Analysis and evaluation of the influence of geological, physical, and technological reservoir characteristics on the performance indicators of geological and technical measures

Full Text

Introductıon

Maintaining field-development efficiency and oil-production intensity requires geological and technical measures that are compatible with the actual reservoir and operating conditions. The same intervention may produce different results in different formations because its effect depends on formation thickness, oil-saturated and opened thicknesses, porosity, permeability, fluid properties, saturation, and pre-treatment production conditions [1–4].

In field practice, the selection of treatment type, candidate well, and operating parameters is often based on the experience of geological and production departments. This experience is indispensable, but an ad hoc choice may not integrate all relevant geological, geophysical, technological, and production indicators. As a result, transferring a measure that was successful in one reservoir to another reservoir or field does not necessarily reproduce the same effect [2–8].

Mathematical models can organize heterogeneous information and compare alternative interventions under uncertainty. Their practical value, however, depends on the transparency of the calibration sample, a genuinely out-of-sample assessment, and explicit limits on transferability. These requirements are especially important for small historical datasets in which reservoir units may repeat and field identifiers may be incomplete.

The aim of this study was to analyse and evaluate the influence of geological, physical, and technological reservoir characteristics on five geological and technical measure (GTM) performance indicators; to document the available sample structure; to evaluate internal out-of-sample performance; and to determine whether the models can support comparative screening of intervention alternatives under incomplete information. The study was not intended to create a universal predictor applicable without local verification.

Materials and Methods

The analysis used three related data components. The Tengiz annual series provided field-development context. The intervention-level matrix contained the cases used to construct and test the performance models. The West Prorva material supplied a target set for scenario calculations. The size, role, and principal limitation of each component are summarized in Tab. 1.

 

Table 1. Structure and analytical role of the source data

Data component	Field / object	Available sample	Analytical role	Qualification
Annual field series	Tengiz	10 annual observations, 2002–2011	Contextual comparison of field-development indicators	Not used to fit intervention regressions
NWZT+P cases	Several fields; identifiers not retained	n = 38	Regression modelling and internal validation	Field-blocked validation cannot be performed
Acid treatment + perforation	Several fields; identifiers not retained	n = 32	Regression modelling and internal validation	Field-blocked validation cannot be performed
Production optimization	Several fields; identifiers not retained	n = 37	Regression modelling and internal validation	Field-blocked validation cannot be performed
Target-field set	West Prorva	9 units described; 5 complete; 15 scenarios	Applicability test and alternative ranking	Observed case-level outcomes unavailable

NWZT+P – near-wellbore-zone treatment plus perforation. The consolidated intervention matrix contains 107 cases in total. A reliable field identifier was not retained; therefore, sample counts by individual field cannot be reconstructed from the available tables.

 

Each intervention case contained 16 inputs: total formation thickness x₁; oil-saturated thickness x₂; opened oil-saturated thickness x₃; sandiness coefficient x₄; porosity x₅; permeability x₆; reservoir-oil viscosity x₇; oil density x₈; gas content x₉; initial oil saturation x₁₀; formation temperature x₁₁; paraffin content x₁₂; sulfur content x₁₃; pre-treatment oil rate x₁₄; pre-treatment fluid rate x₁₅; and pre-treatment water cut x₁₆. Units were standardized as m, fraction, 10⁻³ μm², mPa·s, t/m³, m³/t, °C, %, and t/d, as applicable.

Five outcomes characterized GTM performance: effect duration Y₁ (d), incremental oil Y₂ (t), oil-rate increment Y₃ (t/d), post-treatment oil rate Y₄ (t/d), and post-treatment water cut Y₅ (%). The three intervention groups were NWZT+P, acid treatment plus perforation, and production optimization.

