Analysis of Simultaneously Measured Pressure and Sandface Flow Rate in Transient Well Testing F, Kutuk, SPE, Schlumberger Well Services L. Ayestaran, SPE, Schlumberger Wel Services ‘Summary New well test interpretation methods are presented that eliminate wellbore storage (afterflow) effects. These now ‘methods use simultaneously measured sandface flow rate and wellbore pressure data. It is shown that formation behavior without storage effects (unit response or in- fluence function) can be obtained from deconvolution of sandface flow rate and wellbore pressure data, The storage-free formation behavior can be analyzed to iden- tify the system (reservoir flow pattern) that is under testing and to estimate its parameters. Convolution (radial multi- rate) methods for reservoir parameter estimation and a few synthetic examples for deconvolution and convoli- tion also are presented Introduction, Well testing with measured sandface flow rate can be ‘traced to the beginning of reservoir engineering. The rate must be measured over time to calculate and/or approx: imate constant rate to obtain oven a single reservoir parameter from pressure measurements. This approximate Constant rate has been sufficient for estimating permeabili- ty, skin, and initial formation pressure during the radial infinite-acting period. During this period, the well should produce at a constant rate at the sandface or at a zero rate if buildup test is conducted. Because of compressible fluid in the production string (wellbore storage effects), it takes a long time to reach the radial infinite-acting period, The effect of outer boundaries also may start before the end of the wellbore storage effects. In general, the storage capacity of the wellbore, wellbore geometry, near-wellbore complexities, and ex- ternal boundaries affect transient behavior of a well. Dur- ing the analysis of pressure-time data, each of these phenomena and its duration must be recognized for the application of semilog and typecurve techniques to deter- mine formation flow capacity (Wh), damage skin, and average formation pressure. The influence of these ‘phenomena on transient behavior of a well progresses over time. For the sake of convenience, the test time can be divided into three periods according to which phenomenon is affecting the pressure. These periods are defined as follows. Early-Time Period. The combined effects of wellbore storage, damage skin, and pseudoskin (which include par- ny 185 Say of Pano Engrs FEBRUARY 1985 tial penetration, perforation, acidizing, fractures, non- Darcy flow, and permeability reduction caused by gas saturation around the wellbore) dominate pressure behavior. The stratification and dual porosity also may affect wellbore pressure during this period. Middle-Time Period, During this period, radial flow is established. Conventionally, semilog techniques are used to determine formation kh and initial pressure and skin. Late-Time Period. During this period, outer boundary effects start to distort the semilog straight line. For ex- ample, the gas cap shows a curve-flattening effect on log- log and Horner plots. ‘Sometimes the separation of these periods from each other is impossible; particularly, the effects of bottom ‘water influx and/or gas cap may start during the middle- time period. Thus, the semilog approach sometimes can- not be applied at all. Furthermore, the drawdown or buildup tests as con- ducted today tend to homogenize the reservoir behavior. In other words, most of the reservoirs behave homogeneously during the storage-free radial infinite ‘acting period because most of the heterogeneous behavior takes place during the early-time period, ‘The type-curve approaches have been introduced 10 ‘overcome some ofthese problems, The theories, applica tions, and elaborations of the type-curve methods, as well ‘as many references, can be found in Ref, 1. In 1979, Gringarten et al.? introduced new type-curves that use different parameterization than the earlier ones, namely Ramey,? Agarwal et al.,4 McKinley, and Earlougher and Kersch® types. All the type curves presented by these authors, and many others, were developed under the assumption that the fluid compressibility (density) in the tubing and annulus remains constant during the test period. During the early time, particularly for buildup tests, shut- in pressure increases very rapidly; thus, the compressibili- ty is usually higher than the compressibility of the fluid inthe reservoir for producing wells. Since the pressure in the wellbore is @ function of the depth, the com- pressibility of the fluid at the wellhead can be 10 or even 100 times greater than the compressibility ofthe fluid at the bottom. Thus, the assumption that the wellbore storage coefficient is constant during the drawdown, and par- ticularly during buildup, may not be correct. A variable ‘wellbore storage coefficient alone makes the application 23of type-curve methods almost impossible. The combina- tion of variable or even constant wellbore storage with wellbore geometry further complicates the type-curve matching process. Moreover, wellbore pressure data alone ‘may not indicate changing wellbore storage ‘Some of the problems inherent with the use of type~ ccurve methods can be eliminated by the simultaneous use fof measured sandface flow rate and pressure data. ‘The purpose of this work is to study the use of the ‘measured sandface flow rate in abroad sense with regard to transient well testing. Furthermore, we explore the use ‘of convolution and deconvolution in the interpretation of pressure behavior of a well with afterflow (buildup case) ‘or wellbore storage (drawdown case). Background ‘The use of the sandface flow rate in transient testing is| not new. To our knowledge, van Everdingen® and Hurst? were the first to estimate and use the sandface flow rate to calculate the wellbore pressure. To do this, they approximated the sandface flow rate by the formula where B is a positive constant. These authors stated that the constant, 8, can be determined from well and reser- voir parameters. Using the above formula and the con- volution integral, van Everdingen’” and Hurst® presented an expression for the wellbore pressure with a variable wellbore storage effect. Gladfelter er al.