Oil reservoir water flooded layer intelligent recovery system based on spiking neural network

Abstract

The invention provides an intelligent reservoir water flooded layer recovery system based on a spiking neural network, which belongs to the technical field of oil exploration and development and comprises a data acquisition module, a data preprocessing module, a model training module, a resistivity recovery module and a water flooded layer interpretation module. The data acquisition module is used for acquiring multi-modal data such as conventional and electrical logging curves; the data preprocessing module is used for carrying out range normalization and Z-Score standardization on the data; the model training module constructs a resistivity recovery model by using a pulse neural network, and the resistivity recovery model is optimized and verified in various modes; the resistivity recovery module recovers the original resistivity accordingly; and the water-flooded layer interpretation module determines and interprets the water-flooded layer in a grading manner based on the calculation related parameters. According to the intelligent recovery system for the oil reservoir water flooded layer based on the spiking neural network, the resistivity of the water flooded layer is recovered with high precision, the water flooding influence is accurately quantified, and a scientific and reliable basis is provided for oil reservoir development and adjustment.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of oil exploration and development, and particularly relates to an intelligent recovery system for watered-out layers of an oil reservoir based on a pulse neural network. BACKGROUND

[0002] With the gradual deepening of the global oilfield development process, many oilfields have entered the high water cut development stage and the production decline period, and therefore, the watered-out layer interpretation technology has become a core link for improving the recovery rate and optimizing the development strategy in the middle and late stages of oil reservoir development. However, the traditional well logging interpretation model is mostly constructed based on the assumption of static formation parameters, and it is difficult to fully adapt to the significant increase in reservoir heterogeneity in the water drive process and the highly nonlinear characteristics of the resistivity response. This inadaptability leads to a deviation between the resistivity curve and the true water saturation, and further makes the interpretation result have significant multiple solutions, and it is difficult to accurately reflect the actual dynamics of the oil reservoir.

[0003] Especially noteworthy is that when the watered-out layer presents the complex geological phenomenon of coexistence of "low-resistance oil layer" and "high-resistance water layer", the traditional static training mode is not up to the task. It cannot effectively capture the time sequence correlation characteristics of the resistivity dynamic evolution in the water drive process, and shows low robustness to small sample and high noise logging data, further limiting its accuracy and reliability in the interpretation of watered-out layers.

[0004] In recent years, the rise of machine learning algorithms has brought new light to the field of well logging interpretation. These algorithms, through nonlinear mapping technology, have broken through the parameterization limitations of traditional models to some extent, and have improved the interpretation ability for complex geological relationships. However, they still have limited ability in modeling dynamic processes, and there are still deficiencies in explaining complex nonlinear relationships, which makes it difficult to meet the high-precision requirements of resistivity dynamic recovery and interpretation of watered-out layers.

[0005] Therefore, developing an intelligent interpretation method with strong dynamic feature extraction capability has become an urgent need in the field of resistivity recovery and interpretation of watered-out layers. SUMMARY

[0006] The present application aims to provide an intelligent recovery system for watered-out layers of an oil reservoir based on a pulse neural network, which can accurately recover the resistivity of watered-out layers and accurately quantify the watered-out effect, and provide a scientific and reliable basis for oil reservoir development adjustment.

[0007] To achieve the above-mentioned purpose, the present application provides an intelligent recovery system for watered-out layers of an oil reservoir based on a pulse neural network, which comprises a data acquisition module, a data preprocessing module, a model training module, a resistivity recovery module, and a watered-out layer interpretation module.

[0008] The data acquisition module acquires multi-modal logging curve data, including conventional logging curve and electrical logging curve;

[0009] The data preprocessing module performs range normalization and Z-Score standardization processing on the acquired logging curve data;

[0010] The model training module constructs a resistivity recovery model based on a pulse neural network, adopts a LeakyIntegrate-and-Fire neuron model as a basic unit, receives logging curve data through a feedforward-cyclic hybrid structure, generates resistivity prediction values, adjusts synaptic weights by combining synaptic plasticity and global error backpropagation, optimizes the performance of the resistivity recovery model, and evaluates and verifies the reliability of the model through the resistivity recovery model;

[0011] The resistivity recovery module recovers the original resistivity of the oil layer according to the trained resistivity recovery model;

[0012] The watered-out layer interpretation module calculates oil saturation and water production rate based on the original resistivity, determines watered-out layers, and quantitatively classifies and interprets the watered-out layers.

