This package provides inverse reservoir-surveillance tools that regress reserves and drive mechanism from a measured pressure-versus-production history. They complement the forward tank model (neqsim.process.equipment.reservoir.SimpleReservoir) and the standard PVT laboratory simulations.

Location: neqsim.pvtsimulation.reservoirproperties.materialbalance and neqsim.pvtsimulation.util.DeclineCurveAnalysis.

Table of Contents

Overview

Class Purpose Package
GasMaterialBalance Gas P/Z straight line (OGIP), Cole plot aquifer diagnostic, Havlena-Odeh gas balance pvtsimulation.reservoirproperties.materialbalance
OilMaterialBalance Havlena-Odeh oil balance (OOIP, gas-cap ratio m, water drive, Pirson drive indices) pvtsimulation.reservoirproperties.materialbalance
GeneralizedMaterialBalance Generalized oil/gas balance with both Rs and Rv, measured net withdrawals, known aquifer/rock support and identifiability checks pvtsimulation.reservoirproperties.materialbalance
CalibratedPvtModel, MaterialBalancePvtHistory Immutable original/tuned model handoff, held-out laboratory residuals and explicitly attributed pressure-aligned PVT pvtsimulation.util
CompositionalReservoirHistory Frozen-EOS component production/injection replay, bounded pore-volume matching, optional segregated aquifer and separate held-out pressure residuals pvtsimulation.util
FetkovichAquifer, ReservoirHistoryEnsemble Physical finite-storage support, implicit EOS coupling and discrete joint Bayesian pressure-history hypotheses pvtsimulation.reservoirproperties.materialbalance, pvtsimulation.util
DepletionPvtHistoryGenerator Dry native closed-mass CCE and depletion handoffs; bounded two-phase CVD/DLE generalized PVT with separator/component-closure evidence pvtsimulation.util
VanEverdingenHurstAquifer Radial aquifer influence functions, Carter-Tracy cumulative water influx, ECLIPSE AQUTAB export pvtsimulation.reservoirproperties.materialbalance
DeclineCurveAnalysis Arps + Duong (2011) decline: forward rate/EUR and least-squares history matching pvtsimulation.util

A typical surveillance workflow is:

  1. Diagnose the drive mechanism (Cole plot / drive indices).
  2. If water drive is present, build the cumulative influx We with Carter-Tracy.
  3. Regress OGIP/OOIP with the matching material-balance method.
  4. Cross-check the reserves against a decline-curve EUR.

Gas material balance

For a volumetric dry-gas reservoir with no water influx the material balance reduces to a straight line (Craft & Hawkins; Havlena & Odeh, 1963):

\[\frac{p}{Z} = \frac{p_i}{Z_i}\left(1 - \frac{G_p}{G}\right)\]

A linear regression of $p/Z$ against $G_p$ gives the intercept $p_i/Z_i$ and the x-intercept $G$ (the OGIP) where $p/Z = 0$.

import neqsim.pvtsimulation.reservoirproperties.materialbalance.GasMaterialBalance;

double[] pressure = {350.0, 315.0, 280.0, 245.0, 210.0}; // bara
double[] z        = {0.95, 0.94, 0.93, 0.93, 0.92};
double[] gp       = {0.0, 10.0, 20.0, 30.0, 40.0};        // any consistent surface volume

GasMaterialBalance.Result r = GasMaterialBalance.fitVolumetric(pressure, z, gp);
double ogip = r.getOgip();          // same unit as gp
double piZi = r.getPiOverZi();      // bara
double r2   = r.getRSquared();

If Z-factors are not supplied, they can be computed internally with the Sutton pseudo-critical correlation and the Hall-Yarborough equation of state:

GasMaterialBalance.Result r =
    GasMaterialBalance.fitVolumetric(pressure, gp, 350.0, 0.7); // tempK, gasGravity

The Cole plot of the withdrawal-to-expansion ratio $F/E_g$ against cumulative production diagnoses drive mechanism — a horizontal trend indicates volumetric depletion, an upward trend indicates water influx:

double[][] cole = GasMaterialBalance.colePlot(pressure, z, gp, 350.0);
double[] gpAxis  = cole[0];  // cumulative gas (from the second history point)
double[] fOverEg = cole[1];  // withdrawal / expansion

