From Straight Lines to Deconvolution: The Evolution of the State of the Art in Well Test Analysis Alain C. Gringarten, Imperial Colege London ‘Summary Wel rest analysis has been used for many years to assess well condition and obtain reservoir parameters. Early interpretation methods (by use of straight lines of log-log pressure plots) were limited tothe estimation of well performance. With the introduc- tion of pressure-derivative analysis in 1983 and the development ‘of complex interpretation models that are able to account for de- tailed geological features, well test analysis has become a very powerful tool for reservoir characterization, Anew milestone has ‘been reached recently with the introduction of deconvolution, De- convolution isa process that converts pressure dala at variable rate into a single drawdown at constant rate, thus making more data availabe for interpretation than inthe original dataset, in which only periods at constant rate can be analyzed. Consequently, itis postible t0 see boundaries in deconvolved date, & considerable ‘advantage compared with conventional analysis, in which bound= aries often are not seen and must be inferred. This has a significant impact on the ability to certify reserves, ‘This paper reviews the evolution of well test analysis tech- niques during the past half century and shows how improvements hhave come in a series of step changes 20 years apart. Each one has increased the ability to diseriminate among patentil interpretation models and to verify the consistency of the analysis, This has increased drastically the amount of information that one can ex= tract from well test data and, more importantly, the confidence in that information. Introduction Results that can be obtained from well testing area function ofthe range and the quality of the pressure and rate data available and of the approach used for their analysis. Consequently, at any given time, the extent and quality ofan analysis (and therefore what can ‘be expected from well tet interpretation) are limited by the state- ‘ofthe-art techniques in both data acquisition and analysis. As data improve and better interpretation methods are developed, more and ‘more usefil information can be extracted from well test dat Early well test analysis technigues were developed indepen- dently fom one another and often gave widely different resus for the same tests (Ramey 1992) This has had several consequences + Am analysis was never complete beeause there always was an ltemative analysis method that had not been tried. “Interpreters had no basis on which to agree on analysis results + The general opinion was that wel testing was useless the wide range of possible results Significant progress was achieved in the late 1970s end early 1980s with the development of an integrated methodology on the basis of signal theory and the subsequent introduction of deriva- tives. It was found that, although reservoirs are all different in terms of depth, pressure, fuid composition, geology, etc, their behaviors in well tests were made of a few basic components that were always the same. Well test analysis was about finding these components, which could be achieved in a systematic way, fol Towing a well-defined process. The outcome was a well ts inter- ‘copa © 20 Sete Pevoam Engineers ‘Th gaa (02070 wa csp esa ath 2006 9° Ao Tes en cm nA Ys aS ae ae ‘SiG menses fr on 13 Seon 208 Revaed mu ‘ese rev e207 Paper par apposed i 37 February 2008 SPE Reservoir Evaluation & Engineering pretation model, which defined how much and what kind of knowledge could be extracted from the data, The interpretation madel also determined which of the various published analysis methods were applicable and when they were applicable. Impor- tantly, the integrated methodology made wel test analysis repeat- able and easy to lear, The evolution of the state-of-the-art tech- nigues in well est analysis throughout these years canbe followed from review papers that have appeared at regular intervals in the petroleum literature (Ramey 1980, 1982, 1992; Gringarten 1986; Ehlig-Economides etal. 1990), 'No major breakthrough occurred during the next 20 yeats, ‘hich instead saw minor improvements in existing techniques and the development of new, more complex interpretation mauels. Ia that period, the word “conventional” shifted in meaning from straightine to derivative analysis. The word “moder,” previously attached to pressure log-log analysis, disappeared, suggesting that ‘well test analysis had become mature. ‘A new milestone has been reached recently with the addition of 1 working deconvolution algorithm to the well test analysis tol Ki. The impact of such a development on well test interpretation and its place in the evolution of well test analysis methods are discussed inthe present paper. History of Well Test Analysis Looking back at the history of well est analysis i the oil industry, itis possible to identify different periods during which particular analysis techniques dominated and specific types of information prevailed (Fig 1). ‘At the beginning, most anlysis techniques came from ground ‘water hydrology, in which they had been used for many yea Examples include “semilog” straighiline analyses, suggested by ‘Theis (1935) and applied by Cooper and Jacob (1946), and type- curve matching, also introduced by Theis (1935), The wel test analysis methods prevaiing during the 1950s and 1960s are described in SPE Monograph | by Matthows and Russell (1967) and SPE Monograph 5 by Earlougher (1977). These tech niques, developed in oil companies and illustrated in the work of Miller etal. (1950) and Horner (1951), are based on straight lines and apply to middle time semilog data (Miller et al, 1950; Hornet 1951; Warren and Root 1963; Odeh and Jones 1965) or to simple boundary effects (Muskat 1937; Homer 1951; Matthews etal 1954; Jones 1956) at late times. The main mathematical technique used in those days was the Lapiace transform as published by Van Everdingen and Hurst (1949). Interpretation techniques were de- signed tobe performed exclusively by hand with pencil and graph Paper. The emphasis was on production operations, and wel test analysis results were usually limited to the determination of res- ervoir permeability, well skin effector productivity index, drain age area, and average reservoir pressu, During the late 1960s and early 1970s, most major develop- ments originated (rom universities, led by HJ, Ramey Jr. The emphasis shifted toward the understanding of earlytime behavior because it became apparent that some ofthe results from straight line analyses could be ambiguous (Ramey 1970). It was relized, for instance, thatthe skin was a global value that didnot iaform fally of the causes of well damage or stimulation and therefore did not provide a sound basis for operational decisions. Specialy, the same negative skin could be obtained from aciizing or from fracturing (Ramey 1970), and the same positive skin could be aInterpretation Tools Emphasis ‘Method 150s | Straight tines Laplace transform Homogeneous reservoir behavior Late 608 | Pressure type-cuve Early 70s | analysis ‘Groen’ functions Near.welibore effects Late 708 | Type curves with independent variables Integrated methodology Steblest algorithm Dual-porosity behavior Early 80s | Derivatives Computerized analysis behavior and boundaries 20s Muttiayered reservoir downhole rata measurements Integration with interpretation models from other data leary 008 Deconvolution Enhanced radius of investigation boundaries Fig. Summary of the history of well test analysis. produced by well damage or result from partial penetration (Brons find Marting 1961) or multiphase flow around the well (Kazomi 1975). Type-curve analysis (Ramey 1970; Agarwal et al. 1970; McKinley 1971; Gringarten and Ramey 1974; Gringarten et al. 1974, 1975; Cinco-Ley ot al. 1978; Agarwal et al. 1979) was introduced by Ramey (1970) to get an insight into the meaning of the skin and therefore on the means to cute it. Particular emphasis was placed on wellbore storage (Agarwal et al. 1970), high conductivity fractures (Gringarten et al. 1975) and low. conductivity fractures (Cinco-Ley etal. 1878). Type-curve mack Ing also provided a way to select the applicable strght line for semilog straightline analysis (Ramey 1970), which had been a ‘major shortcoming inthe past, New mathematical