The original 16-dimensional description was reduced to nine composite variables using principal-component reasoning and physically interpretable combinations [9–12]. The final transformations were:

$Z_1 = x_3 / \left(x_1-x_2\right)$ (1)

$Z_2 = x_6*x_3 / x_7$ (2)

$Z_3= 0.397065x_4+0.602935x_5$ (3)

$Z_4 =0.460477x_{12}+0.539523x_{13}$ (4)

$Z_5=0.402704x_{14}+0.597296x_{15}$ (5)

$Z_6 = x_8$ (6)

$Z_7=x_9$ (7)

$Z_8=x_{10}$ (8)

$Z_9=x_{11}$ (9)

Z₁ describes the relationship among total, oil-saturated, and opened thicknesses; Z₂ represents hydraulic conductivity. Z₃ combines sandiness and porosity, Z₄ combines paraffin and sulfur contents, and Z₅ combines pre-treatment oil and fluid rates. The coefficients in the weighted pairs were retained from the source calculation scheme.

For each intervention type and each outcome, a separate multiple linear regression was fitted:

$Y-hat\left(k\right) = a_0\left(k\right) + sum\left[j = \mathrm{1...9}\right] a_J\left(k\right) Z_j$ (10)

Internal out-of-sample performance was evaluated separately within every intervention group by shuffled five-fold cross-validation with random seed 42. In each iteration, four folds were used to refit the model and the held-out fold was predicted. Held-out predictions from all folds were pooled. We report the coefficient of determination (R²) and median absolute percentage error (MdAPE); observations equal to zero were excluded only from the percentage-error calculation. Calculations used the rounded values in the source dissertation tables [8], so minor differences are possible if the original unrounded operational database is restored.

The West Prorva calculation is treated as an applicability test rather than independent external validation. Complete input vectors were available for five reservoir units, and three intervention scenarios were calculated for each unit. Observed case-level outcomes for these scenarios were not available; consequently, the exercise tests whether the model can process the target conditions and rank alternatives, but not whether its numerical predictions are accurate on a new field.

For each scenario, the five predicted outcomes were converted to normalized membership values μ₁–μ₅ according to fuzzy-set theory [13]. Increasing functions were used for the outcomes to be maximized, whereas water cut was minimized. With equal weights, the decision membership (μD) was:

$\mu D = min\left(\mu \mathrm{1,} \mu \mathrm{2,} \mu \mathrm{3,} \mu \mathrm{4,} \mu 5\right)$ (11)

For each reservoir unit, the intervention with the largest μD was ranked first. This conservative max-min rule prevents an alternative with one very weak outcome from being selected solely because it performs well on the remaining criteria.

Results

Analysis and Synthesis of the Main Well-Intervention Efficiency Indicators

The source matrix represents three intervention types with 32–38 cases per group and nine composite predictors. Five-fold cross-validation was therefore used to distinguish descriptive agreement from the ability to predict held-out observations. The verified results are presented in Tab. 2.

 

Table 2. Five-fold cross-validation by intervention type (R² / MdAPE, %)

Intervention	n	Y1	Y2	Y3	Y4	Y5
NWZT+P	38	0.530 / 15.2	0.296 / 23.5	-0.307 / 31.9	0.522 / 24.0	-0.002 / 22.0
Acid treatment + perforation	32	0.247 / 18.4	0.261 / 20.4	-0.174 / 39.0	0.667 / 23.3	-0.325 / 31.1
Production optimization	37	-6.843 / 60.2	-18.589 / 14.8	-97.549 / 24.7	0.235 / 26.0	-66.266 / 39.6

Negative R² means that held-out predictions were poorer than prediction by the corresponding sample mean.

 

For NWZT+P, the most stable held-out relationships concerned effect duration and post-treatment oil rate. Incremental oil showed only limited predictive ability, while oil-rate increment and water cut were not predicted reliably. For acid treatment plus perforation, post-treatment oil rate was the only outcome with a clearly useful positive R². Production-optimization models were unstable for four outcomes; the model for post-treatment oil rate retained only weak positive ability.

