A Hybrid Neural–Statistical Framework for Pipeline Lifespan Prediction under Spatiotemporal Dependence

Abstract

This study addresses pipeline failures in Nigeria’s oil and gas sector, which result in significant economic losses, environmental damage, and operational disruptions. A hybrid modelling framework is proposed, integrating deep learning architectures—Artificial Neural Networks (ANN), Long Short-Term Memory (LSTM), Convolutional Neural Networks (CNN), and Graph Neural Networks (GNN)—with Generalised Additive Models (GAM). The framework captures complex spatial and temporal dependencies whilst maintaining interpretability. The methodology incorporates feature selection, model training, and spatiotemporal cross-validation to evaluate predictive performance across different locations and time periods. Results demonstrate improved accuracy and robustness compared with traditional approaches. The model provides practical insights for predictive maintenance, risk assessment, and real-time anomaly detection, supporting safer and more efficient decision-making in the oil and gas industry.