Intelligent prediction and fitting method for dynamic yield of unconventional reservoir fractured horizontal well

Abstract

The invention relates to an intelligent prediction and fitting method for the dynamic yield of an unconventional reservoir fractured horizontal well, and the method comprises the steps: S1, building an anisotropic seepage mathematical model of a coupling gas-rock dual compression mechanism, and constructing a productivity numerical model; s2, determining variable ranges of geological parameters and fracture network parameters, and determining yield data; s3, a geological parameter and fracture network parameter combination is generated through sampling and substituted into the productivity numerical model for calculation, a geological parameter-fracture network parameter-productivity data set is obtained, and a productivity prediction agent model is constructed; and S4, by taking the productivity prediction agent model as a forward simulator, realizing uncertainty inversion of geological parameters and crack parameters, and substituting an inversion result into the numerical simulation model to perform productivity prediction. According to the method, a machine learning agent model is introduced to accelerate forward modeling calculation, geological priori knowledge and dynamic production data are deeply fused, multi-dimensional automatic matching of geological parameters, seepage rules, fracture network parameters and production dynamics is achieved, and fracturing fracture network multi-parameter automatic fitting and yield prediction are achieved.

Description

Technical Field

[0001] This invention relates to the field of oil and gas exploration and development technology, specifically to an intelligent prediction and fitting method for dynamic production of horizontal wells in unconventional reservoirs. Background Technology

[0002] Unconventional reservoirs are often developed using fractured horizontal wells, and production prediction plays a crucial role throughout the entire project lifecycle. The development effectiveness of unconventional reservoirs is comprehensively controlled by multiple factors, including geological parameters, permeability anisotropy, and fracture network parameters. Among these, reservoir permeability anisotropy significantly affects gas migration paths, while the complex fracture network formed by fracturing further alters the local permeability field distribution. Simultaneously, gas desorption and expansion during production, along with changes in effective rock stress, lead to dynamic evolution of porosity and permeability, resulting in a highly nonlinear production characteristic. Therefore, accurately inverting reservoir geological properties (such as anisotropic permeability) and fracture network parameters (hydraulic fracture length, width, height, and conductivity) is of great significance for production prediction of fractured horizontal wells in unconventional reservoirs.

[0003] Current automatic fitting methods for multi-parameter fracturing can be divided into numerical simulation methods based on physical mechanisms and data-driven methods constrained by mechanisms. While mechanistic simulation methods can describe the gas-rock coupling process, they suffer from the following drawbacks: First, model solving requires manual adjustment of numerous parameters to achieve historical fit, resulting in low efficiency and a tendency to get trapped in local optima. Second, existing models generally simplify seepage anisotropy, using homogeneous permeability approximations and neglecting the directional guiding effect of rock bedding planes and natural fractures, leading to large deviations in pressure propagation predictions. Third, when actual production dynamics deviate from initial predictions, it is necessary to reconstruct the mesh and iteratively solve the problem, resulting in long computation times and an inability to support real-time decision-making. These shortcomings severely restrict the application of mechanistic models in dynamic production capacity optimization.

[0004] In recent years, data-driven methods with mechanistic constraints have emerged, which can fit production curves in the short term by introducing mechanistic constraints (Arps decreasing equations) into time-series neural network models (such as LSTM and Transformer). However, their essential drawbacks are twofold: firstly, the model treats geological and fracture network parameters as black boxes, failing to correlate them with the intrinsic variation patterns of reservoir properties, leading to the accumulation of errors in extrapolation predictions; secondly, key physical mechanisms such as seepage anisotropy are abstracted as statistical noise, resulting in a significant decrease in accuracy when the prediction window is long. Although such methods improve computational efficiency, they neglect the multi-parameter coupling mechanism of seepage, failing to support the rationality of the inversion results. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent prediction and fitting method for dynamic production of unconventional reservoir fracturing horizontal wells based on a mechanism-data fusion driven framework.

