A method, apparatus, system, and storage medium for geothermal field parameter inversion calculation.

Abstract

This invention discloses a method, apparatus, system, and storage medium for geothermal field parameter inversion calculation, comprising: establishing a geophysical layered model; calculating temperature values ​​by Monte Carlo sampling of the preset parameter space of crustal heat generation rate and thermal conductivity, combined with the heat conduction equation, and constructing a Gaussian likelihood function to evaluate the matching degree between the model's predicted temperature and the measured temperature; obtaining the optimal estimate, confidence interval, and correlation matrix between parameters based on the posterior probability distribution; using the high-confidence results generated by Monte Carlo inversion as prior knowledge for a physical guided neural network (PINNs) based on the correlation matrix, and finally outputting the predicted spatial distribution of thermal parameters in the target area through parameter initialization constraints, output layer range limitations, and physical regularization loss terms. The technical solution of this invention overcomes the shortcomings of existing technologies, such as the scarcity of measured geothermal field parameters that cannot characterize regional features, and the inability of existing technologies to utilize geophysical joint inversion of geothermal field parameters.

Description

Technical Field

[0001] This invention belongs to the field of geothermal resource exploration technology, and particularly relates to a method, device, system, and storage medium for geothermal field parameter inversion calculation. Background Technology

[0002] Geothermal resources, as an important clean and renewable energy source, play a crucial role in the transformation of the energy structure. The outward transfer of heat from the Earth's interior is the primary source of geothermal resources; therefore, the Earth's internal thermal structure is a key factor influencing their distribution. However, the Earth's internal geological structure is complex and variable, and traditional single geophysical inversion methods for analyzing thermal structure have significant limitations. Therefore, accurately obtaining geothermal field parameters such as crustal heat generation rate, thermal conductivity, and temperature distribution is a core step in assessing the development potential of geothermal resources and has significant geological implications for geothermal resource exploration.

[0003] Commonly used gravity exploration can obtain information on subsurface density distribution, but it is difficult to directly correlate it with thermal conductivity characteristics; magnetic exploration focuses on the analysis of subsurface magnetic structures, and its intrinsic connection with geothermal parameters is weak; electromagnetic exploration is susceptible to electromagnetic interference and lacks accuracy in the inversion of deep geothermal parameters. Existing methods all have shortcomings in quantifying parameter uncertainties, leading to unreliable geothermal resource assessment results. In recent years, although multi-geophysical joint inversion and data-driven inversion technologies have made some progress, how to efficiently integrate multi-geophysical information under complex geological conditions, fully consider physical laws and parameter uncertainties during the inversion process, and address the problems of low computational efficiency remain critical technical challenges in the field of geothermal exploration. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a method, device, system and storage medium for geothermal field parameter inversion calculation, overcoming the shortcomings of the existing technology that cannot utilize geophysical joint inversion of thermal structure parameters.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] A method for inverting and calculating geothermal field parameters includes:

[0007] Step S1: Obtain topographic maps, basement depth maps, Moho depth maps, magnetic basement depth maps, and Curie depth maps of the target area, and establish a geophysical layered model;

[0008] Step S2: Based on the geophysical layered model, the temperature value is calculated by Monte Carlo sampling of the preset parameter space of crustal heat generation rate and thermal conductivity, combined with the heat conduction equation. A Gaussian likelihood function is constructed to evaluate the degree of matching between the model's predicted temperature and the measured temperature. The optimal estimate, confidence interval, and correlation matrix between parameters are obtained based on the posterior probability distribution.

[0009] Step S3: Based on the correlation matrix, the high-confidence results generated by Monte Carlo inversion are used as prior knowledge for the physical guided neural network PINNs. Through parameter initialization constraints, output layer range limitations, and physical regularization loss terms, the final output is the predicted result of the spatial distribution of thermal parameters in the target area.

[0010] Preferably, in step S2, the Gaussian likelihood function is constructed by combining the heat conduction equation calculation with the measured temperature data to evaluate the degree of matching between the model's predicted temperature and the measured borehole temperature value, i.e.:

[0011] ;

[0012] in, This is the measured temperature vector from the borehole. These are parameters for the geophysical layered model. The number of data points. This is the data covariance matrix , This is the temperature residual vector.

