Accident prediction modeling by artificial neural network in petroleum industry: a case study of National Iranian Oil Products Distribution Company

Abstract

This study aims to predict the number of accidents in the National Iranian Oil Products Distribution Company (NIOPDC) as a case study in 2022 according to the database between 2012 and 2021. Artificial Neural Network (ANN) is used for modeling using curve fitting (Multi-Layer Perceptron-MLP) and time series (Nonlinear AutoRegressive exogenous -NARX) networks. The network parameters are adjusted by optimal architecture values to create a successful model with coefficient determination (R 2) and Mean Square Error (MSE) performance criteria. Also, mathematical methods of Root Mean Square Error (RMSE), Average Invalidity Percentages (AIP), Average Validity Percentages (AVP), and Mean Absolute Error (MAE) are checked out to evaluate the proposed model’s robustness. The results show acceptable R 2 values of 0.98 and 0.99 for MLP and NARX networks, respectively, demonstrating that NARX has more prediction accuracy than the MLP network. By the best model (NARX) results, falls, CNG, and LPG facility accidents will be half. Nevertheless, the aviation center and loading rack accidents will rise 3 and 1.5 times, respectively. The findings will be helpful in systematic accident prevention for decision-making authorities which have been done in the Iranian petroleum industry for the first time.