® presented a ‘method to determine the formation Kh from pressure and afterflow data, The afterflow data were obtained by measuring the rise in the liquid level in the wellbore. Ramey! applied the Gladfelter approach to gas well buildup tests. ‘A considerable amount of work also has been done on. multirate (Variable) rate tests during the last 30 years. However, these are basically sequential constant-rate drawdowns; only transient pressure is measured and rate is assumed constant during each drawdown test, The techniques related to this type of multirate tests also can be found in Ref. 1. All the work mentioned so far deals ‘with the direct problem, In other words, the constant-rate solution (the influence or the unit response function) is convolved (superimposed) with the time-dependent inner ‘boundary condition to obtain solutions to the diffusivity ‘equation. This process is called “‘convolution.”” Hutchison and Sikora, "! Katz et al.,'? and Coates et al. presented methods for determining the influence function directly from field data for aquifers. The proc- css of determining the influence function is called ““decon- volution.”” Jargon and van Poolen'* were perhaps the first to use the deconvolution of variable rate and pressure data 10 ‘compute the constant-rate pressure behavior (the influence function) of the formation in well testing. Bostic er al. "8 used a deconvolution technique to obtain a constant-rate solution from a variable rate history with a known pressure history, They also extended the deconvolution technique to combine production and buildup data as a single test Pascal ® also used deconvolution techniques to obtain a constant-rate solution from Variable rate (measured at the surface) and pressure measurements of a drawdown test me More recently, Meunier ef al.'7 have used sandface flow measurements with pressure data for buildup test analysis. This has been the first successful attempt to use direct measurements of sandface flow rate data in well testing. They showed that the Horner method can be modified to reach a semilog straight line earlier than type- ‘curves or the 1%-cycle rule indicates. ‘Theoretical Developments During the last 4 or 5 decades, many solutions have been developed for transient fluid flow through porous media. ‘The superposition theorem (Duhamel’s theorem) has been used to derive solutions for time-dependent boundary con- ditions from time-independent boundary conditions. For ‘example, the multiple-rate testing is a special application of the superposition theorem. In their classic paper on unsteady-state flow problems, ‘van Everdingen and Hurst'® presented the dimensionless wellbore pressure for a continuously varying flow rate as Prottn)=40P0(tD) + PP app aplty— At. sees Qa) ° ‘An alternative form to Eq. 2a can be obtained by an in- tegration by parts as Pavttn)=40tto)P.v 7" +f ante sotto—nr, 2.) é where = ——[Pi-P Pao Fah 9 gay? Pl sp 1026378 Oo Guerre Pup = Pot, Poltp) = the dimensionless sandface pressure for the constant-rate case without wellbore storage and skin effects, S = steady-state skin factor, Potty) = dppltp)/dtp, v(t) = ayft0)!4rs alt) = AgottpYatp, feference flow rate—if the stabilized % constant rate is available, then 4, should be replaced by 4x. aft) = variable sandface low rate (lowmeter readings), and qoltp) limensionless sandface rate. JOURNAL OF PETROLEUM TECHNOLOGYFig. 1—Dimensionloss sandface flow ratos for constant walloore storage and exponential dectine cases, Although traditionally the skin effect is considered a dimensionless quantity different from the dimensionless formation pressure, the skin effect will be treated here as part of the inner boundary condition for the solution Of unit rate production case. This boundary condition {s known as the homogeneous boundary condition ofthe third ind. should be emphasized that gs. 2a and 2b can be ap- plied for many reservoir engineering problems. The linearity of the diffusivity equation allows us to use Eqs. 2a and 2 for fractured, layered, anisotropic, and hneterogeneous systems as long as the flu im the reser- voir is single phase. Eqs. 2a and 2b can be applied to both drawdown and buildup tests if the intial conditions are known, For a reservoir with an initial constant and uni form pressure distribution [pp(0)=0], Eqs. 2a and 2b can be expressed as Pwolto} [Pace Pwltp— NAF... e.e-- Ga) 3 aptto)+ | ap(*ps vlty Xr. Gb) Furthermore, Eq. 3a also can be expressed as PP actt>—np viride. 3 Pwotto} - G0) In Eqs. 3a and 3c, itis assumed that gp(¢p) exists. If o(tp) is constant, then Eq. 36 must be used, Eqs. 3a and 3¢ are known as a Volterra integral equation of the first kind and the convolution type. ‘Although itis assumed that pp is a constant-rate solu- tion without storage effect, mathematically and physical- ly it can be a solution of a constant-storage case. In practice, pp always will be affected by the wellbore fluid that occupies the volume below the flowmeter unless the ssandface rate is measured through perforations. However, the volume below the flowmeter will be small for most wells, since the flowmeter and the pressure gauge usual- ly can be placed just above perforations, Therefore, ‘throughout this paper, we will assume that pp is not al- fected by the fluid volume below the flowmeter, FEBRUARY 1985 ‘TABLE 1_HOMOGENEOUS RESERVOIR ROCK ‘AND FLUID DATA , boUSTE. 49 ep. pal 107? cy 41000 non ‘00 Kms 40 Pat. Pla 2,006.94 STB 3,000 aft 025, hours 10,000 e o76%10-* men 08 3 02 ‘The purpose of the test well interpretation, as stated by Gringarten et al.,? is to identify the system and deter- ‘mine its governing parameters from measured data in the wellbore and at the wellhead. This problem is known as the inverse problem, The solution of the inverse problem usually is not unique. As Gringarten et al.? pointed out, if the number and the range of measurements increase, the nonuniqueness of the inverse problem will be reduced. ‘Thus, combining sandface flow rate with pressure ‘measurements will enhance the conventional (including. type-curve) well test interpretation methods. ‘As an inverse problem, the sandface pressure, Pzp(tp), thas to be determined by the deconvolution of the integral in Egs. 3a, 3b, and 3c. As stated previously, p,o(¢) is the solution for the constant-flow-rate (sandface) case. ‘Taking the Laplace transform of Eq, 3a and solving for Pav(s)* yields wo(s) “ where s is the Laplace transform variable. ‘The Laplace transform of pp in Eqs. 3b and 3c will be the same as Eq. 4, keeping in mind that p,p(0) ‘Thus, we have only one operational form of the convolu- integral given by Eqs. 3a, 3b and 3c, The superposi: tion theorem in this case is nothing more than the convolution of pzp(tp) and qp(fp). The Laplace trans- form of the convolution integral allows us to express the convolution integral in many different forms. Further- ‘more, the kernel solution can be a solution of constant- rate of constant-pressure case for the convolution integral If the wellbore storage is constant, the dimensionless sandface flow rate can be expressed as*!8 dp wplty) oe ar avtto) o where avlto)=9960l4 ‘As a special application of Eq. 4, the dimensionless ‘wellbore pressure solution for the constant-storage case cean be written directly from Eqs. 4 and 5 as Pivls) . 6 14 Cps°B.0) ° Pots) seo te papa. te fncn Fl cla Lap rom fhe ‘tron 2sFig, 2—Shut.in pressures for constant wellbore storage and ‘exponential decline case where P.(3) is the dimensionless sandface pressure for the constant-rate case without storage effect but including, skin. ‘Van Everdingen and Hurst" presented an equation similar to Eq. 6, and Agarwal et al.