With a supplied cumulative water influx We, the Havlena-Odeh balance $F = G\,E_g + W_e$ is regressed for the OGIP:

GasMaterialBalance.Result rHO =
    GasMaterialBalance.fitHavlenaOdeh(pressure, z, gp, we, 350.0);

Oil material balance

The general Havlena-Odeh material balance (1963, 1964) is

\[F = N\,(E_o + m\,E_g + E_{fw}) + W_e B_w\]

with reservoir withdrawal, oil/dissolved-gas expansion, gas-cap expansion and connate-water/rock expansion terms

\(F = N_p\,[B_o + (R_p - R_s)B_g] + W_p B_w\) \(E_o = (B_o - B_{oi}) + (R_{si} - R_s)B_g, \qquad E_g = B_{oi}\left(\frac{B_g}{B_{gi}} - 1\right)\)

The static helpers build these terms from black-oil PVT, then the fit* methods regress the OOIP N and gas-cap ratio m:

import neqsim.pvtsimulation.reservoirproperties.materialbalance.OilMaterialBalance;

// F, Eo, Eg arrays built with withdrawalF(...), eo(...), eg(...), efw(...)
OilMaterialBalance.Result dep = OilMaterialBalance.fitDepletionDrive(f, eo);       // OOIP (no gas cap / aquifer)
OilMaterialBalance.Result gc  = OilMaterialBalance.fitGasCapDrive(f, eo, eg);      // OOIP + gas-cap ratio m
OilMaterialBalance.Result wd  = OilMaterialBalance.fitWaterDrive(f, eo, we, bw);   // OOIP with known We

double ooip = gc.getOoip();
double m    = gc.getM();

The fractional Pirson drive indices — depletion (DDI), segregation / gas-cap (SDI), water (WDI) and connate-water/rock expansion (EDI) — sum to unity and quantify each mechanism’s contribution:

double[] di = OilMaterialBalance.driveIndices(n, m, eoTerm, egTerm, efwTerm, we, bw, fTerm);
// di = {DDI, SDI, WDI, EDI}

Generalized balance, measured history and tuned PVT

GeneralizedMaterialBalance adds Walsh’s generalized material balance (1995, DOI 10.2118/95-01-07). It retains oil vaporized in gas (Rv) as well as solution gas (Rs). Dropping Rv can bias gas-condensate and volatile-oil estimates. This is a two-surface-component tank model, not a full compositional reservoir simulator or MBAL file reader.

Original fluid, laboratory calibration and historical observations

  1. Preserve the original composition, characterization, sample/laboratory revision, EOS, mixing rule, and separator/stock-tank basis.
  2. Follow the PVT calibration workflow: tune only supported, bounded parameters against the actual laboratory experiment protocols. PVTRegression supports CCE, CVD, DLE, separator and viscosity observations; saturation-pressure and swelling regression identifiers remain unsupported. Validate on withheld observations and keep untuned-versus-tuned residuals.
  3. Freeze the accepted tuned fluid before estimating reservoir volumes. Evaluate Bo, Bg, Rs, Rv and Bw along the applicable depletion/injection path. Black-oil conversion can generate pressure tables from EOS flashes; a pressure table is not automatically equivalent to a CVD/DLE path or a changing-composition injection history. Validate that approximation rather than silently substituting it for the tuned EOS.
  4. Construct GeneralizedMaterialBalance with an initial BlackOilPVTTable.Record. Call addObservation for each aligned average pressure/PVT/production observation. Pass cumulative produced minus injected oil, gas and water, all referenced to the original state. Oil and gas are total stock-tank/separator components, including oil recovered from produced gas and gas released from produced oil.
  5. Supply known aquifer influx in reservoir m3 and the dimensionless rock/connate-water expansion fraction. They are inputs, not fitted parameters. Use zero only when the assumed physical model justifies neglecting them.

