From Forecasting to Prevention: Operationalizing Spatiotemporal Risk Decoupling in Gas Pipelines via Integrated Time-Series and Pattern Mining

Accurate prediction of gas pipeline incidents through risk factor interdependencies is critical for proactive safety management. This study develops a hybrid SARIMA–association rule mining (ARM) framework integrating time-series forecasting with causal pattern decoding, using 60-month U.S. pipeline incident records (2010–2024) from the Pipeline and Hazardous Materials Safety Administration (PHMSA) database, covering leaks, mechanical punctures, and ruptures. Seasonal Autoregressive Integrated Moving Average (SARIMA) modeling with six-month rolling-window validation achieves precise leak forecasts (MAPE = 14.13%, MASE = 0.27) and reasonable mechanical damage predictions (MAPE = 31.21%, MASE = 1.15), while ruptures exhibit pronounced stochasticity. Crucially, SARIMA incident probabilities feed Apriori-based ARM, revealing three failure-specific mechanisms: (1) ruptures predominantly originate from natural force damage, with underground cases causing economic losses (lift = 3.70) and aboveground class 3 incidents exhibiting winter daytime ignition risks (lift = 2.37); (2) leaks correlate with equipment degradation, where outdoor meter assemblies account for 69.7% of fire-triggering cases (108/155 incidents) and corrosion dominates >50-year-old pipelines; (3) mechanical punctures cluster in pipelines <20 years during spring excavation, predominantly occurring in class 2 zones due to heightened construction activity. These findings necessitate cause-specific maintenance protocols that integrate material degradation laws and dynamic failure patterns, providing a decision framework for pipe replacement prioritization and seasonal monitoring in high-risk zones.