* presented the same equation as an integro-differential form for radial systems. Cinco-Ley and Samaniego"? (for fractured reservoirs), ‘and Kuéuk and Kirwan™ (for partially penetrated wells) presented the same expression for the dimensionless wellbore pressure as in Eq. 6. (On the other hand, the dimensionless sandface flow rate ‘can be obtained directly from Eqs. 5 and 6 in terms of the dimensionless formation pressure as 1 S1+Cps"P,00)1 Got “O Fig. 1 presents values of qp calculated from gs. 1 and as a function of realtime fora buildup test using reser- voir and fluid properties given in Table 1. As can be seen from this figure, for the exponential decline case, the sand- face flow rate declines faser than the constant-wellbore- storage case. Ramey and Agarwal" also presented values ‘of qp{tp) as function of ty for various skin and storage constants ‘Another important application of the convolution in- tegral given in Eqs. 3a and 3 was presented by van Ever- ingen,” Husst,® and Ramey'® for calculating the wellbore pressure by using Eq. 1 and the line source solution, For finite wellbore radius, the dimensionless wellbore pressure solution also can be written directly from Eqs. 1 and for the exponential sandface rate decline case as Pv) 8 ® Buvls)= ‘The exponential constant, 8, is given by van Everdingen” xg ner w?/0,000264 k, where a can be determined from wellbore pressure data, such as the wellbore storage constant. Note that 8 is a dimensionless constant like Cp, m6 As Ramey! noted, cannot be estimated as readily as the wellbore storage constant, Cy. However, in prin- ciple, Eq. 9 has a much more immediate connection with the real systems. In fact, the pressure increases more rapidly during the early-time buildup tests; then it slows ddown during the transition period and builds up very slow~ ly during the semilog period. Because of the rapid change (of pressure, the wellbore storage will decrease continuous- ly except in the case of phase redistribution In many cases, it is difficult to recognize changing wellbore storage effects because itis a gradual and con- tinuous change. Furthermore, the finite closing time of the wellhead valve also will affect pressure at the same time. “Most of the work thus far in early-time analysis has been directed toward the construction of type curves from the solutions of Eq. 7 for a constant wellbore storage for dif ferent wellbore geometries, such as fractured wells, par- tial penetration, etc. Type curves for Eq. 8 for different ‘values of 6 and skin also can be developed and used for ‘gradually decreasing wellbore storage cases to determine skin and kh, Fig. 2 presents a semilog plot of pys(At) vs. At, calculated from Eqs. 7 and 8 by using reservoir and fluid parameters given in Table 1. As shown by this figure, the exponential decline case approaches a semilog straight line earlier than the constant-wellbore-storage case. ‘The most important point to be made from the above discussion and buildup data presented in Fig, 2is that the principle limitation ofthe type-curve analysis stems from the lack of information about the sandface flow rate behavior. Thus, the type-curve analysis usually is used for qualitative answers and supported by the semilog analysis, For quantitative analyses of the early-time data, it is necessary to measure the sandface flow rate. Fur- thermore, the use of measured sandface flow rate also can improve the semilog analysis. In the following sections, the convolution and deconvolution of simultaneously measured sandface flow rate and pressure data will be showa to obtain the formation pressure and parameters, Convolution (Superposition) Continuous Multirate Method. The simplest approach to solving the convolution integral is to assume a Psp function in Eq. 3a. This pap function could be a tine source solution, infinite conductivity vertical fractured solution, etc., for the constant-flow-rate case or the con- stant-pressure case. The chosen pp function can be con- volved with qp (sandface flow rate) by using the convolution integral (Eqs. 3a through 3c) to modify the time function or pyp. For example, the convolution (uperposition) of variable rate with the log approxima- tion for the pp function and wellbore pressure common- ly is used for the analysis of multirate tests. The same technique also can be used for the analysis of buildup or drawdown tests with measured sandface flow rate data, ‘Using the log approximation for psp in Eq. 3a, the change in measured pressure in oilfield units can be ex- pressed as Ap t=m{ qo(oiloge—1) +3147, 9) JOURNAL OF PETROLEUM TECHNOLOGYwhere AP wlO=Pi—Puf Ds uci Eq, 9 can be rewritten as mf’ gptelost—ridr AP AD +b, ao qo qo) where b= Sm. Let us approximate the integral in Eq. 9 by the Riemann, sum, which yields Apuftn) gpl) m5 atin ante Mow) 4, ayy Gln) where fy is the measured (discrete) time point. This equae tion has been presented elsewhere for multiple-rate analysis, Eq, 10 gives the continuous form ofthe mulirate (Warlable-rate) equation. Any integration techniques, as well as the Riemann sum used in Eq. 11, ean be used to evaluate the integral given in Eq, 10. Apt of the left Side vs, the first term of the right side of Eqs. 10 or 11 will yield a straight ine with a slope m and an intercept b. For buildup tests, gp(®) should be replaced by ap lh) and Api AD by AP yx(At)=P ws (A!) ~pagi&t=0) in Eqs. 9 through 11. ‘The major advantage of continuous variable-rate test (rate measure just above the perforation) over the con- ventional multirate tests that the wellbore storage effects ‘re minimized, The wellbore storage effect has not been discussed in the literature for the conventional multirate tests, The wellbore measured pressure used in the con- vetional analysis should be taken from the storage-free infinite-ating period, Thus, the end ofthe storage effect should be determined, In other words, the sandface rate should be equal 0 the constant-surface rate for each Moors ‘Moor = Slope of correct Horner semiclog straight Tine, slope of computed Horner semilog straight line, and . tpp = dimensionless producing time. Fora given ¢ (relative error), 8, skin, and producing time, zeros of Eq. 15 with respect to Afp will give the start of the Homer semilog straight line. It is interesting to 0 JOURNAL OF PETROLEUM TECHNOLOGY‘observe that the start of the semilog Horner straight line is a function of the producing time, as expected. AA simple formula cannot be derived for the start of the Horner semilog straight line from Eq. 15 because itis a transcendental function. Fig. 4 presents values of Eg. 15’as a function of dimensionless time for tp B=10~*, S=0.0, and 0.01. As can be seen from Fig. 4, Bq, 15 has two roots (zeros) for the values of tpo, By Sand ¢ given previously. The first zero results from the ceerly-time period, which can be seen easily from Fig. 2 ‘The second zero results from the radial infinite-acting period. The upper curve in Fig. 5 presents dimensionless time for the start of the Horner semilog straight line as 2 function of 8 for typ =10, S=0.0, and ¢=0.1. As