For the existing simplified-reservoir example, reservoir.getSurveillance().generalizedMaterialBalance(pvtRecords, waterInflux, expansionFraction) builds the same calculation from recorded SimpleReservoir history. Arrays include the original state, which must have zero cumulative production/support; every PVT pressure must match its recorded pressure. The tank’s surface-production basis is 288.15 K, 1.01325 bara. Use that same basis for its PVT records. The adapter rejects injection streams because cumulative injection is not recorded by SimpleReservoir. Measured net injection/production histories can instead use addObservation directly. Measured field data do not need to be inserted into a simulated tank’s recorder.

Fit Initial-phase assumption Reported volumes
fitInitiallyOil() No initial free gas Total OOIP and dissolved-gas-inclusive OGIP
fitInitiallyGas() No initial free oil Total OGIP and vaporized-oil-inclusive OOIP
fitKnownGasCap(m) Known initial free-gas/free-oil reservoir-volume ratio Both total inventories
fitOilAndGas() Both initial free phases; independent expansion signals required Both total inventories and inferred gas-cap ratio

Immutable calibration and pressure-path handoff

For dry oil-only DLE or pre-dew gas-only CVD rows, use the native compositional handoff. For actual closed-mass laboratory CCE, use Path.CCE: the complete sample remains in every native state, with zero protocol withdrawal and the original/tuned EOS and separator identities preserved. These absolute states do not infer saturation or populate a measured reservoir production history. Keep laboratory calibration and whole-experiment holdouts separate from reservoir matching and holdouts. It retains actual native EOS states and separator products without absent-phase formation-factor extrapolation; it does not create a generalized black-oil table. Native CVD refines and validates a scan-selected upper dew boundary; the declared independent binary reference now passes. Legacy/default and modified-black-oil paths retain the scan precision limitation. Closed-mass wet CCE preserves actual gas/oil/aqueous cells and separator products; wet CVD and critical-degenerate qualification remain open. After a sampled gas/oil stage, dry gas-only CVD tails retain the fixed cell volume through equilibrium gas withdrawal; an unbracketed single-phase schedule does not start tail withdrawal; no absent oil PVT is extrapolated.

Before accepting laboratory tuples, use the opt-in gas-density interval consistency gate when same-state gas p/T/Z/molar-mass/density data are available. The CalibratedPvtModel overload accepts immutable same-sample evidence, rejects any incompatible row and freezes it in checkpoint/holdout provenance. The original constructor retains its behavior without making this consistency claim. Passing the identity check does not establish measured accuracy or independent provenance.

CalibratedPvtModel freezes serialized original and tuned fluids, sample and model revision references, the laboratory calibration report reference and the identities of complete experiments used for calibration. Fluid getters restore independent copies, including nested EOS parameters. Neither laboratory replay nor reservoir matching changes those checkpoints. Store the NeqSim/software revision in the model references; native checkpoints require compatible software to deserialize.

Use compareHoldout for a complete CCE, CVD or DLE pressure schedule. Select a normalized PVTFluidStudy property key, such as relativeVolume for CCE, liquidRelativeVolume for CVD, or boM3PerSm3 for DLE. Supply one measured value and positive measurement standard deviation per pressure, in the property’s units. Both models run the full experiment, preserving gas removal and the final DLE reference. The returned JSON retains original and tuned predictions, raw residuals and residuals divided by the supplied measurement standard deviations. Residuals are measured minus predicted. These are laboratory-property residuals, not reservoir-pressure likelihoods or parameter confidence intervals. Evidence JSON declares schemaVersion 1.0 and a schemaUri pointing to the versioned schema. Readers must reject unsupported major versions rather than interpreting them as v1.

compareSeparatorHoldout replays an entire multi-stage separator/stock-tank train, including reservoir conditions and every stage’s pressure and temperature. Available comparison properties are total GOR (totalGORSm3PerSm3), oil FVF (boM3PerSm3) and stock-tank oil density (stockTankOilDensityKgPerM3). The supported standard basis is 288.15 K and 1.01325 bara, also used by the existing solvers and recorder. The final stage must use that basis. Other standard conditions require a separately qualified conversion; relabeling the units is insufficient.