[0006] The present invention achieves the above objectives through the following technical solutions:

[0007] A method for intelligent prediction and fitting of dynamic production of unconventional reservoir fractured horizontal wells includes the following steps:

[0008] S1. Introduce the gas and rock elastic state equations to describe the gas expansion and pore compression effects, establish an anisotropic seepage mathematical model coupled with the gas-rock dual compression mechanism, and construct a numerical simulation model of unconventional reservoir gas production capacity.

[0009] S2. Determine the range of geological parameters and joint mesh parameters based on geological and construction data, and determine the production data based on actual production data;

[0010] S3. Use super Latin cube sampling to generate a combination of geological parameters and fracture network parameters, substitute them into the production capacity numerical simulation model to obtain the calculated production capacity data, obtain the geological parameters-fracture network parameters-production capacity dataset, and construct a production capacity prediction proxy model based on the geological parameters-fracture network parameters-production capacity dataset.

[0011] S4. Using the production capacity prediction proxy model as a forward simulator, a Bayesian inference framework is constructed by combining the prior of normal distribution geological parameters. The likelihood function is defined as the sum of squared errors between the predicted production capacity and the measured production capacity under Gaussian noise. By Monte Carlo sampling, the model converges to the posterior distribution of the parameters, thereby realizing the uncertainty inversion of geological parameters and fracture parameters. The inversion results are then substituted into the numerical simulation model for long-term production prediction.

[0012] In a preferred embodiment, the anisotropic seepage mathematical model in step S1 is:

[0013]

[0014] In the formula, k x Let m be the permeability in the x-direction. 2 ;k y Let m be the permeability in the y-direction. 2 ;k z Let m be the permeability in the z-direction. 2 μ is the gas viscosity, Pa·s; φ0 is the initial porosity, dimensionless; C t The overall compressibility factor is expressed in MPa. -1 p represents pressure (Pa); t represents time (d).

[0015] In a preferred embodiment, step S2 further includes:

[0016] The Kalman filter method is used to smooth the actual production data.

[0017] In a preferred embodiment, step S2 further includes:

[0018] The Z-score method was used to standardize the geological parameters, fracture network parameters, and production data.

[0019] In a preferred embodiment, the likelihood function expression in step S4 is:

[0020]

[0021] In the formula, σ ν δ represents the variance of the observation noise; y represents the number of data points; i,act Let f(θ) be the i-th actual production capacity data point; i ) represents the predicted value of the surrogate model under parameter θ.

[0022] The beneficial effects of this invention are as follows: This invention provides an intelligent prediction and fitting method for dynamic production of unconventional reservoir fractured horizontal wells. First, it couples the constitutive equations of gas expansion and rock compression, and introduces the permeability tensor to characterize the anisotropy of seepage, thus solving the simulation bias caused by the simplification of directional seepage in traditional mechanism models. Second, it replaces numerical simulation with a Gaussian process regression surrogate model, and combines Kalman filtering and standardized preprocessing to achieve efficient adaptive fitting of dynamic data, significantly improving computational efficiency. Finally, based on a Bayesian framework, it integrates prior geological parameters and production dynamic data, and simultaneously inverts geological properties and fracture network parameters, overcoming the shortcomings of existing technologies that do not fully consider geological parameters, seepage mechanisms, fracture network parameters, and production dynamics. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the method flow of the present invention;

[0024] Figure 2 This is a schematic diagram of a three-dimensional spatial mesh node in a specific embodiment of the present invention;

[0025] Figure 3 This is a flowchart illustrating the inversion process in a specific embodiment of the present invention;

[0026] Figure 4 The historical output Kalman filter noise reduction results are shown in a specific embodiment of the present invention.

[0027] Figure 5 This is a comparison chart of the output of the proxy model and the numerical simulation model in a specific embodiment of the present invention;

[0028] Figure 6 This is a diagram showing the multi-parameter fitting results of the hydraulic fracturing network in a specific embodiment of the present invention;

[0029] Figure 7 This is a comparison chart of numerical simulation results and actual production data under the optimal fitting parameters in a specific embodiment of the present invention. Detailed Implementation

[0030] The following specific examples illustrate the implementation of this disclosure. Those skilled in the art can easily understand other advantages and effects of this disclosure from the content disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of this disclosure, and not all of them. This disclosure can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this disclosure. It should be noted that, in the absence of conflict, the following embodiments and features in the embodiments can be combined with each other. Based on the embodiments in this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.

