Sensitivity analysis method for accelerating recognition of parameters, space and physical fields in multi-field coupling system based on deep learning

Abstract

The invention discloses a sensitivity analysis method for accelerating recognition of parameters, space and physical fields in a multi-field coupling system based on deep learning. The method comprises the following steps: constructing a multi-field coupling numerical model; establishing a multi-physical field coupling model of a field scale based on geological data; generating a training data set based on Monte Carlo simulation: generating a plurality of groups of parameter instances through Latin hypercube sampling, and executing numerical simulation to store an output result; training a deep learning agent model: adopting an improved ResNet-18 architecture, optimizing model parameters by taking R2 and RMSE as indexes, and establishing regression mapping of the parameters and output; analyzing three-level sensitivity: calculating a global average first-order Sobol index, and quantifying parameter sensitivity; physical field parameter contribution values are accumulated, dominant physical fields are distinguished, and a decoupling strategy is formulated; and drawing a spatial sensitivity distribution diagram, and optimizing the monitoring network layout. According to the method, key parameters, physical fields and space regions can be quickly and efficiently identified.

Description

Technical Field

[0001] The present invention belongs to the technical field of underground engineering resource development and utilization, and in particular relates to a sensitivity analysis method for accelerating identification of parameters, space and physical fields in a multi-field coupling system based on deep learning. Background Art

[0002] Temperature-seepage-stress-chemistry (THMC) multi-field coupling is a common phenomenon in underground space and resource development. In areas such as geothermal energy development, carbon dioxide geological storage, and underground nuclear waste disposal, accurately understanding and mastering this multi-field coupling process is crucial for optimizing resource extraction, assessing environmental risks, and ensuring the long-term safety of projects. However, the strong nonlinearity and complex multi-physics interdependencies of multi-field coupling present numerous challenges for existing models. On the one hand, models struggle to fully capture the complexities of the real world, often requiring simplifications that introduce uncertainty and reduce predictive accuracy. On the other hand, model parameter uncertainty, such as measurement error, scale effects, and unclear parameter relationships, further exacerbates modeling challenges. To address these issues, sensitivity analysis has become a key tool, identifying key sources of model uncertainty and providing guidance for model improvement and system regulation. Traditional methods for sensitivity analysis have limitations, such as: experimental methods offer high accuracy but are costly; mechanistic modeling methods are computationally complex and ignore higher-order coupling effects; and traditional machine learning methods fail to capture nonlinear relationships. Monte Carlo-based sampling uncertainty analysis plays a key role in characterizing and quantifying uncertainty and is considered the most reliable technique. Deep learning, an emerging prediction technology, can effectively capture nonlinear relationships in data and construct high-precision, highly generalizable regression prediction models. Therefore, using deep learning to accelerate the prediction and convergence of Monte Carlo global sensitivity analysis is crucial for quantifying uncertainty in complex multi-field coupled processes. Summary of the Invention

[0003] To solve the above technical problems, the present invention proposes a sensitivity analysis method based on deep learning to accelerate the identification of parameters, space and physical fields in multi-field coupling systems. It can quickly and efficiently identify key parameters, physical fields and spatial areas, and provide support for underground engineering decision-making.

[0004] To achieve the above objectives, the present invention provides a sensitivity analysis method for parameters, space, and physical fields in a multi-field coupled system based on deep learning acceleration, including:

[0005] Constructing multi-field coupled numerical models: Building field-scale multi-physics coupled models based on geological data;

[0006] Generate training data sets based on Monte Carlo simulation: Generate multiple sets of parameter instances through Latin hypercube sampling, perform numerical simulations, and store output results;

[0007] Training deep learning agent models: using the improved ResNet-18 architecture, with R 2 and RMSE as indicators to optimize model parameters and establish regression mapping between parameters and outputs;

[0008] Analyze the three-level sensitivity: calculate the global average first-order Sobol index to quantify parameter sensitivity; accumulate the contribution values ​​of physical field parameters, distinguish the dominant physical fields and formulate decoupling strategies; draw a spatial sensitivity distribution map and optimize the monitoring network layout.

