Ahmed H. El-Banbi

Sampling Volatile Oil Wells

Recombined surface samples are usually used for volatile oil laboratory fluid property studies. A procedure for stabilizing and surface sampling of volatile oil wells is currently used in the industry. However, no investigation of the quality of the samples resulting from this procedure has ever been published. Typically, during surface sampling, bottom-hole flowing pressure is less than the bubblepoint pressure of the original reservoir oil. This causes gas to form in a cylinder of the reservoir around the wellbore. Understanding the dynamics of this cylinder of gas saturation is critical to obtaining a recombined surface sample representative of original reservoir oil. It is possible to obtain a representative sample if this cylinder is stable. This paper presents the results of a study of the sampling procedure. The effects of production rate prior to and during the sampling process were quantified using radial compositional simulation. The buildup and stability of the ring of gas saturation were examined. Guidelines for sampling volatile oil wells is presented. It is based on comparisons of the compositions of recombined surface samples with the compositions of original reservoir oils for various producing schemes. These guideline are expected to give the best chance of obtaining a representative sample from a volatile oil well. Introduction Several authors discussed resevoir fluid sampling. A study on gas condensate reservoir sampling has recommended that sampling should be done early in the life of the reservoir. The usual procedure of reducing the rate before sampling may be useful in increasing the chance of obtaining a valid fluid sample in gas condensate reservoirs. In this paper, we used compositional reservoir simulation to investigate sampling in volatile oil wells. Simulation Model We used a radial compositional simulation model to investigate the changes in composition for volatile oils and to understand the effect of these changes on fluid sampling. The results reported here are those obtained for fluid sample “Oil 2” of Coats and Smart. PVT Modeling. We used an EOS model to match the PVT behavoir of a volatile oil sample. The iso and normal components for C4 and C5 were lumped together and the C7+ fraction was split into three components using the Whitson’s method. This resulted in an eleven-component fluid system. We then used the Peng and Robinson EOS to match the PVT data of the fluid sample. Following Coats and Smart procedure, We used ΩA, ΩB for C1 and the three heavy components, accentric factors for the three heavy components, and the binary interaction coefficients for the three heavy components with C1 as regression variables. The match with the laboratory data was satisfactory. Figs. 1-3 show the match between some simulated and actual PVT properties for differential liberation and constant volume depletion (CVD) data. Radial Compositional Model. We constructed a radial simulation model and used it to investigate the near-wellbore compositional changes. The model had twenty-two grid blocks in the radial direction. The block sizes increased logarithmically from 0.5 ft (the wellbore) to 100 ft. and then uniforamlly to a reservoir radius of 1490 ft. (160 acres). Gasoil relative permeability are shown in Fig. 4. Other reservoir and fluid data for the base case are given in Table 1. Simulation Results Several runs were made to investigate the compositional changes that can occur at different production rates and to study the effect of the common procedure of reducing the production rate before sampling. In the following sections, we discuss the results of our compositional simulation experiments for five different cases. These cases show the SPE 67232 Sampling Volatile Oil Wells Ahmed H. El-Banbi, SPE, Cairo University/Schlumberger Holditch-Reservoir Technologies, and William D. McCain, Jr., SPE, Texas A&M University 2 A. H. EL-BANBI AND W.D. MCCAIN, JR. SPE 67232 effect of producing the at high rate, producing at low rate, reducing the rate from the high rate case before sampling, reducing the rate from the low rate case before sampling, and shutting-in the well before sampling. We used the mole fraction of C7+ 11 in the well stream as indicator of compositional changes between the recombined surface sample and the original reservoir fluid. Effects of compositional changes are also reported. Case 1: Production at High Rate. The well was produced at high rate of 1,000 STB/D. After 220 days of production at the high rate, the well could not maintain its rate because it reached a minimum bottom-hole pressure of 1,470 psia. The average field pressure, first model block pressure, and well bottom-hole flowing pressure are shown in Fig. 5. The change in slope of the field average pressure shows that the bubble point pressure was reached around 50 days. Accordingly, an increase in the producing gas-oil ratio (GOR) can be seen after 70 days of production (Fig. 6). Because of the high production rate, the GOR increased to very high levels. Fig. 7 shows the mole fraction of C7+ versus time. The original fluid C7+ mole fraction is also indicated on the plot. The figure shows that the C7+ mole fraction in the well stream is nearly the same as the original fluid C7+ mole fraction for at least the first 50 days of production. This suggests that a fluid sample taken early in the life of the reservoir (even when the bottom-hole pressure is slightly less than the bubble point pressure) will almost represent the original reservoir fluid. The sample will not be representative after depletion occurs in the reservoir. Gas saturation builds up near the wellbore and in the reservoir as pressure declines (Fig. 8). The gas saturation can build up immediately around the wellbore if the bottom-hole pressure around the wellbore is less than the bubble-point pressure. This gas saturation reduces the relative permeability to oil and increases the relative permeability to gas, reducing the oil productivity index. Case 2: Production at Low Rate. We produced the well at a lower rate this time (500 STB/D). This case has similar results to Case 1 except for the effect of lowering the pressure below the bubble-point pressure is delayed. Fig. 9 shows the C7+ mole fraction for the produced well stream. Although the pressure near the wellbore goes immediately below the bubble-point pressure (and gas saturation builds up), there is a better chance of obtaining a representative sample than the case of high production rate. Other simulation runs, at even lower rates, supported this observation. Case 3: Reducing the High Production Rate Before Sampling. In this case, the production rate was reduced from 1,000 STB/D to 200 STB/D after 180 days of production. Fig. 10 shows the average reservoir pressure, bottom-hole flowing pressure, and the first simulation cell pressure. The near wellbore pressure is affected by the reduction in production rate. At 180 days, the near wellbore pressure jumps to around 3,800 psia and shows a more gentle decline at production rate of 200 STB/D. The effect of reducing the oil production rate can be also seen as sudden decrease in the producing GOR (Fig. 11). The GOR will go back to its normal increasing trend after the production rate is stabilized at 200 STB/D. Fig. 12 (mole fraction of C7+) shows that when the well production rate is suddenly decreased, a spike of C7+ can be detected in the well stream. A fluid sample taken at this time will not be representative of the reservoir fluid. Fig. 13 shows the gas saturation developing near the wellbore and far in the reservoir. The figure indicates that the gas saturation around the wellbore will be affected by the reduction of rate. Case 4: Reducing the Low Production Rate Before Sampling. In this case, the production rate was reduced from a low rate of 500 STB/D to a lower rate of 200 STB/D. Fig. 14 is the C7+ mole fraction for the well stream fluid. At 180 days, the spike can be seen but with a lower magnitude when compared with Case 3 (Fig. 12). This suggests that production at low rate is desirable if a representative fluid sample is to be obtained. Case 5: Shut-in Before Sampling. This case shows the effect of shutting-in the well before fluid sampling. The simulation model was run at production rate of 1,000 STB/D for 30 days, followed by a shut-in period for 10 days, then produced again at a reduced rate of 200 STB/D. Fig. 15 shows the behavoir of C7+ mole fraction. The figure indicates that shutting the well in before sampling has a minimal effect on the quality of the sample. Discussion Obtaining a representative fluid sample is important to estimate the fluid PVT properties. These PVT properties are essential to almost all reservoir and production engineering calculations. Fluid sampling of volatile oil wells can be affected by the conditions of the well before sampling. In general, fluid samples should be taken before considerable depletion occurs in the reservoir. Ideally, the fluid sample will be representative of the original reservoir fluid if the pressure (both in the reservoir and near the wellbore) is not allowed to drop below the bubble-point. If the near wellbore pressure goes below the bubble point, a representative sample may still be obtained. However, if the reservoir pressure drops below the bubble-point, the fluid sample will not be representative of the original reservoir fluid. Compositional Changes. In volatile oil reservoirs, compositional changes affect the production behavoir. We used Case 2 simulation to show some of these effects. Fig. 16 compares the relative permeability in the first grid block for oil and gas. Oil relative permeability goes down with time while gas relative permeability goes up. This is a direct result of the saturation changes occuring near the wellbore with production. With in the increase in gas saturation, more gas SPE 67232 SAMPLING VOLATILE OIL WELLS 3 passes into the wellbore. As a result, pr