can bbe scen from Eq, 15, the dimensionless time for the start of the Horner semilog straight line is a very weak func- tion of skin. It is basically a function of the production time and 8. ‘This had been observed by Chen and Bingham” for the constant-wellbore-storage case. Eq, 15 also can be used for the constant-wellbore-storage case by substituting 1/Cp for 8. However, 1/Cp is a very crude approximation for 6. Eq, 15 can be quite useful for the design of buildup tests {or an optimal value of a producing time to achieve a cer- tain accuracy for the Horner semilog straight line. Par- ticularly, Eq. 15 will be useful for drillstem tests, since production time is limited by production facilities. For large producing time, tp, Eq. 15 can be simplified to 1 Atp=—. a6) 268 Eq, 16 also can be derived from drawdown solutions bby using the same principle given in Appendix B. IF is approximated by 1/Cp, Eq. 16 then can be writ. ten as Atp =p. an 2% ‘As noted previously, Eq. 17 will yield very optimistic values for the start of Horner semilog straight line if in- {deed the wellbore storage remains constant during the test. ‘The dimensionless time for the start of the modified Horner semilog straight line (details of derivations are given in Appendix B) also can be given by 2BtpAD +p +A) p]-B7Atp*Ulty +AN p] +e7Pito(—In B-2y+1n 4425)! ag) AAsinthe Homer ease, the second root of Eq. 16 wil give the dimensionless time for the start ofthe modified Homer semilog straight line asa function of 8. Fig. 4 presents values of Eq. 18 as a function of dimensionless time for f,=10°, B=10~*, $=0.0, and e=0.01 “The lower curve in Fig. 5 presents dimensionless time forthe start ofthe modified Horner semilog straight line a8 funtion of 6 for yp =10*, $0.0, and e=0.01. The start of modified Homer semilog straight line is also a ‘weak function of skin. For lage producing time, typ, Ea. 18 can simplified as 1 Ap. ar ‘The modified Homer semilog straight lie starts atleast one-half eycle earlier than the Horner straight line. In an FEBRUARY 1985, ‘other words, the time required for the start of the modified Horner straight line is half of the time required for the Horner straight line, as can be seen from Eqs. 16 and 19. Depending on the formation and fluid parameters, hours ‘could be saved on the testing time. Very simple gp and pp functions are used to explore the effect of afterflow on the Horner analysis. It must be recognized that the sandface flow rate (afterflow) has to bbe measured to obtain accurate and reliable results from the modified Horner analysis. Even though the modified Horner analysis improves the semilog analysis, it cannot be applied to very early-time pressure data, The other methods, such as continuous ‘multirate, require prerequisite pp functions. Thus, decon- ‘olution methods will be used to obtain pp functions (in- fluence functions) and formation parameters in the following section, Deconvolution Determining wellbore geometries and reservoir types fractured, layered, composite, etc.) is an important part of well testing, The reservoir engineer must have suffi- cient information about the system being analyzed. For example, if type curves for fully penetrated wells are used for partially penetrated wells, both Kh and damage skin will be underestimated. Thus, the system identification becomes a very important part of well testing. For in- stance, identifying a one-half slope on a log-log plot of the pressure data will indicate a vertically fractured well, ‘as two parallel straight lines on a Horner graph will in- dicate a fractured reservoir. However, either wellbore storage or afterflow usually dominates these characteristic behaviors of wells and reservoirs during the early-time period, Thus, the pressure behavior of formation without the wellbore storage effect must be calculated or the wellbore storage effect on the formation pressure should bbe minimized for the conventional identification of the system, This is merely the deconvolution of Eqs. 3a through 3c to caleulate pap(tp) from Pyp{tp) and p(t). Several graphs of p.p(tp) ¥s. fp, such as linear, spherical, etc., will provide information about a given wellbore geometry and reservoir. ‘This approach to system identification is not general, but it uses our conventional knowledge about well and reservoir behaviors. In general, the problem of the system identification is much more complex because often we do rot know the governing differential equation. It is worth repeating that the fluid flow in the forma- tion is described by the linear diffusivity equation for the deconvolution methods given next. ‘There are several methods for the deconvolution of Eqs, 3a through 3c. These methods will be discussed in the following section, Linearization of the Convolution Integral. In this sec- tion, the convolution equation, Eq. 3c, will be solved directly by using the linearization method. Eq. 3c can be discretized as a aie Pwottone= Dy fae ons) “Paplayr. (20)Fig. 6—Calculatd formation pressure crop using linearization ‘method and wellbore pressure drop. By using the trapezoid rule for integration in Eq. 20, Pwotton+= Dy Wepltoi+1)40Uton+ 1 —#pie1) a +P evltodgoltn+1 ~to tiv ~tDi/2.«- 21) Eg. 21 gives a system of linear algebraic equations. The ‘coefficient matrix of this system of equations is @ lower triangular matrix; that is, all its nonzero elements are in the lower right of the matrix. The system of equations can be solved easily by forward substitution. For field data, Py should be replaced by (P:—Pw) and (P vz ~Prg) for drawdown and buildup test, respec- tively; and pyo should be replaced by (pi~Pyp)y and Gar pup which are sandface pressure differences without Sofage effects (formation pressure drop). ¢p for drawdown and (ap) for buildup must be used to replace qp in Eq, 21 for field case data. If the flow is racial, formation & can be determined from the slope of the straight line ofthe (Pyuj~Pw Vs. log At plot. Skin factor also can be determined from the conventional skin formula. However, the trapezoidal method used in Eq 21 gives oscillatory results. As can be seen in Fig. 6, at very early times, the calculated values of (Py) —P w/e oscillates, Furthermore, higher-order methods, including the Simpson rule, yield divergent results for the integral in Eq, 20. Hamming? suggested a stable integration scheme for evaluation ofthe convolution integrals. He also showed that direct integration (discussed as the linearization ‘method) of convolution integrals usually wil result in an oscillation. “The integral in Eq. 20 can be approximated as JP abttons -Aantedr oa soltore4)) Goltoms1 747. y ‘The right side of Eq, 22 can be integrated directly Substitution of the integration results in Eq, 20, and solv inf pg 2 Punttons)= Qolton+1 20 Fig. 7—Calculated formation pressure drop using Hamming mathed and welloore pressure drop. where “sum” mt © Putte uigolt ones ~t0s) fe is equal to —a0(tonss tise and the first value of pip is given by Puotor) gto) ‘As can be