Investigation of CO₂ Displacement Characteristics of Highly Water-Sensitive and Low-Permeability Oil Reservoirs: A Case Study of Block A in Eastern China

Abstract

Highly water-sensitive, low-permeability reservoirs are difficult to develop by water flooding because clay activation and pore-throat plugging impair injectivity. This study combined X-ray diffraction, scanning electron microscopy, water-sensitivity tests, long-core CO₂ flooding, and response surface methodology to investigate Block A in eastern China. The sandstone contained 17.0% clay minerals, mainly illite and kaolinite. After water flooding, clay accumulation, fines migration, and pore-throat bridging produced cumulative permeability damage of 63.09%, confirming substantial water-sensitive impairment. CO₂ flooding comprised stable displacement, gas breakthrough, and stable gas channeling. Increasing the injection pressure from 20 to 35 MPa increased oil recovery from 40.78% to 57.91%. The measured CO₂–oil minimum miscibility pressure (MMP) was 32.70 MPa. Accordingly, the experiments at 20, 25, and 30 MPa were conducted below the MMP, whereas 35 MPa was slightly above the MMP, indicating that the CO₂–oil system had the thermodynamic conditions required to develop multiple-contact miscibility. In contrast, increasing the initial water saturation from 36.59% to 62.04% reduced the oil recovery from 52.13% to 44.21%. Permeability had a minor effect, while stronger heterogeneity reduced recovery to 41.67%. RSM identified the permeability ratio as the dominant factor and revealed significant pressure–saturation and saturation–permeability-ratio interactions. The model predicted a maximum oil recovery of 62.33% at an injection pressure of 35.00 MPa, an initial water saturation of 46.71%, a permeability of 89.58 mD, and a permeability ratio of 8.89. A confirmation experiment conducted near the predicted optimum conditions yielded an oil recovery of 60.42%, corresponding to a relative error of 3.05% from the predicted value. These results support CO₂-flooding design for highly water-sensitive, low-permeability reservoirs.

1. Introduction

In Block A of the East China oil region, the reservoirs are deeply buried, exhibiting strong heterogeneity and water sensitivity. The porosity ranges from 5.2% to 24.2% (averaging 11.2%), permeability from 0.58 to 132.71 mD (averaging 54.56 mD), and the original oil saturation between 40% and 65%. Due to the high water sensitivity of the formations, water injection encounters high and unstable pressures, leading to a low injection-production ratio and poor effectiveness of water flooding. Gas flooding is commonly employed for low-permeability, ultra-low-permeability, and tight oil reservoirs because the injected gas offers superior diffusivity and injectability compared to water, and it avoids the continuous introduction of external water and therefore reduces further disturbance of water-sensitive formations, although it cannot eliminate the effects of residual water or pre-existing water-sensitive damage. Among various gas injection media, CO₂ flooding demonstrates the most remarkable performance and provides a potentially effective option for improving injectivity and oil recovery in water-sensitive, low-permeability reservoirs [1]. Carbon-dioxide-enhanced oil recovery (CO₂-EOR) has been widely applied due to its capability to effectively enhance oil recovery while simultaneously reducing greenhouse gas emissions [2,3,4]. As a tertiary recovery technique, it has garnered significant attention driven by global warming concerns and the demand for greenhouse gas storage [5,6]. Additionally, CO₂ flooding has the twin advantages of permanent CO₂ sequestration and economic benefits, while also cutting greenhouse gas emissions [7,8,9].

Abedini and Torabi investigated CO₂ sequestration under different pressure scenarios via CO₂ horizontal displacement experiments, concluding that near miscible conditions were optimal for CO₂ sequestration and EOR when gravity was neglected [10]. Thus far, most studies on CO₂/oil flow behavior have focused on the core scale [11,12]. These efforts have been crucial for analyzing the multiphase seepage characteristics of oil and CO₂ in reservoir rocks (such as the reduction in oil-effective permeability) and for revealing the phase behavior of CO₂–oil mixtures [13]. Jin et al. [14] also discovered that CO₂, when it comes into contact with oil, combines with the oil’s light components, rendering them soluble, and subsequently permeates the oil. This interaction results in decreased oil viscosity and surface tension, improved oil flowability, and, ultimately, an increased recovery rate. Behnoud et al. [15] noted that upon CO₂ injection into the reservoir, gaseous CO₂ comes into contact with liquid oil, forming a contact surface that generates a gas drive effect, thereby propelling the oil toward the production well. Ma et al. [16] noted that for immiscible and near-miscible flooding, optimizing the CO₂ injection velocity is key. High velocity can accelerate oil recovery without reducing immiscible flooding efficiency, and low velocity can boost recovery under near-miscible conditions. Chen et al. [17] noted that gravity segregation impacts CO₂ sequestration and oil production differently across flooding conditions, with significant effects under immiscible conditions but minimal influence under miscible conditions. Wang et al. [18] noted that CO₂ miscible flooding reduces residual oil saturation to half of immiscible flooding through pre-breakthrough preferential displacement paths and post-breakthrough miscible zone expansion, and optimized WAG injection can elevate immiscible recovery to miscible levels. Su et al. [19] proposed that in tight or low-permeability cores, CO₂–oil IFT reduction and CO₂ extraction play the most important role in CO₂ immiscible flooding.

Beyond conventional CO₂ flooding, CO₂-based technologies have also been investigated for other subsurface energy-extraction and reservoir-stimulation applications. In gas-hydrate reservoirs, CO₂ replacement can combine methane production with carbon sequestration, although its feasibility depends strongly on the bottom-hole temperature–pressure operating window and hydrate phase-equilibrium conditions [20]. CO₂ fracturing represents another reservoir-stimulation approach for which the transport and sedimentation stability of solid particles are governed by fluid viscosity, flow conditions, reservoir temperature and pressure, and fracture-surface properties [21]. These studies highlight the importance of reservoir-specific thermodynamic and transport conditions in CO₂-assisted subsurface energy recovery.

Although there have been some achievements in CO₂ flooding, comprehensive quantitative research under diverse physical and geological conditions is still insufficient, and most existing studies are limited in scope. For special reservoirs like water-sensitive low permeability ones, the potential of CO₂ flooding has not been fully explored. Most studies focus on single conditions like temperature and pressure, lacking analysis of various influencing factors and mechanism exploration, so they fail to reveal the dynamic phase evolution and stage-dominant mechanisms of CO₂ flooding under multi-field coupling.

To address these problems, this study first investigated the pressure-dependent CO₂-flooding characteristics in Block A of eastern China and divided the displacement process into three stages. Single-factor experiments were then conducted to evaluate the key factors influencing CO₂ flooding in highly water-sensitive, low-permeability oil reservoirs, focusing on CO₂ injection pressure, initial water saturation, absolute permeability, and reservoir heterogeneity represented by the permeability ratio. Additionally, response surface methodology (RSM) was employed to quantify the interaction effects among these factors and determine the optimal combination of displacement parameters. This provides a mechanistic understanding and a benchmark for scheme design in CO₂-enhanced oil recovery (EOR) for similar reservoirs. RSM collects data based on the concept of Design of Experiments (DoE), identifies the factors and their interactions that significantly impact the process response, and constructs mathematical models that capture the causal relationships between factors and response variables. The method optimizes the causal model as the objective function, aiming to determine the optimal combination of factor levels within the planned range of experimental design through analysis of the regression equations [22,23]. This method has demonstrated significant effectiveness across various disciplines, including chemistry, biosciences, and food science [24,25]. In recent years, RSM has gradually gained favor in the field of oil and gas field development [26].