tools, such 25 ‘the ones based on Green's functions (Cringarton and Ramey 1973) ‘were also developed, which enabled new interpretation models (Gringarten and Ramey 1974; Gringarten etal. 1975; Cinco-Ley et al. 1978; Agarwal etal. 1979) co be generated, These Improved further the understanding of early-time data as described In SPE Monograph 5 (Earlougher 1977). Analysis, however, was stil mostly manual ‘Starting in the late 1970s, most new developments came from service companies. Type-curve analysis was significantly en- hhanced when the concept of Independent variables was introduced by Gringarten ea. (1979) and Bourdet and Gringarten (1980) ‘This and the integrated well test analysis methodology that was, developed at the same time (Gringartn et al. 1979; Gringarten 1984) made the analysis process easier. It also provided more consistent and more reliable analysis results. This perlod marked the beginning of the end of manual analysis, because the full application ofthe new, incegrated methodology requlred the use of ‘computers. With these anc: new numerical techniques such as the Stehfest’s algorithm for Laplace inversion (Stehfest 1970), new {interpretation models were developed that made it possible to ide. tify more complex well behaviors such a5 double porosity (Crin garten et al. 1979; Bourdet and Gringarten 1980; Gringarten etal. 1981; Gringarten 1984), As a cesult, well test analysis started Dbecoining more useful 2s @ reservoir description ool, both during exploration and for reservoir simulation. At the same time, the usefulness of well test analysis in production operations was re- emphasized with the practical development of NODAL™ (Gchlumberger) analysis (Mach etal. 1979) ‘Well test analysis bocame a true reservoir characterization tool with the Introduction of derivatives by Bourdet et al (19832, 2 1983). Derivatives have revolutionized well est analysis by mak- ing it possible to: " Understand and recognize heterogeneous reservoir behav: tors, such as double permeability (Bourdet 1985; Joseph et al. 1986) and composite (Chu and Shank 1993) * Identity parcial penetration or limited entry (Kuchuk and Kir ‘wan 1987) and ather near-wellbore effects * Analyze horizontal wells (Daviau etal. 1988), + Handle a wide range of boundary effects (Clark and Van Golf-Racht 1985). ‘The power of well test analysis has been further extended re cenely with the introduction of an effective algritim for decon- volution by van Schroeter etal. (2001). Deconvolution converts varlablerate pressure data into a constant-rate single drawdown ‘with duration equal to the total duration of the test. This makes more data avetlable for interpretation and helps greatly in the identification of the interpretation model. For instance, decoavo- Tution enables boundary effects to be seen although they may not Appear In individual flow periods at constant rte ‘The improvements in analysis techniques listed above are closely tied with improvements in data. Unil the early 1970s, pressure messurements were performed with Bourdon-type me: Chanical gauges and were limited in resolution and accuracy. The ‘overall quality of pressue data improved dramatically in the late 1970s and early 1980s with the advent of electronic gauges. the ability to easily design tests to ensure that specific Information ‘could be obtained by use of sophisticated well rest analysis soft ware packages, and the possibilty to monitor botomhole pressure at the sutface with surface pressure readout equipment, New mod: tls were also required to accommodate new testing oF production procedures, such as horizontal wells (Daviau et al. 1988) and si ‘multaneaus downhole pressure and rate measurements (Kuchuk and Ayestaran 1985) Well Test Analysis Methodology ‘The most significant breakthrough in well test analysis since SPE ‘Monograpi 3 (Earlougher 1977) remains the development in the lace 1970s and early 1980s of a general and systematic approach to the analysis of well tests by Gringarten et al. (Gringaren et al 1979; Gringarten 1982, 1984, 19852, 1986). This approach unified the various techniques previously described inthe Itrature, which had been used independently and often gave conflicting results (Ramey 1992), into @ single methodology on the basis of signal Fetruary 2008 SPE Reservoir Evaluation & Engineeringtheory. It pointed out inconsistencies inthe way well test analyses ‘were performed and provided answers to many fundamental ques ‘Hons, which today ae taken for granted but were far from obvious atthe time, such 3s, ‘+ What type of results can realistically be obtained from well resting? + What is the best method to obtain these results? + How does well testing actually contbute to the character tration of a reservoir as compared to ather sources of information such as geophysics, geology, oF petrophysics? ‘The Fundamental Problem of Well Testing. The emphasis of the integrated approach was on the well test "behavior," which refers {0 the response of the well to changes in production conutions. ‘The behavior enables identification of the applicable well test Interpretation model, which controls the maximum number of parameters that can be obtained from a test and the meaning of these parameters, It was shown thatthe process to obtain the well test interpre tation model was @ special application of the general dheoty of signal analysis (Jowarina and Fras 1979). By considering well test- sng and well test analysis within the context of signal theory (Grin garten et al. 1979), it became easier to understand the scope and limitations of well test analysts. Tn signal theory, signal processing is schematically described as (Gringarten 1985a): T9890, ceeseseee sn which 51s an operator: Zan input signal aplied to Sand Q. an ‘ouput signal ceslting from the application of Fito 5. O repre. sents the dyramic response of the system to the input signal J Several types of problems are associated with Ea, depending on ‘which ofthe three quantities, 1,0, oF 5s unknown and rust be Calculated while the eter two are known, both the input signal and the system Sare know, Ocan be calealated without ambiguity, and the solution is unique. This is known asthe direc problem or convoltion. An example of direct problem isa fllows (Ramey 1992): The input 1s (1,2, 3), the pertor S's the adition operation, the ouput Os 6, There isa unique answer. In wel esting and petroleum engeering, uss sed in foword modeling or tes design or preiton (oecasirg). ‘Aernaively the spat signal / and the ouput signal O cou both be knovi, the unknown Ding the system 5 Tiss an inverse problem. In petroleum engineering, Ue Inverse problem f saved faring the Identcaion of an interpretation model. Unlike the dliect problem the solution of the Inverse problem Is non-unlgue Several different systems may exst which, subjected idenial {nut signals, provide kenical ouput signals. By use ofthe same example as forthe diet problem. an inverse problem formulation ‘would be: The input signal Tis (i, 2,3) the output signal O's 6, ‘What is the operator 5? There i not a unique answer: t could be an addition (11236) ora mulipliation (1x23 =). This non Uniqueness isa property of the inverse problem dat cannot be avoided. thas significant implications on the design ofan efficient ‘methodology for well est analysis Finally, the system Sand the ouput signal O may be known, the unknown being the Input signal This problem is known as dleconvolaion an also yields © nox-unigue answer (6 canbe ob tained by adding 5 and I, 4 and 2, or 3 and 3) [n well esting, deconvolution i involved when converting vrlble rate draw: down pressure response int a constantate one Input and Output Signals In well test analysis, the sytem represents the unknown eese- voit the characteristics of which are fo be determined. The input signal is usually a step function in rae ereated by closing a flowing well o an injection well (blldup of faa, respectively); bby opening a well previously shut in (drandowea): or by injecting in a well previously closed (injection. “Toe carresponding output signal Os the change in pressure created by the change in rate and measured in the same well (exploration or production testing) or in a different well (interfer cence testing). Aematively, the Input signal could be the wellhead ‘or bottomiole pressure; the output signal would then be the change in the well production rate. In