigates the use of machine learning to forecast the wellhead pressure of geothermal injection wells using only historical data of wellhead pressure, fluid injection rates, and temperatures. These datasets will be used to construct several machine learning (ML) algorithms, including Multiple Linear Regression (MLR), Random Forest (RF), eXtreme Gradient Boosting (XGBoost), Support Vector Machine (SVM), as well as Artificial Neural Network (ANN) such as Multilayer Perceptron (MLP) and Long Short-Term Memory (LSTM). Water injection data from four injection wells underwent rigorous data cleaning and outlier removal to ensure the fidelity of the results. The performance of the six algorithms was then validated and examined using metrics like the coefficient of determination (R squared), the root mean squared error (RMSE), and the symmetric mean absolute percentage error (sMAPE). Results show that LSTM outperforms other tested algorithms with a sMAPE of 8.02%, followed closely by XGBoost with 8.28% and MLP with 8.55%. RF and SVM achieved a sMAPE of 10.18% and 10.48%, respectively, while the MLR, used in this thesis as a way to better gauge the upper limit of the models, got a sMAPE of 20.56%. It's essential to note that these results are subjective on the specific dataset used in this study, as well as the particular preprocessing, feature engineering, and hyperparameter tuning. Future studies might consider implementing other ML techniques or hybrid learning models, offering valuable insights for geothermal optimization. As ML becomes increasingly important in the geothermal energy sector, such exploration is of significant interest.

KW - Maschinelles Lernen

KW - Geothermie

KW - Wassereinspritzung

KW - Bohrkopfdruck

KW - machine learning

KW - geothermal energy

KW - water injection

KW - well head pressure

U2 - 10.34901/mul.pub.2023.201

DO - 10.34901/mul.pub.2023.201

M3 - Master's Thesis

ER -