An experiment used in calibration cannot be labeled as held out. Identify the whole experiment, not individual CVD/DLE rows, to avoid sharing depletion history between training and validation. Identity checks cannot verify an external laboratory record or establish independence when the same experiment is given different names. The caller must review that provenance and set acceptance tolerances; a finite residual report is not automatic model acceptance.

MaterialBalancePvtHistory then retains the checkpoint bundle, isothermal reservoir temperature, ordered separator path, a path-validation evidence reference, and pressure-aligned BlackOilPVTTable.Record values including the original state. Declare CVD, DLE, COMPOSITIONAL_HISTORY or CONSTANT_COMPOSITION explicitly. Records must be evaluated from the accepted tuned model on that path, either externally or through the bounded DepletionPvtHistoryGenerator route below. The container checks numeric validity, sequence and basis; it does not generate generalized CVD PVT or independently verify caller-supplied records against the EOS.

Its materialBalance method accepts measured average pressures and cumulative net oil/gas/water withdrawals, reservoir-volume influx and dimensionless expansion support. All arrays must include the original zero-withdrawal/support state and match the PVT pressures. CVD/DLE paths decrease strictly; externally evaluated compositional paths can include repressurization. Constant-composition tables are rejected for depletion matching. The result is the existing GeneralizedMaterialBalance builder; laboratory calibration and EOS parameters remain separate from volume fitting. Substantial injection composition changes still require compositional replay.

The handoff tests compare full protocol replay with separately executed existing PVT solvers and recover an analytically specified synthetic gas inventory. These establish software consistency and analytical conservation, not independent laboratory/field accuracy. The bounded native-depletion generator, physical aquifer and discrete uncertainty extensions are described below. Independent laboratory/field qualification and broader compositional/phase-path coverage remain open under issue #4740.

Conservation equations and diagnostics

Let $O$ and $H$ denote total surface oil and gas components. Their combined reservoir volume is $aO+bH$, where

\[a=\frac{B_o-R_sB_g}{1-R_sR_v},\qquad b=\frac{B_g-R_vB_o}{1-R_sR_v}.\]

Let $N_{foi}$ and $G_{fgi}$ denote the original free-oil and free-gas surface components. With net cumulative withdrawals $N_p,G_p,W_p$, aquifer influx $W_e$ in reservoir volume, and dimensionless rock/water support $\epsilon$:

\(aN_p+bG_p+B_wW_p-W_e=N_{foi}E_o+G_{fgi}E_g,\) \(E_o=a+R_{si}b-B_{oi}+B_{oi}\epsilon,\qquad E_g=R_{vi}a+b-B_{gi}+B_{gi}\epsilon.\)

For linear rock/water compressibilities and initial water saturation $S_{wi}$, $\epsilon=(c_f+S_{wi}c_w)(p_i-p)/(1-S_{wi})$. Compressibilities are in 1/bar. Thus $N=N_{foi}+R_{vi}G_{fgi}$ and $G=G_{fgi}+R_{si}N_{foi}$ are the total OOIP and OGIP. Do not equate the initially free-phase components to total volumes. Do not multiply the reservoir-volume aquifer influx by Bw a second time.

The implementation solves the undivided equations using column-scaled EJML QR least squares and SVD conditioning. It rejects non-finite PVT, non-positive FVFs, $1-R_sR_v\leq10^{-10}$, negative fitted phase inventories and scaled condition numbers above $10^8$. Residuals and RMS are reservoir m3, not bar; the condition number is a numerical identifiability diagnostic, not a confidence interval. A small residual does not establish unique reservoir/aquifer parameters. These are in-place volumes, not recoverable reserves.

The generalized-balance tests check independent component inventories, condensate, gas caps, water injection, known influx/rock support, scale changes and the zero-Rv black-oil limit. The tank/EOS integration test uses original PR-EOS PVT to recover the known gas inventory from a synthetic SimpleReservoir history. These synthetic checks establish conservation and software behavior; they are not a field or laboratory validation.