seen from Fig. 7, Eq. 23 gives a stable and ‘nonoscillatory integration scheme for the convolution in- tegral given in Eq, 20, Furthermore, the derivative of the sandface flow rate data, q((p), is not needed in Eq, 23. However, for Eg. 21, qi) must be known, A finite dif- ference appromixation also can be used for gp(¢p), but some accuracy would be los. The semilog plot of (Pn; Puy ¥S. At, given in Fig 7 for radial synthetic buildup test data, yields a straight line starting from a very early time (Ar—0 05 hours) with ‘correct slope. The lower curve in the same figure is the plot of (Pye —Puj)s Which includes the wellbore storage effect. Pooltow. ay Laplace Transform Deconvolution. The convolution of Eq. 4 yields “0 Pivtto)= J KP wot —nAr, --@5) where 1 Ktp)=£"! ] ween eee ee eee (26) ” law (ip) canbe computed either from the Laplace transforms of qp(¢p) data or a curve-fitted equation of 4p(tp) data. However, it would be time consuming to iver all the ¢p(¢p) data in Laplace space and transform it back to real space in accordance with Eq, 26. Thus, approximation functions must be used for q(‘p) data. ‘Once an approximation is obtained, it wll be easy to com pute K(¢p) and integrate Eq. 25 to determine pyp(¢p). 'A few types of approximation functions can be used to approximate qp(tp). We have tried rational functions, power series, and exponential functions. Exponential func- tions give @ good representation of gp(tp) data. Ex: JOURNAL OF PETROLEUM TECHNOLOGYTABLE 2—FRACTURED RESERVOIR ROCK ‘AND FLUID DATA 19 10"? 1000 00 400 4216.94 3,000 ‘0.35 10 fp, hours 10,000 Xephorical matrix used) to? ne 08 * 02 . 0.08 ponential functions approximation for gp can be ‘expressed as 18D +epe82!D +...cyetn'D, ...27) ‘After determining ¢; and 8; from qp(¢p) data, apply- ing the Laplace transformation to Eq. 27 and substituting the resulting expression in Eq. 26 yields fo vette ee ees (28) sXe a Bits ‘As suggested by Sneddon, if the inverse transforma- tion does not exist, Eq. 28 should be divided by p until its inverse transform is possible. For each division of p, Pwoltp) in Eq. 25 has to be differentiated. "We have seen from van Everdingen’s? and Hurst’s® works that qp can be approximated by an exponential function as ap=1~e-*D, a) ‘which is the simplest form of Eq. 27. For this case, Pp can be writen directly from Eq. 3 as Puolt) tp Pwwlto)= +P oto). 30) From Eq. 30 itis very simple to calculate pp if qn data ccan be approximated successfully by using Eq. 29) 44p(fp) data also can be approximated by piecewise ex- ponential functions for a selective interval for Eq. 26. Curye-fit Approximations of pyp. p.p(tp) in Eqs. 3a and 3c can be approximated by choosing Suitable func- tional approximations. These functional approximations. ‘could be power series, continued functions, rational func tions, or exponential functions. The success ofthis method ‘depends on how well the approximation function FEBRUARY 1985 ‘TABLE COMPARISON OF COMPUTED AND ANALYTICAL PRESSURE FOR A FRACTURED RESERVOIR eae APs 13196 65.39, 13378 98.32 43486 115.90 yesee 12487 19625 129.44 338.27 197.03, 19955 198.07 44083 129.98 vai33 14085, azz 141.59 14262 142.26 343.15 14281 yaa 149.32 yacor 143.78 200 14094 14898 14884 200 14868 148.70 148.60, 400 14092 149.02 149.85, 500 15088 150.88 150.80, 600 15165 15163 15159 700 15230 15229 18224 200 15285 15285 152.81 800 15933 15995 15352 represents the behavior of pp. As noted earlier, the log approximation was used to approximate P,p(t)- In this, ‘case, the natural choice will be the power series of In fp such that? Pwlta)= Sy eidln tp. eee GD Substitution of Eq. 31 into Eq. 20 yields n number of equations and m number of unknown parameters, (¢1,€2-m)". In fact, obtaining these paramters, ¢7, becomes an unconstrained optimization problem. We have applied a fourth-degree polynomial, Eq. 31, to wellbore pressure and rate data (synthetic) that were obtained from a fractured reservoir forthe reservoir and fluid parameters given in Table 2, The second column of Table 3 presents the calculated values of Ap y from curve-fit approxima- tion for the fluid and reservoir data given in Table 2, while the third column of Table 3 presents the values of Ap. calculated directly from an analytical solution without storage effect. The fourth column in Table 3 presents APs With the wellbore storage effect. ‘AS seen from Table 3, differences in Apy from the ccurve-fit approximation and the analytical solutions are almost identical, The relative error decreases for large times. Conclusions 1. The convolution integral (superposition theorem) is used for the analysis of continuously varying wellbore flow rate and pressure. This analysis is very similar to the conventional multirate methods. 2. Wellbare storage (afterflow) effects can be present to. significant degree in the Horner semilog straight lin. ‘The Horner analysis is modified by using measured sand- face flow rate data to obtain a correct semilog straight line, The modified Horner semilog straight line starts at least one-half cycle earlier than the conventional Horner astraight line, Approximate formulas are presented for the start of the modified Horner and Horner semilog straight lines as a function of sandface rate decline, production time, and the relative error between the correct and com- puted slopes. 3. The formation pressure (influence function) can be calculated from the deconvolution of measured wellbore pressure and sandface flow rate data. Some new decon- ‘olution techniques are introduced to compute the forma- tion pressure (influence function) without wellbore storage (afierflow) effects. Wellbore and reservoir geometries can be identified from this computed formation pressure, Fur- thermore, the conventional methods can be used to analyze this computed formation pressure to determine formation parameters, 44. Deconvolution of synthetic data from a homogeneous and fractured reservoir shows that it is possible to com- pute the formation pressure from the beginning ofthe test. ‘This computed pressure reveals the characteristic behavior ‘of homogeneous and fractured reservoirs. Acknowledgment ‘We thank Schlumberger Well Services for its permission to publish this paper and acknowledge the encouragement and help of Gerard Catala of Services Techniques Schlumberger during the initial stage of this work. We also thank H.J. Ramey Jr. for providing a solution for the integral in Appendix A. Nomenclature B = oil formation volume factor, RB/STB [res m3/stock tank m?] ¢, = system total compressibility, psi~! kPa") C = wellbore storage coefficient, Bblpsi Im? /kPa] Cp = wellbore storage constant, dimensionless ‘h = formation thickness, ft [m] k = formation permeability, md K = kernel of the convolution integral Pb = differential of pp Pos = shut-in pressure drop, dimensionless Poltp) = formation pressure, dimensionless pi = initial pressure, psi [kPa] = Pp +S = formation pressure including ‘skin, dimensionless pressure drawdown, dimensionless bottomhole flowing pressure, psi {kPa} bottomhole shut-in pressure, psi [kP) stabilized constant rate, STB/D Istock-tank m?