Enterprise AI for Upstream Gas Lift Optimization
Project Challenge
A major Southeast Asian oil & gas company has recently launched an extensive digital transformation program across its operations. A core objective of the program is to leverage artificial intelligence (AI) to help improve operational productivity, asset integrity, and safety. Gas lift is a common artificial lift method in high output wells. The main challenge with gas lift optimization is to determine in a continuous manner, the most efficient gas-liquid-ratio (GLR) for each well to maximize the real-time liquid production while minimizing gas compression requirements.
The lack of real-time production visibility across individual wells make it challenging to manage and optimize the production of multiple wells. The industry has developed various physics-based methods to estimate production rates (e.g., nodal analysis) that replicate the behavior of the production system (wells, flowlines, separators, artificial lift systems) to estimate the production rates. These models are typically complex to maintain and update in a continuous operating environment.
A common alternative to physics-based models is to use virtual flow metering (VFM) that leverages data-driven algorithms to provide continuous estimates of well production rates to determine—in real-time—the optimum artificial lift and network parameters to maximize production. Once accurate VFM is developed, tested, and validated against historical and current well tests, single well and gas lift network can benefit from real-time multiphase estimations to perform data-driven optimization and scenario analysis.
About Global Oil & Gas Company
- Over $100B in 2020 annual revenue
- 10,000+ employees
- Global operations across 15+ locations
Project Highlights
- 16 weeks from kick-off to production-ready application
- 3 billion rows of minute-level operational data ingested from 4 disparate data sources for each of the 2 fields
- Created unified object model to represent asset hierarchy, telemetry data, and shutdown and events logs
- 70+ machine learning features extracted from subjected matter experts and tested
- 4 machine learning model configured per field to predict liquid, oil, water, and gas rates
Approach
BHC3 Production Optimization addresses the constraints posed by physics-based upstream production optimization techniques. The application enables collaboration between production and surveillance engineers by providing, in near-real time, all operational and telemetry data augmented with AI-predicted virtual flow rates, AI-recommended strings for gas lift optimization, and metrics-based anomaly insights with actionable recommendations.
Over 16 weeks, a team of Baker Hughes and C3 AI experts collaborated with project managers and subject matter experts from the oil & gas company to configure the BHC3 Production Optimization application that provides, in near-real time, all operational and telemetry data augmented with AI-predicted virtual flow rates, AI-recommended strings for gas lift optimization, and metrics-based anomaly insights with actionable recommendations. Throughout the the configuration effort, the team addressed five core components:
- Unified data image with analytics-ready data and robust data integration capabilities to automatically normalize raw telemetry feeds (e.g. remove outliers, resolve missing values, handle different units of measure, etc.)
- Virtual flow metering (VFM) models for each field in scope that predict flow rates between well tests with 200% higher precision than the physics-based baseline
- Gas lift optimization (GLO) classifiers to identify and rank top candidate strings for gas lift injection rate optimization with over 80% precision
- Anomaly insights based on metrics to detect strings that are not responsive to gas lift injection and identify potential root causes
- Workflow enabled UI that allows production engineers to identify underperforming wells based on VFM, assess top recommended strings for GLO, and collaborate across teams to maximize production
Using BHC3 Production Optimization, production and surveillance engineers can quickly investigate and prioritize production opportunities per string, through a dedicated string detail screen, run extensive ad-hoc analysis, and act by applying or discarding such opportunities, and review the outcomes.