2. Materials and Methods

The experimental program combined mineralogical and microscopic characterization, evaluation of water-sensitive permeability damage, high-temperature and high-pressure CO₂ displacement tests, and response-surface optimization. Mineralogical composition was determined by X-ray diffraction (XRD) using a [DX-2700] diffractometer (Dandong Haoyuan Instrument Co., Ltd., Dandong, China). Pore morphology was examined by scanning electron microscopy (SEM) using a [KYKY-EM6200] microscope (KYKY Technology Co., Ltd., Beijing, China). These analyses characterized the mineralogical and pore-structural features associated with water sensitivity.

Water-sensitive permeability damage was evaluated using an [LD-3] core-flow testing apparatus (Haian Huada Petroleum Instrument Co., Ltd., Haian, China). The test quantified the hydraulic impairment caused by water–rock interaction. CO₂ displacement tests were conducted on assembled long-core models under reservoir-representative temperature and pressure. A [DQ-1] core-flooding apparatus (Jiangsu Huaan Scientific Instrument Co., Ltd., Haian, China) was used for these tests.

Response-surface optimization was performed using Design-Expert version 8.0.6 (Stat-Ease, Inc., Minneapolis, MN, USA).

2.1. Reservoir Fluids and Core Samples

The experimental fluids and core models were selected according to the reservoir conditions of Block A in eastern China. The formation-water salinity is approximately 25,000 mg/L, the initial reservoir pressure is 36.8 MPa, and the reservoir temperature is 121.6 °C. The surface crude-oil density is 0.8550 g/cm³. At 50 °C, the crude-oil viscosity ranges from 12.10 to 38.74 mPa·s, with an average value of 21.5 mPa·s. The initial boiling point ranges from 115.3 to 167.7 °C and averages 143.6 °C. Formation water was used for core saturation and initial-water-saturation preparation, whereas commercially supplied CO₂ was used as the displacement gas.

The CO₂-flooding models were assembled from four cylindrical short cores, each approximately 25 cm long, to produce a composite core approximately 100 cm in length and 4 cm in diameter. Homogeneous models were assembled from short cores with similar permeability, whereas heterogeneous models were assembled from two permeability classes. The permeability ratio was defined as the ratio of the higher permeability to the lower permeability in the composite model. Representative companion specimens from the same core-material batch were used for XRD, SEM, and water-sensitivity testing so that the mineralogical, microscopic, and displacement measurements described the same reservoir-rock system. Representative core segments used to assemble the long-core models are shown in Figure 1.

2.2. Methods for Water-Sensitivity Characterization and Permeability-Damage Evaluation

Water-sensitivity damage fundamentally arises from a sequence of clay hydration, dispersion, detachment, and migration upon contact with injected water, followed by particle retention, bridging, and plugging within micron-scale pore throats. The type and abundance of clay minerals, their modes of occurrence, and pore-throat connectivity are the key factors governing fines release and transport and the subsequent deterioration of seepage capacity. Quantitative mineralogical analysis by X-ray diffraction (XRD), comparative scanning electron microscopy (SEM) observations before and after water flooding, and pore-size redistribution analysis were therefore combined to reveal the mineralogical basis and pore-structural evolution of water-sensitivity damage. These observations were further integrated with permeability data to clarify how pore-scale throat plugging translates into a macroscopic loss of flow capacity [27,28,29]. The overall experimental workflow is summarized in Figure 2.

2.2.1. Mineralogical Characterization by X-Ray Diffraction (XRD)

Representative untreated core fragments were dried at 60 °C for 24 h and then crushed and ground to a particle size below 40 μm. The powdered sample was packed into a flat sample holder to minimize preferred orientation and was analyzed by powder XRD. Mineral phases were identified by matching the measured diffraction-peak positions with standard reference patterns. The relative contents of quartz, feldspars, clay minerals, and accessory minerals were calculated from the integrated diffraction intensities using the quantitative phase-analysis program of the diffractometer. Particular attention was given to illite and kaolinite because their hydration, detachment, and migration directly control water-sensitive pore-throat damage in low-permeability sandstone.

2.2.2. Scanning Electron Microscopy (SEM) Characterization of Pore Structure

Paired companion fragments were prepared from the untreated core material and from the core after the water-sensitivity flooding test. The specimens were dried at 60 °C to constant mass, gently fractured to expose fresh surfaces, mounted on conductive stubs, and coated with a thin conductive layer before observation. Secondary electron images were acquired at an accelerating voltage of 3–5 kV and a working distance of approximately 8–11 mm. Low- and high-magnification fields were selected at comparable observation scales to characterize grain contacts, intergranular pores, clay coatings, lamellar aggregates, fine-particle detachment, bridging, and pore-throat filling.

2.2.3. Water-Sensitivity Permeability-Damage Test

Independent natural core plugs were used for the water-sensitivity tests at 45 and 120 °C. After solvent cleaning, drying at 60 °C, and dimensional measurement, each core plug was mounted in a core holder, evacuated for 5 h, and fully saturated with mimicked formation brine (MFB) with a salinity of approximately 25,000 mg/L. The core holder was then maintained at the target test temperature until thermal equilibrium was reached. Under both temperature conditions, the confining pressure was maintained at least 3 MPa above the inlet pressure, and the MFB was initially injected at a constant flow rate of 0.15 mL/min until the pressure drop across the core varied by less than 1% over a continuous 30 min period. The initial water-phase permeability, Ki, was calculated from the stabilized pressure drop. Subsequently, under the same temperature, confining-pressure, and flow-rate conditions, the injected-fluid salinity was reduced stepwise following the sequence MFB, 75% MFB, 50% MFB, 25% MFB, deionized water ($\mathrm{D}\mathrm{W}$). At each salinity stage, injection was continued until the pressure drop restabilized, after which the corresponding water-phase permeability, $K$, was calculated. The final stabilized permeability measured during DW injection was defined as the damaged permeability, $K_d$. The inlet pressure, outlet pressure, and pressure drop across the core were continuously monitored and recorded throughout the experiment [30].

$$K=1.688\times {\displaystyle \frac{\mu QL}{A\Delta p}}$$

(1)

where $K$ is the water permeability (mD), $\mu$ is the fluid viscosity (mPa·s), $Q$ is the volumetric flow rate (mL/min), $L$ is the core length (cm), $A$ is the core cross-sectional area (cm²), and $\Delta p$ is the stabilized pressure difference across the core (MPa). The water-sensitivity permeability-damage degree was calculated as follows:

$$D_w={\displaystyle \frac{K_i-K_d}{K_i}}\times 100\%$$

(2)

where $D_w$ is the permeability-damage degree (%), $K_i$ is the initial permeability measured with MFB, and $K_d$ is the stabilized permeability measured during the final DW stage. The fluid viscosity corresponding to each injection stage was used in Equation (1); therefore, the calculated permeability change primarily reflects the reduction in core flow capacity caused by clay hydration, fines migration, and pore-throat blockage.

