A resistivity logging constrained reservoir property parameter prestack seismic inversion method

Abstract

The application provides a reservoir physical parameter prestack seismic inversion method constrained by resistivity logging, which cuts observation data and depth; normalizes seismic data, resistivity data and travel time to form network input; constructs a neural network framework, substitutes obtained porosity and saturation into an elastic rock physics model to calculate elastic parameters; substitutes the elastic parameters into a Zoeppritz equation to calculate underground reflection coefficient; generates a Ricker wavelet, and performs convolution calculation on the reflection coefficient to obtain predicted prestack seismic data; substitutes the porosity and saturation into an Archie formula to calculate predicted resistivity; constructs an objective function; performs automatic differentiation to update network parameters, and further optimizes network output porosity and saturation. The application realizes reservoir parameter inversion driven by physical information, and provides an intelligent scheme for reservoir parameter prediction and evaluation in oil and gas exploration and development.

Description

Technical Field

[0001] This invention provides a pre-stack seismic inversion method for reservoir physical parameters constrained by resistivity logging, belonging to the field of seismic exploration technology. Background Technology

[0002] Seismic and electromagnetic methods play important roles in oil and gas exploration, reservoir identification, and carbon dioxide sequestration monitoring (Constable, 2010). Seismic exploration has always been a primary means of oil and gas exploration and development. However, under complex geological conditions, severe seismic wave scattering and shielding effects result in weak illumination energy of the underlying strata, leading to ambiguity in inferring subsurface structures using only seismic data (Colombo et al., 2013; Takam et al., 2015). In reservoir characterization and monitoring, in cases with limited velocity differences (such as formation water and oil), changes in oil (gas) saturation have little impact on seismic wave velocity (Hu et al., 2009; Lien et al., 2013). Therefore, it is difficult to accurately predict reservoir fluid type and fluid saturation using seismic data alone. Electromagnetic exploration is considered an effective auxiliary means of seismic exploration (Colombo et al., 2012; Peng et al., 2018). Electromagnetic signals are highly sensitive to high-resistivity hydrocarbon reservoirs; therefore, reservoir rock physics models can be used to correlate reservoir resistivity with porosity and fluid saturation. Seismic waves and electromagnetic waves exhibit different degrees of attenuation and scattering, with electromagnetic waves showing relatively stronger attenuation and scattering, resulting in relatively lower resolution for electromagnetic exploration. However, when seismic data has established the reservoir structural framework, electromagnetic exploration can be used to predict fluid types and saturation within the reservoir, thereby reducing exploration risks (Chen et al., 2007; Chen et al., 2012). Therefore, seismic and electromagnetic data can complement each other, overcoming the limitations of single methods and effectively improving the imaging resolution of subsurface geological structures and the accuracy of reservoir characterization, leading to better detection of subsurface resources and monitoring of oil and gas.

[0003] The fusion of seismic and electromagnetic data is mainly based on two strategies: structural similarity and the relationship between rock physical parameters. The first inversion method based on structural similarity is achieved by applying structural similarity constraints on electrical and elastic parameters in the target area. Gallardo and Meju (2003) proposed a joint inversion method based on resistivity and seismic travel time using the cross-gradient method. Wiik et al. (2015) proposed a structural smoothing regularization scheme for marine controlled-source electromagnetic inversion by incorporating seismic stratigraphic properties, and verified the improvement of inversion results by the structural information contained in seismic data in synthetic data and field measured data. Guo et al. (2017) used structural tensors extracted from seismic images to modify the regularization inversion parameters, allowing discontinuous changes in parameters on both sides of the geological structure during the smoothing process, thereby improving the inversion accuracy. Li et al. (2019) introduced a joint total variation constraint of seismic-electromagnetic data into the inversion objective function and proposed an alternating inversion method of marine controlled-source electromagnetic and seismic data. Compared with separate electromagnetic or seismic inversion, the joint inversion can significantly improve the inversion results. Kong et al. (2024) introduced a cross-gradient function to achieve mutual coupling of seismic electromagnetic property parameters, significantly improving the characterization of anomaly morphology, structural features, and the recovery of physical property values. The above application examples demonstrate that joint inversion based on local correlation constraints can effectively improve the convergence and multiple-solution problems of single geophysical inversion. However, this method can introduce artificially introduced boundaries in some cases, where the boundary of one physical model depends on the boundary of another. Furthermore, EM has poor resolution for deep anomalies, and inaccurate resistivity information conveyed by structural constraints often misleads velocity models, introducing errors into the seismic inversion structure.