/d] sandface flow rate, dimensionless reference flow rate, B/D [m?/d] reservoir flow rate, B/D {m3/d] sandface flow rate, B/D [m3/d] wellbore radius, ft {m} Laplace transform variable skin factor time, hours time, dimensionless a2 470, = dimensionless production time TT = transpose ‘Ar = running testing time, hours Alp = shut-in time, dimensionless 0.00026448/6yc 7 2 = & positive constant 0.5772... = Euler's constant = viscosity, ep [Pas] rate-pressure convolved time function dummy integration variable = porosity, fraction = Laplace transform of l@amezuard References 2 rr 15, Earlougher,R.C. Je: Advance in Well Test Anas, Monograph Sees, SPE, Richarson, TX (1977), 5 Geingarten, A.C. ef a “A Comparison Between Different Skin and Wellbore Storage ‘Type Curves for Early Time Transit ‘Analysis, paper SPE 8205 presented atthe 1979 SPE Anoual ‘Technical Conference and Exhibition, Las Vegas, Set. 23-26, Ramey, H.J Jr: "'Sbor-Time Well Test Data Inerpreation in the Presence of Skin Etact and Wellbore Storage," Pt. Tech Ja. 1970) 97-104; Trans, AIME., 249, Agarwal, RG, ALHussainy, Rand Ramey, HU. Je: An In ‘estgation of Wellbore Storage and Skin Effect in Unsteady Lig Ud Flow: 1, Analyeal Treatment,” Soc: Pe Eng. J (Set. 1970) 129.90; Trane, AIME, 289. Mekinley, R.MC.: “Welbore Transmissibility from Afterfow Dominated Pressure Buildup Data," J. Pet. Tech. uly 1971) 863-72; Tans, AIME, 281 svlougher, RC. Je_ and Kerich, K.M.: “Analysis of Shor -Time ‘Trasint Test Dat by Type-Curve Matching, "J. Pet. Tech. uly 1974) 793-800; Trans. AIME, 287 an Evesdingea, A-F- “The Skin Etfect and Its Iniunce onthe Praducive Capacity ofa Wel," J. Pe. Tech, Gune 1953) 171-76, Trans, AIME, 198, Hara, W.:""Exablishment ofthe Skin Effect an Its impediment 1p Fluid Flow into a Well Bore," Pet Eng. (Oct. 1953) BO-BI6, Gadi, RE. Tracy, G.W., and Wikey, L-E,:“Selocing Well ‘Which Wil Respond to Production Stimltion Treatment,” Dil ‘and Prod, Prac, API, Dallas (1955) 117-29, Ramey, HJ Ir. :*"Nor-Darcy Flow and Wellbore Storage Efects in Pressure Bulld-Up and Draidown of Gas Wells." J. Pet. Tech (Feb, 1965) 223-33: Trane, AIME, 234 Hatchinion, 1S. and Sikora, V.: "A Generalized Water-Deive Analysis," J. Pet. Tech. Guy 1989) 169-77; Trans, AIME, 216 Katz, Di, Tek, MR. and Toes, S.C. ""A Generalized Node for Predicting the Performance of Gas Rescrvcite Subject to Water Drive," paper SPE 428, presened at 1962 SPE Annual Meeting, oe Angeles, Oct, 7-10 (Coats, Kc. et: "Determination of Agale Influence Functions From Field Data," J. Pet Teoh. (Dee. 1968) 1417-24: Tram, AAIME, 231, “Jargon, Rand van Pollen, H.K.: “Uni Response Function Frm Varying-Rate Data,” J. Pet Tech, (Rup. 1965) 968-69: Trans, AIME, 234 Bostic, LN. er al: "Combined Analysis of Postactring Perfor mance and Pressure Buildup Data for Evaluating an MHF Gas Well," J. Pet Teck, (et. 1980) 1711-19. Pascal, H.: “Advances in Evaluating Gas Well Delivrailiy Us: ing Vatiable Rate Tere under Non-Darcy Flow,” paper SPE 9881 presented atthe 1981 SPE/DOE Low Permesbilty Symposiem, Denver, May 27-29 Meunier, D., Wittmann, MLJ., and Steward, G.: “laerpetation of Pressure uildp Test Using n-Sts Meassremen of Afterflw,” J. Pet. Tech. Jan. 1985) 143-52 van Bverdingen, AF, and Hurt W.: “Applicaton ofthe Laplace ‘Transformation io Flow Problems in Reservoirs,” Trane, AIME (1985) 6, 305-24 Cinco:Ley, H. and Samaniego, F “Pressure Transient Analysis for Naturally Fractured Reservoty paper SPE 11026 preseated st the 1982 SPE Anmval Technical Conference and Exubiion, New Orieans, Sep. 26-29. JOURNAL OF PETROLEUM TECHNOLOGY20, Kofuk, F and Kirwan, P.A. ‘Type Curves for Parally Penetated Well,” paper SPE 11676 presented athe 1983 SPE California Regional Meeting, Vents, March 23-25, 21, Ramey, HL, Jr and Agarwal, R..: ““Aaaulus Unloading Rates 1 lfuened by Welltore Storage and Skin fe (Oot 1972) 483-62. 22, Chen H.K, and Brigham, W.E.:“Pressare Buildup fru Well with ‘Storage and Skin in a Closed Square,” J. Pet. Teoh, Uan, 1978) 146, 23, Hamming, RLW.: Numerical Method for Slensts and Engineers, MeGraw Hu, New York City (1973) 375-77, 24, Spon, EM: The Use of Incegral Transforms, McGraw-Hill, New York City (1972) 207-18 25, Stefest, HL: “Numerical Inversion of Laplace Transforms," Com- ‘munications ofthe ACM Gan. 1970) 13, No.1, Algor 368 APPENDIX A. ‘Substitution of the exponential integral solution for pp and Eq. 13 for gp in Eq. 12 yields “New Skin and Wellore Storage Pool +90)=— HE] ~ camel ita : +S¢p(atp), 2Btp—9 where =(A2) 40) ‘Thus, the two forms of gp will be used interchangeably. ‘The integral in Eq. A-I cannot be integrated readily and, unfortunately, van Everdingen” and Ramey’ did not present the details of integration. Ramey* provided me ‘a heuristic derivation that was edited out of his paper. Ramey's integration method is given below. The Laplace transform of the integral given in Eq. A-l is tot ()=—_Ku(ve 3 ko: sap (a3) ‘The long-time approximation for Ko(vs ) is vs rool y= -(0 +). (4) Substitution of Eq. A-4 in Eq. A-3 yields 5 L Fo=————dn sn 4429). (as) 2648) ‘The inverse Laplace transform of Eq. A-S yields (Arp) Ye~* [In B—xi-2y+1n 4 +EKBAtp), (A) “Ray. 2: pool smmuncton Suro! U, Sard CA arch 10) FEBRUARY 1985 where 0.572 and Fi(aas) ‘Neglecting the imaginary term sr, the van Everdingen” and Hurst* forms are obtained as E(Atp)= ~ Ye [In B—2y+In 4 +Ei(BAtp)) (AD ‘The values of integral £(Ar>p) from Eq. A-7 were com- pared with the values computed from the Stehfest?* Laplace transform inversion technique using Eq. A-3. When Arp >30, the difference between two values of &(Arp) becomes less than 1%. The difference becomes smaller as At increases. Substitution ofthe log. approximations for the exponential integrals given in Eqs. A-I and A-7 forthe integral yields (in practical units) **) sean]. Pi~Pws(At =m [loe(e AS) 81 —In 8—2y-+In 4+ Ei(adt) +28) (a ‘The long-time approximation for E(t) may be obtained by substituting limiting forms of certain terms in Eq. A-7 for long times. For larger values of Ar, the term e7P8(—In B~2y+1n 4+25) in Eq. A-9 approaches zero. The term e~°41Fi(aas) can be approximated as 1 eM EK(ad=—, .. os (AI0) ar when Are 1. ‘Substituting Eq. A-10 in Eq. A-9 yields 1 aye fan ..30260Ar (At), Further substitution of Eq, A-I1 in Eq. A-8 gives Eq. 14 in the main text. APPENDIX B. ‘The dimensionless form of Eqs. A-8 and A-9 can be writ- ten as .sfin[ SetaP2 | +e}. ap er 333where E(Atp)= Ye“ [—In B—2y-+n 44 Bi(GAtp) +428) B2) ‘The differentiation of Eq. B+ Inf, +40 pV/Atp} yields with respect to a Mom =0.5+ [Ate 3) 3g EAron, 3) where mcom i8 the computed dimensionless slope of the Horner semilog straight line and x equals Inf(tp +40 p/Atp}. Let us define a relative error for the computed slope as Coe eee @4) 05 where Mam >0.5 Substitution of Eq. B-t in Bq, B- yiekls (BS) Taking the derivative of E(Arp) with respect to x and ating in Eq, B-S yields lltp +400) 2p tessa) s25-}. i(BAtp)+2S}- Wp eo Eq. B-6 gives the dimensionless time for the start of the Horner semilog straight line as a function of ¢, 8, tp. and S. ‘A formula for the dimensionless time for the start of the modified Horner semilog straight line can be found a from Eqs. A-I and A-2 as follows. Let us rewrite Eq. ‘ACL in terms of the modified Horner time: rovtes-0sfn[('22*)] Jeep), + 27 2BAtp en where E(Atp)= he Po (—In B~2y+In 4425). ...