layered reservoirs there are t60 ‘output signals: the pressure, and the rates from each individual layer, which must be processed together. ‘Nate input signal canbe created at the surface by shutting or ‘opening the master valve o atthe bottom ofthe well with a special downhole shut-in device. Wellhead shat‘in is commonly used in ‘wells already in production, whereas botiomhole shut-in i san- dard practice after drling [a dilltem test or (DST}]. The way the ‘ate signal is created is not Important as far as well test analysis s concerned. The interpretation methods that are described hereafter are valid for both production tests and DST and also for the analjsis of wireline formation tests, What is most important for analysts i the quality ofthe rte input signal, which must be ofthe proper shape and uation, and the quality of the measured pres sure ouput signa. mn practice, one must differentiate berween the fist drawdown {na reservoir a stabilized pressure (Fig. 2) and a subsequent flow period (Fig 3) Inthe fist case the output pressure signal Ap isthe Aiference between the inal pressure 2 and the pressure p, (0 at n elapsed time AV in the drawdown Pi pe (At) ‘Ap = p,~ p, (At) Time From Start of Drawdown Fig. 2—Pressuro response toa stop rato change, first dramdown after stabilization. February 2008 SPE Reservoir Evaluation & Engineeringon Rate Pressure py(At) 0% rar Time from the start of the test Fig. 3—Pressure response in a subsequent flow period, p= PI PAD. oso pedeaesett aC In the case ofa subsequent flow period in a multirate test, on the other hand, there Is a choice of te output signals (Fig. 3). One can select, as before, the difference between the Initial Pressure p, and the pressure (Mat an elapsed time Ain the flow period of interest (Buildup Flow Period 2, or Drawdown Flow Period m in Fig, 3): pr pdda. - @ Because p, ts usually not known, the signal s actually (M0. This sigoal is analyzed with the Horner method (Homer 1951) and its exiension to multirate (Odeh and Jones 1968). Altemnatvely, ane can select the difference between the pres sure atthe sar of the flow period, p,(Ae=0), and the pressure ‘p.(Qd at an elapsed time Arin the flow period of interest. Ap= lnfr=0) = pA0) a ‘This signal is analyzed by log-log analysis (Ramey 1970) and by specialized analysis (Gringarten etal. 1979), ‘Well Test Analysis Process. Finding he well test interpretation model involves & thre: step process Identification of the Interpretation Model (Inverse Problem). First, ane must identify @ model ofthe actual reservoir § say 3, the behavior of which Is Identical to the behavior of S. dential be hhavior in this case means thatthe observed output signal O ob- tained from the reservoir 5 and the output signal O' calculated from the model 8 exhibic the same qualitative characteristics (2 show similar shapes) 19350... o Identifying the model isthe most important step of the analysis process: if the wrong model is selected, all reservolr parameters Aerived from the analysis will be incorrect, and the subsequent fengineering decisions on the basis ofthese parameters wil ikely be inappropriate. For instance, mistaking a double porosity bebav: Jor fora depletion effect (which was not uncommon before type curve analysis and derivatives became available) has led operators {© abandon wells that were perfectly viable. Finding © implies solving the Inverse problem, which requires an Sdentifcation or pattern-recognition process. By definition, the Solution is not unique. The degree of non-unigueness tends t0 “4 Increase with the complexity ofthe reservoir behavior and to de ‘crease with the amount of information available on the well and reservoir being tested, One must therefore try to reduce the non ‘uniqueness of the solution by using as much information as pos- sible, In practice, this means * Increase the amount and quality of input and output infor ‘mation used directly in the analysis (Le, the amount and quality of both rate and pressure-tes data). + Perform a series of specifically designed verification tests on the model. + Verify the consistency of the well tes interpretation model ‘with addtional, notesting information from geophysics, geology. petrophysis, drilling, production logging et ‘The need for more complete pressure and rate test data has not always been obvious, although it is clear fom Eq. 2 that both pressure and rate information are required for signal processing. ‘This is because at any given tine, the understanding of the inter pretation process and the limitations of measuring devices dictate the requirement for data. Measuring devices and data-equisition requirements in tur fond tobe limited tothe needs ofthe dominant analysis techniques. Progress in measurement devices and test de- sign usually takes place only when new interpretation techniques are developed that require new measurements. For many years, ‘emphasis mainly has been on pressure-bulldup data. Rates often ‘were reported only as average wellhead values before the buildup, [New advanced techniques now require drawdown pressure data as. ‘well as bulldup data and accurate flow rate as function of tie. Inthe same way, early-time pressure data ether were not measured for were not read from recorder chats until required by the early time analysis techniques discussed in SPE Monograph 5 (Ear- Tougher 1977). Accurate measurement of these data was made possible by the subsequent development and routine use of elec tronic gauges. Now, the curren rend is toward longer tests, helped by downhole permanent pressure gauges, to take advantage of new Interpretation models that enable identification of heterogeneities ‘and boundary effects in the reservoir away from the wellbore. Te must be stressed that non-uniqueness is not specific to well test analysis. All interpretation and modeling processes give non- Lunigue answers. This holds tne in geophysical Interpretation, in geological interpretation, in log interpretation, and in the reservoir ‘modeling aspect of reservoir simulation, The problem of non: uniqueness is now well recognized inthe ol industry, Its the main February 2008 SPE Reservolr Evaluation & Engleering Ooreason forthe increasing use of stochastic modeling techniques, ‘which aim at providing allerative ecu probable representations Of the reservoir © capture the uncertainty associated with predic tions (Hemet! 1985; Suro: Peres etal 1991. Tn identifying a well test Inerpetation model from well test data, wo are not limited by ave silty to mathomatiealy represent Jnterpretation models, ether analytically or numerically fe, by cour abit to solve the direct problem), but by our ability (0 solve the inverse problem (Le. by the curent state-of the-art techniques in model denticaion). As densification techniques become mare powerful [as with derivatives (Bourdet tal. 1885) and econo: Jaton (von Sebroeter etal. 2001) and the rsolurion of measure- ‘ents improves, te numberof Behavior components that can be ‘Sentified increases, resulting in more detailed Imerpretaton modes Calculation of the Interpretation Model Parameters (Direct Problem). Once the interpretation model has been Identified, Is response mast be generated (either analytically or numerical), tnd the parameters ofthe model must be adjsteé ntl the model ives the same quantitative response as the actal reservoir. This s in addition to providing the same qualitative response (eg, sme shape), a condition thet controlled the selection ofthe mode inthe fist place. The adjusted numerical values ofthe model parameters are then said to represent the values ofthe corresponding reser ‘oir parameters. ‘At this stag of the interpretation process, the problem to be solved isthe direct problem, because the model i now known. Because the solution of the diect problem is unique, there Is @ unigue set of model parameter values that can provide a best it ‘with the observed date, This means that once the interpretation model is selected, the reservoir parameters corresponding to that model are defined uniquely, and the numerical values of these parameters ae independent ofthe method wsed to calculate them. Results must be the same whether reservole parameters are calcu. lated by vse of straight lines, loglog type-curve matching, or nonlinear regression techniques (Rosa and Hmne 1983). The only fccepabl differences ace those cased by te differences in 