A machine learning‐based approach to the multiobjective optimization of CO2 injection and water production during CCS in a saline aquifer based on field data
Abstract The presence of carbon capture and storage (CCS) projects is important due to the growing production of greenhouse gases, especially carbon dioxide (CO2). Our target functions have been chosen because of the importance of CO2 storage in CCS projects and the requirement for producing less wa...
| Main Authors: | , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Wiley
2023-05-01
|
| Series: | Energy Science & Engineering |
| Subjects: | |
| Online Access: | https://doi.org/10.1002/ese3.1412 |
| _version_ | 1797829852424830976 |
|---|---|
| author | Pouya Vaziri Behnam Sedaee |
| author_facet | Pouya Vaziri Behnam Sedaee |
| author_sort | Pouya Vaziri |
| collection | DOAJ |
| description | Abstract The presence of carbon capture and storage (CCS) projects is important due to the growing production of greenhouse gases, especially carbon dioxide (CO2). Our target functions have been chosen because of the importance of CO2 storage in CCS projects and the requirement for producing less water in projects requiring water production. As a proxy for reservoir simulations, support vector regression, artificial neural network (ANN), and multivariate adaptive regression spline (MARS) have been used. It was determined that MARS had higher accuracy based on examining these three data‐driven models with the available field data. It was, however, very close to the accuracy of the ANN. MARS gave root mean square of error (RMSE), mean absolute error (MAE), and R2 values of 2.78%, 1.95%, and 0.998, respectively, for predicting CO2 storage values in test data, yet 3.73%, 3.53%, and 0.995 for blind data. The RMSE, MAE, and R2 to evaluate the machine learning (ML) model for predicting water production were 3.58%, 2.81%, and 0.997, respectively, while the results for blind data were 3.94%, 2.83%, and 0.997, respectively. By reducing the amount of computational load and time taken to reproduce the simulation data by MARS, it will be beneficial for optimization. This study highlights the application of the ML approach to coupling with genetic algorithms and optimizing CO2 storage and water production. With the proposed frameworks, preprocessing, feature selection, two‐stage validation of the data‐driven model, and optimizer are performed in the aquifer and a repository containing a series of optimal solutions is developed to be used in projects. |
| first_indexed | 2024-04-09T13:26:44Z |
| format | Article |
| id | doaj.art-bcb6a5c86b9546f1bb6e5a487640d335 |
| institution | Directory Open Access Journal |
| issn | 2050-0505 |
| language | English |
| last_indexed | 2024-04-09T13:26:44Z |
| publishDate | 2023-05-01 |
| publisher | Wiley |
| record_format | Article |
| series | Energy Science & Engineering |
| spelling | doaj.art-bcb6a5c86b9546f1bb6e5a487640d3352023-05-10T07:56:55ZengWileyEnergy Science & Engineering2050-05052023-05-011151671168710.1002/ese3.1412A machine learning‐based approach to the multiobjective optimization of CO2 injection and water production during CCS in a saline aquifer based on field dataPouya Vaziri0Behnam Sedaee1Institute of Petroleum Engineering, School of Chemical Engineering, College of Engineering University of Tehran Tehran IranInstitute of Petroleum Engineering, School of Chemical Engineering, College of Engineering University of Tehran Tehran IranAbstract The presence of carbon capture and storage (CCS) projects is important due to the growing production of greenhouse gases, especially carbon dioxide (CO2). Our target functions have been chosen because of the importance of CO2 storage in CCS projects and the requirement for producing less water in projects requiring water production. As a proxy for reservoir simulations, support vector regression, artificial neural network (ANN), and multivariate adaptive regression spline (MARS) have been used. It was determined that MARS had higher accuracy based on examining these three data‐driven models with the available field data. It was, however, very close to the accuracy of the ANN. MARS gave root mean square of error (RMSE), mean absolute error (MAE), and R2 values of 2.78%, 1.95%, and 0.998, respectively, for predicting CO2 storage values in test data, yet 3.73%, 3.53%, and 0.995 for blind data. The RMSE, MAE, and R2 to evaluate the machine learning (ML) model for predicting water production were 3.58%, 2.81%, and 0.997, respectively, while the results for blind data were 3.94%, 2.83%, and 0.997, respectively. By reducing the amount of computational load and time taken to reproduce the simulation data by MARS, it will be beneficial for optimization. This study highlights the application of the ML approach to coupling with genetic algorithms and optimizing CO2 storage and water production. With the proposed frameworks, preprocessing, feature selection, two‐stage validation of the data‐driven model, and optimizer are performed in the aquifer and a repository containing a series of optimal solutions is developed to be used in projects.https://doi.org/10.1002/ese3.1412CCSdata driven modeldeep saline aquifergenetic algorithmmachine learning modelsmultiobjective optimization |
| spellingShingle | Pouya Vaziri Behnam Sedaee A machine learning‐based approach to the multiobjective optimization of CO2 injection and water production during CCS in a saline aquifer based on field data Energy Science & Engineering CCS data driven model deep saline aquifer genetic algorithm machine learning models multiobjective optimization |
| title | A machine learning‐based approach to the multiobjective optimization of CO2 injection and water production during CCS in a saline aquifer based on field data |
| title_full | A machine learning‐based approach to the multiobjective optimization of CO2 injection and water production during CCS in a saline aquifer based on field data |
| title_fullStr | A machine learning‐based approach to the multiobjective optimization of CO2 injection and water production during CCS in a saline aquifer based on field data |
| title_full_unstemmed | A machine learning‐based approach to the multiobjective optimization of CO2 injection and water production during CCS in a saline aquifer based on field data |
| title_short | A machine learning‐based approach to the multiobjective optimization of CO2 injection and water production during CCS in a saline aquifer based on field data |
| title_sort | machine learning based approach to the multiobjective optimization of co2 injection and water production during ccs in a saline aquifer based on field data |
| topic | CCS data driven model deep saline aquifer genetic algorithm machine learning models multiobjective optimization |
| url | https://doi.org/10.1002/ese3.1412 |
| work_keys_str_mv | AT pouyavaziri amachinelearningbasedapproachtothemultiobjectiveoptimizationofco2injectionandwaterproductionduringccsinasalineaquiferbasedonfielddata AT behnamsedaee amachinelearningbasedapproachtothemultiobjectiveoptimizationofco2injectionandwaterproductionduringccsinasalineaquiferbasedonfielddata AT pouyavaziri machinelearningbasedapproachtothemultiobjectiveoptimizationofco2injectionandwaterproductionduringccsinasalineaquiferbasedonfielddata AT behnamsedaee machinelearningbasedapproachtothemultiobjectiveoptimizationofco2injectionandwaterproductionduringccsinasalineaquiferbasedonfielddata |