Abstract:

Carbon dioxide (CO2) corrosion is one of the major concerns in oil and gas industry. This work attempted to apply Machine Learning method - Artificial Neural Network (ANN) to predict CO2 corrosion rate in pipeline. After collecting, selecting features, pre-processing, a dataset of 40 data with 9 features of pipeline operating parameters has been used for research. Applying newest optimizer RMSprop with algorithm Early-Stopping increases accuracy and reduces the effect of small dataset. An Artificial Neural Network is developed, which has 2 hidden layers with 18 nodes and 9 nodes with activate functions ReLU and Sigmoid in sequence. The empirical model Norsok M-506 was applied to compare performances of models. Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), coefficient of determination (R2) were used as evaluating indicators. The predicted corrosion rates of artificial neural network model R2 = 0.938, RMSE = 0,014, MAE = 0,011 provided higher performance than empirical model Norsok M-506.