Sensitivity analysis in using of artificial neural network models to determine infill well locations in a mature oil field

This paper describes a method to rank potential infill well locations using Artificial Neural Networks (ANN) from existing well data. Sensitivity test was conducted for training and testing data used with comparison 2:8, 4:6, 5:5, 6:4 and 8:2 for each data. Root Mean Square Error difference between training and test data show that the best results obtained from the ratio of training data and testing data 8: 2. Two ANN models were built. The first model predicted top sand depth, resistivity, gamma-ray and density-neutron from infill well location (chosen from structural position and good oil rates from offset wells). The second model predicted initial oil rate from outputs from the first model. Predicted initial oil rates from the ANN model were compared with those from the 3D reservoir simulation model. They shows similar prediction of oil rates which gave high confidence in the predicted oil rate. Very different oil rate prediction between the two models can be used as consideration to improve the simulation model.