Virtual Metering and Subcool Control in SAGD Wells

Executive Summary

IntelliDynamics successfully implemented for a major oil sands producer an advanced subcool control system with integrated virtual metering for SAGD (Steam Assisted Gravity Drainage) operations, transitioning from reactive manual control to proactive automated closed loop control of 3 phase production. The solution uses hybrid model-predictive control to maintain optimal subcool values while providing real-time production rate estimates between physical well tests, reducing the risk of steam breakthrough while maximizing production and minimizing operator intervention during process upsets.

Challenge

In SAGD operations, maintaining proper subcool—the temperature difference between the producer well and saturated steam at the producer’s pressure—is critical to safe and efficient production. Steam entering or flashing within a producer can cause significant damage to equipment and reduce production efficiency.

Key Operational Issues:

  • Subcool typically ranges from 10-50°C during normal operations
  • Below 10°C, risk of steam in the producer becomes significant
  • Manual control requires constant operator attention, especially during process events and upsets
  • Conservative setpoints reduce production optimization to maintain safety margins
  • Variable response times between control actions (choke, steam, lift gas) and subcool changes complicate manual control
  • Limited visibility of production rates between periodic well tests, making it difficult to optimize operations in real-time

The operator faced the challenge of running lower subcools for optimal production while avoiding low subcool events that could damage producers, all while having incomplete information about actual production rates.

Solution Approach

A multi-phase implementation was designed to develop predictive capabilities before implementing automated control:

Phase 1: Subcool Prediction Development

Continuous-Valued Virtual Sensors

  • Developed multivariate non-linear machine learning models relating control handles (steam, choke, lift gas) to subcool
  • Created separate models for each well pair to account for individual well characteristics
  • Incorporated physics-based thermal diffusivity algorithms to estimate steam effects on the producers to guide empirical model development

Low Subcool Probability Predictors

  • Built probability models to estimate the likelihood of subcool falling below critical thresholds
  • Provided early warning indicators before actual low subcool events

Phase 2: Virtual Metering Implementation

Production Rate Estimation

  • Developed virtual meters to model liquid and gas production rates in real-time
  • Auto-calibrated the models upon detecting new well tests
  • Enabled continuous visibility of production performance between periodic well tests
  • Provided operators with real-time understanding of production response to control actions

Heat Quantification

  • Virtual metering capabilities designed to quantify heat taken from the reservoir in real-time
  • Enhanced understanding of thermal efficiency and recovery processes
  • Provided additional data for optimizing steam-to-oil ratio

Phase 3: Control System Implementation

Hybrid Model-Predictive Control Strategy

  • Model-based optimizer searches predictive models to find choke settings that minimize deviation from subcool setpoint
  • Proportional control algorithm adjusts model recommendations based on real-time subcool error
  • Linear adjustment capability to remove systematic bias
  • Constraint management to enforce minimum/maximum limits and rate-of-change restrictions
  • Setpoints written to Distributed Control System (DCS) to directly control the assets

Technical Implementation

System Architecture

The solution operates on IntelliDynamics’ Intellect Server platform with multiple coordinated tasks:

Data Acquisition

  • OPC DA interface extracts real-time DCS values
  • Key tags archived to Intellect’s onboard data historian
  • Independent task frequencies for different data, prediction and control operations

Virtual Metering System

  • Multivariate genetically optimized machine learning models predict liquid and gas production rates continuously
  • Models trained on historical well test data and operating conditions
  • Real-time estimates provide visibility between physical tests
  • Prototype virtual meters demonstrated favorable results for both liquid and gas production

Multiple Setpoint Model-Based Optimization

  • Operates on settable periodic basis
  • Current operating conditions obtained from historian and smoothed over 1-hour period to reduce sensor noise
  • Ensemble of predictive models searched to find optimal choke position and lift gas rates
  • Results archived for downstream processing

Proportional Control Enhancement

  • Retrieves model-based choke recommendation
  • Adjusts based on current subcool error from setpoint
  • Applies bias correction and enforces constraints
  • Outputs final choke position and lift gas rates to DCS

Adaptive Model Maintenance

  • Compares estimated vs. calculated subcool
  • Computes average error over time
  • Fractionally adapts models to maintain accuracy
  • Maintains virtual meter performance autonomously over time

Data Processing Techniques

Smoothing Strategy:

  • Data extracted from the customer’s data historian, synchronized, and bucketed to 1-minute intervals
  • 50-period simple moving average (50-minute window) applied to most variables
  • Exception: Thermal variables computed hourly and not smoothed due to inherent stability in thermal diffusivity equations

Model Inputs: The predictive models incorporate:

  • Steam injection rates (long-term effect, up to 900 hours for notable thermal impact)
  • Choke position (fast effect with some decay over time)
  • Lift gas injection rates (fast effect with some decay over time)

Results

Prediction Performance

The subcool virtual sensors demonstrated strong predictive capability across all well pairs:

Model Validation:

  • Models validated on 1.5 years of historical operational data per well pair
  • Accurately predicted subcool values during normal operations
  • Successfully captured complex non-linear dynamics and behavioral shifts based on steam rate

Early Warning Capability:

  • Probability indicators began rising well before actual low subcool events
  • In one documented case, the system detected conditions leading to low subcool 3 hours before the event manifested in actual subcool measurements
  • Consistent leading indication across multiple well pairs for various threshold levels (10°C, 12°C, 15°C)

Virtual Metering Performance

Production Visibility:

  • Prototype virtual meters for liquid and gas production showed favorable results
  • Enabled continuous monitoring of production rates between physical well tests
  • Provided real-time feedback on production response to operational changes

Operational Value:

  • Eliminated blind periods between well tests
  • Allowed operators to see immediate impact of control adjustments
  • Enhanced ability to identify and respond to production anomalies
  • Improved understanding of reservoir heat extraction in real-time

Control Performance

Operational Benefits:

  • Achieved subcool values closer to setpoint with reduced variability
  • Automatic adjustment during process upsets eliminated need for constant operator attention
  • Conservative safety margins reduced while maintaining equipment protection
  • Real-time production visibility enabled better optimization decisions

System Reliability:

  • Models successfully adapted to changing well conditions over time
  • Control system handled edge conditions and unusual operating scenarios
  • Seamless integration with existing DCS and data historian infrastructure
  • Virtual meters maintained accuracy over extended operation

Key Findings

Process Dynamics Understanding

Thermal Behavior:

  • Thermal diffusivity approximately 0.0018 m²/hr
  • Injector-producer spacing of 5-8 meters
  • Steam thermal effects can take up to 900 hours for notable impact
  • Thermal diffusivity simulator results aligned with field observations

Control Handle Characteristics:

  • Steam: Long-term effect requiring integrated moving averages for modeling
  • Choke: Fast response with time-limited prediction horizon
  • Lift Gas: Fast response with time-limited prediction horizon
  • Complex interactions between variables requiring non-linear modeling approach

Virtual Metering Insights:

  • Production rates can be reliably estimated from operating conditions
  • Virtual meters provide valuable visibility between physical well tests
  • Real-time production data enhances optimization decision-making
  • Heat quantification aids in understanding reservoir thermal performance

Implementation Insights

Well-Specific Models:

  • Each well pair exhibited unique behavioral characteristics for both subcool and production
  • Separate models essential for accurate prediction and control
  • Models required periodic retraining to capture evolving edge conditions

Hybrid Control Strategy:

  • Combination of model-predictive and proportional control provided superior performance
  • Model-based component anticipated subcool changes based on operating conditions
  • Proportional component corrected for real-time deviations from setpoint

Integrated Virtual Metering:

  • Virtual meters complemented subcool control by providing production context
  • Enabled operators to balance production optimization with subcool management
  • Real-time rate estimates supported faster decision-making

Business Benefits

  1. Risk Reduction: Decreased frequency and severity of low subcool events, protecting producer equipment from steam damage
  2. Operational Efficiency: Reduced operator workload during routine operations and process upsets
  3. Production Optimization: Enabled operation at lower subcool values safely, optimizing production while maintaining equipment protection
  4. Enhanced Visibility: Real-time production rate estimates between well tests eliminated blind periods and enabled faster response to changing conditions
  5. Improved Decision-Making: Integration of subcool control with virtual metering provided operators with comprehensive real-time understanding of well performance
  6. Predictive Capability: Early warning system allows proactive intervention before critical conditions develop
  7. Thermal Efficiency Monitoring: Real-time quantification of reservoir heat extraction supports steam-to-oil ratio optimization
  8. Scalability: Proven methodology applicable to additional well pairs with documented procedures for model development and deployment

Lessons Learned

Technical

  • Integration of physics-based understanding (heat equation) with empirical modeling produces robust predictive models
  • Smoothing strategies must account for different update frequencies and inherent stability of different process variables
  • Adaptive model maintenance essential for long-term performance in evolving process conditions
  • Virtual metering adds significant value when integrated with control systems, providing operational context for control decisions
  • Production rate models benefit from same fundamental approach as subcool models—combining physics understanding with data-driven techniques

Operational

  • Pilot implementation on limited well pairs allows validation before broader deployment
  • Operator confidence builds through demonstrated early warning capability before full automation
  • Virtual meters provide immediate value even before automated control implementation
  • Real-time production visibility enhances operator trust in automated control recommendations
  • Clear documentation of cloning procedures enables efficient scaling to additional wells

 

Conclusion

This project demonstrates that advanced subcool control integrated with virtual metering in SAGD operations is achievable through a combination of physics-based process understanding, empirical modeling, and hybrid control strategies. The solution provides predictive early warning, automated control capabilities, and continuous production visibility, enabling operators to optimize production while protecting critical equipment.

The phased approach—starting with prediction and probability indicators, adding virtual metering for production visibility, then implementing full automation—allowed validation of the technology and building of operational confidence. The system’s ability to detect low subcool conditions hours before they manifest provides significant value even in manual operation mode, while the automated control capability reduces operator workload and enables operation closer to optimal setpoints.

The addition of virtual metering eliminated blind periods between well tests, providing operators with real-time understanding of production response to control actions and enabling more informed optimization decisions. This integrated approach to subcool control and production monitoring represents a significant advancement in SAGD operational excellence.

Technologies Utilized:

  • Model-predictive control
  • Virtual metering / soft sensors
  • Committees of genetically optimized non-linear multivariate machine learning models
  • Thermal diffusivity modeling
  • OPC DA communication
  • Real-time data historian
  • Adaptive model maintenance

Application Areas:

  • SAGD operations
  • Heavy oil production
  • Thermal recovery processes
  • Advanced process control
  • Production optimization

Key Functional Features:

  • Subcool virtual sensors
  • Low subcool probability prediction
  • Automated subcool control
  • Liquid production virtual meters
  • Gas production virtual meters
  • Real-time heat extraction quantification
  • System self-maintenance through adaptive modeling