A Machine Learning Approach to Predict Gas Oil Ratio Based on Advanced Mud Gas Data

Heavy oil fields are characterised by complex chemistries and exhibit a wide range of physical properties (especially oil viscosity). This paper describes the development of two innovative and cost-effective data acquisition methods: the optimised use of mud gas data and the geochemical characterisation of cuttings extracts, both as an alternative data source for oil viscosity determination. Applications of these methods can lead to the cost reduction of data acquisition programs and potentially improvements in oil recovery. The mud gas provides real-time surface logging data. Due to the nature of heavy oil in the case study wells, the mud gas compositions contain mainly C1-C3. By comparing with the reservoir oil samples from the exploration and appraisal wells, we investigated the possibility of using mud gas while drilling to identify if the reservoir oil belongs to the high viscosity or medium viscosity group. For cutting analysis, we thoroughly investigated solvent extracts from cuttings from the exploration and appraisal wells and correlated them with the oil viscosity from the corresponding PVT samples. When such a correlation is established, we can then apply the correlation to predict oil viscosity from the cuttings of new production wells. We achieved excellent results for both real-time oil viscosity identification from mud gas data and post well oil viscosity prediction from the extraction of cuttings. The C1/(C2+C3) is the key parameter for oil viscosity identification from the standard mud gas data in the field. For the past wells, the mud gas interpretation has high accuracy for determining if the reservoir oil belongs to the high viscosity or medium viscosity regions. We established a strong correlation between the biomarkers and the oil viscosity from the exploration and appraisal wells. When we applied the correlation to new production wells, the oil viscosity prediction was very close to the viscosity measurement from the wellhead PVT samples. The geochemical techniques provide oil viscosity for critical decisions regarding well placement and reservoir management modelling. The new source of reservoir fluid data from mud gas and cuttings provides real-time fluid interpretation while drilling and close to PVT sample quality oil viscosity prediction after drilling. This new approach is cost-efficient and has a large potential to replace the downhole fluid sampling or expensive downhole logging tools for production wells.

Downhole sampling in gas condensate reservoir is well known to be challenging due to the nature of the near critical fluids. Reservoir fluid properties can change dramatically with slightly changes in reservoir pressure and temperature. As a result, accurate and representative PVT data are essential for reservoir fluid modeling and field development purpose. This paper presents the first downhole gas condensate sampling in a deep gas condensate field, offshore East Malaysia. Samples collected from the previous surface tests showed large variation in Condensate Gas Ratio (CGR), from 50 to 200 stb/mmscf. This resulted in large uncertainty in the dew point pressure, condensate yield, reservoir fluid productivity, and also reservoir fluid typing, i.e. rich or lean gas condensate reservoirs. As a result, there are strong needs to acquire high quality downhole samples to reduce these uncertainties, which may affect the entire field development plan. This paper illustrates several key challenges in this particular case study; for example, the risk of stuck tools required us to minimize stationary time during clean up process. Typically in the same area, the usual time allowed per station is less than 2 hours. However, since this gas condensate sampling job is much more complex in its fluid behaviour than other types of reservoir fluids, the wireline formation tester tool were allowed to be slightly longer at each station. Furthermore, best possible downhole fluid samples were requested for PVT modeling. As a result, minimizing OBM contamination and controlling drawdown pressure drop in low mobility reservoir were added into operation challenges. The use of new technology Wireline Formation Tools (WFT) such as focused sampling probe and in-situ fluid analyzer were introduced to ensure that job objectives can be met with limited time per station. The focused sampling approach was used to collect representative PVT sampling in several stations. The direct comparison between conventional and focused sample probes in the same zone will also be presented in this paper. During the pump-out period, the guard and sampling flowline pump rates were controlled and pressure drop between each flowline was monitored closely in the real time to ensure that the clean single phase reservoir gas is flowing though the sample flowline. When the clean gas condensate with minimum OBM contamination is detected from different measurements such as stabilized GOR, live density and composition using the in-situ fluid analyzer tool, PVT samples were collected at each station. It seems like the focused sampling technique procedure is straight forward. However, when different types of mud, i.e. WBM and OBM, was used in different hole sections, the two-phase flowing behaviour, i.e. gas condensate in OBM and in WBM behave differently, and therefore, this two-phase flowing effect requires different operation procedures for this particular reservoir in such a limit time frame.