Reservoir-Fluid Classification

This chapter introduces why practitioners and engineers are interested in classifying fluids for practical purposes. It identifies how similar-behavior fluids are grouped under the same classification. A set of various classifications are introduced to show how different classifications vary on the basis of limits of reservoir and production parameters. The chapter classifies reservoir fluids into six different fluid types including low gas–oil ratio (GOR) oils, black oils, volatile oils, gas condensates, wet gases, and dry gases. The difference in reservoir performance is explained for all six fluid types. The pressure, volume, and temperature (PVT) work requirements are also highlighted for each fluid type. Producing GOR and C7+ mole % are used to identify the fluid type.

Artificial Neural Network Models for PVT Properties

This chapter introduces the artificial neural network (ANN) applications for estimation of pressure, volume, and temperature (PVT) properties. It starts with an introduction on the basics of ANN models and ANN model types. Then it explains the structure of typical ANN models and defines their components: (1) input nodes; (2) hidden layers and node(s); (3) activation function for the hidden node(s); (4) transformation function for the output node(s); (5) output node(s); (6) objective function; (7) optimization algorithm; and (8) training algorithm. This chapter also discusses the previous ANN models that were developed for predicting PVT properties for oil and gas fluids. It provides a comprehensive review of these models. The calculations of ANN are explained with numerical example. This chapter ends with expectations on how improvements in ANN models will yield better models for predicting PVT properties.