[0013] Preferably, the total loss function of the neural network includes: a data fitting loss term, used to measure the difference between the temperature predicted by PINNs and the temperature of the high-likelihood sample in Monte Carlo; a physical regularization loss term, used to ensure that the prediction parameters satisfy the residual constraints of the heat conduction equation; and a geological prior regularization loss term, used to force the correlation of the prediction parameters to match the Monte Carlo results.

[0014] The present invention also provides a geothermal field parameter inversion calculation device, comprising:

[0015] The first processing module is used to acquire topographic maps, basement depth maps, Moho discontinuity depth maps, magnetic basement depth maps, and Curie depth maps of the target area, and to establish a geophysical layered model.

[0016] The second processing module is used to calculate temperature values ​​based on the geophysical layered model by Monte Carlo sampling of the preset parameter space of crustal heat generation rate and thermal conductivity, combined with the heat conduction equation, constructing a Gaussian likelihood function to evaluate the degree of matching between the model's predicted temperature and the measured temperature; and obtaining the optimal estimate, confidence interval, and correlation matrix between parameters based on the posterior probability distribution.

[0017] The third processing module is used to take the high-confidence results generated by Monte Carlo inversion as prior knowledge of the physical guided neural network PINNs based on the correlation matrix. Through parameter initialization constraints, output layer range limitations and physical regularization loss terms, it finally outputs the predicted results of the spatial distribution of thermal parameters in the target area.

[0018] Preferably, the second processing module combines the heat conduction equation calculation with measured temperature data to construct a Gaussian likelihood function to evaluate the degree of matching between the model's predicted temperature and the measured borehole temperature value, i.e.:

[0019] ;

[0020] in, This is the measured temperature vector from the borehole. These are parameters for the geophysical layered model. The number of data points. This is the data covariance matrix , This is the temperature residual vector.

[0021] Preferably, the total loss function of the neural network includes: a data fitting loss term, used to measure the difference between the temperature predicted by PINNs and the temperature of the high-likelihood sample in Monte Carlo; a physical regularization loss term, used to ensure that the prediction parameters satisfy the residual constraints of the heat conduction equation; and a geological prior regularization loss term, used to force the correlation of the prediction parameters to match the Monte Carlo results.

[0022] The present invention also provides a geothermal field parameter inversion calculation system, comprising: a memory and a processor, wherein the memory stores a computer program executed by the processor, and the computer program executes a geothermal field parameter inversion calculation method when executed by the processor.

[0023] The present invention also provides a storage medium storing a computer program, which executes a geothermal field parameter inversion calculation method when running.

[0024] This invention acquires the basement depth map, Moho depth map, magnetic basement depth map, and Curie depth map of the target area, and establishes a geophysical layered model with a clear thermal boundary. Through Monte Carlo sampling of parameters such as crustal heat generation rate and thermal conductivity, and combined with forward modeling of the heat conduction equation, temperature is calculated. Considering the spatial correlation of parameters, a Gaussian likelihood function is constructed to evaluate the matching degree between the model's predicted temperature and the measured temperature. Based on the posterior probability distribution, the optimal estimates, confidence intervals, and correlation matrices between parameters of temperature and thermal parameters are obtained. The results generated by Monte Carlo inversion are used as prior knowledge for Physically Guided Neural Networks (PINNs). Through parameter initialization constraints, output layer range limitations, and physical regularization loss terms, the final output is a probabilistic prediction of the spatial distribution of geothermal parameters in the target area. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0026] Figure 1 This is a flowchart of the geothermal field parameter inversion calculation method according to an embodiment of the present invention. Detailed Implementation

[0027] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0028] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0029] Example 1:

[0030] like Figure 1 As shown, this embodiment of the invention provides a method for inverting and calculating geothermal field parameters, including:

[0031] Step S1: Obtain topographic maps, basement depth maps, Moho depth maps, magnetic basement depth maps, and Curie depth maps of the target area, and establish a geophysical layered model;

[0032] Step S2: Based on the geophysical layered model, the temperature value is calculated by Monte Carlo sampling of the preset parameter space of crustal heat generation rate and thermal conductivity, combined with the heat conduction equation. A Gaussian likelihood function is constructed to evaluate the degree of matching between the model's predicted temperature and the measured temperature. The optimal estimate, confidence interval, and correlation matrix between parameters are obtained based on the posterior probability distribution.