These results modify the interpretation of the original in-sample comparisons. Small deviations between calculated and observed values in the calibration table do not by themselves establish forecasting ability. The low case-to-predictor ratio, repeated reservoir units, rounding, and heterogeneous operational practice increase the uncertainty of every equation. The models should therefore be used together as a comparative screening system rather than as five independent unconditional forecasts.

Forecasting Performance Indicators under Different Field Conditions and Selecting Further Technological Decisions

The constructed equations were applied to West Prorva reservoir units whose geological and technological conditions differ from the calibration cases. Of nine described units, five had the complete inputs needed for calculation: VIII-2, VIII-3, IX-2, T-II, and T-III. Three intervention alternatives were evaluated for each unit, producing 15 scenarios. Tab. 3 reports the highest-ranked alternative for each unit.

 

Table 3. Highest-ranked intervention for West Prorva reservoir units

Reservoir unit	Highest-ranked intervention	μD	Interpretation
IX-2	Production optimization	0.48	Highest membership among the three scenarios
T-III	NWZT+P	0.35	Relative preference; local verification required
T-II	Acid treatment + perforation	0.33	Relative preference; local verification required
VIII-2	Production optimization	0.29	Low membership; recommendation is tentative
VIII-3	NWZT+P	0.15	Very low membership; insufficient for direct deployment

Values are relative scores within the three alternatives tested for each reservoir unit; they are not probabilities of success.

 

Production optimization ranked first for IX-2 and VIII-2. NWZT+P ranked first for T-III and VIII-3, whereas acid treatment plus perforation ranked first for T-II. The decision memberships were modest. This agrees with the source observation that some target conditions produced excessively high or low numerical forecasts because the target and calibration data came from different fields.

The ranking therefore preserves the original purpose of the approach: to compare alternatives when information is incomplete. At the same time, the low membership values are warning signals. They indicate that the target object does not closely reproduce a confident calibration pattern and that the numerical results should be checked by field-specific engineering analysis before implementation.

Discussion

The analysis confirms that geological, physical, and technological variables contain useful information for GTM assessment, but the strength of the relationship differs markedly among outcomes and intervention types. This outcome-specific behaviour is physically plausible. Treatment response depends not only on the reservoir variables retained in the matrix but also on treatment volume, chemical concentration, perforation design, well integrity, equipment condition, execution quality, downtime, and contemporaneous operating constraints. These variables were not represented in the consolidated dataset.

The fuzzy max-min rule remains useful because it organizes a multi-criteria decision and makes the limiting outcome visible. Nevertheless, fuzzy ranking cannot repair weak underlying regressions. When an outcome model has poor held-out performance, its membership value also carries substantial uncertainty. A responsible workflow should use the ranking to shortlist alternatives, then apply local simulation, economic screening, integrity checks, and expert review.

The West Prorva exercise demonstrates computational transfer, not predictive transfer. The procedure accepts a new input vector and produces a comparative order, but the absence of observed post-treatment outcomes prevents independent external validation. Prospective application should record the selected candidates, predicted indicators, treatment design, and actual results so that the model can be recalibrated and tested on genuinely independent field data.

Limitations of the approach

The consolidated intervention matrix does not retain a reliable field identifier. Representativeness by individual field and field-blocked validation therefore cannot be established from the available material.

The group sizes are small relative to nine predictors, and repeated reservoir units may reduce the effective number of independent observations. Five-fold cross-validation limits optimistic interpretation but does not remove this structural uncertainty.

Calculations were reproduced from rounded tabulated values. If the original unrounded database becomes available, the regressions and validation metrics should be recalculated.

West Prorva supplied complete inputs for only five units and no observed outcomes for the calculated alternatives. The ranking is therefore an applicability test, not an independent external validation.

Intervention-design variables, execution quality, well integrity, downtime, and economic indicators were not included. Equal criterion weights may not correspond to local operational priorities. The equations should be used only within the observed input ranges and for intervention definitions consistent with the calibration data. Materially different geological or operational conditions require local recalibration and prospective validation.