[0009] Optionally, the multi-field coupling model is based on TOUGHREACT-FLAC 3D Software construction and parameter calibration are based on the geostatistical characteristics of the target reservoir.

[0010] Optionally, the deep learning proxy model comprises 4 stages and 6 residual units, capturing spatial heterogeneity through convolutional layers.

[0011] Optionally, the decoupling strategy includes: simplifying the CO2-fluid response to a seepage-stress coupling model; and adopting a seepage-stress-chemistry coupling model for the mineral response.

[0012] Optionally, the optimizing the monitoring network layout includes optimizing the monitoring point layout according to the spatial sensitivity distribution.

[0013] Optionally, the deep learning model training includes: 2 >0.98 and root mean square error RMSE are the optimization goals.

[0014] An electronic device, characterized in that the device includes: a processor and a memory storing computer program instructions; when the processor executes the computer program instructions, it implements the sensitivity analysis method for accelerating the identification of parameters, space and physical fields in a multi-field coupling system based on deep learning.

[0015] A computer storage medium, characterized in that computer program instructions are stored on the computer storage medium, and when the computer program instructions are executed by a processor, the sensitivity analysis method of parameters, space and physical fields in a multi-field coupling system based on deep learning acceleration is implemented.

[0016] Technical effect of the invention: The present invention discloses a sensitivity analysis method for parameters, space and physical fields in a multi-field coupling system based on deep learning acceleration identification. By constructing an integrated analysis framework based on deep learning, the efficiency and accuracy of uncertainty quantification in a multi-field coupling system are significantly improved, and a clear decision-making basis is provided for engineering optimization. First, in order to address the pain point of time-consuming calculations of traditional numerical models, in the sensitivity analysis process based on Monte Carlo simulation, the present invention adopts a deep learning proxy model based on an improved ResNet-18 framework to replace the traditional numerical model, thereby improving the computational efficiency by 4 orders of magnitude. At the same time, the deterministic coefficient R is guaranteed. 2 >0.98, effectively capturing the high-order coupling effects between parameters and the nonlinear mapping of parameters to responses. Furthermore, the present invention innovatively proposes a three-level sensitivity analysis system: quantifying parameter sensitivity through the global average first-order Sobol index, identifying key parameters such as injection flow rate and fracture spacing; decoupling the multi-field coupling mechanism through the cumulative contribution value of the physical field, and simplifying the full coupling to a seepage-stress coupling model for CO2-fluid response, and using a seepage-stress-chemistry coupling model for mineral reaction to avoid the error accumulation caused by traditional simplified methods; finally, optimizing the monitoring network layout through the spatial sensitivity distribution map, such as in supercritical carbon dioxide geothermal systems, deploying pressure sensors near production wells based on sensitive area identification. The present invention realizes full-chain optimization from efficient calculation to precise decision-making, and provides a standardized solution for the safe regulation of multi-field coupling systems in underground engineering. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The accompanying drawings, which constitute part of this application, are intended to provide a further understanding of this application. The exemplary embodiments and descriptions of this application are intended to explain this application and do not constitute an improper limitation on this application. In the accompanying drawings:

[0018] Figure 1 This is a flow chart of a sensitivity analysis method for accelerating identification of parameters, space, and physical fields in a multi-field coupling system based on deep learning according to an embodiment of the present invention;

[0019] Figure 2 A scatter plot of the Min-Max normalized results comparing the multi-field coupling numerical simulation and the alternative model prediction for an embodiment of the present invention;

[0020] Figure 3 Schematic diagram of the Sobol index of the first three parameters of all response variables in the embodiment of the present invention;

[0021] Figure 4 Schematic diagram of the first-order Sobol index (S1) distribution of the first three contributing parameters of CO2(aq) concentration in an embodiment of the present invention. DETAILED DESCRIPTION