2.3. CO₂-Flooding Apparatus and Experimental Conditions

The high-temperature and high-pressure CO₂-flooding system consisted of an injection unit, a long-core displacement unit, a backpressure-control unit, a pressure-monitoring unit, a gas-acquisition unit, and a temperature-control unit (Figure 3). The injection unit comprised an ISCO pump and intermediate fluid containers. The pump provided a flow-rate resolution of 0.01 mL/min and a maximum working pressure of 120 MPa. The long-core holder was rated to 200 °C and 60 MPa. The outlet pressure was controlled using a backpressure valve, and the inlet pressure, outlet pressure, and core differential pressure were recorded continuously by pressure transducers connected to the data-acquisition system. The produced fluids were separated in a gas–liquid separator; the liquid volume, gas volume, gas–oil ratio, and produced-oil density were measured during displacement. The core holder and fluid containers were maintained in a constant-temperature oven.

2.4. CO₂-Flooding Experimental Procedures

2.4.1. Baseline CO₂-Flooding Experiment

(1)

Four 25 cm short cores were assembled into a composite long core of approximately 100 cm. The length and cross-sectional area were measured, and the apparent core volume was calculated. The assembled model was installed in the core holder and evacuated for 5 h.

(2)

Formation water was introduced under vacuum until the core was fully saturated. The pore volume was determined from the saturated-water volume, and the absolute permeability was measured under steady flow. Crude oil was then injected at 1.0 mL/min to displace formation water. Oil injection was continued until the produced water volume reached the value required for the target initial water saturation and no further water was produced. The initial oil saturation was calculated from the pore volume and the retained water volume.

(3)

The core holder was heated to 120 °C and maintained at this temperature until thermal equilibrium was reached. To investigate the effect of pressure on CO₂ flooding performance, the system displacement pressure was set to 20, 25, 30, and 35 MPa by adjusting the outlet backpressure valve. After pressure stabilization, CO₂ was injected at 0.5 mL/min. The injected pore volume, oil production, gas production, pump reading, inlet pressure, outlet pressure, differential pressure, gas–oil ratio, and produced-oil density were recorded at every 0.1 PV of cumulative CO₂ injection.

(4)

The experiment was terminated when the produced gas–oil ratio exceeded 5000 mL/mL, cumulative oil production exceeded 90%, or no additional oil was produced. The apparatus was then depressurized, cleaned, and prepared for the next run.

The cumulative oil-recovery factor was calculated from the produced oil volume and the initial oil volume in the core:

$$R_o={\displaystyle \frac{V_{op}}{PV{S}_{oi}}}\times 100\%$$

(3)

where $R_o$ is the cumulative oil-recovery factor (%), $V_{op}$ is the cumulative produced-oil volume (mL), $PV$ is the pore volume of the long-core model (mL), and $S_{oi}$ is the initial oil saturation.

2.4.2. Single-Factor Comparative CO₂-Flooding Experiments

Four single-factor comparative series were conducted to investigate the effects of injection pressure, initial water saturation, permeability, and permeability ratio. Different natural long-core models were used in the four single-factor series, and the model assignments and initial properties for individual tests are summarized in

Table 1. Basic parameters of the assembled long-core models.

Experimental Series	Long-Core Model ID	Model Type	Pressure (MPa)	Initial Water Saturation (%)	Total Length (cm)	Radius (cm)	Porosity (%)	k1 (mD)	k2 (mD)	Rk
Injection pressure	1	Homogeneous	20	56.19	100.12	1.92	16.32	2.84	—	1.00
	2	Homogeneous	25	53.62	99.93	2.15	14.78	3.59	—	1.00
	3	Homogeneous	30	50.85	100.05	1.83	17.95	6.43	—	1.00
	4	Homogeneous	35	55.25	99.84	2.08	15.23	4.32	—	1.00
Initial water saturation	5	Homogeneous	25	36.59	100.18	1.97	13.49	4.19	—	1.00
	6	Homogeneous	25	44.24	99.97	2.19	18.67	3.75	—	1.00
	7	Homogeneous	25	51.83	100.08	1.81	16.04	2.67	—	1.00
	8	Homogeneous	25	62.04	99.81	2.12	14.11	5.47	—	1.00
Permeability	9	Homogeneous	25	58.17	100.19	1.94	18.32	0.62	—	1.00
	10	Homogeneous	25	53.91	99.88	2.07	15.87	11.81	—	1.00
	11	Homogeneous	25	54.17	100.14	1.88	13.92	86.13	—	1.00
	12	Homogeneous	25	48.78	99.95	2.20	17.15	125.84	—	1.00
Permeability ratio	13	Homogeneous	25	49.91	100.03	1.99	16.59	3.67	—	1.00
	14	Heterogeneous	25	56.31	99.89	2.05	14.45	4.14	26.16	6.32
	15	Heterogeneous	25	51.11	100.07	1.89	18.43	4.38	86.02	19.64
	16	Heterogeneous	25	47.18	99.98	2.03	15.08	3.86	126.29	32.74

Notes: k₁ and k₂ denote the lower- and higher-permeability classes used to assemble the heterogeneous long-core models, respectively, and R_k = k₂/k₁. For homogeneous models, k1 represents the model permeability and k₂ is not applicable.

. Within each series, the target factor was systematically varied, while the experimental temperature, CO₂ injection rate, fluid-preparation procedure, measurement interval, and termination criteria were maintained consistently. Oil recovery and cumulative CO₂ injection were normalized by the original oil in place and pore volume, respectively, and residual differences in the non-target core properties were considered when interpreting the results of each experimental series.

(1)

Injection-pressure series: The displacement pressure was set to 20, 25, 30, and 35 MPa. The mean core permeability was approximately 4.30 mD, and the CO₂ injection rate was 0.5 mL/min.

(2)

Initial-water-saturation series: The CO₂ injection pressure was maintained at 25 MPa, the CO₂ injection rate was 0.5 mL/min, and the mean core permeability was approximately 4.02 mD. The target initial water saturations were 36.59, 44.24, 51.83, and 62.04%. These saturations were established by controlling the volume of formation water displaced during the oil-saturation step.

(3)

Permeability series: The CO₂ injection pressure was maintained at 25 MPa, and the CO₂ injection rate was 0.5 mL/min. Long-core models with permeabilities of 0.62, 11.81, 86.13, and 125.84 mD were tested.

(4)

Heterogeneity series: Homogeneous and heterogeneous long-core models with permeability ratios of 1.00, 6.32, 19.64, and 32.74 were tested at a CO₂ injection pressure of 25 MPa and a CO₂ injection rate of 0.5 mL/min. The permeability ratio was calculated as kmax/kmin for the two permeability classes used in each heterogeneous model.