The evaluation of petroleum resources requires a good understanding of the reservoir fluids and their behavior. Specifically, PVT (pressure-volume-temperature) data are required at the earliest possible stage to develop economic evaluations, interpret well test data, design production equipment, and predict future reservoir performance. A fluid sample obtained with an openhole wireline formation tester tool provides the engineer with the first (and potentially the only) chance to analyze a representative reservoir fluid sample before significant decisions are made. The fluids must be characterized by direct measurement of certain properties and the quality of the PVT laboratory results depends almost entirely on the validity of the sample used for the analysis. This paper compares the fluid PVT data obtained using samples from wireline formation tester tools and from surface production test equipment. The PVT laboratory procedures and the accuracy of the analysis methods are discussed. The compositional analyses of the hydrocarbon samples are compared, from methane to heptanes-plus chemical structure, along with the key volumetric parameters such as density, bubblepoint pressure, viscosity, formation volume factor, and gas-oil ratio. Different types of reservoirs (with fluids ranging from low gas-oil ratio oils to dry gas) were examined to determine the effects of mud filtrate invasion in the openhole environment, type of fluid, and reservoir characteristics. In general, the results indicate that samples obtained with wireline formation tester tools can provide representative reservoir fluid parameters from compositional analyses to basic volumetric behavior as a function of pressure and temperature. Critical reservoir parameters such as the fluid density, bubblepoint pressure, formation volume factor, viscosity, and gas-oil ratio are consistent within about 15 percent for most oil reservoir fluids. Gas condensate reservoir fluids present the most difficult sampling conditions for both the downhole and surface sampling operations. Careful sampling techniques and detailed monitoring of the downhole and surface fluid conditions are required to reconstruct useful results for the reservoir fluid properties.

Downhole fluid analysis (DFA) has successfully been used to delineate reservoir connectivity and fluid properties and to understand the origins of many complexities in oil reservoirs with the Flory-Huggins-Zuo equation of state (FHZ EOS). Equilibrium asphaltene gradients strongly imply reservoir connectivity, with fluid equilibration often taking tens of millions of years. However, reservoir fluids often demonstrate complicated compositional gradients from undergoing dynamic processes such active late gas charges, biodegradation, and water washing. In this paper, a simple 1D two-component diffusive model with analytical solutions is developed for taking into account dynamic processes in oil reservoirs. Two field applications of the developed diffusive model are active gas charging and biodegradation to better understand these dynamic processes. In the first field application, an active late gas charge results in huge nonequilibrium gradients in several fluid properties including API gravity, gas/oil ratio (GOR), saturation pressures, and asphaltene content (color optical density) measured by DFA and surface laboratory analysis. Because the reservoir is not in equilibrium, these fluid property gradients cannot be modeled by the equilibrium EOS model. Thus, a diffusive model is needed to account for the large GOR gradient. The diffusive model predicts the large variations of GOR gradients successfully. The asphaltene gradient is evidently under control of convective currents induced by the gas charge. The exact nature of this process is currently under investigation. The asphaltene gradient is consistent with the predictions by the FHZ EOS with the assumption of asphaltenes locally in equilibrium with the local fluids described by the diffusive model. The second field application is biodegradation occurring at oil/water contact (OWC). Compared to the predictions of the FHZ EOS, it is observed that the upper half of the oil column follows an equilibrium asphaltene distribution well. However, a much larger asphaltene gradient is found in the lower half of the reservoir, which gives rise to a huge (8×) viscosity variation and affects production. The measured asphaltene distribution in the whole oil column can be described nicely by the diffusive model along with the FHZ EOS. The diffusion of alkanes to the OWC (where they are rapidly consumed) is a control step. The consumption of alkanes at the OWC reduces the oil volume and increases asphaltene content and viscosity. Petroleum system modeling predicts initiation of reservoir charging in the Eocene matching the diffusive times required in the FHZ EOS modeling.

