HM Background
History matching involves constructing numerical models (representing a reservoir) that align with observed, measured data. Throughout the model calibration process, it's crucial to keep the following in mind:
- Reservoir models are simplifications of reality, inevitably containing errors and approximations inherent in any approximation of physical phenomena.
- The purpose of history matching is decision-making; it serves no standalone purpose.
- Model inputs are uncertain, with uncertainties that are inevitably underestimated.
- Observable data always includes a degree of error and approximation.
- History matching is to be viewed within a broader framework of uncertainty quantification. Proper decision-making requires assessing the uncertainty of outcome, and history matching aids in constructing a model of uncertainty for informed decision-making processes.
Generalized Reservoir Modeling Workflow
Most, if not all, reservoir modeling workflows share several fundamental elements.
First, one or multiple reservoir models are constructed, each requiring a set of model parameters. The engineer or geoscientist then adjusts these parameters through sensitivity runs. Sensitivity analysis helps identify which parameters significantly influence model responses and which do not.
Screening is the process of selecting models, often iteratively, and requires generating additional models by resampling parameters. The selected models then undergo refinement before reaching the final stage of the workflow and objective: to forecast future reservoir production scenarios with a reliable degree of uncertainty.
The general modeling workflow consists of key elements: Parameters, Sensitivity Runs & Analysis, Screening, and Model Refinement.
At its core, history matching involves selecting reservoir models that best align with both static and dynamic reservoir data. This process primarily consists of screening, followed by model refinement.
Traditionally, screening is performed by a geoscientist, who selects a single "best" model based on static data, with little or no consideration to dynamic historical data, such as well phase rates and well pressures. The reservoir engineer then modifies this model to match historical production data—a difficult and time-consuming process. Most often, this refinement results in a static model that differs significantly from the one originally selected during screening.
Some advocate for sophisticated (and costly) history matching software that claims to integrate sensitivity analysis, screening, and model refinement into a single black-box approach. However, no single method is universally suitable. The most critical aspects are not just the history matching technique but also how the reservoir model is parameterized and the workflow used to construct models.
Classical HM Workflow
Traditional (manual) history matching involves modifying the fluid flow model during the model refinement stage and is still commonly used in assisted history matching methods. The workflow, illustrated in the diagram below (left), follows these steps:
- Geological Model Creation – The geological model serves as the foundation for the fluid flow model, often requiring an upscaling step.
- Flow Simulation – The flow model is simulated based on the geological input.
- History Matching Modifications – Adjustments are applied directly to the flow model to improve the match with historical data.
Integrated G&G Workflow
Only in the past decade have researchers and software developers sought to integrate the geological modeling process within the history matching loop. In this enhanced workflow, the history matching algorithm can adjust uncertain parameters in both the geological model and the flow model.
This approach addresses some of the challenges of traditional methods and is expected to produce more reliable history-matched models. However, its implementation is significantly more complex.
Regardless of the history matching workflow, certain critical tasks should remain the responsibility of the reservoir engineer and geomodeling team rather than relying on an "automated" approach. These tasks include:
- Interpreting reservoir data
- Parameterizing the reservoir model(s)
- Determining the uncertainty of input parameters and reservoir data
These are inherently subjective decisions that significantly impact the quality of the model—and, consequently, the history match. The interpretations, assumptions, and decisions made by the geomodeling team ultimately shape the numerical reservoir model and should be carefully examined and questioned throughout the modeling process. More often than not, failures in history matching stem from incorrect assumptions and interpretations rather than flaws in the history matching methodology itself.
The key question, then, is: How can we critically evaluate these interpretations, assumptions, and decisions—collectively referred to as the "prior model"—before initiating history matching? One proposed solution involves the use of metric space methods.
Model Refinement
For complex, reservoir models with many wells (100's), it is inevitable to have a significant fraction of wells do not match historical data even after significant history matching efforts. The problem is particularly challenging for secondary or tertiary displacement processes, such as water, polymer, or CO2 floods, where inter-well geological properties play a crucial role in fluid flow.
To improve the history match in such cases, the reservoir engineer will be forced to adjust the petrophysical properties between injector-producer pairs but presents several challenges:
- How to identify specific grid blocks where modifications should be applied
- Determine the magnitude of property adjustments
- Ensure the modifications remain consistent with the geological data used to construct the model in the first place, including spatial correlations, facies interpretations, histograms, and seismic data.
Balancing these factors is challenging in achieving a history-matched model that is both geologically and physically plausible. This final constraint is often overlooked in practice, well knowing that disregarding geological data can lead to poor predictive models.
Geologically-Consistent History Matching
Geologically-consistent history matching seeks to integrate production data with other key geological inputs—such as core samples, geological intuition, seismic data, and well tests—ensuring a more consistent and hopefully more reliable reservoir model. Geologically-consistent history matching is principally a constrained data integration problem rather than a pure optimization problem. The key idea in geo-consistent history matching is the use of probability models that account for all data including well logs, cores, seismic, geological scenarios, dynamic production data, tracers, etc.