Real-Time Early Warning System for Hydraulic Slugging and Flow Cessation Using Unsupervised Analysis of Wellhead Pressure Transients

Résumé fourni par la source

Abstract Hydraulic slugging and flow ceasing remain stubborn operational problem in onshore well production. Pressure swings damaging the equipment, disrupt separation, force unplanned shutdowns. Traditional monitoring such as steady-state flow maps, reactive alarms can't keep up with how these instabilities actually evolve. This work develops an automated early warning system that catches slugging and predict ceasing using only wellhead pressure data. No multiphase flowmeters, no labeled datasets. The framework runs on unsupervised learning since labeled data doesn't exist. upstream and downstream pressure readings directly from SCADA. We establish stable flow baseline for each individually using change point detection then anomaly layer detection to flag meaningful deviations the subtle pattern shift that signal slugging before conventional alarms notice anything. For ceasing prediction pressure changes trend get analyzed through hybrid pattern recognition combined . the architecture processes the streaming data continuously, pushing alerts when conditions shift. Tests used months of historical data from multiple wells, compared against operator logs and production sheets. Pressure signatures alone carried enough signal to separate stable flow from slugging from dying wells. Classification accuracy landed above 75% with warning firing well before high pressure trips. Honestly, better than expected. More valuable was catching gradual energy depletion the slow fade before a well stops. That advance notice gives you the opportunity to schedule intervention instead of scrambling after shutdown. The system proved that routine wellhead pressure measurements contain sufficient diagnostic informations for reliable state classification. It successfully identified incipient instabilities and ceasing trends, providing early warnings that preceded conventional alarm systems and enabled proactive operational response. Most research targets offshore sever slugging or builds physics models from scratch. We are targeting onshore hydraulic slugging using only field data and unsupervised techniques. A gap largely ignored. The framework make from the low cost pressure data an actionable intelligence without requiring additional instruments. For mature assets operating on tight budgets, this matters. Real-time change-point detection combined with transient pressure analysis offers practical surveillance capability that extend asset life and cuts deferred production.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.