With the increased exploration and development of unconventional reservoirs, the complicated production mechanisms of unconventional wells have gradually become a hot topic among the oil and gas industry. Due to the ultra-low permeability and porosity, the fluid phase behavior in shale reservoirs significantly differs from the conventional fluid phase behavior, increasing the production forecasting complexity. A substantial effort to better understand the mechanisms is the ability to characterize the unconventional well gas-oil ratio (GOR) behavior. The GOR always plays a critical indicator to help predict long-term oil/gas production trends and develop appropriate production strategies. In this paper, GOR behavior was discussed based on an unconventional parent-child horizontal well set in the Eagle Ford shale formation. Subsequently, fracture hit intensity can be determined through the producing GOR characterization. Afterward, the historical production data were well matched. The long-term GOR trends (20 years) were then predicted with the calibrated reservoir model. Based on the simulation results, an interpretation of the fracture hit impact on GOR behavior, and the well productivity was established. This study provides some key insights into GOR behaviors, especially for the parent-child well GOR trends with considering the impact of fracture hits. The Eagle Ford GOR is strongly influenced by the flowing bottomhole pressure. Meanwhile, the GOR trends of both parent and child wells are extremely sensitive to fracture hits, strong correlations between GOR and fracture hits are observed. Compared to the parent well, the flat GOR period of the child well is much shorter due to pressure depletion. The existence of a child well also reduces the rising speed of the parent well with a lower plateau. In addition, the long-term production prediction shows that fracture hits negatively influenced both well performances, where the child well has a more severe production loss than the parent well. Through the findings presented in this work, a better understanding of the unconventional well GOR behaviors can be obtained. The analysis approaches proposed in this paper provide valuable insights into GOR characterization and contribute to the production forecasting from unconventional plays. The results can help to improve the efficiency of reservoir management, field development, and economic valuation in future projects.

The Norman Wells reservoir was depleted primarily by solution gas drive until 1984 when a field-wide pattern water flood was implemented. The resultant reservoir repressurization has had an impact on the PVT properties. This paper presents the results of a 1988 study undertaken to determine the current reservoir fluid properties. The results indicate that:the current methodology for tuning the Peng-Robinson equation-of-state could not simultaneously match the bubble point pressure (Pb) and the corresponding solution gas-oil-ratio (Rsb) of these samples;neither the 1965 properties nor Standing's correlation could be used to match the PVT behavior of the new samples; andthe bubble point pressure varied widely while Rsb was nearly constant, It was found that preferential evolution and subsequent redissolution of light gases (CH4, N2 and CO2) caused major changes to the bubble point pressure but very little change to Rsb. The new PVT properties were used to:illustrate that reservoir fluid withdrawals are now balanced by injected volumes;improve the gas production proration factor by 15%; andprovide more accurate data for facility design and volume metering considerations. Introduction The Norman Wells Pool is located directly under the Mackenzie River, about 80 km south of the Arctic Circle in the Northwest Territories of Canada (Fig. 1). It is a layered Devonion limestone reef complex with natural fractures and matrix permeabilities of less than 10 mD(1). The pool was discovered in the 1920s and underwent its first major development during the 1940s when wells were drilled on the mainland and on two natural islands (Fig. 2). Production levels were determined by the demand for petroleum products in the North. In the early 1980s a waterflood operation was started with the mainland wells. Later, the pool was fully developed for a pattern waterflood using directional drilling and artificial islands. Waterflood operations began in 1985. It was recognized that the history of Norman Wells, which included intermittent injection of water and petroleum products, solution gas drive resulting in high producing gas-oil ratios (GOR), and a really inconsistent depletion of a tight reservoir, would cause changes in the reservoir fluid properties. Once the reservoir was repressurred from pool wide waterflood operations, reservoir fluid sampling and analysis could be carried out to quantify the changes. After first reviewing the relevant historical aspects. The sampling procedures and measured properties will be presented. Then the analysis and conclusions which enabled the practical application of the results will be discussed. Historical Operations Upon discovery in the 1920s, the reservoir pressure was 5500 kPa. In the 1940s some 60 wells were drilled under the terms of the Canol Agreement between the United States Army, Canada and Imperial Oil(2). A field-wide pressure survey done in 1945 showed a slight decline in pressure to 5200 kPa. As shown in Figure 3, from this time on through to the early 1980s the active well count ranged from 40 to 50 and the pool oil rate was 300 m3/d to 400 m3/d.