0 luton of the various methods. Th other words, different interpretation methods that use the same interpretation model must produce the seme parameter val- tes awhen applied properly. This was not universally understood before the development of the integrated methodology, because Straightline methods [MDH (Miller et al. 1950) and Horner (1951) and type-curve analysis—with different ype curves rep resenting the same model (Agarwal tal, 1970; McKinley 1971) often gave diferent results Verification ofthe Taterpretation Model. Because ofthe now uniqueness, one must verify the interpretation model found during the identification step. Consistency checks are made among al characteris inferred by the model and the coresponding known Information from the actal erervor and measured data. Ifthe ‘model satisfies all the checks, tis deemed to be “consistent” and to represent a valid solution tothe problem. Ifthe model fils any check, ts considered Invalid “The interpretation process must be repested to deny all pos sible consistent models, which can be raked in terms of decreas Ing probably, If needed, anew well est can then be designed to confirm the mest probable model Well Test Interpretation Model (One important ingredient of the integrated methodology was the realization from experience that although reservoirs are diferent in terms of physical description type of rock, depth, pressure, size type of fluid, uid content, et), the numberof possible dynamic behaviors ofthese reservoirs during a well tet are limited. This is Lbecause a reservoir acts as a low-resolution filter so that only high Contrasts in reservoir properties can appear in the output signal (Perez-Rosales 1978). Furthermore, these dynamic behaviors are ‘obtained from the combination of three components (Gringarten et al. 1879; Gringarten 1982, 1985a) that dominate at different times during the test, namely +The basic dynamic bebavior of the eservoir during middle tds, which i usually the sare forall he wells In a given reservoir February 2008 SPE Reservoir Evaluation & Engineering + Near wellbore effects at early times cesulting from the well completion that may vary from well 0 well oc from test to test ‘Boundary effects at late times, determined by the nature of| the reservotr boundaries, whieh Is the same for all the wells in a given reserve, and by the distance from the well fo these bound ries, which may differ from well to well, Basic Reservoir Behaviors. The bast reservoir dynamic behavior reflects the number of porous media of different mobilities (As) and storativties (by) that participate inthe flow process (Grin garcen 1984, 1986). These basic well test behaviors are illustrated in Fig 4 ‘Homogencous Behavior. I there is only one mobility and one storatvity involved, the behavior i called “homogeneous.” Ho- mogeneous behavior means that variations in mobility (kia) nd storativity (bef) throughout the reservoir are too sinall co be Sn in well test data, Tn terms af flow, ther s essentially only one single porous medium, As 2 result, the permeability measured in 2 test corresponds tothe same permeability system as that described by core data. The respective values of permeability could be dif ferent, but only because the conditions of the measurements are different. Akhough uniformly homogeneous properties are as- sumed in the derivation of the analytical representations of the Interpretation mode! from the diffusivity equation, the word “ho- mogeneous” associated here to the word "behavior" does not i ply tha the actual reservoir has homogeneous properties throughout. Heterogeneous Behavior, “Veterogencous” behavior, on the other hand, means two oF more mobiles and storalvities are fnjeractng. These may be uniformly distributed or segregated, but thelr main characteris Is that their values are noticeably diferers. (One example of heterogeneous behavior isthe double-porosity behavior (Warren and Root 1963). Double-porosity behavior in volves two media with widely different permeabilities, and only the most permeable medium can produce flud into the wel. The other acts as a recharge for that most permeable mediuen, Double porosity behavior combines two successive homogeneous behav: fors, which only differ by their porosities—or more correctly, by ‘hie storatvitis, The fist homogeneous behavior is controlled by the mobility and storatvity ofthe most permeable porous medium at early middle-times. The second homogeneous behavior is con trolled by the same mobility and the sum ofthe stoativities ofthe constitutive media at late middle-vimes, Double-porosty behavior ‘occurs generally in naturally fractured reservoirs, in multilayered reservoirs with high permeability contrast between the layers, and Jn single layered eservoirs with high permeability variation along the reservoir thickness. Double-porosity behavior is typically ound in carbonate reservols, and in carbonate, limestone, granite, basalt, and unconsolidated sand formations (Gringarten 1984). Ha 1. HOMOGENEOUS BEHAVIOR One mobility — Khe One storativty ech 2. HETEROGENEOUS BEHAVIOR More than one mobility, storativity x x 12 z a. 2poresty 2Pemesbilty Composite fesse, mined ay Mayet goer nat Fig. 4Basic woll test interpretation-model reservoir behaviors. 6Another example of heterogeneous behavior is the double petmeability behavior (Bourdet L985), which refers to two distinct porous media as in double porosity, bu each medium can produce Into the well. Examples of double-permeability behavior can be found in multilayered reservolrs with relatively low permeability contrast between the layers. Commingled reservoirs are @ special «case of double-permeablity behavior with no interlayer crossflow. Contrary to homogeneous behavior, double-porosity and double: permeability behaviors imply thatthe permeability measured in a test and the permeability measured in a core may correspond to different porous media, ‘A third example of heterogeneous behavior is the composite behavior, which implies one set of mobility and strativity values aroure the well ae a differen one at some distance from the wel, ‘Composite behavioe may be caused by a change in reservoir thick ‘ess or porosity, a variation of facies, o a change in Td mobility In the reservoir. Examples of composite behaviors are found in such circumstances as low-permeabllity oll reservoirs when pres sute around the wellbore drops below the bubblepoine pressure, in low permeability gas condensate reservoirs when pressure is less ‘han the dewpoint pressure (Chu and Shank 1993), in carbonate reservoirs after acidification, and in oll reservoirs surrounded by an aquifer, Near-Welibore and Outer-Boundary Effects. To be complete, a wel test interpretation model must include the applicable near- wellbore and reservoir outer-boundary effets in aition to the basic reservoir behaviors. As with basle behaviors, the number of possibilities fs limited. They are listed in Fig. 5. “The near-wellbore conditions include wellbore storage (Van Everdingen and Hurst 1049; Ramey 1970), skin effect (Van Ever: ingen 1953; Hurst 1953), a single (usually hydraule) facture (Rus sell and Truitt 1964; Gringarten etal. 1975; Cinco-Ley et al. 1978; ‘Agarwal et al. 1979), partial penetration or limited entry (Brons and Marting 1961), ad a horizontal wel (Reis and Giger 1982). (Outer boundaries can be ofthe types: prescribed rate (eg, no flow 2s inthe case ofa sealing fault), prescribed pressure (eg for Instance, constant pressure, as inthe ease of gas cap or en active aquifer) or leaky (Le. semipermeable), as in the case of @ nom sealing feult, No-flow and constan-pressure boundaries can also be created in a developed reseevoir by near-by production oF in Jection wells, respectively. Because of the low resolution of the ‘well test signals currently available, itis difficult in some cases to ‘obtain much detail on the shape ofthe boundaries from well test analysis. For instance, itis dificult to distinguish a circular res ervoir from a square reservoir with the same area when the well is atthe center. Boundaries that can be diagnosed in the harizontal direction with current well test analysis techniques ae single linear faults, intersecting fauks (wedges), parallel faults (channels), oper rectangles (Le, three boundarles intersecting at right angles), ret- angular reservirs, or circular reservoirs. In eacs case, distinction