Investigasi Performa Cyclic Steam Stimulation Terhadap Peningkatan Perolehan Minyak Dengan Menggunakan Artificial Neural Network

Junastri, Ade (2022) Investigasi Performa Cyclic Steam Stimulation Terhadap Peningkatan Perolehan Minyak Dengan Menggunakan Artificial Neural Network. Other thesis, Universitas Islam Riau.

Text 173210338.pdf - Submitted Version Restricted to Repository staff only Download (1MB) | Request a copy

Abstract

Currently, more and more heavy crude oil reserves are found, so it is necessary to apply enhanced oil recovery (EOR) technology which is considered effective in increasing oil recovery. Cyclic Steam Stimulation (CSS) method is a thermal EOR method with a mechanism that goes through the steam injection, immersion, and oil production cycle. Therefore, it is necessary to conduct research on the performance of the cyclic steam stimulation parameter to increase oil recovery. Steam is an efficient medium for heating the subsurface layer and the liquid reservoir contained therein, because much of the energy available in the steam is in the form of latent heat which is released at a constant temperature when it condenses and is associated with a relatively cold subsurface. Analysis of the performance of the cyclic steam stimulation parameter on the increase in oil recovery in this study was carried out by simulation research using CMG-STARTS software and Python. The aim is to investigate the performance of cyclic steam stimulation to increase oil recovery from several operational parameters tested, namely injection volume, steam quality, injection rate, soaking time, injection temperature, and injection pressure. This ANN method is a deep learning method from input data and produces output data. By using 516 data ratio 80% of the RF and SOR calculation model results from CMG software for training data and 20% of the model results for testing. In order to get prediction results using the ANN method optimally, trial and error will be carried out on the number of hidden layer nodes. The optimal and stable hidden layer nodes are obtained at nodes 14 with a value of R2 train 0.9976 and R2 test 0.9882. RMSE train 0.0501 and RMSE test 0.1126 and MAPE train 0.1767 and MAPE test 0.2697. It can be concluded in this study that using ANN in rf and sor prediction using 14 nodes hidden layer proved to be very good and successful.

Actions (login required)

View Item