Pressure transient analysis has been known as an important tool to properly characterize a reservoir. The scales of measurement for different testing methods depend mainly on their radius of investigation, i.e. how far pressure changes/disturbance can be seen into reservoirs. In a large reservoir, a fullscale testing method such as a Drill Stem Test (DST) is required to test enough reservoir volume for flow assurance purposes. To test for the reservoir extent, a production test is required to reach reservoir boundaries. In recent years, there have been tremendous developments for short period testing methods. The Formation Tester (FT) is one of the tools that are used to evaluate reservoir fluid properties and acquire pressure transient data from both dual packer and single probe configurations. With the current Downhole Fluid Analyzer (DFA) technology, reservoir fluid properties such as composition, Gas-Oil-Ratio (GOR), in-situ density, and viscosity can be obtained in real time. These insitu fluid properties can be used to analyze pressure transient data for reservoir permeability, skin factor, and reservoir heterogeneity; and as a result, the DFA serve to push IPTT applications to another step forward. This paper will specifically focus on the application of DFA for IPTT purposes and also a continuation of several published papers which discussed the use of different scales of pressure transient data. The results from this paper will mainly focus on the use of single probe and dual packer IPTT to obtain reservoir properties in thinly bedded reservoirs with waxy crudes from the South East Asia region. The viscosity variation from oil based mud filtrate to reservoir fluid during the pump-out and its effect on the pressure derivative will be discussed using actual examples to demonstrate this point. The end results may contradict traditional testing methods on the use of virgin reservoir fluid properties for pressure transient analysis. Hopefully, this paper might open another option when FT is deployed together with the current DFA tool for quick and effective reservoir evaluation in thinly bedded reservoirs which are often encountered in many parts of the world.

Calculated wellbore pressures were obtained for parameters of radius ratio and permeability. In all cases but two, after-production was allowed to occur for one day. The calculated pressure build-up data were compared with actual pressure build-up data from a condensate well. The paper discusses the comparison, gives reasons for, inability to match calculated and actual data, and presents conclusions derived from the comparison. Introduction Considerable information has been published on the decline in wellbore pressure for oil and gas fields resulting from production of fluids. Considerable information also has been published on pressure build-up for oil fields, but relatively little has been published for gas and condensate fields. Perrine summarized the equations derived for pressure build-up in oil wells and showed how the different methods could be applied. On the other hand, very little theoretical information has been published on pressure build-up in gas or gas-condensate wells. Tracy, by pointing out the similarity between the equation describing oil wellbore pressure decline and the equation describing gas wellbore pressure decline, showed how methods of pressure build-up analysis could be applied to a gas well. In analyzing pressure build-up data for gas or gas-condensate wells, it would be very desirable to compare actual pressure build-up data with calculated pressure build-up data. Such a comparison would lead to a better feeling for the quantitative picture of gas-well build-up curves and would help to establish a degree of confidence in a given build-up curve. This paper discusses the comparison and presents conclusions made from it. Calculated wellbore pressures are given for parameters of radius ratio and permeability. In all cases but two, after-production was allowed to occur for one day. The work was done in an attempt to evaluate the reserves of a gas-condensate field. As will be discussed later, it was not possible to make an evaluation. METHOD OF CALCULATION The method used for solving the depletion and pressure build-up behavior for a gas-condensate well was that developed by West, Garvin and Sheldon. For the IBM 704 computer program used, it was assumed that liquid condensing out of solution would have only a minor effect on the flow behavior. Therefore, the field was treated as having single-phase flow, with the gas saturation Sg remaining constant at unity minus the connate-water saturation Sw. As will be discussed later, this may not have been a good assumption. It also was assumed that oil production could be considered as equivalent gas production by using a conversion factor based on an analysis of the liquid produced. An outer boundary condition of no-flow was assumed. BASIC DATA The data required for the study were reservoir properties, fluid properties and production data. Reservoir properties included the following: thickness h, 179 ft; porosity, 0.194; connate-water saturation, 0.43; gas saturation, 0.57; permeabilities k, 1.0, 0.5, 0.25 and 0.15 md; radius ratios re/rw, 800, 1,000 and 2,000; wellbore radius rw, 5 in.; initial reservoir pressure Pi, 6,529 psia; and reservoir temperature, 272 degrees F. The selection of permeabilities was based on core-analysis data showing an average air permeability of about 2 md. With a relatively high connate-water saturation, the average effective gas permeability would be about one-half, or less, of the air permeability. The selection of radius ratios was based partly on the observed rapid decline in wellbore pressure during production and partly on the fact that the well was believed to be located in a relatively small fault block. Reservoir fluid properties were obtained from a reservoir fluid study on a recombined sample of gas and condensate. Gas formation-volume factors were calculated from pressure-volume measurements made at reservoir temperature. Gas viscosities were calculated by the method of Carr, et al, from an analysis of the recombined sample. These data are shown as a function of pressure in Table 1. JPT P. 1131^