can be made with reasonable confidence between constant pres sure and no flow. Leaky conditions ean aso be identified ifthe cost Is long enough (Yaxley 1987). Nonrectangular boundaries and meanders in fluvial channels can also be seen in well test data (Zambrano et al. 2000; Mijiryawa and Gringarten 2008) In addition, the boundary type in the vertical direction can be Identified if the well is partially penetrating or horizontal. This Includes a constan-pressure upper-boundary effect caused by a {5 cap oF a constant lower-pressure boundary effect resulting from an active bottamhole waterdsive, The Complete Interpretation Model. The complete intrprota- tion model is made of the combination of the individual compo- rents described above. Although the number of interpretation ‘model components are limited (five near-wellbore effects, wo basic reservoir behaviors, and three types of outerboundary ef- fects) their combination can yield several thousand different in- terpretation models to match all observed well behaviors. ‘The challenge of the well est interproter is to diagnose from the ‘observed well behavior which ofthe components described above shouldbe included in the interpretation model. This is achieved by Identifying the flow regimes assoclated with these components. ‘The identification process relies on the fact that these various flow regimes (linea, bilinear, spherical, radial, etc) yield different transient pressure behaviors during a test and occur at different times. A schematic ofthe complete interpretation process is shown in Fig 6. Evolution of Well Test Analysis Methods The extent to which the identification process of Fig. 6 can be performed effectively is a direct function of the anslysis tech: niques being used ane particularly of their ability to diagnose and verify an interpretation model efficiently (Gringarten 1987). This 's summarized in Fig 7 Tn terms of diagnosis and verification, the derivative method is ‘much beter than the log-log pressure type-curve matching method. Both are significantly better than the straight line techniques, es pecially f they are performed with software that can generate the ‘model directly rather than relying on matching with published type NEAR-WELLBORE RESERVOIR BOUNDARY EFFECTS BEHAVIOR EFFECTS Wellbore Homogeneous Specified storage rates a Heterogeneous Fractures Specified +2-Porosity pressure oe + 2Permeabilty penetration *Composite Horizontal Leaky well boundary EARLY TIMES MIDDLE TIMES LATE TIMES Fig. 5—Components of the woll test interpretation model February 2008 SPE Reservoir Evaluation & Engineeringpenrirication |__| VERIFICATION | ] |) woourmmes re tse | | Homogeneous spocind rate | Hntogencevs Spectiod pressure | | Co [EeSeethy Leaky boundary || “ CONSISTENT | 7 WELLTEST RESERVOR | erreors | hoes NEAR WELLBORE Eres. WELL TEST INTERPRETATION MODEL Fig. 6 Well test interpretation-madel identification process, curves, Specifically, the straight-line techniques, hough simple corresponding fo the low regime being analyzed). This Is why. to use, are poor at selecting the very straight Hines on which they when powerful personal computers became availabe, the deriva are to be applied. And once a straight line has been selected, there tive approach superseded log-log pressure analysis, which before {is no rule to indicate if i is indeed the right one, (Le, the one had superseded straight-line techniques ANALYSIS METHOD | IDENTIFICATION | VERIFICATION Pressure type curves Fair (limited) Fair to good Deconvolution Much better | Same as derivative ‘7-Ranking of well test interpretation methods. February 2008 SPE Reservolr Evaluation & Engleering adensification has also greatly improved recently with the de ‘velopment of a stable algorithm for deconvolution (von Schroeter al, 2001), By converting pressure at varlable rate into pressure at constant fate, deconvolution transforms atest ito single drew down with ration equal that ofthe est thus increasing the Amount of data tat ean be analy with “conventional” analyses. ‘The gain clearly greater in long tes, such as with permanent downhole pressure gouges, in which the total test duration is one octwo orders of magnitude greater than the duration of the longest Tow period at constant rate, Deconvolution, however, aso use- ful inshore tess such a5 DSTs because i neeases ie rds of Investigation and enables differentiation beween te et behavior an artifacts of the derivative calculation Fig. algo provides e clear erection fr future development in wells analysis, Any farther improverent in interpretation tech nology can come ony from further significant improvements in the dencaton and validation steps. Any new retin hat does roe achieve these goals is uniikelyto havea lasting impact on well ‘es analysis technology (Blasingame et 1989; Onur and Rey ods 1988; Duong 1989) Straight-line Analyses. Straightline analysis techniques rely on the existence ofa straight line on plo of the pressure response vs. some function of the elapsed tine when a particular low regime dominates (Fig 8). The straight-line slope and intercept provide the well and reservoir parameters that control this flow ragime. To ‘deretythe complet interpretation model, straight fine analyses must bbe applied to all the flow regimes presen in the pressure behavior. Straight-line analyses include “specialized” analysis methods (Gringarten et al. 1979; Gringarten 1985a) based on the signal defined by Eq. 4 and superposition analyses (Oden and Jones 1965) based on the signal defined by Eq, 3. In specialized pls, the Wellbore storage High K fracture change in pressure during a glven flow pesod, Ap from Ea, is plowed gaint lw repime-speifc function ofthe elepsed te {AQ0, on a Cartesian graph. Ad) comes from the equations de. Seribing te various low regimes. Kis eqal to: for wellbore storage (Ramey 1970) and pseudosteady-state flow in closed res- frvoes (Jones 1956), ¥Ar for high-conductivty fracture, (Clark 1968) and channel Hlnear flows (Miller 1962; Milfbein and CCichowicr 1968), 437 forlow- conductivity fracture and bilineat flow (Cinco-Ley and Samaniego 1981), 1/V¢ for spherical flow (Moran and Fines 1962), and log(A0 for radial flow in reservoirs ofinfnte extent (ills ea. 1880) of bounded by a sealing fault (Gomer 1991) oF by two no-flow intersecting faults (van Pollen 1965; Prasad 1975) Homer and superposition analyses, onthe other hand, quite eAAd tobe plated agaist a low-reglme-speciic superposition tine (also called generalized Horner ne} Pillar a.)MGes 49) Saysar) 100. +6) con a Cartesian plot. 4d isthe same as for specialized analyse. Horner and superposition pls cannot be used if d= (ue. for wellbore storage and psudosteady stat flan). The permet thickness products ebrained from the aca fow regime aig line slope (ilee etal, 1950; Homer 1951), whereas the skin fect i obtained from the intercept. The shapes of the dts also provide information onthe skin: Pressure data each the stalght line from below in damaged wells and from above in stimulated wells (Miler el. 1960). The mai advanage of the staighline rmathods is their ease of implementation, because they were de- Signed trough simplifying assumptions tobe performed with only 4 plece of graph paper, a pene ruler, and simple ealeulaons Low K fracture Spherical flow Early times 4 Early times "| [Eany times 4 ev feat)=(at'2 fat=(aty' Aeaty=(aty12 g Radial flow Skin Double porosity Composite “A [Middle times Pere Middle times Early or i . late times 5 = a once te 3 soe Mat)= log At] flat)= log At One sealing fault Channel boundaries _Intersecting faults Closed reservoir Late times Late times Late times Late times, J] a POO 7 Drawdown only f(AN)= log At. Aat=(aty’? f(At= log At| flAt= At RAL) oF E' (G.-GaVGn-1-An)] AB! AG + At) - flat) Fig, Straightline analyses. 