2.4.3. Slim-Tube MMP Experiment

Before interpreting the long-core displacement results, high-pressure slim-tube displacement experiments were conducted to independently determine the miscibility conditions between CO₂ and crude oil. The experimental temperature was 120 °C, close to the reservoir temperature of 121.6 °C. CO₂ was used as the injection medium, and the crude oil used in the slim-tube experiment was obtained from the same target subarea as that used in the long-core flooding experiments. The experiments were conducted in accordance with SY/T 6573-2016, Test Method for Minimum Miscibility Pressure-Slim Tube Method. The experimental apparatus is shown in Figure 4, and the main test parameters are summarized in

Table 2. Main parameters of the high-pressure slim-tube minimum miscibility pressure test.

Parameter	Value
Experimental temperature	120 °C
Injection gas	CO2
Model dimensions	Φ3.8 × 300 cm
CO2 injection rate	0.5 mL/min
Pressure points	20, 25, 30, 35, 40, and 45 MPa
Porosity	15.6%
Pore volume	530.7 mL
Permeability	50 mD
Final injection volume	1.2 PV

.

(1)

Before each experiment, the slim-tube model and connecting lines were flushed with petroleum ether until no visible crude oil was present in the effluent; the system was then dried with nitrogen and pressure-tested for leaks.

(2)

The model temperature was maintained at 120 °C, and the model was saturated with the selected crude oil from Block A. Crude-oil injection was continued beyond 2 PV until no detectable difference remained between the properties of the effluent and those of the injected crude oil. The system was then adjusted to the target pressure and maintained until temperature and pressure equilibrium was reached.

(3)

CO₂ was injected using a high-pressure ISCO pump at a rate of 0.5 mL/min. Oil production was recorded at intervals of 0.1 PV. Each experiment was terminated when the cumulative CO₂ injection reached 1.2 PV. After completing one pressure point, the model and lines were cleaned and resaturated before the next pressure-point experiment.

(4)

The experiment was repeated at 20, 25, 30, 35, 40, and 45 MPa. The relationship between oil recovery and pressure at an injection volume of 1.2 PV was plotted. According to the slim-tube criterion, the data below and above 90% oil recovery were separately subjected to least-squares linear fitting, and the intersection of the two fitted straight lines was determined as the MMP.

The relationship between oil recovery and CO₂ injection pressure for the selected Block A crude oil, together with the piecewise linear determination of the MMP, is shown in Figure 5.

2.5. Response Surface Experimental Design and Statistical Analysis

2.5.1. Box–Behnken Experimental Design

A four-factor, three-level Box–Behnken design was used to quantify the combined effects of injection pressure (A), initial water saturation (B), permeability (C), and permeability ratio (D) on the CO₂-flooding oil-recovery factor. The low, center, and high levels were selected from the ranges covered by the single-factor experiments (

Table 3. Factors and coded levels used in the Box–Behnken design.

Factor	Symbol	−1	0	+1
Injection pressure (MPa)	A	20.00	27.50	35.00
Initial water saturation (%)	B	36.590	49.315	62.040
Permeability (mD)	C	0.62	63.23	125.84
Permeability ratio	D	1.000	16.855	32.710

). The design contained 29 runs, including five replicated center points used to estimate the experimental error. The measured cumulative oil-recovery factor was selected as the response variable Y.

2.5.2. Regression Analysis and Model Validation

The Box–Behnken data were fitted using a second-order polynomial model containing linear, quadratic, and two-factor interaction terms:

$$Y={\beta }_0+{\displaystyle \sum _{i=1}^4}{\beta }_i{X}_i+{\displaystyle \sum _{i=1}^4}{\beta }_{ii}{X_i}^2+{\displaystyle \sum _{i=1}^3}{\displaystyle \sum _{j=i+1}^4}{\beta }_{ij}{X}_i{X}_j$$

(4)

where $Y$ is the predicted oil-recovery factor, ${\beta }_0$ is the intercept, ${\beta }_i$ is the linear coefficient, ${\beta }_{ii}$ is the quadratic coefficient, ${\beta }_{ij}$ is the interaction coefficient, and $X_i$ and $X_j$ are the coded independent variables. Analysis of variance was used to evaluate model and term significance through the F-value and p-value. Terms with p < 0.05 were considered statistically significant. The model fit was assessed using the coefficient of determination (R²) and adjusted R². Agreement between the measured and predicted recovery factors was evaluated using a parity plot. The validated regression equation was then used to construct response surfaces and to determine the factor combination that maximized the predicted CO₂-flooding recovery.

3. Results and Discussion

3.1. Water-Sensitivity Characterization and Permeability-Damage Evaluation

3.1.1. Mineral Composition and Mineralogical Basis of Water Sensitivity

The XRD pattern of the target core is dominated by the characteristic quartz reflection at d = 3.3410 Å. K-feldspar, plagioclase, illite, kaolinite, and pyrite were also identified. As shown in Figure 6, the well-resolved phase matches indicate that the core consists of stable framework minerals together with clay minerals capable of hydration, dispersion, and migration.

The quantitative phase composition is listed in

Table 4. Mineral composition of the tested core.

Core ID	Mineral Content (%)
Core ID	Illite	Kaolinite	Quartz	K-Feldspar	Plagioclase	Pyrite
1	9.0	8.0	57.4	10.7	12.2	2.7

. Quartz accounts for 57.4%, while K-feldspar and plagioclase account for 10.7% and 12.2%, respectively. The combined quartz–feldspar framework content is therefore 80.3%, indicating that the core is a quartz-rich feldspathic sandstone. Pyrite accounts for 2.7% and is not a dominant contributor to clay activation or pore-throat plugging during water flooding. Although the high framework-mineral content maintains the overall load-bearing stability of the rock, flow capacity in low-permeability sandstone is controlled by a limited number of connected pore throats. Consequently, localized clay destabilization can cause a pronounced permeability loss even when the bulk rock volume changes only slightly.

Illite and kaolinite account for 9.0% and 8.0%, respectively, giving a total clay–mineral content of 17.0%. This value exceeds the 6.78–14.14% range reported by Zhou et al. [28] for natural low-permeability sandstone cores. This comparison indicates that the target core contains a substantial abundance of water-sensitive clay minerals. However, the damage severity is governed not only by the total clay content but also by the location of clay particles on pore walls and within pore throats and by their stability upon water contact. Based on the combined XRD results and SEM morphologies, the book-like and stacked aggregates shown in Figure 7a are assigned to kaolinite, whereas the thin, curled particles shown in Figure 7b are assigned to illite. Illite has a relatively high specific surface area and readily undergoes hydration and dispersion upon contact with low-salinity water. Kaolinite is less expansive, but weakening of the attachment between its lamellae and the framework surface promotes detachment and the generation of mobile fines. Together, these two clay minerals establish a coupled water-sensitivity mechanism involving hydration and dispersion, particle detachment, migration, and pore-throat plugging.