Conclusion

The original research logic – from geological and technological indicators, through dimensionality reduction and regression, to fuzzy comparative selection – has been retained. The available intervention matrix contains 107 cases distributed among three GTM types and supports exploratory modelling and comparative analysis.

Five-fold cross-validation showed that predictive ability is outcome-specific. The most useful results were obtained for effect duration and post-treatment oil rate after NWZT+P, and for post-treatment oil rate after acid treatment plus perforation. Most production-optimization equations were unstable on held-out observations.

The fuzzy-set procedure provides a transparent ranking of alternative GTMs. For the five complete West Prorva units, the preferred alternatives reproduced the source calculations, but the decision memberships were low to moderate.

Accordingly, the method may be used for preliminary screening and comparative decision support. It should not be used for unconditional numerical prediction outside the calibration domain. Field-specific recalibration and validation against observed post-intervention results are required before operational deployment.

ADDITIONAL INFORMATION

Funding source. This research was carried out independently by the authors, without external financial support.

Competing interests. The authors declare that they have no competing interests.

Authors’ contribution. All authors confirm that their authorship complies with the international ICMJE criteria (each author made a significant contribution to the study conception, research implementation, and manuscript preparation, read and approved the final version before publication). The main contributions are distributed as follows: Zhomart K. Zhanturin – text writing, analytics; Murat N. Abishev – work with sources, manuscript editing; Yessengeldi U. Arystanaliyev – overall supervision of the preparation; Zhanylsyn K. Zaidemova – source research, preparation of the literature review; Shokhan M. Medetov – preparation of formulas and reference data.

ДОПОЛНИТЕЛЬНО

Источник финансирования. Авторы заявляют об отсутствии внешнего финансирования при проведении исследования.

Конфликт интересов. Авторы декларируют отсутствие явных и потенциальных конфликтов интересов, связанных с публикацией настоящей статьи.

Вклад авторов. Все авторы подтверждают соответствие своего авторства международным критериям ICMJE (все авторы внесли существенный вклад в разработку концепции, проведение исследования и подготовку статьи, прочли и одобрили финальную версию перед публикацией). Наибольший вклад распределён следующим образом: Жантурин Ж.К. – написание текста, аналитика; Абишев М.Н. – работа с источниками, редактирование рукописи; Арстаналиев Е.У. – общее руководство подготовкой, структурирование текста; Зайдемова Ж.К. – работа с источниками, подготовка литературного обзора; Медетов Ш.М. – подготовка формул и справочных данных.

About the authors

Zhomart K. Zhanturin

Atyrau Oil and Gas University named after S. Utebaev

Author for correspondence.
Email: aing-zhomart@mail.ru
ORCID iD: 0009-0007-4944-1850

Cand. Sc. (Engineering), Associate Professor

Kazakhstan, Atyrau

Murat N. Abishev

Atyrau Oil and Gas University named after S. Utebaev

Email: m_abishev_nik@mail.ru
ORCID iD: 0009-0001-5793-3800

Cand. Sc. (Engineering), Associate Professor

Kazakhstan, Atyrau

Yessengeldi U. Arystanaliyev

Atyrau Oil and Gas University named after S. Utebaev

Email: esen-65@mail.ru
ORCID iD: 0000-0001-7219-2038

Cand. Sc. (Engineering), Professor

Kazakhstan, Atyrau

Zhanylsyn K. Zaidemova

Atyrau Oil and Gas University named after S. Utebaev

Email: b.n.m.99@list.ru
ORCID iD: 0000-0002-6628-024X

Cand. Sc. (Engineering)

Kazakhstan, Atyrau

Shokhan M. Medetov

Atyrau Oil and Gas University named after S. Utebaev

Email: medetov.76@mail.ru
ORCID iD: 0009-0002-0137-228X

Cand. Sc. (Engineering)

Kazakhstan, Atyrau

References

Supplementary files