48 Fett 2008 SPE Reservoir Evaluation & EngneringSpecialized plots are the easiest 0 use, followed by Homer plats. Superposition was usually considered too cumbersome co be dane bby band until off-the-shelf well test analysis software became available on personal computers in the mid-1980s. Until then, straightline methods were routinely applied only tothe analysis of the radial flow regime in buildups {the corresponding MDH (Miller etal, 1950) and Horner (1951) analyses were the main emphasis of SPE Monograph | (Matthews and Russell 1967). All ‘ow periods before the bulldup being analyzed in a multirate test hhad (0 be approximated by a single drawdown with a duration equal to: 4V,/9. a Jn which 1 the “equivalent” Homer production time, V, the ccumolative production since the lst pressure equalization, and q the last rate before the buildup (such an approximation introduces significant errors in the analysis (Horner 1951), as discussed later inthis paper]. Eq, 6 then reduces to the radial-flow Homer time for the case ofa single drawdown of duration followed by a bull: eae a ® ‘The main limitation of straight-line techniques Is thelr inability t| Identify with confidence the proper straight line to be used in an analysis, as indicated in Fig. 7 An apparent straight line through ‘sei of data does not prove the existence ofa specific flow regime, and If the selected straight line is not a real straight line oF is a straight line corresponding to a different flow regime from chat expected, an analysis on the basis ofthat straight line would yield erroneous results. Consequently, straight lines cannot be used with confidence to identify an interpretation model. The knowledge of the applicable interpretation model is actually required to identify the straight lines usable for analysis An addtional problem, which affects specialized plots only. Is imustated in Fig. 8 I shows a radal-low specialized plot [MDH (Miler tal. 1950)| for a buildup following an inital constant rate drawdown of duration ¢,, Although radial flow in this example ‘Specialized Plot starts at 45 hours and lasts through che end ofthe bulldup at 2 hours, the corresponding buildup points are on the radial flow straight line if , =720 hours only. For smaller values of ¢, buildup data frst follow the radial flow semilog straight line, then fall Delow it, The time during which the semilog straight line exists ‘through the pressute points (the “Iength” of the straight line) Is, Clearly 2 function of the production time. The reason is that spe- cialized analyses strictly apply only to the initial drawdown in a stabilized reservoir (Gringorten etal. 1979). They also can be used ina subsequent flow period as long as the elapsed time inthe Flow period being analyzed is small compared with the duration of the previous flow period. If ths 's no longer the case, data points eviate from the straight line even though the flow regime of Interest still dominates. The risk for an interpreter is that the later part of the data set can be (and often is) mistaken for the MDH straight line (Ramey and Cobb 1971), thus ylelding erroneous analysis results. This problem does not exist with Homer and superposition plots, because the only condition forthe existence ofa straight line for a given flow rogime is that data exist within the range of valiity of the corresponding flow regime. As shown in Fig. 9, there is no restriction on the magnitude ofthe production time 4 Because ofthe production time dependency, specialized plots are mainly used for the analysis of near-weilbore effects, whereas Horner and superposition analyses ae used for reservoir behavior and boundary effects Log:Log Pressure Analysis. Type-curve o log-log analysis meth- fods were inirodaced in the petroleum erature by Ramey (1970) Jn an attempt to overcome the limitations of staight-ine-based analysis methods (Matthews and Russell 1967; Earlougher 1977) ‘The intial objective was to identify the correct infinite-acting ra dbal-Flow straight line on an MDH (Miller etal. 1950) or a Horner (1951) semilog plot and to pecmit analysis of test data when such a radial-flow siraight line had not yet been produced (Ramey 1970). Log-log analysis was subsequently expanded into a process for identifying the various components of the inerpretation model (Gringarten et a. 1979; Bourdet and Gringarten 1980). Horner Plot 3900, Pressure Change, Ap ( psi) 3800) 700) Pressure , p(dt) ( psia) 8 a TO Elapsed time, At ( hours) Fig, February 2008 SPE Reservoir Evaluation & Englneering 3300 10 70 10" (yay sat 1—Spacialized vs, Horner plots ”Although the type-curve method had been introduced as supplementary 0 staightline techniques (Ramey 1970), there was rch arguing in the well testing literature from the early 1970s t0 the mid 198ts abou the relative merits ofthe two approaches. A rumber of interpreters were confused by the lack of clear meth- Odology on how to select the “right” «ype curve among the many that were published during that time (Agarwal et al. 1970: Me- Kinley 1971; Earlougher and Kersh 1974; Gringarten etal. 1975 Cinco-Ley and Samaniego 1978) and by the fact that diferent type curves published by different authors (Agarwal etal. 1870; Me Kinley 1971 forthe same wellbore storage case often gave cfferent resus wien applied tothe same data (Ramey 1980). The conso- \versy ever led to an early SPE board decision (Ramey 1992) not to include full-scale type curves in the Earloughee SPE Monograph 5 (Eerlougher 1977), and it was recommended in Monograph 3 that type-curve analysis be only used in an emergency or as @ checking device ater “conventional” (1.2, straight fine) methods had failed. After the systematic approach othe analysis of well tests wes established (Gringarten etal. 1979; Gringeten 1982, 1985a, 1986). the differences among published wellbore storage ‘ype curves (Agarwal etal, 1970: MeKintey 1971: Earlougher and Kersh 1974) were explained (Gringarten eta, 1979), an industry- standard type curve emerged for wellbore storage and skin (Grin garten etal 1979), and the early SPE board decision was reversed For the purpose of log-log analysis, the change in pressure ducing a given flow period inthe test, Ap from Eq, 4, is plated vs the elapsed time, on a log-log graph. Such a graph scales Ap and iin exactly the same way for both interpretation model and field data and isthe only graph t do so. It permits model dent fication by emphasizing characteristic shapes for different flow regimes (Fig, 10). For this reason, a log,log plot is called a diag nostic plot (Gringaten etal. 1979), Because the constitutive flow regimes ae also associated with specialized and superposition pls, Tog: log dlagnostie plots and specialized or superposition plots ean be used together to identify and verify the various flow regimes that dominate during atest (Ramey 1870; Gringarten etal. 1979) ‘Although quite powerful compared with straight-line methods, identification fiom fog-og pressure analysis has its limitations In particular, the lack of resolution in pressure change makes It dif- Flcut to diagnose flow regimes that oceur at late times, Even ‘lyme anal milecime low regimes canna be identified eas- ily i they do not yield a log-log straight ine. This is iustrated in Fig. 10. Fig. 10 shows the 1oglog shapes of the various flow regimes that can be identified by log-log analysis in the case of the first drawdown in a sablized reservoir. Although mostly theoret- ical, this case yields the true log-log characteristics of the flow rugimes, whereas subsequent flow periods are affected by the rate history (Raghavan 1980) in the same way specialized plots are (Gringarten et al. 1979). ‘Wellbore storage yields a straight line of unit slope (Le. one log eyele Ap, for one log cycle 40) (Ramey 1970) at early dimes, bbeeause Ap is proportional to Az (Van Everdingen and Hirst 1949), A high-conductivityfrecture communicating with the well bore exhibits an early-time log:log straight line of half-unit slope (one log eyele Ap for two log cycles 44, because Ap is propor tional tothe squaze root of At during 1D flow from the matrix into the fracture (Clark 1968). A low-conductvity fracture yields @ ‘quarter-unit slope (ane log eycle Sp forfour log cycles A) (Cinco. Ley and Samaniego 1981), which corresponds to bilinear flow in the fracture, On the other hand, other possible near-wellbore ef fects cannot be identified because ofthe lack of resolution in the pressure change. Patil penetration with positive mechanical skin, for Instance, is undistingulshable from a damaged, fully penetat ng well (Kuchuk and Kirwan 1987). ‘Radial flow is also difficult diagnose because tt does not ‘yield a staight line. Ir instead exhibits 2 nondescript log-log shape, Which corresponds to the linear relationship between Ap and Tog(Ad (Van Everdingen and Hurst 1949) characteristic of that flow regime, Heterogeneous behavior yields an S-shaped curve, ‘which corresponds to two distinct homogeneous