3.1.2. Pore-Structure Evolution During Water Flooding

Before water flooding, Figure 8a shows open intergranular pores and interconnected dark pore spaces at the 10 μm scale. At the 2 μm scale, Figure 8c shows that platy clay occurs mainly as discontinuous coatings or localized aggregates attached to grain surfaces and pore walls, rather than as continuous fillings across the pore space. Thus, although the overall pore size of the core is small, a limited number of critical pore throats remain effectively connected. These narrow pathways principally control the flow capacity of the low-permeability core.

Figure 8b,d show that, after water flooding, the original open pore network was transformed into a heterogeneous structure characterized by clay coverage, particle accumulation, and local compaction. Platy and curled clay aggregates extensively covered the grain surfaces and occupied the intergranular pores, dividing continuous dark pore spaces into smaller isolated voids and markedly reducing both pore boundaries and the effective flow cross-sectional area. At higher magnification, clay platelets were densely stacked at pore-throat constrictions and bridged opposing pore walls, directly filling the narrow passages. The structural alteration induced by water flooding was therefore not uniform compaction but preferential damage at the critical pore throats that control connectivity. Consequently, limited local plugging can produce a permeability loss that is disproportionate to the change in porosity.

These morphologies reveal a continuous water-sensitivity damage pathway. Low-salinity water first disrupts the original water–rock ionic equilibrium, causing illite aggregates to hydrate and disperse while weakening the attachment of kaolinite lamellae and other fine particles to the sandstone framework. The released particles subsequently migrate through larger pores and undergo mechanical retention, deposition, and bridging where the particle size approaches the local throat width. Continued particle capture ultimately converts open pore throats into constricted channels filled with fines, thereby reducing the effective pore-throat radius and network connectivity [28,29]. Thus, locally exposed grain surfaces produced by particle removal and downstream pore-throat plugging may occur simultaneously, representing the release and redeposition ends of the same fines-migration process.

3.1.3. Pore-Size Redistribution After Water Flooding

Figure 9 compares the measured pore-size distributions of the water-sensitive core before and after water flooding. Before water flooding, the effective flow space was mainly composed of 2–3 μm intergranular pores and their connecting pathways. After water flooding, clay hydration, fines migration, and localized bridging subdivided part of the larger connected pores into residual microvoids of 0–0.5 μm and reduced the contribution of pores larger than 2 μm to connected flow. The primary consequence of water flooding was therefore not a simple reduction in total pore volume but a change in pore-throat size distribution and connectivity toward a greater abundance of small pores, fewer effective large pores, and poorer network connectivity. Consistent with the SEM observations, the original 2–3 μm flow pathways were converted by clay filling and particle bridging into more numerous but less connected residual voids. Under these conditions, the core may retain appreciable pore volume, while the number of effective pore throats spanning the inlet and outlet decreases substantially. This behavior explains the typical response of low-permeability sandstone, in which water injection produces only a limited porosity change but a marked permeability loss.

3.1.4. Permeability-Damage Response

The temperature of 120 °C was selected to approximate the reported reservoir temperature of 121.6 °C. To link the mineralogical and pore-throat changes to the macroscopic flow response, Figure 10 compares the results of stepwise salinity-reduction flooding tests conducted at 45 and 120 °C. The permeability measured with MFB was taken as the baseline. As the injected fluid was sequentially changed to 75% MFB, 50% MFB, 25% MFB, and DW, the cumulative permeability-damage degree increased progressively at both temperatures.

At 45 °C, the cumulative permeability-damage degrees after the injection of 75% MFB, 50% MFB, 25% MFB, and DW were 27.64%, 44.73%, 58.36%, and 63.09%, respectively. The corresponding damage degrees at 120 °C were 35.66%, 55.11%, 65.29%, and 68.89%, respectively. At both temperatures, the first reduction in salinity produced the largest increment in permeability damage. This behavior was consistent with the rapid disruption of the ionic equilibrium at clay–mineral surfaces, followed by clay hydration, fines release, and particle migration. As the salinity decreased further, the released fines were retained, deposited, and bridged at pore-throat constrictions, progressively reducing the effective pore-throat radius and connectivity.

At all diluted-brine and DW stages, the permeability-damage degrees measured at 120 °C were higher than those measured at 45 °C. Previous high-temperature sensitivity experiments have also shown that elevated temperature may promote pore-throat compression under confining pressure, intensify clay–mineral hydration and dispersion, accelerate fines detachment and migration, and promote mineral dissolution, precipitation, and transformation. Temperature-induced changes in rock wettability and water-film thickness may further restrict effective flow pathways [31]. These mechanisms provide a reasonable explanation for the greater permeability damage observed at 120 °C. Therefore, water-sensitivity test results obtained at lower temperatures may underestimate the actual degree of permeability damage under reservoir-temperature conditions.

Combined with the XRD and SEM observations, these results confirm that the target sandstone is highly susceptible to water-sensitive permeability damage. The combined content of illite and kaolinite reached 17.0%, providing the mineralogical basis for water sensitivity, while the SEM images obtained after water flooding revealed clay-particle detachment, migration, accumulation, and pore-throat blockage. Consequently, continued injection of low-salinity water is unfavorable because it increases flow resistance, reduces injectivity, and restricts the effective swept volume. CO₂ flooding does not continuously introduce an external aqueous phase and therefore reduces further disturbance of the water–rock ionic equilibrium and additional fines-induced pore-throat damage. However, CO₂ flooding cannot eliminate the effects of residual water or pre-existing water-sensitive damage. Therefore, CO₂ flooding can be considered a potentially more suitable displacement strategy for the development of this highly water-sensitive, low-permeability reservoir.

3.2. Division of Oil Displacement Process

Figure 11 illustrates the stagewise evolution of CO₂ flooding in the low-permeability reservoir of Block A, East China. Based on the oil production rate, produced-oil density, and gas–oil ratio (GOR), the displacement process was divided into three stages. Stage I represents initial stable displacement, Stage II gas breakthrough, and Stage III stable gas channeling. The blue vertical lines mark the boundaries between these stages. Previous CO₂ core-flooding studies used a producing GOR of approximately 500 mL/mL as an operational indicator of gas channeling, and gas breakthrough and channeling were identified from the abrupt increase in GOR and the corresponding changes in oil-production dynamics [32,33]. In the present experiments, the first sustained increase in GOR to 400–600 mL/mL, observed consistently across multiple dynamic curves, was taken to indicate gas breakthrough. Stable gas channeling was identified when the GOR remained above 600 mL/mL, the oil-production rate remained low, and the produced-oil density gradually approached a plateau. As shown in Figure 11a, the oil-production-rate profiles at 20 and 30 MPa exhibited similar trends, with no pronounced difference in the transition points between displacement stages. In contrast, at 35 MPa, a relatively high oil production rate was maintained over a larger cumulative CO₂ injection volume. Figure 11b shows that the produced-oil density exhibited stagewise variations corresponding to the oil-production-rate response. During the initial stable-displacement stage, the produced-oil density remained nearly constant, and oil production was primarily driven by the pressure gradient. After CO₂ breakthrough, the produced-oil density decreased continuously as CO₂–oil contact and mass transfer intensified. This behavior was consistent with the preferential extraction and transport of lighter hydrocarbon components by CO₂. During the stable gas-channeling stage, the density declined gradually slowed and eventually approached a plateau, while the oil production rate remained low. The produced-oil-density profiles at 20 and 30 MPa were similar. At 35 MPa, however, the density declined more gradually, while both the pronounced density decreased and the rapid oil-rate decline shifted toward larger injected pore volumes. These results suggest that 35 MPa prolonged the stable-displacement stage and delayed gas breakthrough and subsequent gas channeling.