(B-8) %E(Atp) in Eq. A-7 is the only remaining term that is rnot included in the modified Horner time. Thus, the relative error of slope of the computed modified Horner semilog straight line (as in the Horner case) can be ex- pressed as SeCAtD) 9) where pba 1 van (2**) + 10) ar /P 1” 26Atp Differentiation of Eq. B-8 with respect to.x and substitu- tion of the result in Eq. B-9 yields — BAB +800 28 PADD +p +N (-In B=2y+In 4425), 1D Eq. B-11 gives the dimensionless time for the start of the modified Homer semilog straight line as a function of , B, tpp, and skin e-8D ‘SI Metric Conversion Factors bbl x 1.589873 E-OL ep x Loe E-03 ft x 3.048 E-01 psi x 6.894 757 E+00 PiSecepc er pcton aes ee van nuscg coed Set ‘ia Pm GP Popov ns Pho Tv Co JOURNAL OF PETROLEUM TECHNOLOGYDiscussion of Analysis of Simultaneously Measured Pressure and Sandface Flow Rate in Transient Well Testing H, Pascal, U. of Albers While the idea of analyzing simultaneously measured pressure and flow rate during transient well testing t0 determine the reservoir properties has been recognized {for about 20 years, itis only in the last several years that a good deal of interest has focused on this problem. '* ‘The objective of this Discussion is to clarify some basic features associated with solutions presented by Kucuk and Ayestaran,® Pascal® described a method of analyzing variable-rate tests by considering rate variation during the pressure drawdown test as & continuous function of time. A mathematical model was formulated in terms of a solu- tion of the Volterra integral equation of convolution type, ‘obtained from Duhamel’s superposition principle, in which a unique pressure response corresponding to con- stant rate was determined analytically from variable-rate drawdown, Since the single-point drawdown test is initiated from conditions of stabilized reservoir pressure, p;, and is flowed continuously for '>0 while the corresponding ‘lowing bottomhole pressure is recorded, the correspond= ing Volterra equation is een, Pi-PwlO=AP WD. “fey, a ; in which 4() is the sandface flow and jt) represents the pressure response of constant flow rate, Obviously, the determination of f¢) from Eq, | requires a knowledge of both functions p,,(0) and q(t). These may be obtained from measurements carried out in wells dur: ing the drawdown test provided that adequate instrumen- tation for sandface flow rate measurement is used. When these flow parameters are expressed by analytical func: tions, using for this purpose Marquardt’s algorithm based ‘on the nonlinear least-squares technique, an appropriate approach to determine function fe) in Eq. 1 analytically is the Laplace transform. Examples based on this approach to validate the analysis of simultaneously measured pressure and sandface rate in determining reservoir properties from transient well tests were given in Ref. 6 and 7. Hutchinson and Sikora,® Katz er al.,? and Coates er al. also have paid special attention to the determina tion of f( from field measurements for aquifers. Their results are of particular interest in predicting future reser- voir performance in the presence of a water drive and are based on a numerical deconvolution of Eq. 1. A numerical approach for determining ¢) under conditions specifically associated with transient well testing also was used by Jargon and van Poolen.! Their field examples revealed the problems associated with the numerical solution of Eq. 1 Ref. 11 was the first attempt to validate the experimental evidence for using the function je) from simultaneous, (sre van MARCH 1985 measurements of pressure and flow rate in a well pro- ducing at a variable sandface rate Tn the numerical solution of integral Eq, 1, several major problems may arise, some of which have been discussed in the papers mentioned previously, While it is outside the scope of this Discussion to address these problems in detail, some that appear essential to an ade- quate appreciation of this new approach in flow test analysis are stressed here. It is evident that in variable-rate drawdown tests, in which sandface rate declines continuously in time, a rate jump appears in a very small interval of ime, 01), ° in which coefficients ag, ay, and a were determined by ‘means of Marquardt's algorithm, Eq. 3 yields the following Volterra integral equation for determining /(0) stsaye Function € Time Fig, 1—Iustration of behavior of function ut) mayan) ened @ pul Field data indicated thatthe pressure drawdown, Ap ¥(?), also may be expressed analytically asa continuous func” tion of time by pal Logie e948, where coefficients 4, B, and r* are determined from Mar- quardt’s algorithm, ‘Since we have Ap,,()=0 at Eq. 5 may be rewritten as ee Spat © A log: P Jargon and van Poolen' also observed that a relation in the form of Eq, 5 was required forthe interpolation proc- ess used in their numerical approach to deconvolve Eq. 1 ‘As recently shown by Pascal,'* an exact analytical solution of Eq. 4, in which Ap,.(0 is expressed by Eq. 5, may be written as 4 AP a, aartry, fo: o ‘Thus, having obtained pressure drawdown corresponding to.a constant sandface rate in its more rigorous form, f), conventional procedures for analysis of drawdown tests are valid. As Pascal and Quillian!? have shown, the permeability and skin may be estimated by using the familiar semilog plot of pressure drawdown vs, time, 46 From this, itis clear that a numerical deconvolution of Eq. 1 requires an adequate quadrature formula and discretization of the pressure drawdown Ap,,(t), Errors ‘associated with this approach could be extremely large in certain cases and difficult to evaluate theoretically. It is always difficult to choose an appropriate criterion for Judging the effectiveness of numerical methods, partic ularly in the case of a numerical solution of the Volterra integral equation of convolution type. As a result, the problem in determining fit) is the accuracy of the ‘numerical approach, which is difficult to solve in a general case (i.e., for any sandface rate profile and pressure drawdown occurring in practice). Pressure response of constant rate determined from Eq 7 is an exact analytical solution of Eq. 1. However, a problem may arise in conditions where pressure drawdown of variable rate Ap,.