This research introduces a fluid characterization tool designed for the oil and gas industry, focusing on improving the understanding of fluid composition by correlating Gas Oil Ratio (GOR) and fluid properties. The tool was developed using a dataset that includes GOR values ranging from 1000 to 6000 scf/stb, bubble point values between 3500 and 5500 psia, and API gravity values from 40° to 55°. This dataset was collected from over 40 wells in the Delaware Basin, ensuring precision and avoiding the manual errors associated with traditional analog tools. Traditionally, the process involved finding an analog well, a step prone to errors due to variations in proximity and fluid properties. The tool, however, establishes correlations between GOR, fluid properties, and relevant data points, creating an algorithm for deriving reservoir fluid composition. This composition seamlessly integrates into the Global Fluid Characterization (GFC) Model. The fluid characterization tool proved to be accurate and reliable in characterizing fluid composition. It provided immediate and trustworthy results through its user-friendly interface. With only GOR values, professionals can get a good estimate of reservoir fluid composition. The tool eliminated the need for laborious and expensive laboratory analyses, making it a valuable time-saving and cost-effective solution. The dataset used for its development, encompassing a wide range of GOR from 1000 scf/bbl to ~5000 scf/bbl. Additionally, the capability of the tool can be enhanced as new wells are incorporated in the Global Fluid Characterization (GFC) Model. It enables us to use compositions for various purposes like making informed decisions in reservoir simulations, well testing, and production activities.

Arps in 1944 developed decline curve equations for analyzing reservoir/well production decline and the estimated decline rate constant is used for production forecast. In making reservoir production forecast, the Arps equations are usually calibrated to match the reservoir historical production decline by adjusting the decline rate constant value. The decline rate constant indirectly influences the reservoir properties responsible for production decline. However, the value assigned to the decline rate constant might either overestimate or underestimate the uncertainties in the reservoir rock and fluid properties responsible for reservoir production decline. The decline rate constant in the Arps equation has no equation relating it to the reservoir properties that are responsible for the reservoir production decline hence; the direction of uncertainty quantification on these reservoir parameters is not explicit, therefore undermining the predictive power of the Arps equations. The impact of reservoir properties uncertainties on the production rate decline and forecast by Arps equations cannot be examined due to the empirical nature of the Arps equations. Also, the decline rate constant is not explicitly related to reservoir rock and fluid properties responsible for production decline. However, the predictive power of the Arps model can be greatly improved if the decline constant can be expressed as a function of the reservoir rock and fluid properties responsible for production decline. Hence, the impact of uncertainty on the production forecast by Arps models can be examined. The objective of this work is to introduce reservoir rock and fluid properties into the Arps decline curve equations. This will enable the imposition of reservoir physics on the decline curve equations and the impact of uncertainties in these properties on matching historical production decline and forecast. In addition, a relationship between the decline rate constant in the Arps equations and the reservoir rock and fluid properties that influence reservoir production decline has been developed. A mechanistic approach using dimensional analysis was used to develop the relationship. The decline rate was assumed to be proportional to the various reservoir rock and fluid parameters, such as permeability, porosity, net-to-gross, drainage area, fluid density, and viscosity, that could influence production decline and by unit dimensional analysis, a relationship was developed between the decline constant and these reservoir rock and fluid properties. The decline rate constant was directly related to the product of reservoir permeability, fluid density, and the square of the pressure difference but inversely proportional to the cubic power of fluid viscosity. The relationship developed for the decline constant in this work will give the modified Arps model strong predictive power and allows for the impact of uncertainties on the production forecast to be evaluated. The results obtained from the modified Arps equations give better reservoir properties dependent outcomes than the traditional Arps equations and enable analysis of the effects of reservoir rock and fluid properties uncertainties on the production decline and forecast.

