ADVERTISEMENT

Compositional Simulation, Artificial Intelligence Optimize Water Injection

You have access to this full article to experience the outstanding content available to SPE members and JPT subscribers.

To ensure continued access to JPT's content, please Sign In, JOIN SPE, or Subscribe to JPT

The complete paper discusses optimization of a development plan involving low-salinity water injection (LSWI). This methodology combines compositional simulation and a mathematical optimization tool that uses artificial intelligence to maximize the net present value (NPV) of the project under evaluation. The adapted methodology allowed an optimal development plan, considering the uncertainty associated in the reservoir, by use of multiple geostatistical realizations and simultaneous history-matched models.

LSWI

LSWI is an enhanced-oil-recovery technique in which the salinity of the injected water is controlled with the objective of increasing the recovery factor. Of the mechanisms proposed for LSWI, wettability alteration is the most widely accepted for describing the increase of the recovery factor. The wettability change in reservoirs with the presence of sandstones mainly is attributed to the multionic exchange between the injected fluid and the clay surface, as is the double layer expansion. From a technical point of view, the LSWI success is related strongly to reservoir lithology.

An integrated analysis and optimization study was performed using numerical simulation to evaluate LSWI as an alternative for increasing the recovery factor in the Namorado field in the Campos Basin of Brazil.

Because LSWI alters the initial chemical equilibrium and induces changes in the system, the modeling of geochemistry is a fundamental part of the simulation in addition to the compositional modeling of fluids. During LSWI, the following three types of reactions can occur simultaneously:

  • Reactions in the aqueous phase
  • Mineral dissolution and precipitation reactions
  • Ion-exchange reactions

Considering that the chemical changes generated in the reservoir during LSWI induce the wettability alteration, this effect must be factored into the modeling. The effect of the wettability alteration is modeled by modifying the relative permeability curves using an interpolant as the equivalent fraction of an ion subject to cation exchange on the surface of the rock—the species molality in the aqueous phase or the mineral volume (fraction). Definition of multiple relative permeability curves for the same type of rock is possible, wherein each group of curves corresponds to a defined interpolant value. Generally, two groups of relative permeability curves are defined, the first representing the conditions of high-salinity water injection (HSWI) and the second those of LSWI.

Model Description

The model used as a reference for the current simulation study corresponds to a synthetic model for development and reservoir management applications that represents the Namorado field. The structural characteristics, facies, petrophysical properties, fluid properties, well information and production history data were obtained from public data for the Namorado field provided by the National Petroleum Agency of Brazil.

The study evaluated HSWI and LSWI as recovery techniques; the reference model data were used to build a model compatible with a commercial compositional simulator.

The simulation model corresponds to a corner-point grid with 81×58×20 cells (100×100×8 m), of which 38,466 are active cells (Fig. 1). The fluid model has seven pseudocomponents that represent 33.18 °API crude. The reservoir has a reference pressure of 32 068 kPa at a reference depth of 3000 m. The water/oil ­contact depth was defined at 3100 m.

Fig. 1—Simulation model.

 

LSWI Optimization Methodology. A software tool that combines advanced statistical analysis, machine learning, artificial intelligence, and data-interpretation techniques was used to maximize project NPV. The optimization project was executed in three stages.

Assisted History Match. For the Namorado field, the simulation start was defined as 1 June 2013. Historical data information were reported for four vertical producing wells for 4 years. With this information, the assisted history-matching process of the simulation model was performed to create combinations of the most-influential parameters to achieve the history match.

For the Namorado field, assisted history-matching parameters are related to rock compressibility. Water/oil relative permeability curve parameters and porosity permeability and net/gross ratio distributions were considered. For the field, 500 equiprobable geological realizations are available in which the attributes of porosity, permeability, and net/gross ratio are defined. The porosity distribution was estimated on the basis of field facies distribution, as well as the net/gross ratio, while the permeability distribution was determined on the basis of porosity.

