| Authors: |
MUCHAMAD Harry, Aditya WAHYUDI, Midat AL ISLAM, Nada AFRA, Jantiur SITUMORANG |
| Keywords: |
injection, injection decline, machine learning, JIWA Flow |
| Conference: |
Stanford Geothermal Workshop |
Session: |
Injection |
| Year: |
2021 |
Language: |
English |
| Abstract: |
Reinjection plays a crucial role in geothermal field operation and management. The sustainability of geothermal power plants depends on how field injection capacities can be maintained, and the reinjection can effectively provide pressure support to producer wells. A good monitoring of injection performance is therefore essential in achieving excellent field management. Currently, analyzing and monitoring injection well issues & performances generally rely on wellbore simulation, which is challenging and very time-consuming. This study aims to develop an easier and faster alternative solution for analyzing geothermal injection well's performance by applying machine learning (ML). The ML model was compiled, trained, and validated using synthetic data generated from a wellbore simulator (JIWA Flow). This study's best algorithm is Feed Forward Neural Network (FFNN) with R Squared and MSE of 0.998 and 3.4. It is proved by the comparison between the unseen data from JIWA Flow and prediction from FFNN. The application of the proposed ML to calculate the exponential annual injection decline from several scenarios are provided. |
| Download: |
|
File Size: |
1505 KB |