[0031] In existing technologies, mechanism-driven methods are limited by computational efficiency and parameter sensitivity, while mechanism-constrained data-driven methods lack physical interpretability and long-term generalization ability. Therefore, it is urgent to construct a mechanism-data fusion-driven framework. Based on anisotropic seepage mechanism models, this framework introduces machine learning surrogate models to accelerate forward modeling calculations and deeply integrates geological prior knowledge with dynamic production data. This enables multi-dimensional automatic matching of geological parameters, seepage laws, fracture network parameters, and production dynamics, solving the challenges of automatic fitting of multiple parameters and accurate long-term production prediction for fracture networks in unconventional reservoirs.

[0032] Based on this, such as Figure 1 As shown, this invention provides an intelligent prediction and fitting method for dynamic production of unconventional reservoir fractured horizontal wells, including the following steps:

[0033] S1. Introduce the gas and rock elastic state equations to describe the gas expansion and pore compression effects, establish an anisotropic seepage mathematical model coupled with the gas-rock dual compression mechanism, and construct a numerical simulation model of unconventional reservoir gas production capacity.

[0034] Influenced by factors such as sedimentary environment and bedding development, the permeability of unconventional reservoirs varies significantly in different flow directions. The differential flow equation considering permeability anisotropy is as follows:

[0035]

[0036] In the formula, ρ is the gas density, kg / m³ 3 ;k x Let m be the permeability in the x-direction. 2 ;k y Let m be the permeability in the y-direction. 2 ;k z Let m be the permeability in the z-direction. 2 μ is the gas viscosity, Pa·s; φ is the porosity, dimensionless; p is the pressure, Pa; t is the time, d.

[0037] In this invention, the gas compressibility coefficient and the rock compressibility coefficient are introduced to characterize the gas expansion and pore compression effects caused by pressure changes during the production process. The gas density and rock porosity can then be expressed as follows:

[0038] ρ=ρ0[1+C g (p-p0)] (2)

[0039] φ=φ0[1+C f (p-p0)] (3)

[0040] In the formula, ρ0 is the initial gas density, kg / m³. 3 C g The gas compressibility coefficient is expressed in MPa. -1 p0 is the initial formation pressure, Pa; φ0 is the initial porosity, dimensionless; C f The rock compressibility coefficient is expressed in MPa. -1 .

[0041] Substituting equations (2) and (3) into equation (1) and rearranging, we get:

[0042]

[0043] in:

[0044] C t =C f +C g (5)

[0045] In the formula, C t The overall compressibility factor is expressed in MPa. -1 .

[0046] At the initial production time t=0, the pressure is the original formation pressure, and the inner boundary conditions are:

[0047] p(x,y,z) t＝0 =p0(x,y,z) (6)

[0048] Set non-permeable outer boundary conditions:

[0049]

[0050] In the formula, L x L represents the length of the study region in the x-direction, in meters. y y is the length of the study area in the y direction, in meters; h is the reservoir height, in meters.

[0051] Setting constant bottom hole flowing pressure production conditions:

[0052]

[0053] The pressure is obtained to the grid in contact with the main fracture through the above process, and the gas production is calculated using Darcy's law:

[0054]

[0055] In the formula, Q i,j,k Let m be the gas production of the i-th grid in the x-direction, the j-th grid in the y-direction, and the k-th grid in the z-direction of three-dimensional space. 3 / d;k wf Main fracture permeability, m 2 A i,j,k The contact area between the mesh and the main crack is m. 2 ;ΔL x / y / z Let m be the grid length in each direction.