Selection of PVT Correlations

This chapter discusses the approaches to select appropriate pressure, volume, and temperature (PVT) correlations. It starts by highlighting the problem of large variation of PVT properties calculated from the same inputs. It then discusses the approaches for PVT correlation selection (e.g., based on geographical location, fluid type, application, data ranges and ranking, and expert systems). The chapter also discusses the limitations of each PVT correlation selection technique. It explains with example calculations different techniques that can be used for correlations selection.

Low Gas–Oil Ratio Oils

This chapter defines low gas–oil ratio (GOR) oils and discusses their occurrence. It explains the practical reasons to distinguish low GOR oils from black oils. This chapter explains how to generate pressure, volume, and temperature (PVT) properties for low GOR oils. It also shows that these fluids are not necessarily heavy oils as they sometimes have reasonably low viscosity and can flow easily under normal conditions of temperature. The sequence of calculations to generate PVT properties for low GOR oils is explained through practical example calculations.

Analysis of multi-layered commingled and compartmentalized gas reservoirs

The evaluation and performance prediction of multi-layered compartmentalized gas systems can be difficult. This is mostly due to the uncertainties related to production allocation either within each commingled well or between interrelated reservoir compartments. This paper presents a model that can provide reliable estimates of the total gas in place for multi-layered commingled and compartmentalized reservoirs. The model is also capable of generating prediction profiles for every well in the production system in addition to forecasting individual layers production for each compartment. The proposed model is based on coupling the layered stabilized flow model for material balance calculation in commingled systems with communicating reservoir model that is used as material balance tool for compartmentalized gas reservoirs. The model has the flexibility to be applied for history matching and prediction purposes. In history matching, the model solves the equations simultaneously using optimization routine to find the best parameters of original gas in place (OGIP), deliverability coefficients and compartment transmissibility coefficients. The model requires the knowledge of initial reservoir pressure in every compartment, some rate production history and bottom-hole flowing pressures. The model can also utilize additional information such as shut-in pressures per layer, repeat formation tester and production logging tool measurements (if available) to improve the history match. For prediction, the model uses the estimated parameters (compartment OGIP, transmissibility coefficients between compartments and flow parameters for each layer) to calculate the production rates and reservoir pressures for every well/tank based on a provided bottom-hole flowing pressure. The model was verified against a commercial reservoir simulator for several synthetic cases. The model was also applied on different field cases to estimate OGIP and flow coefficients for every layer as well as compartment transmissibility coefficients. Moreover, calculation of cumulative gas transferred across the communicating reservoirs allows detection of poorly drained compartments, which could be included in future redevelopment plans.

Development of an expert system for selection of multiphase flow correlations

AbstractOur target was to develop an expert system to help petroleum engineers in selecting the most suitable multiphase flow correlation in the absence of measured flowing pressure. A large database of pressure points was collected and analyzed using many multiphase flow correlations. The expert system was developed with a set of rules to identify the best correlation for variety of well, flow, and PVT conditions. The expert system is based on the idea of clustering the data and finding the best multiphase flow correlation(s) for each cluster. The error associated with the selected correlation is also quantified for every correlation in each data sub-cluster so the engineer would expect the accuracy of pressure drop prediction when utilizing this approach. Over the entire database, if one multiphase flow correlation is selected, the overall mean absolute percent error ranges between 12.7 and 57.5%, while the range of errors for best correlation(s) in different data sub-clusters range from 0.01 to 3% for most cases with accurate PVT. The expert system was validated by a new set of data. It succeeded in identifying the best correlation(s) 70% of the times, and the calculated pressure was more accurate than using one correlation by a factor of 2. Use of the expert system in the validation database gave a mean absolute percent error of 8.8%. This represents approximately one-third of the error value when any single correlation is used over the entire validation dataset. (Error for using single correlation ranges from more than 21 to 29%.)

Sobol and Halton Sequences: Efficient Experimental Design Techniques towards Improved Uncertainty Quantification of Petroleum Reservoirs

