Efri Lastuti Manurung, Veronica (2022) Prediksi Tingkat Keberhasilan Injeksi Polimer Pada Reservoir Minyak Ringan Menggunakan Metode Artificial Neural Network. Other thesis, universitas islam riau.

Text 153210094.pdf - Submitted Version Restricted to Registered users only Download (4MB) | Request a copy

Abstract

Polymer gel injection can overcome the conformance problem after the water flooding work is done. There are many aspects to determine the success rate of the polymer gel injection both in terms of reservoir parameters and technical aspects. To give good pressure results, it is necessary to do modeling that can predict the effect of the parameters used on oil recovery results and water cut reduction results. In this study, 8 parameters will be considered and several scenarios need to be carried out so that they can be used to predict the value of the recovery factor and water cut. These parameters include reservoir pressure, reservoir temperature, oil density, oil viscosity, polymer concentration, xlinker concentration, brine salinity and injection rate. Because modeling and experiments to predict the success of polymer gel injection using simulation methods and laboratory tests require detailed data, are expensive and take a long time, an artificial intelligence method, namely the Artificial Neural Network, is used. The Artificial Neural Network method can predict the successful results of a job without requiring a detailed data set. By using 2000 data with a ratio of 80% training data and 20% testing data, trial and error were carried out on 3 ANN hyperparameters, namely the value of 5 variations in the number of hidden layer nodes, namely on the number of nodes 3, 5, 9, 15, and 20, activation function solver method. From the research results, it was obtained that the best ANN model architecture for this study was 8-20-1 (8 inputs, 20 hidden layers and 1 output) with an R 2 value for training data of 0.98 for RF output and 0.93 for WC output. Meanwhile, for the data testing results, the value of R is 0.98 for RF output and 0.90 for WC output. The RMSE value for the training data is 1.92 for the RF output and 3.6 for the WC output. For data testing, the RMSE value is 2.19 for RF output and 4.33 for WC output. The MAPE value for training data is 3.9 for RF output and 3.8 for WC output and for data testing results obtained a MAPE value of 4.60 for RF output and 4.23 for WC output. In addition, the results obtained from the ANN model when compared with the actual data from the CMG simulation results are of course very close so that it can be concluded that the ANN results are quite accurate because the R 2 value is almost close to value of 1 with an average value obtained of 0.9 and an error percentage of less than 10% seen of the MAPE value obtained on average is less than 5%. 2

Actions (login required)

View Item