[0056] Assuming all main fractures are connected to the wellbore and all gas in the main fractures is extracted, the overall gas production can be obtained by summing the gas production of all main fracture grids. In this case, it is only necessary to determine the pressure change of the grid in contact with the main fractures at each time step to calculate the total gas production.

[0057] The three-dimensional computational domain is discretized into control volume elements, with each node corresponding to a control volume V surrounding itself. For example... Figure 1 As shown, a node has 6 adjacent nodes, located on the east (e), west (w), south (s), north (n), top (u), and bottom (b) grid faces respectively.

[0058] When integrating over the control volume, it is assumed that the solution has an interpolation distribution relationship between nodes. First, integrating the differential seepage equation over the control volume and time period yields:

[0059]

[0060] By using Gauss's formula to transform the volume integral into a surface integral on the upper boundary of the control volume, and simultaneously by the central difference of the interface values ​​for the pressure diffusion term, we can obtain:

[0061]

[0062] Solving the left-hand side requires providing (p). P (p) E (p) W (p) N (p) S (p) U (p) B The relationship between pressure and time is unknown. This relationship is typically determined using the pressure (p) at time t. P,t ) and the pressure (p) at time t+Δt P,t+Δt The weighted combination constitutes the average pressure over this time interval, which is then integrated to calculate:

[0063]

[0064] In the formula, λ is a constant between 0 and 1.

[0065] The fully implicit scheme is unconditionally stable for any Δt. It is fully implicit when λ = 1, in which case it is stable with respect to (p). P Time integral I p It can be written as:

[0066]

[0067] When solving using the fully implicit scheme, equation (11) becomes:

[0068]

[0069] In the formula, (k x ) e 、(k x ) w 、(k y ) n 、(k y ) s 、(k z ) u 、(k z ) b The value is not at a node; it needs to be approximated based on the value at the node, and simplified using the central difference format.

[0070]

[0071] Set a time step Δt, take the pressure distribution of the previous time layer as the new boundary condition, and use the Thomas algorithm to solve the nodal pressure at time t+Δt. First, solve the solution for all grid nodes in the spatial domain at time t. After the iteration is completed, advance with a time step Δt, and then perform a new round of iteration in the space until the final time is reached. After the pressure is determined, the output at each time step is solved by equation (9).

[0072] S2. Determine the range of geological parameters and joint mesh parameters based on geological and construction data, and determine the production data based on actual production data;

[0073] Kalman filtering is an algorithm that combines previous state information with current observations to estimate the state using all observation data, resulting in a smoother curve. First, the curve is modeled as a state-space model, assuming the curve is a time-varying sequence of states, with the state equation evolving over time as:

[0074] x t =A t xt-1 +ω t (16)

[0075] In the formula, x t Let A be the state vector at time t; t ω is the state transition matrix; t This is process noise.

[0076] The observation equation describing the relationship between the observed values ​​and the state is:

[0077] y t =H t x t +υ t (17)

[0078] In the formula, y t H is the observation vector at time t; t υ is the observation matrix; t To observe noise.

[0079] Kalman recursive filtering includes a prediction phase and an update phase. The state prediction formula is:

[0080] x t|t-1 =A t x t-1|t-1 (18)

[0081] In the formula, x t|t-1 This is a state prediction value based on the previous time step.

[0082] The formula for predicting covariance is:

[0083]

[0084] In the formula, P t|t-1 To predict the covariance matrix; Q t Let be the covariance matrix of the process noise.

[0085] The Kalman gain formula is:

[0086]

[0087] In the formula, R t Let be the covariance matrix of the observation noise.

[0088] The state update formula is:

[0089] x t|t =x t|t-1 +K t (y t -H t x t|t-1 ) (twenty one)

[0090] In the formula, xt|t This is the state estimate after fusing the observation data.

[0091] The covariance update formula is:

[0092] P t|t =(IK t H t )P t|t-1 (twenty two)

[0093] In the formula, P t|t This is the updated covariance matrix.

[0094] By incrementing the time step t and repeating the prediction and update phases at each time point until all data is processed, the daily production curve can be smoothed, abnormal fluctuations caused by institutional adjustments in the early stages of mining can be eliminated, and more stable input data can be provided for subsequent production capacity prediction.