Literature-based development priorities

Method and reference Current evidence and next implementation
Walsh generalized balance (1995), 10.2118/95-01-07 Implemented above; retain Rv, consistent surface basis and phase-inventory distinctions.
Whitson & Torp CVD analysis (1983), 10.2118/10067-PA CVD and EOS tuning exist; qualify depletion-path generation of generalized PVT against public measured data, rather than treating unrelated constant-composition flashes as a CVD experiment.
Fetkovich finite aquifer (1971), 10.2118/2603-PA FetkovichAquifer now supplies physical reservoir-volume storage and aquifer depletion; optional compositional replay couples segregated water support to EOS volume closure. Analytical/RK4 and interval-refinement checks are synthetic qualification; independent field matching remains required. AquiferDrive remains a separate gas-equivalent surrogate.
Walsh pressure uncertainty (1999), 10.2118/56691-MS Undivided volume residuals above avoid dividing by small expansion terms, but do not handle pressure error statistically. CompositionalReservoirHistory provides bounded fixed-volume replay of measured component amounts with independent pressure-sigma residuals and withheld pressures. Phase-selective replay now predicts produced composition and separator GOR; independent measured produced-phase and public field qualification and correlated errors remain open.
Ogele et al. Bayesian integration (2006), 10.2118/100257-MS ReservoirHistoryEnsemble now combines explicit discrete joint geological volume/phase-fraction/aquifer/PVT/history prior masses with the pressure-history likelihood, correlations and prior shifts. Refine the supplied joint grid; this is not continuous posterior or field qualification.
Emerick & Reynolds ES-MDA (2013), 10.1016/j.cageo.2012.03.011 Candidate uncertainty layer after forward matching is qualified; propagate PVT/volume/aquifer priors and report posterior correlations and forecast uncertainty. Reuse external ensemble tooling where appropriate.
Tran et al. rescaled CVD/density-based analysis (2025), 10.1016/j.jgsce.2025.205551 Research candidate for near-critical gas condensate with variable BHP under boundary-dominated flow. Obtain the full method and reproduce its numerical cases before implementation; it is not a general substitute for average-pressure material balance.

Compositional measured-history replay

CompositionalReservoirHistory replays measured production and injection using SimpleReservoir component inventories and its isothermal, fixed-volume TV flash. Each trial restores the accepted CalibratedPvtModel tuned checkpoint at the supplied initial pressure and temperature. Initial gas/oil/aqueous reservoir-volume fractions are fixed geological inputs. fitVolume changes only their total initial pore volume; it cannot adjust the EOS to hide PVT error. Preserve the original and tuned checkpoints and independently review the laboratory hold-out evidence before reservoir fitting.

Supply all EOS component names, in any order, with nonnegative produced and injected mol per interval in that order. These are amounts, not mol/s, Sm3 or accumulated surface volumes. Convert measured phase rates and compositions to component amounts using the accepted separator/stock-tank basis and record that conversion in the history reference. An injected species must already exist in the accepted EOS component basis. No oil/gas/water surface-volume conversion or generalized CVD table is invented here.

For component $i$ and interval $k$, the ledger is

\[n_{i,k+1}=n_{i,k}+n_{i,k}^{\mathrm{inj}}-n_{i,k}^{\mathrm{prod}}.\]

Each accepted step must close this ledger and the supplied reservoir volume. Injection is mixed before withdrawal for the interval inventory check; this does not resolve within-interval transport or mixing. Negative inventories, absent requested initial phases, non-finite pressure and failed TV-flash volume closure reject the trial. The standalone SimpleReservoir.advanceComponentAmounts API requires a dedicated tank with no wells and setRecordProductionHistory(false). A rejected interval changes neither fluid, time nor calculation identity. After successful molar replay, legacy surface-production recording and stream-transient accounting cannot be resumed on that tank because its separator-dependent surface counters are unavailable. Use a new tank for those workflows. The usual stream production pressure cutoff is not applied to measured component replay.

Every interval may include measured average end pressure and a positive pressure standard deviation, both in bara. Use NaN for both to request prediction only. The standardized residual is measured minus predicted pressure divided by the supplied sigma. Calibration and held-out sums of squared standardized residuals are separate; held-out pressures never enter volume selection. A zero held-out sum with no held-out observations is not validation; inspect the interval observation flags. The initial pressure is fixed and is not a fitted observation. Amounts and their component compositions are prescribed, so produced GOR/composition in this measured-component mode is not an independent validation prediction. For independently predicted composition, use opt-in equilibrium phase withdrawals and explicit separator-path outputs; prescribed phase molar amounts remain inputs, and independent measured validation is still required.