3.3. Impacts of Different Factors on Recovery Efficiency

3.3.1. Injection Pressure

Figure 12 shows the effect of injection pressure on CO₂ flooding performance in the long-core models. As the injection pressure increased from 20 to 35 MPa, oil recovery increased from 40.78% to 57.91%. Relative to the measured MMP of 32.70 MPa, the flooding experiments at 20, 25, and 30 MPa were conducted below the MMP, whereas 35 MPa was slightly above the MMP, indicating that the CO₂–oil system had the thermodynamic conditions required to develop multiple-contact miscibility. According to Table 1, the permeability and initial water saturation of the model used at 30 MPa were 6.43 mD and 50.85%, respectively, whereas the model used at 35 MPa had a lower permeability of 4.32 mD and a higher initial water saturation of 55.25%. Nevertheless, higher oil recovery was obtained at 35 MPa. This comparison indicates that although residual differences in core properties may have affected individual oil-recovery values, the overall recovery trend was primarily governed by the increase in injection pressure when considered together with the MMP results. Below the MMP, increasing pressure enhanced CO₂ dissolution in oil, promoted oil swelling, reduced oil viscosity, and strengthened the extraction of light components [34]. At 35 MPa, these effects were further combined with the contribution from the transition toward multiple-contact miscible flooding, thereby improving the final oil recovery.

The GOR curves in Figure 12b show that pronounced gas production occurred after CO₂ breakthrough at all pressure levels. Higher pressure increased CO₂ dissolution in oil during the early stage of flooding and reduced free-gas production during the initial displacement stage. As injection continued, gas channeling developed and caused a rapid increase in the GOR. As shown in Figure 12c, the differential pressure generally decreased with increasing cumulative CO₂ injection volume, and the pronounced decreases observed in several experiments approximately coincided with the rapid increase in the GOR. However, the differential pressure did not vary monotonically with injection pressure, indicating that residual differences in core permeability, porosity, initial water saturation, and water-sensitivity damage also affected the measured pressure response.

3.3.2. Water Saturation

The influence of initial water saturation on oil recovery of CO₂ flooding is investigated, and the results are shown in Figure 13. In Figure 13a, low-water-saturation cores show slow initial recovery efficiency but reach a higher final recovery of 52.13% at 36.59% water saturation. High-water-saturation cores have faster initial recovery but drop to 44.21% at 62.04% water saturation. The lower recovery at higher initial water saturation is consistent with a smaller oil-filled pore volume and potentially reduced oil-phase connectivity, which may have limited effective CO₂–oil contact and displacement. As shown in Figure 13b, the GOR increased after CO₂ breakthrough in all the experiments. The higher-water-saturation models exhibited a faster early GOR increase but a lower final GOR, whereas the lower-water-saturation models showed a relatively delayed GOR response. These differences may reflect the combined effects of the initial fluid distribution and core-specific flow pathways and should not be regarded as direct evidence of high-permeability water channels. Figure 13c shows that the initial differential pressure generally increased with initial water saturation, reaching 2396 kPa at the highest water saturation before decreasing with continued CO₂ injection. The higher differential-pressure response may be associated with increased water-related flow resistance, reduced oil-phase connectivity, and possible water-sensitive pore-throat restriction. The approximate coincidence between the differential-pressure decline and the rapid GOR increase is consistent with CO₂ breakthrough and the subsequent development of preferential gas-flow pathways. Overall, increasing the initial water saturation reduced the final oil recovery, although residual differences among the natural cores may also have affected the individual dynamic responses.

3.3.3. Permeability

The influence of permeability on oil recovery of CO₂ flooding is investigated, and the results are shown in Figure 14. In Figure 14a,d, when permeability increases from 0.62 mD to 125.84 mD, the recovery factor rises slightly from 53.23% to 55.81%. However, significant differences exist in the recovery dynamics across permeability levels. At 0.62 mD, water sensitivity causes clay expansion, blocking pores and trapping residual oil as adsorbed films or isolated droplets. Injected CO₂ is compressed at the core inlet. As permeability increases, the gas compression phase shortens. At 125.84 mD, larger pore throats enhance CO₂–oil contact, almost eliminating gas compression. Despite initial production differences, similar final oil recovery occurs due to the good mobility of light oil. In Figure 14b, gas channeling happens earlier as permeability rises. Lower-permeability cores show steeper GOR increases, while higher-permeability cores exhibit gentler GOR rises. However, the final GOR tends to be similar across permeability levels. High-permeability cores have wider pore throats, allowing easier CO₂ flow and viscous fingering, shortening gas channeling time. In contrast, low-permeability cores with complex pore structures require CO₂ to overcome higher capillary and viscous forces, slowing gas propagation and delaying channeling. Figure 14c illustrates that lower-permeability cores have higher CO₂-flooding pressure differentials. At 0.62 mD, the initial pressure differential reaches 4446.22 kPa. Lower permeability indicates poorer reservoir fluid-flow properties. As permeability increases, the pressure differential decreases since CO₂ encounters less resistance in high-permeability reservoirs [35].

3.3.4. Heterogeneous Cores

Figure 15 compares the CO₂-flooding responses of the homogeneous model and heterogeneous models with different permeability ratios. As shown in Figure 15a,d, the final oil recovery generally decreased with increasing permeability ratio. When the permeability ratio increased from 1.00 to 32.74, the final oil recovery decreased from 54.51% to 41.67%. However, this relationship was not strictly monotonic. Oil recovery reached a minimum at a permeability ratio of 19.64 and then increased slightly at 32.74 while remaining below that of the homogeneous model. For the two models with higher permeability ratios, the early period of slow recovery growth shortened markedly and nearly disappeared at ratios of 19.64 and 32.74. This behavior may be attributed to the greater permeability contrast, which intensified displacement nonuniformity. Consequently, CO₂ migrated more readily through high-permeability zones, whereas the oil in low-permeability zones remained poorly swept, resulting in limited incremental recovery during the later displacement stage.

As shown in Figure 15b, the gas–oil ratio (GOR) increased rapidly in all the models after CO₂ breakthrough. However, the onset of the pronounced GOR increase did not vary monotonically with permeability ratio. The heterogeneous models generally exhibited higher late-stage GOR values, indicating more pronounced gas production after CO₂ breakthrough. Figure 15c shows that the pressure drop across the core decreased with increasing permeability ratio. At a permeability ratio of 32.74, the initial pressure drop was approximately 410.64 kPa. This result indicates that a greater permeability contrast altered the internal flow-resistance distribution and facilitated CO₂ transport through the high-permeability zones, thereby reducing the overall displacement resistance [36].