[0095] To avoid the impact of dimensional differences between geological and fracture parameters on the accuracy and computational efficiency of the surrogate model, Z-score standardization is used to standardize the input features, eliminating the influence of dimensional differences in input parameters. The standardization equation is as follows:

[0096]

[0097] In the formula, x′ i The parameter value is the standardized value; μ zs The characteristic mean; σ zs The characteristic standard deviation is denoted as .

[0098] S3. Use super Latin cube sampling to generate a combination of geological parameters and fracture network parameters, substitute them into the production capacity numerical simulation model to obtain the calculated production capacity data, obtain the geological parameters-fracture network parameters-production capacity dataset, and construct a production capacity prediction proxy model based on the geological parameters-fracture network parameters-production capacity dataset.

[0099] Hyper-Latin cubic sampling generates uniformly distributed samples in each dimension and ensures the dispersion of sample points in the multidimensional space through permutation and optimization algorithms. Suppose we need to generate n sample points in a d-dimensional parameter space, where each parameter x... i The range of values ​​for is [x i,min ,x i,max Then, hyper-Latin cube sampling first divides the range of values ​​for each parameter into m non-overlapping intervals:

[0100]

[0101] In each interval M i,j Random value selected from within:

[0102]

[0103] The sample points are rearranged using the permutation matrix D so that the samples satisfy the minimum maximum spacing optimization criterion in multidimensional space:

[0104] x i,c =r i,D(i,c) ,i=1,...,d,c=1,...,m (26)

[0105] In the formula, D(i,c) is the interval index of the c-th sample point in the i-th dimension.

[0106] The optimization criterion for minimizing the maximum spacing is:

[0107] min D max 1≤i＜j≤m ||x i -x j ||2 (27)

[0108] In the formula, x i and x j Let be a d-dimensional vector of two sample points.

[0109] Geological and fracture parameter samples generated by hyper-Latin cube sampling are substituted into the production capacity numerical simulation model to calculate the corresponding production curves. Based on the geological parameters, fracture parameters, and production curves, a Gaussian process regression surrogate model for production capacity prediction is constructed. The Gaussian process model is a non-parametric regression method suitable for handling time series with smoothness and noise.

[0110] Gaussian process regression consists of two parts: mean prediction and variance prediction. The squared exponential kernel function is:

[0111]

[0112] In the formula, K(x) i ,x j ) represents the elements of the covariance matrix K; σ f n is the signal variance, which is dimensionless. tol L represents the total dimension of the parameters, which is dimensionless. d Let be the length scale parameter of the d-th dimension, which is dimensionless.

[0113] For the new input sample x * The mean of the predicted output is:

[0114]

[0115] In the formula, This is the transpose of the covariance matrix of the new input samples; Let be the observation noise variance; I be the identity matrix; and y be the predicted value matrix.

[0116] The variance of the predicted output is:

[0117]

[0118] Maximum likelihood estimation is used to optimize the kernel function hyperparameters and construct a surrogate model for capacity prediction. Subsequently, automatic fitting is carried out based on the surrogate model to improve computational efficiency.

[0119] S4. Using the production capacity prediction proxy model as a forward simulator, a Bayesian inference framework is constructed by combining the prior of normally distributed geological parameters. The likelihood function is defined as the sum of squared errors between the predicted production capacity and the measured production capacity under Gaussian noise. By Monte Carlo sampling, the model converges to the posterior distribution of the parameters, thereby realizing the uncertainty inversion of geological parameters and fracture parameters. The inversion results are then substituted into the numerical simulation model for long-term production capacity prediction.

[0120] In this invention, a capacity forecasting proxy model constructed using Gaussian process regression replaces traditional numerical simulation, reducing computational costs from hours to seconds and enabling large-scale sampling of Bayesian inversion. This method not only retrieves the optimal estimates of the inversion parameters but also fully characterizes the uncertainties of the parameters through posterior distributions and transfers them to capacity forecasting, providing a more reliable decision-making basis. Simultaneously, Monte Carlo sampling can automatically identify high-probability regions, avoiding traditional inversion methods from getting trapped in local optima, and demonstrating better applicability to highly nonlinear systems such as capacity forecasting.