[0022] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments in this application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0023] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0024] Solutions in the prior art include:

[0025] Experimental methods: Physical experiments (e.g., core flooding) directly measure the impact of multi-field coupling model parameters (e.g., porosity, adsorption coefficient, displacement pressure) on system responses (e.g., porosity, permeability). These methods provide reliable and accurate data, making them suitable for small-scale mechanism verification. However, they are expensive, time-consuming, and difficult to replicate complex subsurface conditions (e.g., heterogeneity and multi-field coupling). In multi-field coupling systems, they are often used to calibrate mechanism models or validate proxy models, but they cannot cover the entire parameter space.

[0026] Mechanistic modeling: Based on physical equations and the universal properties of rocks, a multi-field coupled equation set is directly constructed and solved by numerical methods. Its advantage is that it has strong physical interpretability and supports long-term predictions (such as the millennial evolution of nuclear waste repositories), but the computational cost is extremely high (a single simulation requires hours of CPU time), and high-order coupling effects (such as chemical-mechanical feedback) are often simplified. Typical tools include TOUGHREACT-FLAC 3D , which is suitable for mechanistic studies but difficult to use for large-scale sensitivity analysis.

[0027] Traditional machine learning (such as support vector machines and random forest models) uses statistical learning to establish a mapping relationship between parameters and responses. This approach is computationally efficient (predictions in seconds) and suitable for parameter screening (such as ranking geothermal-sensitive parameters). However, it relies on large amounts of data, is inadequate for capturing high-dimensional nonlinear relationships, and suffers from poor physical interpretability.

[0028] Computer Experiment Design and Analysis: This approach uses sampling designs (such as Latin Hypercube Sampling) and surrogate models (Kriging, PCE) to efficiently explore parameter space, combined with the Sobol index to quantify sensitivity. This approach balances efficiency and accuracy, clearly separating main and interaction effects, and is suitable for medium-dimensional problems. However, it faces limitations such as the curse of high dimensionality (sample size increases dramatically when parameter dimensions exceed 20), insufficient capture of nonlinearities, and difficulty analyzing multi-field coupling (requiring separate-field modeling).

[0029] The goal of this paper is to provide an integrated framework that combines deep learning-based surrogate models with global sensitivity analysis to address the computational inefficiencies and poor accuracy of existing techniques for quantifying uncertainty in multi-field coupled systems. This framework aims to efficiently analyze the sensitivities of model parameters, physical fields, and spatial domains, thereby identifying key parameters, optimizing system performance, and providing a more reliable basis for engineering design and decision-making.

[0030] like Figure 1 As shown, this embodiment provides a sensitivity analysis method for parameters, space, and physical fields in a multi-field coupling system based on deep learning acceleration, including:

[0031] Constructing multi-field coupled numerical models: Building field-scale multi-physics coupled models based on geological data;

[0032] Generate training data sets based on Monte Carlo simulation: Generate multiple sets of parameter instances through Latin hypercube sampling, perform numerical simulations, and store output results;

[0033] Training deep learning agent models: using the improved ResNet-18 architecture, with R 2 and RMSE as indicators to optimize model parameters and establish regression mapping between parameters and outputs;

[0034] Analyze the three-level sensitivity: calculate the global average first-order Sobol index to quantify parameter sensitivity; accumulate the contribution values ​​of physical field parameters, distinguish the dominant physical fields and formulate decoupling strategies; draw a spatial sensitivity distribution map and optimize the monitoring network layout.

[0035] Furthermore, the multi-field coupling model is based on TOUGHREACT-FLAC 3D Software construction and parameter calibration are based on the geostatistical characteristics of the target reservoir.

[0036] Furthermore, the deep learning proxy model contains 4 stages and 6 residual units, capturing spatial heterogeneity through convolutional layers.

[0037] Furthermore, the decoupling strategy includes: simplifying the CO2-fluid response to a seepage-stress coupling model; and adopting a seepage-stress-chemistry coupling model for the mineral response.