[0013] Preferably, the conventional logging curve includes natural gamma, acoustic time difference, neutron porosity, density, and natural potential; and the electrical logging curve includes deep lateral resistivity and shallow lateral resistivity.

[0014] Preferably, the range normalization expression is:

[0015]

[0016] wherein x norm is the normalized data, x min is the minimum value in the original data, x max is the maximum value in the original data, x max -x min is the range of the data.

[0017] The Z-Score standardization expression is:

[0018]

[0019] wherein x std is the standardized data, μ is the mean of the original data, and σ is the standard deviation of the original data.

[0020] Preferably, the dynamic expression of the LeakyIntegrate-and-Fire neuron model is:

[0021]

[0022] where V is membrane potential, τ m is membrane time constant, V rest is resting potential, I syn (t) is synaptic input current;

[0023] The resistivity prediction is that when the membrane potential exceeds the threshold value V th , the neuron fires a pulse and resets to V reset , adopts a feedforward-circulation hybrid structure, the input layer receives well logging curve data, the hidden layer contains pulse neuron nodes, and the output layer generates resistivity prediction values through pulse frequency coding.

[0024] Preferably, the expression for adjusting synaptic weights is:

[0025] Δw ij = η·∑ t δ j (t)·S i (t-Δt);

[0026] where η is the learning rate, δ j (t) is the error gradient, Δw ij is the change in synaptic weight, S i (t-Δt) is the pulse state before Δt time;

[0027] The expression for evaluating the resistivity recovery model is:

[0028]

[0029] where RMSE is the root mean square error, MAE is the mean absolute error, R 2 is the determination coefficient, is the predicted value of the i-th resistivity, is the true value of the i-th resistivity.

[0030] Preferably, the expression for oil saturation is:

[0031]

[0032] where S o is the current oil saturation of the reservoir, RT original is the original resistivity, RT current is the current resistivity, a, m, n are coefficients related to the pore structure and saturation of the rock, φ is the effective porosity of the formation, σ w is the viscosity of water;

[0033] The expression for water production rate is:

[0034]

[0035] wherein, K rw is the relative permeability of the water phase, K ro is the relative permeability of the oil phase, μ w is the viscosity of the water phase, μ o is the viscosity of the oil phase.

[0036] Preferably, the quantitative grading is performed using ΔS o The change in oil saturation, ΔRTresistivity change, is used to quantitatively grade and interpret the watered-out layers, which are divided into four levels: unwatered, weakly watered, moderately watered, and strongly watered.

[0037] Therefore, the present application adopts the above-mentioned oil reservoir watered-out layer intelligent recovery system based on a pulse neural network, and the technical effects are as follows:

[0038] 1. High-precision resistivity recovery: By constructing a dynamic resistivity recovery model based on a pulse neural network, high-precision recovery of the resistivity of watered-out layers is achieved. The model combines the spatiotemporal dynamic characteristics of pulse neurons with rock physics constraints, effectively capturing the dynamic changes in resistivity during water flooding, and improving the accuracy of resistivity recovery.

[0039] 2. Solving the problem of electrical confusion: In view of the electrical confusion problem caused by the coexistence of "low-resistance oil layers" and "high-resistance water layers" in watered-out layers, the SNN model effectively distinguishes between low-resistance oil layers and high-resistance water layers through high-precision modeling and dynamic feature extraction, improving the accuracy of watered-out layer interpretation.

[0040] 3. Enhancing interpretation accuracy and reliability: By combining multi-modal logging curve data, the SNN model exhibits high precision and strong interpretability in resistivity recovery tasks. Model evaluation combines RMSE, MAE, R 2 and other indicators to verify its reliability and accuracy in watered-out layer interpretation.