The bounded search evaluates 17 logarithmically spaced volumes, then refines the best grid basin by golden-section search. At least two calibration pressures and informative feasible trials are required. Inspect isAtBound(), getRejectedTrials() and all residuals. The search bounds, objective and sensitivity are not confidence intervals; the search does not prove a global optimum. An infeasible trial is reported, never replaced by a plausible pressure. The basic no-aquifer, single-volume fit does not represent pressure covariance, rock support, thermal effects or spatial displacement. Optional segregated aquifer support and explicit joint PVT/history uncertainty scenarios are described below.

The complete example below is executed by CompositionalReservoirHistoryTest.documentedWorkflow. Use the CompositionalReservoirHistory.Step and .Fit nested types. It generates a synthetic 1000 m3 history and recovers that known volume while withholding the third pressure. Identical original/tuned models are intentional in this software fixture; this is not evidence of laboratory calibration or independent field accuracy.

SystemInterface original = new SystemPrEos(400.0, 200.0);
original.addComponent("methane", 0.9);
original.addComponent("ethane", 0.1);
original.setMixingRule("classic");
original.init(0);
CalibratedPvtModel model = new CalibratedPvtModel(original, original.clone(), "synthetic sample v1",
    "original PR/classic", "accepted PR/classic", "synthetic calibration fixture", new String[] {"CCE-1"});
Step first = new Step(86400.0, new double[] {400000.0, 40000.0}, new double[] {10000.0, 100000.0},
    Double.NaN, Double.NaN, false);
CompositionalReservoirHistory forward = new CompositionalReservoirHistory(model, "synthetic molar history v1",
    new String[] {"methane", "ethane"}, 400.0, 200.0, new double[] {1.0, 0.0, 0.0},
    new Step[] {first, first, first});
double[] pressure = forward.replay(1000.0).getPressuresBara();
Step[] measured = new Step[3];
for (int i = 0; i < measured.length; i++) {
  measured[i] = new Step(86400.0, first.getProducedMoles(), first.getInjectedMoles(), pressure[i + 1],
      0.5, i == 2);
}
CompositionalReservoirHistory inverse = new CompositionalReservoirHistory(model, "synthetic pressure history v1",
    forward.getComponents(), 400.0, 200.0, new double[] {1.0, 0.0, 0.0}, measured);
Fit fit = inverse.fitVolume(500.0, 2000.0);
double poreVolumeM3 = fit.getResult().getInitialVolumeM3();
double heldOutChiSquare = fit.getResult().getHoldoutChiSquare();

Results expose times, pressures, initial and remaining component moles, standardized pressure residuals, separate calibration/hold-out objectives and an independent final EOS state. Inventories are in-place quantities, not recoverable reserves. No unqualified separator basis is used to relabel pore volume as OOIP/OGIP. The existing generalized balance remains available when accepted path-dependent surface PVT is supplied.

Synthetic tests cover off-grid volume recovery, held-out pressure exclusion, error weighting, bound and infeasibility reporting, named-basis mapping, changing injection composition, oil/gas appearance, aqueous injection, interval refinement, ownership and serialization. Independent checks use the low-density ideal-gas limit and direct EOS volume evaluation at predicted pressures. These establish analytical/software consistency, not independent laboratory or field validation. Remaining work under #4740 includes independent field/laboratory cases, broader depletion phase-path coverage, independently predicted produced-phase validation, multi-tank transmissibility and the separately gated rescaled-CVD research candidate. The bounded PVT generator, finite-aquifer and discrete posterior extensions follow below.

