Analysis of effective factors on gas leakage of polyethylene pipes for the intelligent forecasting of leakage degree

It is essential to prevent gas pipeline leakages to protect the environment and avoid financial losses and casualties, especially in densely populated areas. For half a century, in gas distribution networks, polyethylene pipes with advantages, including corrosion resistance properties, ease of implementation, and lower operation cost, are considered a replacement for metal pipes. In the present study, the degree of leakage in polyethylene pipes is predicted by collecting effective data on the leakage in these types of pipes. First, a logistic regression model is designed to forecast. Then three nonlinear models, including the Multi-layer Perceptron Neural Network (MLP), the Radial Basis Function (RBF), and the Support Vector Machine (SVM), are used to improve the accuracy. In addition, a hybrid model is proposed to improve the classification accuracy that models linear and nonlinear patterns simultaneously. Different architectures are considered and examined for each of these models to identify the best models' structure. It is concluded from the empirical results that the best single model is the MLP with an accuracy of 88.80%, the sensitivity of 88.19%, and the MSE of 0.111, and the hybrid model with the improvement that results in an error reduction and increase in sensitivity, i.e., the accuracy of 88.85%, a sensitivity of 89.86%, and the MSE of 0.080 forecasted degrees of leakage. Therefore, the hybrid model had a higher efficiency for forecasting the leakage degree compared to individual models. © The Author(s), under exclusive licence to Springer Nature Switzerland AG 2022.