(2) cannot be expressed accurately by Eq. 5. Obviously, the imitations associated ‘with deconvolution of Eq. 1 may appear in certain situa- tions of practical interest in transient well testing, The field ‘examples could better show conditions in which accurate values of reservoir properties may be obtained Naturally, we now ask if itis possible to obtain an ade- {quate representation of the variable sandface rate effects ‘on the pressure drawdown such that the deconvolution of | Eq, | is no longer required in determining reservoir prop- erties from variable-rate tests, For this purpose, we focus fon some analytic properties of Eq, 2 Examining the shape of the kernel function in Eq. 2, ent Re Paty ® ‘one can see that it is a postive function for 01 This behavior of the sandface flow rate may occur in specific situation. In general, the behavior of the sand: face flow rate is controlled by choke and tubing size, for- mation fluid, rock properties, etc. In fact, it is difficult to regulate the sandface flow rate behavior during drawdown tests. Fig. R-1 presents measured drawdown, sandface flow rate data from four different reservoirs; each sandface flow rate data set exhibit different behavior from others. For example, the sandface rate presented by Curve A reaches a maximum value at 1=0.024 hour, declines, and then increases slightly with time. On the other hand, the sandface rate presented by Curve B in- ‘ceases with time and then remains almost constant after 1=0.1 hour. Probably, the sandface flow rate data presented by Curve C can be described crudely by Pascal’s Eq. 1. However, the maximum occurs at 1=1,440 see- fonds, not between 5 and 60 seconds. ‘We will not discuss Pascal's approximation for the well- bore pressure behavior (Pascal's Eq. 5 or 6) in detail here; however, none of the wellbore pressure data measured simultaneously with the sandface flow rate data presented by Fig. 1 can be approximated by Pascal's Eq. 5 or 6. As discussed in our paper, if we have to approximate the sandface flow rate data for deconvolution, general- ized functions (power series, rational or exponential poly rnomials, etc., depending on the behavior) should be used. Furthermore, we do not have to approximate the well- bore pressure data; they can be used directly for decon- volution. Ifthe behavior ofthe sandface flow rate can be Cory 1865 Sot f Pea Engines OCTOBER 1985 pinner) data from various fields. Fig. R-2—Synthetle drawdown pressure and flow rate data approximated by a single exponential function, then our Eq. 30, as well as Pascal’s Eq. 7, can be used without approximating the wellbore pressure, ‘We agree with Pascal that deconvolution is a difficult, inverse integral operation. That is why we suggested a few different techniques to handle a variety of situations Without elaborating further, let us use Pascal's method [4pw(t/g( vs. F(t) plot) to estimate skin and permea- bility. To compare Pascal's method with other methods, wwe designed a drawdown test using the formation fluid and rock properties given in Table 1 in our paper. We ‘made the sandface rate decline exponentially after it reaches a maximum value, as shown in Fig. R-2. Fig, R-2 also presents the wellbore pressure and shows that the sandlace flow rate behavior ofthis drawdown test can 1867Fig. R-3—MDH plot of the synthetic drawdown test. ‘TABLE R-1—INTERPRETATION RESULTS OF ‘THE SYNTHETIC DRAWDOWN TEST Start of Semitog ‘Straight Ling Method ta ims) _s True values = 40.00 10.00 MOH’ 36 427 1107 Pascal 12 aoae 026 Giacfoter os 40.20 982 Convolution (rmltrate) 0008 40.30 10.25 bbe described by Pascal's Eq. 3. Although this is a syn- thetic test, its behavior is similar to the field case presented by Curve C. ‘The sandface flow rate given in Fig. R-2 can be ap: proximated as 4) =2623.6+714.8e 2-6, where 1) >0.24 hour. ‘We have analyzed the drawdown test given in Fig, R-2 using Miller-Dyes-Hutchinson (MDH) (Fig. R-3), Pascal! (Fig. R-4), Gladfelter er al,? (Fig. RS), and Fig. R-4—Pascal's plot of the synthetic drawdown test. convolution (Fig. R-6) methods. For new time and pressure function computations for the Pascal, Gladfelte, and convolution methods, we have used the dimensionless sandface flow rate [99()=4(/q ref] for convenience. ‘The results of the analysis from each method are listed in Table R-1. We have also listed the start of the straight line for each method. In other words, the data presented at each plot in Figs. R-3 through R-6 cannot be analyzed by these four methods before the start of the straight line. ‘The estimated values of permeability and skin from each method are acceptable for all practical purposes for this particular drawdown test. However, the start of the straight line is different for each method, For example, it is 3.6 hours for MDH, 1.2 hours for Pascal, 0.8 hour for Gladfelter, and 0.005 hour for convolution. However, the Gladfelter method, the simplest of the methods with the exception of MDH, indicates a transitional straight line (as shown in Fig. R-S, it is between 0.05 and 0.3 hour) with a slope 25% larger than the correct straight line. The existence of a transitional straight line makes the Gladfelter method less desirable than the convolution method. If this test were run for 0.3 hour, if the permea- bility of the reservoir were much lower than 40.0 md, fo if the skin factor were much greater than 10, the Gladfelter method would have indicated wrong reservoir parameters. Fig. R-6—Sandface rate convolution plot of the synthetle JOURNAL OF PETROLEUM TECHNOLOGY‘We found that MDH yields higher skin and permeabil- ity values than true valves. In our opinion, if the Gladfelter ‘method has to be used instead of the convolution method, it should be used with MDH to avoid picking up the tran- sition straight line as a correct one. If MDH does not in- dicate an apparent semilog straight line, then Gladfelter may not yield the correct straight line. In wel testing, estimating formation parameters as early as possible is desirable to minimize outer boundary ef- fects, such as gas cap, barriers, discontinuity, etc. This is probably one of the main reasons for measuring the sandface flow rate along with pressure. Thus, the best method is the method that estimates formation parame- ters, such as permeability and skin, as early as possible. According to these criteria, Pascal's method is slightly better than MDH and much worse than both the Gladfelter and convolution methods. In terms of computation, both the Gladfelter and convolution methods are simpler than Pascal’s method. In addition to lengthy computation of ‘OCTOBER 1985 F and F (Pascal's Eqs. 18 and 19), Pascal's methods also require use of a nonlinear minimization algorithm such as Marquardt’s? t0 compute a, and « for Eq. 1 References 1. Pascal, H: "Discusion of Analysis of Simultaneously Measured Pressure and Sandface Flow Rate in Transient Well Testing, J et. Tech. (March 1985) 565-58 2. Glatfter, REL, Tey, G.W., and Wise, LE “Selecting Well ‘Which Will Respond io Produstion Stimulation Treatment, ° Dr dnd Prod. Prac, APL Dallas (1958) 117-29. 3, Marguard, D.W.: "An Algorithm Fr Leas Squares Estimation ‘of Nonlinear Parameters,” Soc. pp. Math ine 1963), M1 No. 2, 81-41 SI Metric Conversion Factors bbIID x 1.589 873, psi X 6.894 757 E-01 E+00 mid kPa ser 1869