3.4. Response Surface Analysis

3.4.1. Model Development and Analysis of Variance

A recovery-prediction model was developed to identify the optimal combination of operating and reservoir parameters. A four-factor, three-level Box–Behnken design was used to quantify the individual and coupled effects of injection pressure (A), initial water saturation (B), permeability (C), and permeability ratio (D) on oil recovery. The design comprised 29 experimental runs, including five replicates at the center point.

Table 5. Experimental design scheme and results.

Std	Run	Injection Pressure (MPa)	Initial Water Saturation (%)	Permeability (mD)	Permeability Ratio	Oil Recovery (%)
1	26	20.0	36.590	63.23	16.855	54.54
2	17	35.0	36.590	63.23	16.855	58.18
3	18	20.0	62.040	63.23	16.855	43.21
4	16	35.0	62.040	63.23	16.855	55.30
5	7	27.5	49.315	0.62	1.000	49.83
6	9	27.5	49.315	125.84	1.000	54.48
7	24	27.5	49.315	0.62	32.710	41.50
8	28	27.5	49.315	125.84	32.710	44.81
9	20	20.0	49.315	63.23	1.000	50.20
10	3	35.0	49.315	63.23	1.000	60.20
11	1	20.0	49.315	63.23	32.710	44.30
12	29	35.0	49.315	63.23	32.710	50.20
13	15	27.5	36.590	0.62	16.855	50.46
14	10	27.5	62.040	0.62	16.855	43.36
15	12	27.5	36.590	125.84	16.855	54.11
16	8	27.5	62.040	125.84	16.855	48.96
17	21	20.0	49.315	0.62	16.855	49.20
18	23	35.0	49.315	0.62	16.855	53.80
19	27	20.0	49.315	125.84	16.855	50.10
20	6	35.0	49.315	125.84	16.855	58.60
21	11	27.5	36.590	63.23	1.000	55.26
22	19	27.5	62.040	63.23	1.000	46.27
23	22	27.5	36.590	63.23	32.710	45.47
24	2	27.5	62.040	63.23	32.710	41.20
25	13	27.5	49.315	63.23	16.855	54.63
26	4	27.5	49.315	63.23	16.855	58.63
27	25	27.5	49.315	63.23	16.855	56.66
28	5	27.5	49.315	63.23	16.855	55.60
29	14	27.5	49.315	63.23	16.855	57.10

presents the BBD matrix and the measured oil-recovery responses; the standard and randomized run numbers are both retained to ensure traceability.

A quadratic model was selected to describe oil recovery. The model was statistically significant (F = 61.62, p < 0.0001), whereas the lack of fit was not significant (F = 0.213, p = 0.9784), indicating no detectable systematic lack of fit relative to the pure experimental error. The model yielded R² = 0.9840, adjusted R² = 0.9681, predicted R² = 0.9518, a coefficient of variation of 1.96%, and an Adeq Precision value of 26.215. The close agreement between the adjusted and predicted R² values, together with the adequate signal-to-noise ratio, supports the use of the model within the investigated factor ranges [37,38]. The fitted regression equation is given by Equation (5). The analysis of variance for the quadratic oil-recovery model is summarized in

Table 6. RSM model for the oil recovery.

Source	Sum of Squares	df	Mean Square	F-Value	p-Value
Model	871.29	14	62.24	61.62	<0.0001
A—Injection pressure	166.73	1	166.73	165.08	<0.0001
B—Water saturation	131.47	1	131.47	130.17	<0.0001
C—Permeability	43.74	1	43.74	43.31	<0.0001
D—Permeability ratio	198.13	1	198.13	196.16	<0.0001
AB	17.85	1	17.85	17.67	0.0009
AC	3.80	1	3.80	3.76	0.0728
AD	4.20	1	4.20	4.16	0.0607
BC	0.95	1	0.95	0.94	0.3484
BD	5.57	1	5.57	5.51	0.0341
CD	0.45	1	0.45	0.44	0.5158
A2	0.031	1	0.031	0.031	0.8632
B2	97.12	1	97.12	96.16	<0.0001
C2	79.83	1	79.83	79.04	<0.0001
D2	192.28	1	192.28	190.37	<0.0001
Residual	14.14	14	1.01
Lack of fit	4.91	10	0.4914	0.213	0.9784
Pure error	9.23	4	2.31
Cor total	885.43	28

.

In terms of coded factors, the fitted oil-recovery model is as follows:

$$\begin{array}{c}\mathrm{Y}=56.524+3.7275\mathrm{A}-3.3100\mathrm{B}+1.90917\mathrm{C}-4.06333\mathrm{D}+2.1125\mathrm{AB}+0.9750\mathrm{AC}\\ -1.0250\mathrm{AD}+0.4875\mathrm{BC}+1.1800\mathrm{BD}-0.3350\mathrm{CD}\\ +0.06925{\mathrm{A}}^2-3.8695{\mathrm{B}}^2-3.50825{\mathrm{C}}^2-5.4445{\mathrm{D}}^2\end{array}$$

(5)

$$\mathrm{A}={\displaystyle \frac{\mathrm{P}-27.5}{7.5}}, \mathrm{B}={\displaystyle \frac{{\mathrm{S}}_{\mathrm{w}}-49.315}{12.725}}, \mathrm{C}={\displaystyle \frac{\mathrm{K}-63.23}{62.61}}, \mathrm{D}={\displaystyle \frac{{\mathrm{R}}_{\mathrm{k}}-16.855}{15.855}}$$

where Y is oil recovery (%); A, B, C, and D are the coded values of injection pressure, initial water saturation, permeability, and permeability ratio, respectively; P is injection pressure (MPa); S_w is initial water saturation (%); K is permeability (mD); and R_k is the permeability ratio.

All four linear terms were statistically significant (p < 0.0001). Based on the F-values of the coded main effects, their relative influence followed the order D > A > B > C, indicating that the permeability ratio exerted the strongest main effect within the investigated ranges. Among the two-factor interaction terms, only injection pressure × initial water saturation (AB, p = 0.0009) and initial water saturation × permeability ratio (BD, p = 0.0341) were significant at the 0.05 level; the AC, AD, BC, and CD interactions were not significant. The significant negative quadratic terms B², C², and D² produced curvature in the corresponding response surfaces, whereas A² was not significant.

Figure 16 compares the measured and model-predicted oil recoveries. The data points were distributed close to the y = x line, indicating good agreement between the measured and fitted responses. Together with the high predicted R² and nonsignificant lack of fit, this result indicates that the quadratic model adequately describes oil recovery within the experimental design space.