[0004] Another joint inversion method is based on the joint inversion of rock physical properties, achieving joint inversion through the relationship between reservoir parameters and electrical and elastic parameters. Rock physics models establish the relationship between reservoir electrical, elastic, and physical parameters, forming the theoretical basis for seismic and electromagnetic forward and inverse retrieval studies of reservoir parameters. Empirical parameters in rock physics models (such as critical porosity, cementation coefficient, saturation coefficient, and tortuosity coefficient in Archie's formula) play a crucial role in the mapping relationship between elastoelectric and physical parameters. Classical rock physics models typically set these empirical parameters to fixed values, making it difficult to accurately predict the elastoelectric parameters and other responses of rocks with realistic and complex pore structures.

[0005] Gao et al. (2012) proposed a Gauss-Newton inversion algorithm combining electromagnetic and full-wavefield seismic data to estimate reservoir rock physical parameters, such as porosity and fluid saturation, overcoming the strong non-uniqueness caused by inversion from single geophysical data. Ren et al. (2017) proposed a joint inversion method using marine controlled-source seismic data with a multi-chain Markov chain Monte Carlo sampler, which improved the accuracy of reservoir saturation assessment.

[0006] The numerical analysis and practical application examples based on traditional methods described above demonstrate the feasibility of the seismic-electromagnetic joint inversion method and its effectiveness in reducing the ambiguity of single geophysical inversion results and improving solution stability. However, traditional numerical methods suffer from various limitations, such as susceptibility to local extrema, excessive reliance on initial parameters, model simplification and assumption limitations, and limited ability to handle nonlinear relationships. In the inversion process, the forward modeling operators for complex porous reservoirs often exhibit high nonlinearity, making the inversion problem difficult to solve.

[0007] With the development of artificial intelligence and machine learning methods, deep learning technology has been widely applied in the field of geophysics. Deep learning focuses on constructing the overall framework of the algorithm rather than precisely designing algorithm parameters, extracting hidden internal connections between different data that are imperceptible to humans, and "seeing" the effective features hidden within the data—something traditional numerical methods struggle to achieve. Therefore, fully utilizing deep learning to explore the correlations between different geophysical data to improve inversion and interpretation capabilities has become a current research hotspot in the field of joint inversion.

[0008] Guo et al. (2020) first utilized deep learning constraints to achieve the conversion and joint inversion of audio magnetotelluric and seismic travel time data, preserving the structural similarity between different inversion models. This method achieves nonlinear transformation between resistivity and wave velocity through deep learning. Sun et al. (2020) proposed a supervised deep learning method for predicting salt geometry using seismic and electromagnetic data and tested the fusion performance of different network stages, providing good inspiration for the fusion of network architectures to achieve joint inversion using deep learning. The above data-driven joint inversion methods directly establish the mapping relationship between input and output, but they require a large amount of geophysical data and do not explicitly follow the physical rules in the inversion problem.

