Identification of Geothermal Reservoir Determination using Artificial Neural Network (ANN)

Geothermal utilization in Indonesia is mostly for electricity generation. Electricity consumption has increased while geothermal production has not increased, so it is necessary to develop geothermal wells. One of the efforts is the prediction of well behavior so that the well performance can be known which a need for well development is. To predict the behavior of geothermal wells temperature prediction (T) and pressure (P) with location parameters (x and y), well depth (z) injection flow rate (qinj) and injection temperature (Tinj) using the Artificial Neural Network (ANN) method. The first is the generation of well production models, M-1, M-2 and M-3, each model has 6 wells. Data is generated during one year of production and data separation is carried out, i.e. data for 11 months is used as ANN training data and data for the last 1 month is used as test data. The results of the prediction with ANN will be compared with the test data. Calculation of errors between the predicted results and the test data on M-1 is 0.0251 for temperature (T) and 0.0303 for pressure (P), while the MSE value is 0.00378. At M-2 is 0.0283 for temperature (T) and 0.0468 for pressure (P), while the MSE value is 0.000795. At M-3 is 0.0445 for temperature (T) and 0.0566 for pressure (P), while the MSE value is 0.0121. Based on the results obtained the error value and MSE are relatively small, so ANN can be used to predict the behavior of geothermal wells. Then the variation in the number of hidden layers is done. H-15 has the best error value and MSE, while h-50 has the best regression value (R).