Aquifer influx

VanEverdingenHurstAquifer implements the constant-terminal-rate dimensionless pressure $P_D(t_D)$ of Van Everdingen and Hurst (1949) for radial aquifers, together with the Carter-Tracy (1960) cumulative water-influx method.

import neqsim.pvtsimulation.reservoirproperties.materialbalance.VanEverdingenHurstAquifer;

double u = VanEverdingenHurstAquifer.aquiferConstant(
    0.20, 1.0e-4, 30.0, 3000.0, 180.0);  // phi, ct (1/bar), h (m), re (m), angle (deg)

// tD (dimensionless time), deltaP = p_i - p (bar), reD = dimensionless outer radius
double[] we = VanEverdingenHurstAquifer.cumulativeInfluxCarterTracy(
    tD, deltaP, u, Double.POSITIVE_INFINITY);  // infinite-acting aquifer

// Export an ECLIPSE AQUTAB influence table for a Carter-Tracy aquifer
String aqutab = VanEverdingenHurstAquifer.exportAqutab(tD, Double.POSITIVE_INFINITY);

The returned We array is the aquifer term consumed by GasMaterialBalance.fitHavlenaOdeh and OilMaterialBalance.fitWaterDrive. Use Double.POSITIVE_INFINITY for an infinite-acting aquifer or a finite reD for a bounded (no-flow outer boundary) aquifer.

Decline-curve history matching

DeclineCurveAnalysis is a static, unit-agnostic Arps + Duong toolkit. In addition to the forward rate, cumulativeProduction, eur and forecast methods it least-squares fits decline parameters to a measured rate-time history.

import java.util.Map;
import neqsim.pvtsimulation.util.DeclineCurveAnalysis;

// t[] in days, q[] in any consistent rate unit
Map<String, Double> arps = DeclineCurveAnalysis.fitArps(t, q); // qi, di, b, rSquared
double eur = DeclineCurveAnalysis.eurFromFit(arps, 50.0);       // EUR to qLimit = 50

// Windowed fit — skip early transient / cleanup points
Map<String, Double> arpsW = DeclineCurveAnalysis.fitArps(t, q, 3, t.length - 1);

For fracture-dominated tight-gas and shale wells, the Duong (2011) model avoids the reserves over-estimate of hyperbolic Arps:

Map<String, Double> duong = DeclineCurveAnalysis.fitDuong(t, q); // q1, a, m, rSquared
double q  = DeclineCurveAnalysis.rateDuong(1000.0, 1.5, 1.2, 100.0);
double gp = DeclineCurveAnalysis.cumulativeDuong(1000.0, 1.5, 1.2, 100.0);

Fit both models, compare rSquared, and prefer Duong when the well is in transient linear (fracture-dominated) flow.

Gray correlation

For gas / gas-condensate vertical wells, the Gray (1974) correlation (neqsim.process.equipment.pipeline.PipeGray) predicts in-situ liquid holdup and a condensate-film effective roughness. It is the vertical-well companion to the material-balance reserves work above. See Flow Assurance and the process pipeline documentation for details.

PipeGray well = new PipeGray("Gray well", inletStream);
well.setDiameter(0.0889);   // 3.5 inch tubing
well.setLength(3000.0);
well.setElevation(3000.0);  // vertical
well.setNumberOfIncrements(10);
well.run();
double dP = well.getTotalPressureDrop(); // bar

Units

Finite aquifer, uncertainty and native depletion PVT

The optional FetkovichAquifer overload of CompositionalReservoirHistory couples segregated immiscible reservoir-volume water support to the existing TV flash. ReservoirHistoryEnsemble integrates explicit joint geological, aquifer, PVT and component-history hypotheses with common pressure observations and independent Gaussian sigmas. See physical assumptions and posterior evidence.

DepletionPvtHistoryGenerator uses opt-in native CVD/DLE snapshots, one multistage separator basis and a separately accepted pure-water EOS to construct limited dry two-hydrocarbon-phase generalized PVT. It rejects absent phases and unacceptable approximate surface-component closure; see the generation and acceptance contract. These extensions do not complete independent laboratory/field qualification or the remaining multi-tank/research acceptance in issue #4740.

Explicit initial occupied-volume experiments can use the transactional native fixed-cell route. It preserves selected original/tuned EOS, sample/experiment and separator provenance, with an explicit reference convention. Its bounded synthetic qualification does not complete wet upper-dew laboratory or independently measured field acceptance.