[0009] Building upon this, embedding physical laws into deep learning networks has become a key challenge in addressing the generalization of joint inversion problems. Under a physics-driven strategy, algorithms incorporate forward modeling processes based on physical laws (such as Maxwell's equations and wave equations). Instead of using the predicted model as the final result in an end-to-end manner, these algorithms assist gradient-based model inversion, for example, by constraining the gradient descent direction. Hu et al. (2023) developed a deep learning-enhanced joint inversion framework that combines deep learning with traditional independent inversion work to achieve joint inversion of two-dimensional DC resistivity data and seismic propagation time. Ren et al. (2024), drawing on Hu's technical theory, conducted research on joint inversion of surface seismic full waveforms and high-density electrical resistivity methods based on deep learning. Liu et al. (2023) proposed a neural network model based on physical information, using Bayesian networks to achieve joint inversion of three-dimensional seismic electromagnetic data, and accurately quantifying undetermined global parameters in the rock physics model, achieving accurate estimation of reservoir properties in the Sleipner carbon dioxide storage monitoring scenario. Zhou et al. (2023) proposed a deep learning-based seismic image-guided magnetotelluric (MT) inversion strategy to achieve multi-physics coupled inversion with embedded geophysical knowledge. Subsequently, Zhou et al. (2024) based on the principle of structural similarity, embedded the seismic image texture operator extracted by the convolutional neural network VGG19 into the magnetotelluric inversion objective function, and realized high-resolution magnetotelluric inversion constrained by seismic images.

[0010] In summary, deep learning-based seismic-electromagnetic joint inversion methods, which fully utilize the rich information contained in multi-source data to achieve high-precision inversion imaging of subsurface reservoirs, have become a current research hotspot. However, existing research incorporating deep learning technology for joint inversion mostly relies on the principle of structural similarity, starting from images and using convolutional neural networks for feature extraction, which is then combined with traditional methods. Research on joint inversion based on rock physical properties is relatively limited. Furthermore, these studies largely depend on large amounts of data or data-physical knowledge-driven approaches. However, given the current situation, especially with the limited amount of electromagnetic data, data-driven methods struggle to meet the requirements of large datasets. Summary of the Invention

[0011] Based on the above problems, this invention obtains porosity and saturation by joint inversion of the acquired one-dimensional pre-stack seismic records and resistivity, establishes a non-mapping relationship between seismic-resistivity data and reservoir physical parameters, improves the accuracy of underground reservoir physical parameter inversion, and reduces inversion ambiguity.

[0012] To address the aforementioned problems in the prior art, this invention provides a resistivity-constrained pre-stack seismic parameter inversion method, which avoids the dependence of traditional methods on the initial model and realizes reservoir parameter (porosity, saturation) inversion driven by physical information, providing an intelligent solution for reservoir parameter prediction and evaluation in oil and gas exploration and development.

[0013] The specific technical solution is as follows:

[0014] A pre-stack seismic inversion method for reservoir physical parameters constrained by resistivity logging includes the following steps:

[0015] S1, Integrating pre-stack seismic records d seis and resistivity logging data d res The observation data and depth were cropped to ensure that their dimensions were equal, and the resistivity data were converted to the time domain.

[0016] S2. Perform maximum-min normalization on the seismic data, resistivity data, and time respectively to form the network input d. input ;

[0017] The maximum-minimum normalization in S2 is shown in formula (1), and the three are combined to form the network input d. input , as in formula (2);

[0018]

[0019]

[0020] Where, d seis d res t and t represent the original seismic data, resistivity logging data, and travel time, respectively; X.min() and X.max() represent taking the minimum and maximum values ​​of X, respectively. t * These are normalized seismic data, resistivity logging data, and travel time, respectively.

[0021] S3. Construct a neural network framework, with network input d. input The output is m = [Φ, S];

[0022] In S3, the network is a five-layer fully connected network with the activation function tanh. Normalization layers are added to the hidden and output layers, and a sigmoid activation function is added after the output layer to limit the network output range to [0,1]. This operation adds prior information about the range of the target attribute values ​​to be inverted, i.e., the porosity and saturation values ​​are both between [0,1].

[0023]

[0024] In formula (3) θ represents the constructed neural network inversion operator; m represents the reservoir physical properties, where Φ and S represent porosity and water saturation, respectively; p Represents the parameters of the neural network.