3.4.2. Response Surface and Interaction Analysis

The pairwise response surfaces are shown in Figure 17a–f, with the two unplotted factors fixed at their center levels. Figure 17a shows that oil recovery generally increased with injection pressure and initially increased and then decreased with increasing initial water saturation, indicating that the oil-recovery response to injection pressure varied with initial water saturation. With permeability and permeability ratio fixed at their center levels, increasing the injection pressure from 20 to 35 MPa increased oil recovery by 3.64 percentage points at an initial water saturation of 36.590%. At an initial water saturation of 62.040%, the same pressure increase raised oil recovery by 12.09 percentage points. These results indicate that the greater pressure effect was associated with the near-miscible and miscible conditions of the CO₂–oil system. Increasing pressure enhanced CO₂–oil mass transfer, oil swelling, and viscosity reduction, thereby partially offsetting water occupancy, reduced oil-phase connectivity, and possible water-sensitive pore-throat restrictions. However, it did not eliminate the effects of residual water or pre-existing water-sensitive damage [28,29,30]. Figure 17b,c mainly reflect the fitted main and quadratic effects of injection pressure, permeability, and permeability ratio. Within the investigated ranges, increasing pressure increased oil recovery, whereas the pressure × permeability and pressure × permeability-ratio interactions were not statistically significant.

The dome-shaped surface in Figure 17d is governed by the quadratic effects of initial water saturation and permeability. At low permeability, high flow resistance and possible water-sensitive pore-throat restriction may have reduced CO₂ injectivity and sweep efficiency. Recovery increased as permeability rose to an intermediate level; at higher permeability, accelerated CO₂ transport may have shortened the effective CO₂–oil contact time and promoted earlier breakthrough. Figure 17e shows a significant interaction between the initial water saturation and permeability ratio. With injection pressure and permeability fixed at their center levels, increasing the permeability ratio from 1.000 to 32.710 decreased oil recovery from 55.26% to 45.47%, corresponding to a reduction of 9.79 percentage points at an initial water saturation of 36.590%. At an initial water saturation of 62.040%, the same increase in permeability ratio reduced oil recovery from 46.27% to 41.20%, corresponding to a reduction of 5.07 percentage points. These results indicate that the effect of permeability ratio on oil recovery varied with initial water saturation. At lower initial water saturation, increasing the permeability ratio intensified preferential CO₂ flow through high-permeability zones and reduced the sweep of low-permeability zones, thereby causing a greater loss in oil recovery. At higher initial water saturation, the water occupancy, reduced oil-phase connectivity, and possible water-sensitive pore-throat restrictions had already lowered the recovery baseline, reducing the marginal loss caused by a further increase in permeability ratio. Figure 17f shows that a moderate permeability increase improved injectivity, while excessive heterogeneity reduced sweep efficiency. Because the permeability × permeability-ratio interaction was not significant, the surface curvature is attributed to their individual quadratic effects rather than to a statistically supported interaction.

3.4.3. Numerical Optimization

Numerical maximization of the fitted model over the investigated factor ranges produced a model-predicted optimum at an injection pressure of 35.00 MPa, an initial water saturation of 46.71%, a permeability of 89.58 mD, and a permeability ratio of 8.89, corresponding to a predicted oil recovery of 62.33%. Because the injection pressure reached the upper boundary of the experimental range, the model does not support extrapolation beyond 35 MPa.

3.4.4. Experimental Confirmation of the Predicted Optimum

To experimentally confirm the RSM-predicted optimum, an additional CO₂-flooding experiment was conducted. The target conditions were an injection pressure of 35.00 MPa, an initial water saturation of 46.71%, a permeability of 89.58 mD, and a permeability ratio of 8.89. Because the properties of natural cores could not exactly match the target values, the parameters used in the confirmation experiment are presented in

Table 7. Experimental confirmation of the RSM-predicted optimum.

Condition	Pressure (MPa)	Initial Water Saturation (%)	Permeability (mD)	Permeability Ratio	Predicted Recovery (%)	Experimental Recovery (%)	Relative Error (%)
Model-predicted optimum	35.00	46.71	89.58	8.89	62.33	—	—
Confirmation experiment	35.00	46.80	86.59	8.64	62.32	60.42	3.05

. The experimental apparatus, fluid-preparation procedure, CO₂ injection rate, temperature, measurement interval, and termination criterion were consistent with those described in Section 2.4.2. Based on the actual factor levels, the fitted RSM model predicted an oil recovery of 62.32%, whereas the experimentally measured oil recovery was 60.42%. The absolute deviation between the two values was 1.90 percentage points, corresponding to a relative error of 3.05%. The agreement between the predicted and experimental values supports the predictive accuracy of the RSM model near the optimum within the investigated design space.

4. Conclusions and Implications

This study systematically investigated the CO₂-flooding characteristics of the highly water-sensitive, low-permeability reservoir in Block A, eastern China, through XRD, SEM, water-sensitivity tests, long-core flooding experiments, and response surface methodology. The main conclusions are as follows:

(1)

XRD analysis showed that the sandstone contained 17.0% clay minerals, comprising 9.0% illite and 8.0% kaolinite. SEM observations showed that water flooding promoted clay-particle detachment, migration, accumulation, and localized pore-throat blockage. During stepwise salinity-reduction flooding, the final cumulative permeability damage was 63.09% at 45 °C and 68.89% at 120 °C, confirming the high water sensitivity of the target sandstone and the greater permeability damage under reservoir-temperature conditions.

(2)

The CO₂-flooding process was divided into three stages: the initial stable-displacement stage, gas-breakthrough stage, and stable gas-channeling stage. During the initial stage, the oil-production rate remained relatively high, while the produced-oil density changed only slightly. After CO₂ breakthrough, the oil-production rate declined, and the produced-oil density decreased progressively because of intensified CO₂–oil contact and mass transfer. During stable gas channeling, the oil-production rate remained low, while the produced-oil density gradually approached a plateau.

(3)

Injection pressure, initial water saturation, permeability, and reservoir heterogeneity affected CO₂-flooding performance in different ways. Increasing injection pressure from 20 to 35 MPa increased oil recovery from 40.78% to 57.91%. Increasing initial water saturation from 36.59% to 62.04% reduced recovery from 52.13% to 44.21%. Increasing permeability from 0.62 to 125.84 mD raised recovery only slightly, from 53.23% to 55.81%. Increasing the permeability ratio produced an overall recovery reduction from 54.51% to 41.67%, reflecting increasingly uneven displacement and reduced sweep of the low-permeability regions.

(4)

RSM analysis showed that the permeability ratio exerted the strongest main effect on oil recovery, followed by injection pressure, initial water saturation, and permeability. Significant interactions were observed between injection pressure and initial water saturation and between initial water saturation and the permeability ratio. The model predicted a maximum recovery of 62.33% at an injection pressure of 35.00 MPa, an initial water saturation of 46.71%, a permeability of 89.58 mD, and a permeability ratio of 8.89. An independent confirmation experiment yielded an oil recovery of 60.42%, compared with the model prediction of 62.32% under the actual confirmation conditions, corresponding to an absolute deviation of 1.90 percentage points and a relative error of 3.05%. Therefore, CO₂-flooding design for highly water-sensitive, low-permeability reservoirs should balance pressure enhancement against water-sensitive damage and heterogeneity-induced gas channeling.