[0041] 4. Improving the robustness of the interpretation model: The SNN model exhibits strong robustness to small sample and high noise logging data, and can maintain stable interpretation performance under complex geological conditions. BRIEF DESCRIPTION OF DRAWINGS

[0042] Figure 1 is a schematic diagram of the original oil layer resistivity recovery process based on machine learning of the present application;

[0043] Figure 2 is a schematic diagram of the SVR algorithm training result of the present application;

[0044] Figure 3 is a schematic diagram of the LightGBM algorithm training result of the present application;

[0045] Figure 4A schematic diagram of a training result of the SNN algorithm of the present application;

[0046] Figure 5 A bar chart of the resistivity recovery REMS, MAE and R2 evaluation indexes of the SNN algorithm, the SVR algorithm and the LightGBM algorithm in the training set;

[0047] Figure 6 A bar chart of the resistivity recovery REMS, MAE and R2 evaluation indexes of the SNN algorithm, the SVR algorithm and the LightGBM algorithm in the validation set;

[0048] Figure 7 A schematic diagram of a well RT prediction result of the present application;

[0049] Figure 8 A flow chart of a watered-out layer logging quantitative interpretation model construction of the present application;

[0050] Figure 9 A schematic diagram of a LUD3245 logging comprehensive interpretation result in the test area of the present application;

[0051] Figure 10 A schematic diagram of a LUD3264 logging comprehensive interpretation result in the test area of the present application;

[0052] Figure 11 A schematic diagram of a resistivity prediction curve of 8 encrypted wells of the present application;

[0053] Figure 12 A map of the medium-strong watered-out layer thickness in the test area of the present application;

[0054] Figure 13 A map of the medium-strong watered-out degree in the test area of the present application. DETAILED DESCRIPTION

[0055] The technical solutions of the present application are further described below through the drawings and examples.

[0056] Unless otherwise defined, the technical terms or scientific terms used in the present application shall have the usual meanings understood by those skilled in the art to which the present application belongs.

[0057] Example One

[0058] As shown in Figure 1 , the present application provides an intelligent recovery system for watered-out layers in oil reservoirs based on a pulse neural network, comprising a data acquisition module, a data preprocessing module, a model training module, a resistivity recovery module and a watered-out layer interpretation module.

[0059] The data acquisition module acquires multi-modal logging curve data, and the following multi-modal logging curves are selected to construct a data set for representing the dynamic evolution law of the resistivity of watered-out layers:

[0060] Conventional logging curves: natural gamma ray (GR), acoustic time (AC), neutron porosity (CNL), density (DEN), spontaneous potential (SP), used to describe the static physical properties of reservoir lithology, porosity and shale content, etc.

[0061] Electrical logging curves: deep lateral resistivity (RT), shallow lateral resistivity (RXO), directly reflecting the dynamic changes of formation resistivity affected by water drive.

[0062] The data preprocessing module performs range normalization and Z-Score standardization on the collected logging curve data. The two normalization methods eliminate the dimensional differences and numerical range deviations of different logging curves, and improve the model's ability to capture physical meaning.

[0063] For the values in the logging curve that are evenly distributed and have a stable range, such as spontaneous potential (SP) and gamma ray (GR), the range normalization method is used to preserve their original distribution pattern, making them more suitable for threshold-based classification tasks. The formula is as follows:

[0064]

[0065] where x norm is the normalized data, x min is the minimum value in the original data, x max is the maximum value in the original data, and x max -x min is the range of the data.

[0066] For values in the logging curve that are approximately Gaussian distributed, such as acoustic time (HAC) and density (DEN) curves, the Z-Score standardization method is used, which is robust to outliers and has a more stable distribution after standardization, making it more suitable for statistical inference scenarios. The formula is as follows:

[0067]

[0068] where x std is the standardized data, μ is the mean of the original data, and σ is the standard deviation of the original data.

[0069] The model training module constructs a resistivity recovery model based on the pulse neural network, uses the Leaky Integrate-and-Fire neuron model as the basic unit, receives logging curve data through a mixed structure of feedforward and loop, generates resistivity prediction values, adjusts synaptic weights through synaptic plasticity and global error backpropagation, optimizes the performance of the resistivity recovery model, and evaluates the reliability of the model through the resistivity recovery model.

[0070] The dynamic expression of Leaky Integrate-and-Fire neuron model is as follows:

[0071]

[0072] Wherein, V is membrane potential, τ m is membrane time constant, V rest is resting potential, I syn (t) is synaptic input current;

[0073] The resistivity prediction is that when the membrane potential exceeds the threshold value V th , the neuron fires a pulse and resets to V reset , adopts a feedforward-cyclic hybrid structure, the input layer receives well logging curve data, the hidden layer contains pulse neuron nodes, and the output layer generates resistivity prediction value through pulse frequency coding.