The relative water/oil permeability curves were modeled using the Corey equation. A Bayesian optimizer was used to obtain the models with the best history match, minimizing the global error between the simulated and the historical data for the cumulative oil, water, and gas production; average reservoir pressure; and well bottomhole pressure for each well. For this study, an acceptable history match occurred when the global error was equal to or less than 5.5%. The results obtained for the global error function present errors in the range of 3.62 to 5.44%.

The history-matching results obtained for the oil, gas, and water cumulative production establish that the Bayesian optimization algorithm obtained models with a good representation of historical data.

Probability Forecast. Using the 50 best cases obtained from the assisted history match, the production probabilistic forecast for a period of 26 years was performed under a primary production scheme (or base case) of the four vertical producing wells in the field. During the forecast, the wells were constrained by the last pressure-drop value, recorded individually for each at the end of the history match. Additionally, a maximum liquid rate constraint of 2000 m3/d and a minimum well bottomhole pressure of 17 632 kPa were imposed. As a well-­monitoring constraint, a minimum oil production rate of 20 m3/d and a gas/oil ratio of 200 m3/m3 were defined.

The primary production scheme was used as a base case to interpret field production behavior if the current development plan is continued; additionally, it was used to quantify the incremental oil-recovery factor during the evaluation of a different development plan. On the basis of the accumulated oil production values, the scenarios P10, P30, P50, P70, and P90 were selected and were subsequently used as base case for the selection and optimization of the recovery strategy.

Robust Optimization. Considering that the robust work flow involves several optimization scenarios simultaneously, two different study categories must be defined, termed as the master study and dependent studies.

The master study is responsible for generating experiments according to the design defined by the optimization algorithm. The experiments created by the master study are executed for all dependent studies. For each dependent study, the values of the objective functions are calculated and then transferred to the master study to calculate the overall objective function.

Dependent studies are used to represent each of the selected realizations in the probabilistic forecast. This allows evaluation of the uncertainty associated with the distribution of properties such as porosity, permeability, and net/gross ratio in the NPV optimization process.

The objective function defined for the dependent studies was the NPV. The NPV in this case is composed of five terms representing the initial investment in surface facilities and new wells, revenues from the oil sales, oil-production costs, and water produced and injected cost. The robust objective function was defined as the average NPV calculated for each of the dependent studies.

The optimization method performed approximately 240 simulation jobs to obtain the optimal NPV. The optimal case shows that the recovery method to be implemented is LSWI, which reaches an NPV of $2,453,000,000 using a total of 18 new wells (12 are producers, and six are injectors).

The values obtained for the NPV in each dependent study are consolidated. For the Namorado field, the NPV for the P10 scenario was $1,807,000,000, while, for the P90, it was $3,315,000,000.

With regard to the recovery factor, the best method to implement is LSWI in comparison with conventional water injection. In the optimal LSWI case, for scenario P10 the recovery factor was 40.75%, while, for P90, it was 42.21%.

This article, written by JPT Technology Editor Chris Carpenter, contains highlights of paper SPE 198995, “Low-Salinity Water-Injection Optimization in the Namorado Field Using Compositional Simulation and Artificial Intelligence,” by Diana Mercado Sierra, SPE, Argenis Alvarez Rojas, and Victor Salazar Araque, Computer Modelling Group, prepared for the 2020 SPE Latin American and Caribbean Petroleum Engineering Conference, 27–31 July, Virtual. This paper has not been peer reviewed.

Compositional Simulation, Artificial Intelligence Optimize Water Injection

01 September 2020

Volume: 72 | Issue: 9

ADVERTISEMENT


STAY CONNECTED

Don't miss out on the latest technology delivered to your email weekly.  Sign up for the JPT newsletter.  If you are not logged in, you will receive a confirmation email that you will need to click on to confirm you want to receive the newsletter.

 

ADVERTISEMENT

ADVERTISEMENT

ADVERTISEMENT

ADVERTISEMENT