[0033] Step S3: Based on the correlation matrix, the high-confidence results generated by Monte Carlo inversion are used as prior knowledge for the physical guided neural network PINNs. Through parameter initialization constraints, output layer range limitations, and physical regularization loss terms, the final output is the predicted result of the spatial distribution of thermal parameters in the target area.

[0034] As one embodiment of the present invention, in step S2, Monte Carlo random sampling is performed on the crustal heat generation rate H and thermal conductivity k:

[0035] (Log-normal distribution, mean) Standard deviation (Determined by geological experience)

[0036] (Uniformly distributed, with the range of values ​​determined by the rock type).

[0037] The steady-state one-dimensional heat conduction equation is solved using the finite difference method:

[0038] ;

[0039] Calculate depth point z j Measured temperature Ti(z) j ) and the calculated predicted temperature Tobs(z) j Residual:

[0040] ;

[0041] Assuming the error follows a Gaussian distribution, a likelihood function is constructed to evaluate the degree of matching between the model's predicted temperature and the actual borehole temperature. The function expression is as follows:

[0042] ;

[0043] in: It is the measured temperature vector of the borehole. These are the parameters of the geophysical layered model (rock thermal conductivity k, basement heat flow Qk, rock heat production A). It is the number of data points. It is the data covariance matrix This describes the uncertainty of temperature observations and the correlation between data points. This is the temperature residual vector.

[0044] Calculate the posterior probability distribution, and based on the posterior probability distribution, obtain the optimal estimate, confidence interval, and correlation matrix between the thermal parameters. The posterior distribution of the thermal parameters based on Bayes' theorem is:

[0045] ;

[0046] in, For the prior distribution of thermal parameters, This is the normalization constant.

[0047] Posterior distributions sampled using Markov chain Monte Carlo (MCMC) sampling. ,

[0048] Posterior mean estimate (minimum mean square error estimate):

[0049] ;

[0050] Maximum a posteriori estimation (MAP):

[0051] ;

[0052] For parameters (such as base heat flow) )of Confidence intervals, calculate posterior samples Quantiles:

[0053] ;

[0054] Calculate the covariance matrix and correlation matrix based on the posterior samples:

[0055] ;

[0056] in, This indicates that the parameters are positively correlated. This indicates that the parameters are negatively correlated.

[0057] A training set is constructed by selecting high-likelihood samples (confidence > 90%) from the posterior distribution obtained from the Monte Carlo inversion. Parameter information includes spatial location and thermal parameters (such as thermal conductivity, heat yield, etc.).

[0058] ;

[0059] in, The spatial coordinates of the sample point (x i y i 、z i ), These are the parameters of the layered model (such as stratigraphic number and lithology code). These are the thermal parameters retrieved from the MC (such as thermal conductivity κi and thermal yield Ai).

[0060] Construct a multilayer neural network (MLP).

[0061] ;

[0062] in, for That is, the input dimension (coordinates + model parameters). for That is, the output dimension (temperature + thermal parameters).

[0063] Using high likelihood samples from the Monte Carlo posterior distribution to extract high likelihood samples from the high likelihood sample set We select samples from the dataset and initialize the network parameters by minimizing the difference between the initial predictions and the high-likelihood samples.

[0064] ;

[0065] in, For the high likelihood sample set is the mean of the posterior distribution.

[0066] The output is constrained based on the 95% confidence interval of the Monte Carlo inversion.

[0067] ;

[0068] Construct the loss function. The total loss function includes: a data fitting loss term, used to measure the difference between the temperature predicted by PINNs and the temperature of the high-likelihood sample in Monte Carlo; a physical regularization loss term, used to ensure that the prediction parameters satisfy the residual constraints of the heat conduction equation; and a geological prior regularization loss term, used to force the correlation of the prediction parameters to match the Monte Carlo results.

[0069] ;

[0070] in, To fit the loss weights to the data, For data fitting loss term, For physical regularization loss weights, This is the physical regularization loss term. As the geological prior loss weight, This is a geological prior loss item.

