Multiagent AI Improves Offshore Production Surveillance and Intervention

_ This article, written by JPT Technology Editor Chris Carpenter, contains highlights of paper SPE 226728, “Offshore Production Surveillance and Intervention Using Multiagent AI, ” by Dushyant S. Shekhawat, Joy Barua, and Karan Bhatia, SPE, SLB, et al. The paper has not been peer-reviewed. _ This paper introduces an agentic artificial-intelligence (AI) framework designed for offshore production surveillance and intervention. Agentic AI is a novel framework comprising a collection of AI models operating autonomously yet collaboratively to achieve a common goal. Each model specializes in performing a certain task to streamline production monitoring, root-cause analysis, predictive maintenance, and optimization workflows. It uses comprehensive data sets, including production history, well coordinates, intervention history, and petrophysical and completion information, to support dynamic decision-making across the asset. Methodology The proposed framework leverages a collection of task-specialized AI agents, each designed to address specific workflows critical to offshore production operations. These agents operate collaboratively within a modular architecture, enabling seamless integration of diverse capabilities such as production monitoring, anomaly detection, optimization, and decision support. A unified conversational interface powered by large language models (LLMs) facilitates human/AI interaction, ensuring accessibility for both technical and nontechnical users. The system architecture follows modern principles of scalable and modular AI development, emphasizing containerized deployment for isolated agent operation and secure knowledge-sharing through a centralized knowledge graph. Input Data and Sources. The agentic AI framework integrates a diverse range of structured and unstructured data sources essential for comprehensive production-system surveillance and optimization. Primary inputs include supervisory control and data-acquisition (SCADA) and sensor data, providing real-time measurements of pressure, temperature, flow rates, and equipment status across wells and facilities. Production-history data sets form the foundation for decline-curve analysis (DCA) and performance benchmarking. Petrophysical logs offer detailed insights into subsurface properties. Additionally, simulated models, both production and reservoir, are used to incorporate physics-based understanding of the asset, aiding in scenario forecasting and sensitivity analyses. Operational reports, encompassing maintenance, intervention records, and well-integrity data, provide essential context for diagnosing anomalies and planning interventions. Finally, vector databases store structured knowledge, including historical AI-agent outputs, engineering interpretations, and user comments, enabling continuous learning and contextual decision-making within the system. AI Agents and Workflows. In the agentic AI framework for production operations, each key petroleum-engineering workflow within offshore production is represented by a dedicated, task-specific AI agent. Orchestrating Agent. The Orchestrating Agent serves as the central control layer within the agentic AI framework, coordinating the interaction between AI agents, user inputs, and final system outputs. This agent acts as the brain of architecture, ensuring that the appropriate workflows and analyses are selected and executed dynamically based on both technical requirements and user-driven queries. At its core, the Orchestrating Agent consists of the following four components: - Planning Agent: This component interprets user inputs, system prompts, and operational context to determine which analytical workflow should be initiated. - Subject-Matter Expert (SME) Agent: Once the appropriate workflow is selected, the SME Agent applies domain-specific knowledge of petroleum engineering to refine and contextualize the analysis. This agent determines which models, diagnostics, or engineering methodologies are most applicable. - Decision Agent: After individual AI agents perform their specialized analyses, the Decision Agent consolidates these outputs into a coherent, technically sound response. - Data-Visualization Agent: To facilitate user interpretation and operational decision-making, the Data-Visualization Agent transforms complex analytical outputs into intuitive graphical formats.