Reservoir simulation models often suffer from significant uncertainties due to the lack and inaccuracies of the measured data. Hence, uncertainty analysis and quantification are considered as major concerns in building robust and predictive simulation models. The conventional approach of the uncertainty analysis is conducted by running a very large number of simulation runs in an attempt to capture all effects of the uncertain parameters. Running a large number or simulation runs is costly and very time consuming and therefore efficient approaches have been arisen. One of these efficient approaches is the Experimental Design technique. The experimental design technique is widely applied in different engineering practices for assessing and quantifying uncertainties. Experimental design is the technique used to guide the selection of the samples within the design search space of the uncertain parameters in order to obtain the maximum amount of information using low number of experiments. Several experimental design techniques are applied in petroleum industry and specially in reservoir simulation assisted history matching. Some experimental design techniques shows more effectiveness than others. The objective of this paper is to introduce two efficient experimental design techniques (Halton and Sobol sequences) to the reservoir simulation assisted history matching workflow. This work is to complete the work done by the authors and published in Shams et al. (2017). Shams et al. (2017) applied and tested the potentiality of the two proposed experimental design techniques through a comparative study between their performance in solving assisted history matching problems of different scale material balance problems and the performance of the most widely used experimental design technique, Latin hypercube. In this paper the comparative study is conducted using numerical reservoir simulation model. A performance indicator is developed to compare between the three studied techniques. The performance indicator represents the relative error between the estimated values of history matching parameters calculated using the studied experimental design methods and their exact solutions. The results of this work validate the previous obtained conclusions and indicate that the Sobol and Halton sequences experimental design techniques are more efficient and superior to Latin hypercube method. Introduction Quantifying the uncertainties of reservoir simulation models has a significant enhancing impact on the output of the study. Experimental design technique has been considered as one of the most effective approaches used for quantifying the uncertainties of the studied parameters within their search space. Although the experimental design technique was first proposed by Ronald Fisher (1925) for agricultural applications, it was first applied in reservoir engineering practices by Damsleth et al. (1992). Damsleth et

A Comparative Study of Proxy Modeling Techniques in Assisted History Matching

Assisted history matching approach has been arisen in the last few decades in an attempt to make the process of history matching faster and easier. Assisted history matching simply involves converting the history matching problem to an optimization problem. One main aspect of the assisted history matching is building proxy model that interpolates the relationship between the objective function and the history matching parameters. Several proxy modeling techniques are introduced in the literature, some are useful and some are not. This paper provides a comparative study between four powerful proxy modeling techniques in assisted history matching; Thin Plate Spline, Radial Basis Function, Kriging, and Artificial Neural Network. Two test problems of different reservoir engineering approaches (material balance and reservoir simulation) are used to test and compare the performance of the studied proxy methods on solving assisted history matching problems. To make the comparison reliable, a performance indicator is developed to compare between the four studied techniques. The performance indicator represents the relative error between the estimated values of history matching parameters calculated using the studied proxy modeling methods and their exact solutions. The results of this work indicate that the Kriging and artificial neural network proxy techniques are more efficient and superior to thin plate spline and radial basis function. Introduction The significant increasing usage of reservoir models poses a lot of challenges in making models deign and calibration easier and achieved in a timely manner. One major challenge in building reservoir analytical or numerical models is the model calibration process. Reservoir models can only be trusted and predictive after good calibration with actual measured data. What makes model calibration, history matching, challengeable is that it is a non-uniqueness solution process. Same history matching can be achieved using several combination sets of the same history matching parameters. As a consequence, the assisted history matching techniques have been arisen in an attempt to make the process easier. The assisted history matching process is always conducted through three main aspects; experimental design, proxy modeling, and optimization. The objective of the experimental design is to explore the uncertainty range of the reservoir history matching parameters. The chosen samples are used to build the scoping runs. Once analytical or numerical simulation runs are conducted, the objective function is calculated. The objective function takes a formula representing the difference between simulated and observed data. After that a proxy model is created. The proxy model is an analytical function that

A Comparative Study of Experimental Design Techniques in Assisted History Matching

Quantifying the uncertainty of reservoir models has always been considered as a major concern. Uncertainty and sensitivity analyses provide us with information about how “incorrect” a proposed prediction scenario is. One effective approach to quantify reservoir uncertainties is to apply the concept of Experimental Design. As the name indicates, experimental design is the technique used to guide the choice of the experiments to be conducted in an efficient way. Samples are selected in the design space of the uncertain parameters in order to obtain the maximum amount of information using low number of experiments. Geoscientists and engineers are building larger models to address reservoir heterogeneity and hence reservoir simulation is usually expensive (takes significant time to run). Experimental design techniques became very useful in assisted history matching. Several experimental design techniques are introduced in literature, some must be more effective than others in typical reservoir simulation problems. The objective of this paper is to compare different experimental design techniques and to introduce two efficient ones (Halton and Sobol sequences) to reservoir engineering problems. To show the potentiality and efficiency of the Halton and Sobol sequences design techniques, a comparative study is conducted to compare between their performance in solving assisted history matching problems and the performance of the most widely used experimental design technique, Latin hypercube. Other conventional techniques were compared too, but found to be less efficient than Latin hypercube. Two different scale reservoir models are used as test problems. A performance indicator is developed to compare between the three studied techniques in terms of the relative error between the estimated values of history matching parameters calculated using the proposed assisted history matching procedure and their exact solutions. The results of this work indicate that the Sobol and Halton sequences experimental design techniques are superior to Latin hypercube method.