[0071] in, Data fitting loss term:

[0072] ;

[0073] Where N is the total number of sampling points, T PINN The geothermal field distribution predicted by the PINN neural network. The temperature field is calculated by drawing the i-th high-likelihood sample from the Monte Carlo posterior distribution.

[0074] in, Physical regularization loss term:

[0075] ;

[0076] in, For the number of configuration points, , To calculate the thermal property parameters of each configuration point j based on the PINNs prediction results, The temperature gradient predicted for PINNs.

[0077] Among them, the geological prior loss item is:

[0078] ;

[0079] in, The KL divergence between the predicted distribution of PINNs and the Monte Carlo prior distribution measures the difference in probability distributions. This represents the predicted value of PINNs for the k-th thermophysical parameter; This represents the prior mean of the k-th parameter obtained from the Monte Carlo inversion. This represents the prior standard deviation of the k-th parameter obtained from the Monte Carlo inversion. and These represent the lower and upper bounds of the confidence interval for the k-th parameter, respectively. This represents a linear rectified function, implementing a penalty for one-sided out-of-bounds errors; This represents the total number of dimensions of the thermophysical parameters.

[0080] The Adam optimization algorithm was used to train the neural network. After training, the network was input into a geological model for prediction.

[0081] Using MC Dropout for multiple sampling, Dropout prediction is enabled, and K samples are performed for each spatial location (x, y, z) for prediction:

[0082] ;

[0083] in Indicates the first The parameter vector obtained from the sampling.

[0084] Calculate the statistics for each grid cell:

[0085] ;

[0086] Generate a spatial distribution raster map of geothermal parameters in the target area, including the mean, standard deviation, and confidence interval of temperature, thermal conductivity, and heat generation rate.

[0087] Example 2:

[0088] This invention also provides a geothermal field parameter inversion calculation device, comprising:

[0089] The first processing module is used to acquire topographic maps, basement depth maps, Moho discontinuity depth maps, magnetic basement depth maps, and Curie depth maps of the target area, and to establish a geophysical layered model.

[0090] The second processing module is used to calculate temperature values ​​based on the geophysical layered model by Monte Carlo sampling of the preset parameter space of crustal heat generation rate and thermal conductivity, combined with the heat conduction equation, constructing a Gaussian likelihood function to evaluate the degree of matching between the model's predicted temperature and the measured temperature; and obtaining the optimal estimate, confidence interval, and correlation matrix between parameters based on the posterior probability distribution.

[0091] The third processing module is used to take the high-confidence results generated by Monte Carlo inversion as prior knowledge of the physical guided neural network PINNs based on the correlation matrix. Through parameter initialization constraints, output layer range limitations and physical regularization loss terms, it finally outputs the predicted results of the spatial distribution of thermal parameters in the target area.

[0092] As one embodiment of the present invention, the second processing module combines the heat conduction equation to calculate and the measured temperature data to construct a Gaussian likelihood function to evaluate the degree of matching between the model's predicted temperature and the measured borehole temperature value, that is:

[0093] ;

[0094] in, This is the measured temperature vector from the borehole. These are parameters for the geophysical layered model. The number of data points. This is the data covariance matrix , This is the temperature residual vector.

[0095] As one embodiment of the present invention, the total loss function of the neural network includes: a data fitting loss term, used to measure the difference between the temperature predicted by PINNs and the temperature of the Monte Carlo high-likelihood sample; a physical regularization loss term, used to make the prediction parameters satisfy the residual constraints of the heat conduction equation; and a geological prior regularization loss term, used to force the correlation of the prediction parameters to match the Monte Carlo results.

[0096] Example 3:

[0097] This invention also provides a geothermal field parameter inversion calculation system, comprising: a memory and a processor, wherein the memory stores a computer program executed by the processor, and the computer program executes a geothermal field parameter inversion calculation method when run by the processor.

[0098] Example 4:

[0099] This invention also provides a storage medium storing a computer program, which executes a geothermal field parameter inversion calculation method during runtime.

