Probabilistic Pipeline Reliability Assessment Using Exponentiated-Exponentiated Distributions Integrated with Hybrid Neural Networks: Evidence from Nigeria

Abstract

Standard reliability models for oil and gas pipelines chiefly the Weibull distribution assume monotonically increasing or constant failure rates, a restriction that is empirically violated in Nigerian pipeline systems, where failure rates fluctuate with operational loading cycles, seasonal soil conditions, and maintenance histories. This paper introduces a novel reliability framework integrating Exponentiated-Exponentiated (E-E) Probability Distributions with Hybrid Neural Networks (HNNs) to provide flexible, uncertainty-aware pipeline reliability assessment. The E-E distribution, constructed by applying the exponentiation operator twice to a baseline distribution, accommodates non-monotonic (bathtub-shaped, unimodal, and reverse-J) failure rate functions. When paired with HNN-generated feature embeddings drawn from Artificial Neural Network (ANN), Long Short-Term Memory (LSTM), Convolutional Neural Network (CNN), and Graph Neural Network (GNN) architectures, the E-E distribution transforms point predictions of Remaining Useful Life (RUL) into full probabilistic failure-time distributions with calibrated confidence intervals. Empirical validation on a 21-year Nigerian pipeline dataset (N = 500 segments, 18 predictors) demonstrated that the GNN-GAM variant achieved the best operationally balanced performance, with R² = 0.9899, RMSE = 0.6032, and MAE = 0.1235 on the held-out test set, and an empirical 95% confidence interval coverage rate of 94.2%, closely matching the nominal target within the accepted ±2 percentage point engineering tolerance. The framework outperforms classical Weibull-based and deterministic neural approaches, enabling risk-stratified maintenance scheduling. These results demonstrate that the proposed framework represents a substantial methodological advance, with direct relevance to Nigeria's oil and gas sector.