behaviors sepa rated by a transition petiod, a characteristic of heterogeneous ss tems, In practice, only double-porasity behavior (Bourdee and Gringarten 1980) can be identified Tn general, boundary offects are dificult to identify except for ‘constant pressure boundetles and closed systems from drawdown data, which respectively show a stabilization or become asymp- tot to unit slope log-log straight line at late times [pis a linear function of Ar (Jones 1956)) “The main limitation of pressure type-curve analysis comes from its use as a manual process before well test analysis software bhecame available. Once the interpretation model had been identi- fled, the data were matched witha dimensionless type curve rep resenting the model behavier, following the matching procedure described in SPE Monograph 3 (Eariougher 1977). Logg analy: sls then yllds all dhe model parameters, the values of which could ‘hen be compared with those obtained feom individual straight-line Wellore storage ——‘High-conductvity fracture Low-conductivty fracture Early mes Early tines Early times 3} Untstope attunt slope ‘utr an 3 | seagntine ‘ight ine Saige tne 3 () Closed (2) Constant 2 Homogeneous Double poreaty «(eed @) Constant 3 3 Rangionee : lunitsiope (1) 5 sesh : S-enaped cure = pe ' ‘Staized Fa pressure iisdte times Late tines Log of Elapsed time, at (hours) Fig. 10-Flow regime logrtog pressure shapes. 80 February 2008 SPE Reservoir Evaluation & Engineering oOanalyses Thre were, however, only alimited number of pblisked {ype curves, covering a limited numberof combinations of near wellbore effects reservoir behaviors, and outer boundaries. In ad- Aion, most publsed type curves. forthe sake of simplicity, were valid only forthe fist drawdown alter fll stabilization of the Feservoir pressure. Finally, experence showed that contrary (0 carly expectations (Ramey 1980), pressure ype-cuve matching Usually was non unique fora given mode f radial flow had not been not reached daring the flow period of interest (Ramey 1952). Log: Log Derivative Analysis, Pressure-dervative functions have bbeen mentioned at various times in the petroleum literature (Van Everdingen and Hurst 1949; Jones 1956; Carter and Tracy 1960; Ramey 1965; van Pollen 1985; Agarwal etal. 1965; Gringaren and Ramey 1971; Lescaboura'et al. 1975), in connection with Water Influx (Van Everdingen and Hurst 1949; Carter and Tracy 1960; Agarwal et al. 1963) interference testing (Lescaboura ea. 1975), reservoir boundaries (Jones 1956; van Pollen 1965), and wellbore storage caleulations (Van Everdingen and Hurst 1949) Ramey 1965). Applications to well test analysis first appeared in the late 1970s: A log-log plot of dA piatd vs. Acwas suggested es an alternative © strlght-line analyses for nterfesence tests (Tia ‘and Kumar 1980a), tess in fractured wells (Tiab and Puthigal 1988), and tess in eservoirs bounded by two parallel faults (Tiab ‘and Kumar 19806) and by multiple faults (Tiab and Crichlow 1979), The advantage of using a derivative on the basis of the natural log of elapsed time, o(Ap)/atlogd), which emphasizes radial flow, was also demonstrated forthe description of hetero ‘geneous reservoirs (Perez-Rosales 1978). The practicality and [power of the derivative approach for well test interpretations, how ‘ever, was recognized only after the 1983 publications by Bourdet etal, (1983a, 1983) of derivative type curves expressed in terms of independent variables for both homogeneous (Cringarten etal 1979) and double-porosity interpretation models (Bourdet and Gringarcen 1980). Taking the derivative with respect tothe natural lg of A emphasizes radial flow, whieh isthe mast common flow regime around a well and yields a stabikization while radial flow dominates. The derivative could be taken with espect to differ ent flow regime to yield a stabilization when that flow regime dominates. For instance, the derivative with respect to A¢ yields @ stabilization during wellbore storage at eafly times and during pseudasteady-state flaw at late times. ‘The major advantage of pressure derivative is that it has greater diagss and verification capabies thn the change in pressure iself with the accuracy of suaight-line meds. Detvaive shapes for various Mow reghnes at eary. mile, an ate times in 2 test are dspayed in Fig. 11 for 414). When wellbore storage dominates, the pressure derivative is proportional tothe elapsed time and is identical to the change in pressure, Consequently, when Apand alg} /inAd) are ploted on the same log-log graph they share the same unk slope log-log staght line at earl times. Dam- aged ells exhuit a axiom a easy tine, fllwtng the bore storage unit slope sraight line the higher thes, the higher the maximum). Nondamaged or stimulated wells, on the other hand, show a small maximum ot ro masinum a al. Incase of & igh conductivity factred well, the ealy-time-dervatve te spores proportional othe square oot of tine. Ona log tog plo, the derivative response follows a hal-unk slope straight line (Al goa and Ayoub 1985), The amphtude ofthe derivative response is fal that ofthe pressure change: When both pressure and derivative curves are ploted onthe same log-log graph, the to early time Straight Ines are parallel an are vertically csplaed by a factor of two. Fora low-conductivityfrcture, daring lies flow at erly tines, the deriva response is properonal othe fourth root of time and exhibits @ stalght line of one-quarter unt slope on a Jogsog plot (Wong et al. 1986). The amplitude of the derivative response is one-fourth tat ofthe pressure change. During partial pertetion or limited entry spherical flow behavior, the derivative Fesponse is proportional tothe inverse of the square cot of time {Moran and Finkle 1962; Cutham 1974: Raghavan ae Clark 1975; Kohiaas etl 1982). On a lgog plot, this ved straight line swith a megane half-ait h Radial flow yields stbilization (Pecez-Rosales 1978; Bour det al 19839, which s inversely proportional to the dominant mobility Aju: the higher dhe stabzation level, the lower te mobility. A change in nobly resuldng fom heterogeneous be- havior Is characterized by two stabilizations on the decivative. A second stabilization ata higher level than the fist one indicates 4 decrease in mobility, wheres a sbilzaton ata lower level fenotesa mobility increase (Tis and Crichlow 1979). A change of storativty, on the other hand, yells « maximum ot min- mum between the Ina and ‘inal stabiizadons. A. maim {5 obtained when storatvitydeereases—a minimum, when stort- ivy increases Wellbore storage High K fracture _Low K fracture _ Spherical flow ary times Early times [Mice times a | Negative 3 Unt ape Hattunt sore |} ovanerantsose | [Harun 3 sage ine tragtee | [| Suatane ope sight ne ‘S__Raclal flow ___ Mobility change Storativty change One sealing fault [ate tines Midale times | [Middle times Late times g wen a Stanton 2 zg o o 5 Sieiaton = 3 ao) oa D channel boundaries Intorsecting faults Closed reservoir Constant pressure S [Late times Late times. Late times Late times: siantzgiss | | coy 2 ae i ep Hatantsepe | | Fatant soe stag ine | engine” | [statin Z Log of Elapsed time, At (hours) Fig. 11—Flow-regime log: February 2008 SPE Reservoir Evaluation & Engineering Hog derivative shapes. 