[0100] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for inverting and calculating geothermal field parameters, characterized in that, include: Step S1: Obtain topographic maps, basement depth maps, Moho depth maps, magnetic basement depth maps, and Curie depth maps of the target area, and establish a geophysical layered model; Step S2: Based on the geophysical layered model, the temperature value is calculated by Monte Carlo sampling of the preset parameter space of crustal heat generation rate and thermal conductivity, combined with the heat conduction equation. A Gaussian likelihood function is constructed to evaluate the degree of matching between the model's predicted temperature and the measured temperature. The optimal estimate, confidence interval, and correlation matrix between parameters are obtained based on the posterior probability distribution. Step S3: Based on the correlation matrix, the high-confidence results generated by Monte Carlo inversion are used as the prior knowledge of the physical guided neural network PINNs. Through parameter initialization constraints, output layer range limitations, and physical regularization loss terms, the final output is the predicted result of the spatial distribution of thermal parameters in the target area. The total loss function of the neural network includes: a data fitting loss term, which measures the difference between the temperature predicted by PINNs and the temperature of the high likelihood sample in Monte Carlo; a physical regularization loss term, which makes the prediction parameters satisfy the residual constraints of the heat conduction equation; and a geological prior regularization loss term, which forces the correlation of the prediction parameters to match the Monte Carlo results. Physical regularization loss term for: Where, N coll κ represents the number of configuration points. PINN A PINN To calculate the thermal property parameters of each configuration point j based on the PINNs prediction results, The temperature gradient predicted for PINNs.

2. The geothermal field parameter inversion calculation method as described in claim 1, characterized in that, In step S2, by combining the heat conduction equation calculations with measured temperature data, a Gaussian likelihood function is constructed to assess the degree of matching between the model's predicted temperature and the measured borehole temperature value, i.e.: Among them, T obs Here, m represents the measured borehole temperature vector, m is the geophysical layered model parameter, N is the number of data points, and C is the temperature vector. d Let the data covariance matrix be N×N, e T This is the temperature residual vector.

3. A geothermal field parameter inversion calculation device, characterized in that, include: The first processing module is used to acquire topographic maps, basement depth maps, Moho discontinuity depth maps, magnetic basement depth maps, and Curie depth maps of the target area, and to establish a geophysical layered model. The second processing module is used to calculate temperature values ​​based on the geophysical layered model by Monte Carlo sampling of the preset parameter space of crustal heat generation rate and thermal conductivity, combined with the heat conduction equation, constructing a Gaussian likelihood function to evaluate the degree of matching between the model's predicted temperature and the measured temperature; and obtaining the optimal estimate, confidence interval, and correlation matrix between parameters based on the posterior probability distribution. The third processing module is used to take the high confidence results generated by Monte Carlo inversion as the prior knowledge of the physical guided neural network PINNs based on the correlation matrix. Through parameter initialization constraints, output layer range limits and physical regularization loss terms, it finally outputs the predicted results of the spatial distribution of thermal parameters in the target area. The total loss function of the neural network includes: a data fitting loss term, which measures the difference between the temperature predicted by PINNs and the temperature of the high likelihood sample in Monte Carlo; a physical regularization loss term, which makes the prediction parameters satisfy the residual constraints of the heat conduction equation; and a geological prior regularization loss term, which forces the correlation of the prediction parameters to match the Monte Carlo results. Physical regularization loss term for: Where, N coll κ represents the number of configuration points. PINN A PINN To calculate the thermal property parameters of each configuration point j based on the PINNs prediction results, The temperature gradient predicted for PINNs.

4. The geothermal field parameter inversion calculation device as described in claim 3, characterized in that, The second processing module combines the heat conduction equation calculations with measured temperature data to construct a Gaussian likelihood function to assess the degree of matching between the model's predicted temperature and the measured borehole temperature values, i.e.: Among them, T obs Here, m represents the measured borehole temperature vector, m is the geophysical layered model parameter, N is the number of data points, and C is the temperature vector. d Let the data covariance matrix be N×N, e T This is the temperature residual vector.

5. A geothermal field parameter inversion calculation system, characterized in that, include: A memory and a processor, wherein the memory stores a computer program executed by the processor, the computer program, when executed by the processor, performs the geothermal field parameter inversion calculation method as described in any one of claims 1-2.

6. A storage medium, characterized in that, The storage medium stores a computer program, which, when running, executes the geothermal field parameter inversion calculation method as described in any one of claims 1-2.

Images