_ This article, written by JPT Technology Editor Chris Carpenter, contains highlights of paper SPE 220733, “Field Study of the Condensate-Banking Effect in an Unconventional Gas-Condensate Reservoir With Hydraulically Fractured Wells,” by Jichao Yin, SPE, Han (Jake) Li, and Cyd Marie Ruidiaz-Santiago, SPE, Occidental Petroleum, et al. The paper has not been peer reviewed. _ This paper investigates condensate-banking effects on well performance by conducting field-modeling studies on Delaware Basin deep Wolfcamp condensate producers using compositional simulation models with hydraulic fractures. The models found condensate banking to be of minimal concern; condensate dropped out mostly in fracture networks and could be efficiently produced through highly conductive fractures. An optimal development plan with an aggressive flowback strategy, validated by actual Permian Basin gas-condensate well performance, is proposed that should improve project value without condensate-banking risk. Introduction In the past decades, the Delaware Basin has been developed with horizontal wells targeting the Bone Spring and Wolfcamp formations. The operator drilled two appraisal wells in late 2023 to test the deep Wolfcamp in New Mexico Field A. Key aims of this study were to analyze how well performance changes with well spacing, stacking scheme, completion designs, and drawdown strategy to improve well productivity and economics, thus optimizing deep-bench drilling-spacing-unit design, which aligns with field-development goals. Geology New Mexico Field A covers more than 14,000 surface acres. Multiple productive zones have been established in the Bone Spring and Wolfcamp formations, ranging from 8,500 to 11,500 ft true vertical depth. Six landing zones (Landings A through F) are identified within the Wolfcamp formation. The upper zones (Landings A through C) have been heavily developed to date in Field A, while deeper zones remained undeveloped until appraisal drilling in 2023. Landings D and E are the deepest zones drilled to date, at 12,000–12,500 ft. With increasing depth, the reservoirs in Field A transition from volatile oil in the Bone Springand upper Wolfcamp to gas condensate in the deeper Wolfcamp as a result of increasing thermal maturity. Landings D and E are unconventional targets, with hydrocarbons generated in place from the organic-rich mudstone and permeability in the nanodarcy range, requiring hydraulic fracture stimulation to achieve economic flow rates. Development Opportunities and Challenges The deep Wolfcamp contains primarily retrograde gas condensate with a high gas/oil ratio (GOR), which poses many challenges for future development. One well-known risk is that posed by condensate-banking effects. When pressures drop below the dewpoint, gas-condensate reservoirs are prone to wellbore-productivity losses. In conventional reservoirs, operators commonly maintain flowing bottomhole pressures close to dewpoint pressures by controlling drawdown or injecting gas. However, using this approach with unconventional reservoirs raises several concerns. First, most unconventional fields are developed with primary depletions without injection for pressure maintenance. Second, the multistage hydraulic fractures effectively extend a well beyond its wellbore and fundamentally alter the pressure distribution. Most importantly, with a rapid pace of unconventional development, a controlled drawdown could significantly delay production, thereby decreasing project net present value (NPV). For these reasons, the severities of unconventional well condensate banking must be reevaluated to provide a more-effective mitigation solution for optimal field development.