[0074] For the resistivity regression task, a supervised learning algorithm based on pulse sequence inner product is adopted, and a multi-pulse error function is constructed by pulse sequence inner product:

[0075]

[0076] Wherein, S target and S output are target and output pulse sequences respectively.

[0077] Combined with synaptic plasticity and global error back propagation, the synaptic weight is adjusted, and the specific formula is as follows:

[0078] Δw ij = η·∑ t δ j (t)·S i (t-Δt);

[0079] Wherein, η is learning rate, δj(t) is error gradient, Δw ij is the change of synaptic weight, S i (t-Δt) is the pulse state before Δt time;

[0080] Taking L sandstone reservoir block as the research object, the logging data of 11 wells are selected for research, of which 80% of the logging curve data is used as the training set, and the remaining 20% is used as the verification set.

[0081] The pulse neural network exhibits high precision and strong interpretability in the resistivity recovery task of watered-out layers through spatiotemporal information coding and biological heuristic optimization strategy. Model evaluation needs to combine multiple indexes such as RMSE, MAE, R 2 , and rely on geological dynamic data to verify its reliability.

[0082] The average deviation of the predicted resistivity and the measured value is reflected by the root mean square error (RMSE), which is evaluated by the formula as follows:

[0083]

[0084] The average absolute error (MAE) is used to reflect the case of heterogeneous reservoirs or local water flooding anomalies.

[0085]

[0086] The determination coefficient (R 2 ) is used to measure the interpretation ability of the model to the change of the formation resistivity, and the formula is as follows:

[0087]

[0088] The closer the value is to 1, the higher the fitting degree of the model to key factors such as lithology and pore fluid.

[0089] Wherein, RMSE is the root mean square error, which measures the deviation between the predicted value and the true value, and is sensitive to outliers; MAE is the average absolute error, which represents the average value of the absolute value of the prediction error, and can directly reflect the actual situation of the prediction error; R 2 is the determination coefficient, which represents the explanation proportion of the model to the variance of the target variable; is the predicted value of the i-th resistivity, is the true value of the i-th resistivity.

[0090] As shown in Figures 2-4 , it is known that the L9 Jurassic reservoir infilling test area in the study area is a low-amplitude anticline structure; the X1 and X2 reservoirs are low-porosity and low-permeability, and the average porosity is 14.5% and 16.6%, and the average permeability is 2.1 mD and 29.9 mD; the X3 reservoir is medium-porosity and medium-permeability, and the average porosity is 18.8%, and the average permeability is 73.3 mD; the oil layer is controlled by structure and lithology, and the thickness is relatively large, with an average of 6 m to 10 m; the oil-bearing property is affected by physical properties, and the X1 and X2 are mainly oil-water layers, and the X3 is mainly oil layer; the reservoir develops edge and bottom water, and different machine learning algorithms are designed to predict and verify the resistivity recovery of the well group in the test area.

[0091] As shown in Figures 5-6 , the prediction accuracy of the SNN model in the training set is R 2 = 0.98, REMS = 0.011, and MAE = 0.08, and in the validation set, R 2 = 0.58, REMS = 0.41, and MAE = 0.40, and the performance is the best. The accuracy of restoring the original oil layer resistivity is improved by 24% compared with the SVR regression model, and by 7% compared with the LightGBM algorithm model.

[0092] For eight test wells, the resistivity prediction curves of SNN and LightGBM algorithm models are fitted with the actual resistivity as shown in the following figure. Figure 7 As shown, from Figure 7 As can be seen, the SNN algorithm model achieved an average accuracy of 82.7% in resistivity recovery for multiple wells. This further demonstrates the significant advantages of spiking neural networks in handling complex, nonlinear problems. Especially in resistivity recovery applications, spiking neural networks offer higher computational efficiency and stronger generalization capabilities compared to other machine learning methods such as support vector machines and the LightGBM algorithm, thereby providing more accurate predictions and recovery results.

[0093] like Figure 8 As shown in the figure, a pulse neural network (SNN) is used to build a resistivity recovery model, and the model outputs the original resistivity curve after recovery (RT original ), providing basic data for subsequent oil saturation calculations. Based on the recovered original resistivity, oil saturation (So) and water production rate (Fw) are further calculated. Water-flooded zones are identified by combining changes in oil saturation, resistivity, and water production rate in the oil-bearing interval. Current oil saturation is calculated using a modified Archie formula, combining the recovered original resistivity (RToriginal) and current resistivity (RTcurrent).