[0121] Bayes' theorem is expressed as:

[0122]

[0123] In the formula, χ(θ|y act ) represents the posterior distribution of the parameters; θ represents the vector of geological and fracture parameters to be inverted; y act This refers to actual production data; χ(y) act |θ) is the likelihood function; χ(θ) is the prior distribution of the parameters; χ(y) act Marginal likelihood constant.

[0124] Assume the prior distribution of the parameters follows a multidimensional normal distribution:

[0125]

[0126] In the formula, μ0 is the prior mean vector; Σ0 is the prior covariance matrix.

[0127] Define the likelihood function as the sum of squared prediction errors under Gaussian noise:

[0128]

[0129] In the formula, σ ν δ represents the variance of the observation noise; y represents the number of data points;i,act Let f(θ) be the i-th actual production capacity data point; i ) represents the predicted value of the surrogate model under parameter θ.

[0130] Substitute into Bayes' formula:

[0131]

[0132] Generate posterior samples using the Metropolis-Hastings algorithm in Markov chain Monte Carlo sampling:

[0133]

[0134] In the formula, α is the acceptance probability; θ * The candidate parameters generated for the proposal distribution; q(·|·) is the proposal distribution.

[0135] The Markov chain Monte Carlo iterative sampling process first generates candidate parameters from the proposal distribution, calculates the acceptance probability of each candidate parameter, and collects the samples at this point as the posterior distribution after the Markov sampling chain stabilizes. The parameter with the smallest fitting error is then substituted into the capacity numerical simulation model to achieve long-term capacity prediction. The overall calculation process is as follows: Figure 3 As shown. Specific Implementation

[0137] A case study was conducted on well J1, a fractured well in a certain block. The geological and fracture basic parameters of well J1 were obtained from the fracturing report, as shown in Table 1. The uncertain parameters in the table were fitted.

[0138] Table 1 Basic parameters of well J1

[0139]

[0140] Well J1 produced for 350 days. Due to factors such as production testing, the data fluctuated significantly in the early stages of production. Kalman filtering was applied to the actual production data to reduce noise and improve data stability. The filtering results are as follows: Figure 4 As shown.

[0141] The region exhibits a significant difference in horizontal principal stresses, resulting in generally straight artificial fractures. Therefore, the average fracture value is simplified and used as the geometric parameter for the hydraulic fractures. Hyper-Latin cubic sampling was conducted based on the fundamental parameters in Table 1, and this sampling scheme was substituted into the production capacity numerical simulation model for a 350-day production capacity simulation. For this case, 100 parameter combinations were sampled for production capacity simulation. 80% of the samples were used as the training set and 20% as the test set to construct a Gaussian process regression production capacity prediction surrogate model. The comparison results between the surrogate model predictions and the numerical simulation values ​​under the same production time are shown below. Figure 5As shown, the average relative error of the training set is 3.4%, and the coefficient of determination is 0.9854. The average relative error of the test set is 5.1%, and the coefficient of determination is 0.9736. This indicates that the model training effect is good, and the surrogate model has strong generalization and robustness, and can effectively replace the numerical calculation model.

[0142] according to Figure 4 After filtering, the actual production data was automatically fitted to the uncertain parameters in Table 1. The multi-parameter fitting results of the hydraulic fracturing network are as follows: Figure 6 As shown. A total of 184 sets of Markov chain Monte Carlo iterations were carried out, with parameter scheme combinations as follows. Figure 6 As shown in Table 2, the dashed lines represent all possible solutions, and the solid lines represent the optimal parameter combination with the smallest error.

