A Rainflow Cycle–Based Machine Learning Framework for Predicting Pressure Surge Severity in Natural Gas Pressure Reduction Skids

Pressure Reduction Skids (PRSs) are essential for maintaining safe downstream pressures in UK natural gas distribution networks; however, they are routinely exposed to upstream pressure fluctuations that can induce cyclic loading, regulator instability, and surge behaviour, even below design pressure limits. This study introduces a combined rainflow cycle–based severity assessment and machine-learning forecasting framework to characterise and predict inlet pressure surge severity using only hourly upstream pressure data. A dataset comprising 50,903 hourly inlet-pressure measurements from a UK PRS, spanning over 2,100 operational days, was analysed. Rainflow cycle counting revealed typical daily peak-to-trough pressure cycles of 3–5 bar, with high-severity days reaching up to 7 bar. Surge-severity events were defined using a 90th-percentile threshold of daily maximum cycle range. Statistical and rainflow-derived features were then used to train Logistic Regression, Random Forest, and Gradient Boosting models to predict next-day surge severity. The Random Forest model achieved the best performance with an AUC of 0.79, demonstrating that upstream pressure patterns contain predictive information on future surge conditions. The proposed approach enables practical SCADA-level surge monitoring, supporting proactive PRS management and integrity-focused operational decision-making.