[0094] The expression of oil saturation is:

[0095]

[0096] Among them, S o is the current oil saturation of the reservoir, RT original is the original resistivity, RT current is the current resistivity, a, m, and n are coefficients related to the rock porosity structure and saturation, usually used to describe the oil-water flow characteristics, φ is the effective porosity of the formation, σ w is the viscosity of water, measured in Pascal seconds (Pa·s).

[0097] The oil saturation evaluation model is:

[0098]

[0099] Among them, a, b, m, n are dynamic correction coefficients, σ w is the conductivity of the mixed solution.

[0100] The water production rate expression is:

[0101]

[0102] Among them, K rw is the relative permeability of water phase, Kro Relative permeability of oil phase, μ w Viscosity of water phase, μ o Viscosity of oil phase.

[0103] The water production rate evaluation model is:

[0104]

[0105] As shown in Table 1, according to the water saturation and water production rate results, the water flooded layers are quantitatively classified and interpreted using ΔSo and ΔRT: the water flooded layers are divided into four levels of non-water flooded, weak water flooded, medium-strong water flooded and strong water flooded. Combined with the resistivity curve shape and production dynamic data, the water flooded patterns (such as bottom water flooded, top water flooded, and intra-layer water flooded) are identified, which provides a basis for analyzing the water drive front advance law.

[0106] Table 1

[0107]

[0108] To verify the generalization ability of the model, two new wells are selected to test and verify the generalization of the model. The logging data and dynamic production data of the new wells are input into the model to restore the original resistivity and calculate the oil saturation and water production rate.

[0109] As shown in Table 1, according to the water saturation and water production rate results, the water flooded layers are quantitatively classified and interpreted using ΔSo and ΔRT: the water flooded layers are divided into four levels of non-water flooded, weak water flooded, medium-strong water flooded and strong water flooded. Combined with the resistivity curve shape and production dynamic data, the water flooded patterns (such as bottom water flooded, top water flooded, and intra-layer water flooded) are identified, which provides a basis for analyzing the water drive front advance law. Figures 9-10

[0110] Table 2

[0111]

[0112]

[0113] Through comparison of the logging interpretation results and the core analysis results, the relative error of porosity is between-3.26% and 3.70%, and the relative error of oil saturation is between 1.04% and 4.25%, which proves that the water flooded interpretation model has high conformity, and the core oil saturation analysis results are basically consistent with the water flooded interpretation.

[0114] Based on the established quantitative interpretation model of water flooded layers, the water flooded layers of 8 infilled wells in the infilling adjustment area are interpreted, and the resistivity prediction curve and the water flooded layer interpretation results are shown in Figure 11

[0115] Figures 12-13 ​​​As shown, the results show that the average thickness of the strong water flooded layer in the encryption test area is 2.8 m. The top is mainly unwatered-weakly watered, and the remaining oil layer thickness is 1.0 m-3.5 m, which gradually transitions downward to medium-strong water flooded and strong water flooded, with a total thickness of 1.0 m-6.0 m. The bottom is the original water layer. The remaining unwatered-medium watered oil layer thickness at the top ranges from 1.5 m to 3.5 m, with an average of 2.8 m. The current oil saturation above the original oil-water interface ranges from 39.0% to 63.0%, with an average of 47.3%.

[0116] The planar distribution of the thickness of the medium-strong water flooded layer is controlled by multiple mechanisms such as differences in sedimentary microfacies permeability, structural driving advantage channels, and reservoir heterogeneity. Weak correlation between porosity and permeability results in local high-permeability strips, causing injected water to preferentially form advantage seepage channels at the high structural position, with a resistivity drop of 30%-50%(2.0<ΔRT≤4.0), and presenting dispersed medium-strong water flooded patches on the plan view.

[0117] Therefore, the present application adopts the above-mentioned intelligent recovery system for water flooded layers in oil reservoirs based on a pulse neural network, adopts a system based on a pulse neural network (SNN), achieves the purpose of high-precision recovery of the original resistivity of water flooded layers and effective interpretation of water flooded layers. The system acquires multi-modal logging curves through a data acquisition module, normalizes and standardizes them through a preprocessing module, and then constructs and optimizes a resistivity recovery model through a model training module. The resistivity recovery module recovers the original resistivity of the oil layer accordingly, and the water flooded layer interpretation module further calculates the oil saturation and water production rate, realizing quantitative classification and interpretation of water flooded layers.