To increase the production and recovery of marginal, mature, and challenging oil reservoirs, developing new inflow control technologies is of great importance. In cases where production of surrounding reservoir fluids such as gas and water can cause negative effects on both the total oil recovery and the amounts of energy required to drain the reservoir, the multiphase flow performances of these technologies are of particular significance. In typical cases, a Long Horizontal Well (LHW) will eventually start producing increasing amounts of these fluids. This will cause the Water Cut (WC) and/or Gas Oil Ratio (GOR) to rise, ultimately forcing the well to be shut down even though there still are considerable amounts of oil left in the reservoir. In earlier cases, Inflow Control Devices (ICD) and Autonomous Inflow Control Devices (AICD) have proven to limit these challenges and increase the total recovery by balancing the influx along the well and delaying the breakthrough of gas and/or water. The Autonomous Inflow Control Valve (AICV) builds on these same principles, and in addition has the ability to autonomously close when breakthrough of unwanted gas and/or water occurs. This will even out the total drawdown in the well, allowing it to continue producing without the WC and/or GOR reaching inacceptable limits. As part of the qualification program of the light-oil AICV, extensive flow performance tests have been carried out in a multiphase flow loop test rig. The tests have been performed under realistic reservoir conditions with respect to variables such as pressure and temperature, with model oil, water, and gas at different WC's and GOR's. Conducting these multiphase experiments has been valuable in the process of establishing the AICV's multiphase flow behavior, and the results are presented and discussed in this paper. Single phase performance and a comparison with a conventional ICD are also presented. The results display that the AICV shows significantly better performance than the ICD, both for single and multiphase flow. A static reservoir modelling method have been used to evaluate the AICV performance in a light-oil reservoir. When compared to a screen-only completion and an ICD completion, the simulation shows that a completion with AICV's will outperform the above-mentioned completions with respect to WC and GOR behavior. A discussion on how this novel AICV can be utilized in marginal, mature, and other challenging reservoirs will be provided in the paper.

Empirical Aids to Reservoir Fluid Simulation: Using Data Driven Models to Augment Compositional Simulation

The variation of temperature profile of the sandface and wellbore offers important indicators for the occurrence of various downhole phenomena, as functions of reservoir formation and fluid properties. This paper aims to investigate the assimilation of temperature and production data for robust and efficient estimation of reservoir formation properties and fluid characterization. Very few previous work in the literature used Distributed Temperature Sensing (DTS) data for the characterization of the reservoir fluid and geostatistical description. The main contribution of this work lies in the proposed methodology for integration of temperature data from DTS systems to obtain a robust estimation of the reservoir fluid description, as well as introducing the integration of temperature data as an alternative and/or complementary data for integrative reservoir characterization. Two synthetic reservoir examples are investigated. In the first example, we focus on resolving the geostatistical description of a 3D multi-phase reservoir under water flooding. An ensemble-based data assimilation algorithm is used to compare the characterization of reservoir formation porosity and permeability by integration of temperature profile and well production data as opposed to the integration of production logging tool (PLT) data. Our results show that the DTS temperature data shows good potential as an alternative to the PLT data for reservoir rock characterization, although there may be a risk of model overfitting due to the large amount of data. In the second example, we attempt to estimate the original composition of the reservoir fluid as well as reservoir rock properties using downhole temperature profile data. A relatively simple multi-layer radial reservoir model is considered in this case. However, in order to honor the physical bounds on the various model parameters, a constrained ensemble-based data assimilation algorithm based on Lagrange multipliers is adopted. The computational results show that the original reservoir fluid composition could be excellently estimated for a case with known reservoir properties. Presence of uncertainty in the reservoir formation properties resulted in an poor estimation of the original reservoir fluid composition.

The paper is devoted to the problem of determining the composition of reservoir gas for gas condensate fields. A methodology for planning gas condensate tests (GCT) is proposed, which allows to assess the possibility of obtaining conditioned samples of reservoir fluid. For the case when it is impossible to take conditioned samples, an approach for their interpretation is developed.At the first step it is proposed to numerically create a set of compositions by adding to the gas of the laboratory “depleted” composition a certain amount of condensate equilibrated to it. The resulting compositions have different condensation onset pressures above the laboratory pressure. Subsequently, for each of the obtained compositions hydrodynamic (HD) modeling should be carried out and the composition for which the obtained values of condensate and gas flow rates and gas oil ratio (GOR) are close to the GCT data should be selected.The analysis of numerical experiments on synthetic flow rate data for a typical well shows a significant dependence of GOR on the component composition and low sensitivity to changes in the main parameters of the HD model. When adjusting the HD model according to the actual data, it is proposed to first select from the set of reconstructed compositions the one for which the model GOR values will be closest to the actual data. The next step, when the selected composition is fixed, is the selection of other parameters of the HD model based on the data of gas and condensate flow rates. This methodology was demonstrated to determine the component composition and main reservoir parameters from GOR data for one real well.