[0143] Table 2. Fitting Results of Optimal Mesh Parameters

[0144] Parameters / Units value Hydraulic fracture length L / m 280 <![CDATA[Hydraulic fracture width w f / mm]]> 2.45 <![CDATA[Hydraulic fracture height h f / m]]> 27.3 <![CDATA[Hydraulic fracture conductivity w f K f / D·m]]> 0.0198 <![CDATA[x-direction permeability k x / 10 -3 mD]]> 5.1 <![CDATA[y-direction permeability k y / 10 -3 mD]]> 4.8 <![CDATA[z-direction permeability k z / 10 -3 mD]]> 0.74

[0145] Substituting the optimal fitting parameters into the numerical simulation model and setting the simulated production time to 500 days, the capacity calculation results are as follows: Figure 7 As shown, the data was thinned and plotted to highlight the differences between the data. The average relative error between the numerical simulation results of the production capacity of the J1 well fitting section and the actual production data was 4.86%. Using the production data from day 350 to 400 as the test section, the average relative error between the numerical simulation results and the actual production data was 2.61%, indicating high production capacity prediction accuracy and proving the effectiveness of the technical solution of this invention.

[0146] Therefore, the method proposed in this invention can achieve rapid and accurate historical fitting of multiple parameters of the fracture network in unconventional reservoir fracturing horizontal wells, and realize long-term production prediction, laying the foundation for the final recovery rate assessment.

[0147] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for intelligent prediction and fitting of dynamic production of unconventional reservoir fractured horizontal wells, comprising the following steps: S1. Introduce the gas and rock elastic state equations to describe the gas expansion and pore compression effects, establish an anisotropic seepage mathematical model coupled with the gas-rock dual compression mechanism, and construct a numerical simulation model of unconventional reservoir gas production capacity. S2. Determine the range of geological parameters and joint mesh parameters based on geological and construction data, and determine the production data based on actual production data; S3. Use super Latin cube sampling to generate a combination of geological parameters and fracture network parameters, substitute them into the production capacity numerical simulation model to obtain the calculated production capacity data, obtain the geological parameters-fracture network parameters-production capacity dataset, and construct a production capacity prediction proxy model based on the geological parameters-fracture network parameters-production capacity dataset. S4. Using the production capacity prediction proxy model as a forward simulator, a Bayesian inference framework is constructed by combining the prior of normal distribution geological parameters. The likelihood function is defined as the sum of squared errors between the predicted production capacity and the measured production capacity under Gaussian noise. By Monte Carlo sampling, the model converges to the posterior distribution of the parameters, thereby realizing the uncertainty inversion of geological parameters and fracture parameters. The inversion results are then substituted into the numerical simulation model for long-term production prediction.

2. The intelligent prediction and fitting method for dynamic production of unconventional reservoir fractured horizontal wells as described in claim 1, wherein the anisotropic seepage mathematical model in step S1 is: In the formula, k x Let m be the permeability in the x-direction. 2 ;k y Let m be the permeability in the y-direction. 2 ;k z Let m be the permeability in the z-direction. 2 μ is the gas viscosity, Pa·s; φ0 is the initial porosity, dimensionless; C t The overall compressibility factor is expressed in MPa. -1 p represents pressure (Pa); t represents time (d).

3. The method for intelligent prediction and fitting of dynamic production of unconventional reservoir fractured horizontal wells as described in claim 1, further comprising step S2: The Kalman filter method is used to smooth the actual production data.

4. The method for intelligent prediction and fitting of dynamic production of unconventional reservoir fractured horizontal wells as described in claim 1, further comprising step S2: The Z-score method was used to standardize the geological parameters, fracture network parameters, and production data.

5. The method for intelligent prediction and fitting of dynamic production of unconventional reservoir fractured horizontal wells as described in claim 1, wherein the likelihood function expression in step S4 is: In the formula, σ ν δ represents the variance of the observation noise; y represents the number of data points; i,act Let f(θ) be the i-th actual production capacity data point; i ) represents the predicted value of the surrogate model under parameter θ.

6. A computer-readable storage medium comprising instructions that, when executed on a computer, cause the computer to perform the method as described in any one of claims 1-5.

7. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the method as described in any one of claims 1-5.

Images