5‘The derivative fora sealing fault yields a lte-time stabilization ata level equal to ewice that for infince acting radial flow (Clark dnd Van GolfRacht 1985). A channel configuration produces @ Tatetime balf-ant slope straight line. Such a straight line appears Immediately aftr the homogeneous infinte-acting radial flow sta- bilizaion if the well is equidistant from the two parallel bound: aries, Ifthe well Is closer to one ofthe boundaries, its preceded by a second stabilization at ice the level ofthe first one. When ‘vo faults intersect, the derivative shows a latelme stabilization at a level equal ta 2 (van Pollen 1963; Prasad 1975) times the radlal-low stabilization level, in which @ Is the wedge angle in radians, This final stabilization is preceded by a half-unit slope Jog-log straight lie. During pseudosteady’state behavior in a closed reservotr, the drawelown pressure derivative exhibit a late time log-log straight line of slope nity (Clark and Van Golf-Rache 1988). This line is reached faster by the derivative than by the pressure (Fig 10) because the slope ofthe derivative is identically Uuity, while the slope of the pressure drop is only approximately tunity, In the case of a constant pressure boundary, on the other hhand, the derivative tends to zero (Clark and Van Golf-Racht 1983) while Ap stabilizes, The rte of decline of the decivaive ccurve depends on the shape of the boundary and is fester for @ circular constantpressure boundary than for a linear constant- pressure boundary. Once an Interpretation model has boon identified, well and reservoir parameters are obtained by matching the pressure deriva tive for that Interpretation model with the derivative of the field data, As with pressure data, the match can be performed numeri cally or manually using a derivative type curve forthe applicable interpretation model. The change in pressure must be matched at the same time to caleulate the skin effect because the derivative Is rot very sensitive to that parameter. For some flow regimes, po ameters can be obtained directly from the derivative for these flow regime, without matching with a complete model. For in- stance, the permeability thickness product can be calculated di- fectly fom the radial flow stabilization line, and the wellbore Storage can be obtained from the intersect of the radlal flaw sta- bilization and the unitslope wellbore storage lines (Gringarten 19856), The same pracedure can be applied to other flow regimes (Tiab 1989, 19932, 19936; Tiab ec al. 1999, “The main drawack of derivatives Is that, contrary to pressure data, they are not measured but must be calculated. The useful ness therefore depends on hov well they are computed. Tho var ‘us derivative shapes shown in Fig. 11 assume thot the data are from an inkl, constant-ate drawdown in anew eesevoir with 10 ‘ior production sty. In prac, this sneer te case, ed the Gervaive must be taken wth respect the superposition die of Eq, 6 with /¢)-log/A) to avoid te inuence ofthe production time on the length of the radial flaw stabilization (Bourdet eal 1062s; Bourdet tl, 196) (oulzate derivative). This transforms the derivative of pressure data from a subsequent flow period into fan equtvaentfestarawdown drivadve except when the end of the previous flaw petiod is not in radial flow. Then dhe malate derivative may differ from the drawdown derivative (Clark and Van Golf Racht 1985) (Fig. 12) depending on the previous rate history [the multirate derivative follows canstion from the draw dowirtistdevatve to the dewdown second derivative (Cinco Ley et al. 1985; Cinco-Ley and Samaniego 1989) The interpreter must be careful not misinterpret ths deviation fre flow cele Iicavior(Gringarten 2005) ‘The mula derivative also difers from the frst drawwdoen desivatve in buldups in closed reservoirs under pseudosteady State flow, Because of depletion, the pressive tends to stabilize to the average reservoir pressure, and buildup derivatives tend to ero, whereas derivatives in drawdowns Yel a unit-slope log-log Straight ine TE must be stressed that that che multe derivative, although taken with respect othe superposition tie, must be ploted 2s 3 function of the elapied tims, Somo well test analysis software routinely pots the multirate decivatve vs. an equivalent time, de Med as (Agarwal 1980) at pear ‘or its malate equivalent. The equivalent ime wes introduced by ‘Agarwal (880) t conver buldup data into equivalent drawdown ‘ata so that they could be matched with published drawéown type ‘rves, To work, the equlvalert tine rete adil flow to have been reached before the baldup being analyzeé. When applied to derivatives the eqtvalent ime creates distortion that makes iden- tification of flow regimes more dificult and can be misinterpreted for sesevol behaviors (Fig. 13). The fat drawdown derivative andthe multe desvaive are proportional to the slope of the MDH nd superposition plo. a o 1 aieup 7 uke Drawdown! x omete oranown | ooeane | ultup je Drawdown derivative [duration Drawdown duration +f {a Wellbore Storage and Skin (©) Double Porosity Drawdown | derivative +h i Log of Pressure Change and Derivative Drawdown, derivative Bulldup 1 dorivative, Drawdown duration i (a) Composite At (hours) Fig. 12—Drandown vs. buildup log-log derivative shapes. 8 Febcuary 2008 SPE Reservoir Evaluation & Engineering10 i oo jane nue g Fol ¢ i & Fro & | 1 fore toe opeed ine — and eualent — finer) Fig. 13—Distortion of log-log derivative shapes because of ‘equivalent ime as a function of the production time (example of ‘channel boundaries). respectively, The slope must be obtained numerically, by use ofan algorithm that must be able to remove as much of the noise as possible without altering the signal. This operation must be cartled ‘ut with care because the shape of the resulting curve depends ‘apon the method used to dlfferentiate the data (Fig. 14). ‘A numberof other factors can affect the shape of the derivative curve and therefore mislead the interpreter. Some can be easly ‘dentifed: sampling frequency ofthe data acquisition, gauge reso. Jution, time or pressure errors a the start of the period, erratic raw data points, of multiphase flow. Others are more difficult 10 see land may affect the analysis. These include end effects (ithe last pressure in a flow period is too high or too low, the derivative Shows an upward or downward trend, which must not be confused ‘with a boundary effect), phase redistribution in the wellbore, and 4 pressure trend in the reservoir (ig, 18). TBut by far the most impact comes from the rate history, Inad= equate description ofthe flow rate history is common in well test analysis. For instance, some flow-rate data may be missing, espe- cially during dailling, stimulation, and the cleanup period. Fluid ray have been injected into the well and not accounted for, oF fates may be allocated and not measured. In addition, the history may have been truncated or simplified. Oversimplifying the flow-rate history can jeopardize the reliability of the pressure ‘derivative a8 a diagnostic too (his holds true also for the Horner land superposition graphs), For instance, truncating the preduction history by Keeping only the latest rates before the period being analyzed yields erroneous buildup or multirate decivatives with ‘upper tends above the correct stabilization line (Fig 16). These ‘ould be mistaken for a decrease in mobility or storativity oF a ‘o-flaw boundary. On the other hand, replacing all preceding flow pericds with a single drawdown With a rate equal tothe last rate before the period of interest and a duration equal tof, ftom Eq, 7 ‘produces a hump on the log-log multirate derivative response (Gop of Fig. 17). This behavior could be mistaken for a com: posite behavior. ‘As rule, the more recent the changes In production rates, the mote detailed the rate history mast be, Describing accurately the rate history during 2 peciod corresponding tothe lst 40% of the ‘cumulative production ofthe wel, and using Eq, 7 to calculate a te forthe frst 60%, provides a correct derivative (Daungkaow eal. 2000) (botom of Fig. 17) Deconvolution Deconvolution has received much attention recently (von Schro- fer etal, 2001, 2004; Gringarten etal 2003; Levitan 2005; Grin garten 2005; Tk etal. 2005; Levitan et al, 2006), folowing the publication of a stable deconvolution algorithm (von Schroeter fetal 2001). As suggested by Fig. 1 itis not a new interpretation ‘method, but a new tool to process pressure and rate data to obtain more pressure deta to Interpret. Deconvolution transforms vari able-rte pressure data into a constant rate initial drawdown with a uration equal tothe tral duration of the test and yields directly the corresponding pressure derivative, normalized to a unit rate ‘This derivative Is therefore fee from the distortions caused by the pressure-derivative calculation algorithm shown in Fig. 12 and rom errors introduced by incomplete or truncated rate histories, ‘Some ofthe benefits of deconvolution ae illustrated in Figs. 18 ‘through 20. Fig, 18 shows pressure and rate datas. ume for a [North Sea well. Downhole pressure is available only for the initial DST and a production test two years later. Surface rates are aval slope =m e = a ‘Superposition time (a) Multipoint regression Superposition time (b) Moving window 2L = 20% of data span tog 4p : cS af ae “ Tog at Tog at Fig. 14—impact of differentiation algorithm (Bourdet et al. 1888) on log-log derivative shapes. February 2008 SPE Reservoir Evaluation & Engineering