[0118] Finally, it should be noted that: the above examples are only used to illustrate the technical solutions of the present application and not to limit them, although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that: it can still modify or equivalently replace the technical solutions of the present application, and these modifications or equivalent replacements also cannot make the modified technical solutions deviate from the spirit and scope of the technical solutions of the present application.

Claims

1. An intelligent recovery system for water-flooded layers in oil reservoirs based on pulse neural networks, characterized in that: It includes data acquisition module, data preprocessing module, model training module, resistivity recovery module and water-flooded layer interpretation module; The data acquisition module acquires multi-modal logging curve data, including conventional logging curves and electrical logging curves; The data preprocessing module performs range normalization and Z-Score standardization on the collected logging curve data; The model training module constructs a resistivity recovery model based on a spiking neural network, adopts a LeakyIntegrate-and-Fire neuron model as a basic unit, receives well logging curve data through a feedforward-loop hybrid structure, generates resistivity prediction values, combines synaptic plasticity and global error back propagation, adjusts synaptic weights, optimizes resistivity recovery model performance, and verifies model reliability through resistivity recovery model evaluation; The resistivity recovery module recovers the original resistivity of the oil layer according to the trained resistivity recovery model; The water-flooded layer interpretation module calculates oil saturation and water production rate based on original resistivity, determines the water-flooded layer, and performs quantitative classification and interpretation on the water-flooded layer.

2. The intelligent recovery system for water-flooded layers in oil reservoirs based on pulse neural networks according to claim 1 is characterized in that: The conventional logging curves include natural gamma, acoustic time difference, neutron porosity, density, and natural potential; the electrical logging curves include deep lateral resistivity and shallow lateral resistivity.

3. The intelligent recovery system for water-flooded layers in oil reservoirs based on pulse neural networks according to claim 1 is characterized in that: The normalized expression of the range is: Among them, x norm is the normalized data, x min is the minimum value in the original data, x max is the maximum value in the original data, x max -x min is the range of the data; The Z-Score normalization expression is: Among them, x std is the standardized data, μ is the mean of the original data, and σ is the standard deviation of the original data.

4. The intelligent recovery system for water-flooded layers in oil reservoirs based on pulse neural networks according to claim 1 is characterized in that: The dynamic expression of the Leaky Integrate-and-Fire neuron model is: Where V is the membrane potential, τ m is the membrane time constant, V rest is the resting potential, I syn (t) is the synaptic input current; The resistivity is predicted to be th When the neuron fires a pulse and resets to V reset , a feedforward-loop hybrid structure is adopted, the input layer receives the logging curve data, the hidden layer contains pulse neuron nodes, and the output layer generates the resistivity prediction value through pulse frequency encoding.

5. The intelligent recovery system for water-flooded layers in oil reservoirs based on pulse neural networks according to claim 1 is characterized in that: The expression for adjusting synaptic weight is: Δw ij =η·∑ t d j (t)·S i (t-Δt); Among them, η is the learning rate, δ j (t) is the error gradient, Δw ij is the change in synaptic weight, S i (t-Δt) is the pulse state before Δt time; The resistivity recovery model evaluation expression is: Among them, RMSE is the root mean square error, MAE is the mean absolute error, R 2 is the coefficient of determination, is the predicted value of the i-th resistivity, is the true value of the i-th resistivity.

6. The intelligent recovery system for water-flooded layers in oil reservoirs based on pulse neural networks according to claim 1 is characterized in that: The oil saturation expression is: Among them, S o is the current oil saturation of the reservoir, RT original is the original resistivity, RT current is the current resistivity, a, m, n are coefficients related to the rock porosity structure and saturation, φ is the effective porosity of the formation, σ w is the viscosity of water; The water production rate expression is: Among them, K rw is the relative permeability of water phase, K ro is the relative permeability of the oil phase, μ w is the viscosity of the water phase, μ o is the oil phase viscosity.

7. The intelligent recovery system for water-flooded layers in oil reservoirs based on pulse neural networks according to claim 1 is characterized in that: The quantitative classification is performed using ΔS o The water-flooded layers are quantitatively classified and interpreted based on the changes in oil saturation and ΔRT resistivity, and divided into four levels: no water-flooding, weak water-flooding, moderately strong water-flooding, and strong water-flooding.

Images