[0025] S4. Substitute the porosity Φ and saturation S obtained from the network output into the elastic rock physics model to calculate the elastic parameters; the specific formulas in S4 are as follows:

[0026]

[0027]

[0028]

[0029]

[0030]

[0031]

[0032]

[0033] μ sat =μ dry (11)

[0034]

[0035]

[0036]

[0037] ρ sat =(1-φ)ρ mat +φρ fl (15)

[0038] ρ mat =(1-v cl )ρ qtz +v cl ρ cl (16)

[0039] ρ cl =(1-S g )ρ w +S g ρ g (17)

[0040] Where P e For effective pressure, v mat n is the Poisson's ratio of the grains. cThe coordination number is 7 in this scheme; φ0 is the critical porosity; φ is the saturated rock porosity; K HM With μ HM These are the equivalent bulk modulus and shear modulus, respectively; K dry With μ dry These are the bulk modulus and shear modulus of dry rock, respectively; K mat With μ mat These are the bulk modulus and shear modulus of the rock skeleton, respectively; K sat With μ sat These are the bulk modulus and shear modulus of fluid-saturated rock, respectively; k fl The bulk modulus of the fluid phase is denoted by the pore water saturation S. w With supercritical gas saturation S g Calculated; V p V s These represent the longitudinal wave velocity and the transverse wave velocity, respectively; ρ sat ρ fl ρ mat ρ w ρ g ρ cl ρ qtz These are, respectively, the density of saturated rock, the density of fluid, the density of rock skeleton, the density of pore water, the density of supercritical gas, the density of clay, and the density of quartz; v cl This represents the volume fraction of clay.

[0041] S5. Substitute the elastic parameters calculated in step S4 into the Zoeppritz equation to calculate the subsurface reflection coefficient r. pp ;

[0042] The equations in S5 are as shown in equations (18)-(21), and the underground reflection coefficient r is calculated. pp ;

[0043] r pp (t,θ)=R+Gsin 2 θ+F(tan 2 θ-sin 2 θ) (18)

[0044]

[0045]

[0046]

[0047] Where t represents the travel time, θ represents the incident angle of the seismic wave; ΔV P ΔV s Δρ and Δρ represent the P-wave velocity, S-wave velocity, and density difference at adjacent travel times, respectively; V p Vs , These represent the average values ​​of P-wave velocity, S-wave velocity, and density at adjacent travel times.

[0048] S6. Generate the Ricker wavelet and perform convolution calculation on the reflection coefficients calculated in step S5 to obtain the predicted pre-stack seismic data.

[0049] Predicted prestack earthquake data in S6

[0050]

[0051] In equation (22), w(θ) represents the seismic wavelet.

[0052] S7. Substitute the porosity Φ and saturation S obtained from the network output into Archie's formula to calculate the predicted resistivity.

[0053] Predicting resistivity in S7 As in equation (23):

[0054]

[0055] Among them, R w ρ is the pore water resistivity; n is the saturation index; m is the cementation coefficient.

[0056] S8. Construct the data terms in the objective function using real earthquake and resistivity data, as well as the predicted earthquake and resistivity data calculated in steps S7 and S8 respectively; at the same time, use the chain rule of neural networks to calculate the gradient of porosity φ and saturation S obtained from the network output, thereby constructing the regularization term in the objective function.

[0057] The objective function in S8 is constructed as shown in equation (24);

[0058]

[0059] The first term is the seismic data fitting term, the second term is the resistivity data fitting term, and the third term is the model fitting term; α1, α2, and α3 are the weights of the three terms, respectively; σ res and σ seis These represent the standard deviations of the actual resistivity data and the seismic data, respectively; m represents the inversion parameter model. This represents the partial derivative of the model.

[0060] S9. Automatic differentiation is performed based on the constructed joint inversion loss function to update the network parameters, thereby optimizing the network output porosity and saturation to fit the real observation data; when the loss drops to the set minimum threshold or the number of training iterations reaches the maximum, the result is output.

[0061] This invention addresses the challenging problems in the exploration of complex oil and gas reservoirs by combining multidisciplinary technologies, integrating physical models with geophysical observation data to achieve reservoir inversion and fluid identification, demonstrating the following unique advantages and beneficial effects:

[0062] 1) The inversion results and reliability are greatly improved.

