Reinventing the Upstream Control Room: An Agentic AI Approach to Early Deviation Detection and Remediation

Abstract

Abstract The upstream control room is where production, injection, chemical, and emissions performance are monitored against target, and where operational deviations must be detected and resolved before they compound into lost output, avoidable emissions, or safety exposure. In practice, this task is hindered by four recurring gaps observed across production operations: no automated system alerts operators to an emerging shortfall before it has already grown; shortfalls are not always reconciled consistently across the control room, well- performance, and subsurface teams; compensation to recover a shortfall is often arranged ad hoc through informal communication, bypassing structured, formal validation and slowing recovery; and diagnosing the cause of a shortfall still depends on individual operator experience and a manual search through troubleshooting documentation. Left unaddressed, these gaps translate directly into slower recovery, inconsistent decisions across shifts, and lost production, chemical, and energy value. This paper presents the design, implementation, and early learnings of a Next-Generation Central Control Room (NextGen CCR), an artificial-intelligence-native decision-support layer deployed in an offshore upstream pilot to enable early operational-drift detection and structured remediation while preserving human decision authority. The solution combines near-real-time operational data, optimized operating envelopes, planned-activity context, and several thousand pages of procedural knowledge within a layered architecture that integrates deterministic analytics, large language models, and an operational knowledge graph. Deviations are detected against targets and envelopes, contextualized against planned activities and equipment status, and investigated by orchestrated, domain-specific agents that produce ranked, evidence-backed diagnoses and prioritized, standards-compliant recommendations. A scoped conversational assistant offers deeper guidance grounded in operating philosophies and equipment manuals, and human validation is embedded throughout the workflow. Early pilot results show improved end-of-day production prediction and faster, evidence-based investigation of critical alarms, with change management recognized as central to sustained adoption.