Artificial neural network model for process monitoring of crude oil distillation column in a petroleum refinery

Abstract

Crude oil fractions are essential products which are major sources of income that influence the economic growth. However, the complex nature of crude oil is calling for manufacturing innovation that will ease its process monitoring. Thus, there is need to design the crude distillation unit (CDU) operation to meet customers’ demand for the crude oil fractions. In this study, artificial neural network was applied for the process design and monitoring of a CDU in a petroleum refinery. A total of 230 data sets was used as experimental data (comprising controllable and uncontrollable variables) from which 88%, 6% and 6% were used for training, validation and testing of neural network respectively. The architectures for the CDU unit design and its neural network controller were 14 inputs, one hidden layer and seven outputs (14-1-7) and; 13 inputs, one hidden layer and six outputs (13-1-6) respectively. Logistic sigmoid transfer function was used as the activation function for the neural network training in the algorithms. The mean absolute error (MAE) and the mean square error (MSE) were used to test the accuracy of the model. After iteration numbers of 396 and 786 for CDU design and its neural network controller respectively, convergence was attained with training error of less than 10⁻⁶. For CDU design, the minimum values of MSE and MAE were 0.0055 and 0.0568 while their maximum values were 0.3517 and 0.4139 between the testing data and experimental data respectively. For the CDU neural network controller, the minimum values of MSE and MAE were 0.0000 and 0.000 while their maximum values were 0.1412 and 0.2922 between the testing data and experimental data respectively. In conclusion, artificial neural network could effectively be used as a machine learning tool for process monitoring of CDU in a petroleum refinery.