[0063] To address the issues of low reliability and multiple solutions in inversion from single geophysical data, this patented solution innovatively introduces physical constraint information such as rock physics models and seismic forward modeling operators. Combined with multi-source seismic and resistivity observation data, it forms a physically constrained neural network joint inversion framework. This method effectively reduces uncertainties in the inversion process, achieving high-precision imaging and fluid identification of oil and gas reservoirs.

[0064] 2) Overcoming the challenges of handling nonlinear mapping relationships

[0065] Traditional pre-stack seismic reservoir inversion methods have limitations when handling complex nonlinear mapping relationships, and are prone to getting trapped in local extrema, affecting the accuracy of the inversion results. The deep learning-based physical constrained resistivity-pre-stack seismic joint inversion method proposed in this patent uses deep learning technology to uncover the coupling relationships between multi-source data. It is not only applicable to highly nonlinear inversion problems, but also significantly improves accuracy, fully demonstrating the combined advantages of resistivity data and pre-stack seismic data.

[0066] 3) Does not depend on the initial model

[0067] Compared with traditional methods, this method does not require an initial model, thus eliminating the dependence of traditional methods on an initial model. Under this condition, the stability and consistency of reservoir parameters (such as porosity and saturation) obtained by neural network inversion are fully verified. Attached Figure Description

[0068] Figure 1 A flowchart illustrating the construction of the neural network framework for this invention;

[0069] Figure 2 The inversion results of the single method and the combined method for one-dimensional reservoir parameters are shown in the embodiments.

[0070] Figure 3 The one-dimensional reservoir parameter inversion results are shown in the example. Detailed Implementation

[0071] The specific technical solution of the present invention will be described in conjunction with the accompanying drawings.

[0072] A pre-stack seismic inversion method for reservoir physical parameters constrained by resistivity logging, the specific steps of which are as follows:

[0073] S1, Integrating pre-stack seismic records d seis and resistivity logging data d res The observation data and depth are clipped (interpolated and expanded) to ensure that their dimensions are equal, and the resistivity data are converted to the time domain.

[0074] S2. Perform maximum-min normalization on the seismic data, resistivity data, and time respectively (as in formula (1)), and combine the three to form the network input d. input (as in formula (2)).

[0075]

[0076]

[0077] S3. Construct a neural network framework, such as Figure 1 The network input is d input The output is m = [φ, S]. The network is a five-layer fully connected network with the activation function tanh. Normalization layers are added to the hidden and output layers, and a sigmoid activation function is added after the output layer to limit the network output range to [0,1]. This operation incorporates prior information about the range of the inverted target attribute values, i.e., porosity and saturation values ​​are both between [0,1].

[0078]

[0079] S4. Substitute the porosity φ and saturation S obtained from the network output into the elastic rock physics model (taking the Gassmann model of unconsolidated sandstone as an example) to calculate the elastic parameters (P-wave velocity, S-wave velocity, and density). The specific formulas are as follows:

[0080]

[0081]

[0082]

[0083]

[0084]

[0085]

[0086]

[0087] μ sat =μ dry (11)

[0088]

[0089]

[0090]

[0091] ρ sat =(1-φ)ρ mat +φρ fl (15)

[0092] ρ mat =(1-v cl )ρ qtz +v cl ρ cl (16)

[0093] ρ cl =(1-S g )ρ w +S g ρ g (17)

[0094] S5. Substitute the elastic parameters calculated in step S4 into the Zoeppritz equation (as shown in equations (18)-(21)) to calculate the subsurface reflection coefficient r. pp .

[0095] r pp (t, θ) = R + Gsin 2 θ+F(tan 2 θ-sin 2 θ) (18)

[0096]

[0097]

[0098] S6. Generate the Ricker wavelet and perform convolution calculation on the reflection coefficients calculated in step S5 to obtain the predicted pre-stack seismic data.

[0099]

[0100] S7. Substitute the porosity Φ and saturation S obtained from the network output into Archie's formula to calculate the predicted resistivity. (as in equation (21)). Where R is...w Ω·m is the pore water resistivity, with a value of 0.17 Ω·m; m is the cementation coefficient, with a value of 1.3; n is the saturation index, with a value of 2.0.

[0101]

[0102] S8. Construct an objective function (as shown in Equation (22)) using real earthquake and resistivity data, as well as predicted earthquake and resistivity data calculated in steps S7 and S8 respectively. The first term is the earthquake data fitting term, the second term is the resistivity data fitting term, and the third term is the model fitting term; α1, α2, and α3 are the weights of the three terms; σ res and σ seis These represent the standard deviations of the actual resistivity data and the seismic data, respectively.

[0103]

[0104] S9. Automatic differentiation is performed based on the constructed joint inversion loss function to update the network parameters, thereby optimizing the network output porosity and saturation to fit the real observation data. The result is output when the loss drops to the set minimum threshold or the maximum number of training iterations is reached.

[0105] Figure 2 The inversion results of the single method and the combined method for one-dimensional reservoir parameters are shown in the embodiments. Figure 2 (a) Porosity; (b) Saturation. Black represents the true value; blue represents the seismic inversion result; green represents the resistivity inversion result; and red represents the resistivity-seismic joint inversion result.

[0106] By comparing with real data (black curve), it can be seen that the joint inversion results are far superior to those of single seismic inversion (blue curve) and single resistivity inversion (green curve), and are better able to reflect the distribution and variation of subsurface reservoir parameters. Therefore, this embodiment can fully demonstrate that the resistivity-constrained pre-stack seismic reservoir parameter inversion effect proposed in this scheme is significantly improved compared with the single method.

[0107] Figure 3 The one-dimensional reservoir parameter inversion results are shown in the example. Figure 3 (a) Porosity; (b) Saturation. Black represents the true value; gray represents the joint inversion result of different random seeds (a total of 10 different random seeds); red represents the average value of different random seeds.

[0108] The inversion results lead to the following conclusions: the resistivity logging-constrained pre-stack seismic reservoir inversion method proposed in this study can achieve high-precision inversion of subsurface reservoir parameters. Furthermore, this method exhibits good robustness; that is, network frameworks constructed using different random seeds can accurately invert reservoir parameters close to their true values.

Claims

1. A pre-stack seismic inversion method for reservoir physical parameters constrained by resistivity logging, characterized in that, Includes the following steps: S1, Integrating pre-stack seismic records d seis and resistivity logging data d res The observation data and depth were cropped to ensure that their dimensions were equal, and the resistivity data were converted to the time domain. S2. Perform maximum-min normalization on the seismic data, resistivity data, and time respectively to form the network input d. input ; Max-min normalization, as shown in formula (1), and the three are combined to form the network input d. input , as in formula (2); Where, d seis d res t and t represent the original seismic data, resistivity logging data, and travel time, respectively; X.min() and X.max() represent taking the minimum and maximum values ​​of X, respectively. t * These are normalized seismic data, resistivity logging data, and travel time, respectively. S3. Construct a neural network framework, with network input d. input The output is m = [Φ, S]; S4. Substitute the porosity Φ and saturation S obtained from the network output into the elastic rock physics model to calculate the elastic parameters; S5. Substitute the elastic parameters calculated in step S4 into the Zoeppritz equation to calculate the subsurface reflection coefficient r. pp ; S6. Generate the Ricker wavelet and convolve it with the reflection coefficient calculated in step S5 to obtain the predicted pre-stack seismic data. S7. Substitute the porosity Φ and saturation S obtained from the network output into Archie's formula to calculate the predicted resistivity. S8. Construct the data terms in the objective function using real earthquake and resistivity data, as well as the predicted earthquake and resistivity data calculated in steps S6 and S7 respectively; at the same time, use the chain rule of the neural network to calculate the gradient of the porosity Φ and saturation S obtained from the network output, thereby constructing the regularization term in the loss function Loss. S9. Automatic differentiation is performed based on the constructed joint inversion loss function to update the network parameters, thereby optimizing the network output porosity and saturation to fit the real observation data and ensure the smoothness of the model. When the loss function Loss drops to the set minimum threshold or the number of training iterations reaches the maximum, the result is output.

2. The pre-stack seismic inversion method for reservoir physical parameters constrained by resistivity logging according to claim 1, characterized in that, In S3, the network is a five-layer fully connected network with the activation function tanh. Normalization layers are added to the hidden and output layers, and a sigmoid activation function is added after the output layer to limit the network output range to [0,1]. This operation adds prior information about the range of the target attribute values ​​to be inverted, i.e., the porosity and saturation values ​​are both between [0,1]. In formula (3) θ represents the constructed neural network inversion operator; m represents the reservoir physical properties, where Φ and S represent porosity and water saturation, respectively; p Represents the parameters of the neural network.

3. The pre-stack seismic inversion method for reservoir physical parameters constrained by resistivity logging according to claim 2, characterized in that, The specific formula in S4 is as follows: m sat =μ dry (11) r sat =(1-φ)ρ mat +fr fl (15) r mat =(1-v cl )r qtz +v cl r cl (16) r cl =(1-S g )r w +S g r g (17) Where P e For effective pressure, v mat n is the Poisson's ratio of the grain size. c The coordination number is 7 in this scheme; φ0 is the critical porosity; φ is the saturated rock porosity; K HM With μ HM These are the equivalent bulk modulus and shear modulus, respectively; K dry With μ dry These are the bulk modulus and shear modulus of dry rock, respectively; K mat With μ mat These are the bulk modulus and shear modulus of the rock skeleton, respectively; K sat With μ sat These are the bulk modulus and shear modulus of fluid-saturated rock, respectively; K fl The bulk modulus of the fluid phase is denoted by the pore water saturation S. w With supercritical gas saturation S g Calculated; V p V s These represent the longitudinal wave velocity and the transverse wave velocity, respectively; ρ sat ρ fl ρ mat ρ w ρ g ρ cl ρ qtz These are, respectively, the density of saturated rock, the density of fluid, the density of rock skeleton, the density of pore water, the density of supercritical gas, the density of clay, and the density of quartz; v cl This represents the volume fraction of clay.

4. The pre-stack seismic inversion method for reservoir physical parameters constrained by resistivity logging according to claim 3, characterized in that, The equations in S5 are as shown in equations (18)-(21), and the underground reflection coefficient r is calculated. pp ; r pp (t,θ)=R+Gsin 2 θ+F(tan 2 θ-sin 2 i) (18) Where t represents the travel time, θ represents the incident angle of the seismic wave; ΔV P ΔV s Δρ represents the longitudinal wave velocity, transverse wave velocity, and density difference at adjacent travel times, respectively. These represent the average values ​​of P-wave velocity, S-wave velocity, and density at adjacent travel times.

5. The pre-stack seismic inversion method for reservoir physical parameters constrained by resistivity logging according to claim 4, characterized in that, Predicted prestack earthquake data in S6 In equation (22), w(θ) represents the seismic wavelet.

6. The pre-stack seismic inversion method for reservoir physical parameters constrained by resistivity logging according to claim 1, characterized in that, Predicting resistivity in S7 As in equation (23): Among them, R w ρ is the pore water resistivity; n is the saturation index; m is the cementation coefficient.

7. The pre-stack seismic inversion method for reservoir physical parameters constrained by resistivity logging according to claim 1, characterized in that, In S8, the loss function Loss is constructed as shown in equation (24); The first term is the seismic data fitting term, the second term is the resistivity data fitting term, and the third term is the model fitting term; α1, α2, and α3 are the weights of the three terms, respectively; σ res and σ seis These represent the standard deviations of the actual resistivity data and the seismic data, respectively; m represents the inversion parameter model. This represents the partial derivative of the model.

Images