[0038] Furthermore, the optimizing the monitoring network layout includes optimizing the monitoring point layout according to the spatial sensitivity distribution.

[0039] Furthermore, the deep learning model training includes: 2 >0.98 and root mean square error RMSE are the optimization goals.

[0040] Specifically, such as Figure 1The implementation process of the present invention shown includes:

[0041] The multi-field coupling model of the present invention is based on the geological and geophysical data of the study area and uses the TOUGHREACT-FLAC 3D The coupled framework constructs a field-scale supercritical CO2-enhanced geothermal system (scCO2-EGS) numerical model, or based on other sites or other software, it only needs to provide a multi-field coupled forward model.

[0042] The training data set of the present invention selects input parameters and prior intervals, performs numerical simulation based on Latin hypercube sampling (LHS), and generates multiple groups of samples as the data basis for subsequent replacement model construction.

[0043] The present invention is based on a proxy model of deep learning (DNN): an improved ResNet-18 architecture is used to construct a deep neural network alternative model. The model consists of 4 stages and 6 residual units. The convolutional layer captures the impact of structural heterogeneity on the spatial characteristics of the output response, and the residual unit alleviates the gradient vanishing and gradient explosion problems, accelerating the convergence of the model. During the training process, the model parameters are optimized by minimizing the regularized L1 norm loss function, so that the model can accurately establish the regression mapping relationship between the multi-field coupling model parameters and the system output. According to the research purpose and the characteristics of the data set, a suitable deep learning algorithm is selected and constructed, and the preprocessed data is input into the model for training to determine the optimal parameter combination. The deterministic coefficient (R 2 ) and root mean square error (RMSE) were used as evaluation indicators of model goodness of fit to ensure that the model prediction effect supported the subsequent Monte Carlo sensitivity analysis.

[0044] The sensitivity analysis module of the present invention performs Sobol sampling and uses a trained surrogate model to predict and output the sampled parameter set. The direct contribution of the parameter to the response variable is quantified based on the average first-order Sobol index across the entire model domain. The physical field that dominates the response is determined based on the cumulative contribution of all parameters in each physical field, and a decoupling scheme is customized accordingly. Further site information is obtained from the regional distribution of sensitivity, and the monitoring network is rationally configured based on the spatial sensitivity differences of the grid.

[0045] Taking the simulation of a multi-field fully coupled supercritical CO2-enhanced geothermal system as an example, this framework is used to conduct a multi-level global sensitivity analysis of the system. The implementation process is as follows:

[0046] (1) Data collection and model construction: The scCO2-EGS conceptual model was designed based on the geological conditions and injection and production scheme of the research site. 3DNumerical simulation software establishes a multi-field coupling model and determines the initial conditions and boundary conditions of the model, such as formation temperature, injection fluid, and the locations of injection wells and production wells.

[0047] (2) Generation of training data sets: 24 key parameters (injection rate, rock specific heat capacity, initial fracture aperture, etc.) and 14 response variables (CO2(aq) concentration, porosity, etc.) of the multi-field coupling model were determined and assigned prior distributions. Subsequently, a Latin hypercube sampling method was used to generate 1,000 parameter instances in the defined multidimensional parameter space. Each set of parameters was then input into the multi-field coupling model, and an integrated simulation was performed using a high-performance computing cluster. The model output results were systematically stored. The 1,000 samples were divided into a training set and a test set in a ratio of 9:1 for subsequent training and evaluation of alternative models.

[0048] (3) Construction and training of alternative models: First, the data set is preprocessed, including logarithmic transformation and Min-Max normalization of the response variable data output by the model to avoid problems such as difficulty in model convergence caused by differences in the order of magnitude between parameters. At the same time, the parameters are decomposed into homogeneous and heterogeneous components according to their spatial distribution characteristics, mapped to images and spliced ​​into a 100×100 matrix to facilitate subsequent feature extraction. The Resnet-18 model is selected and improved for deep learning training of the preprocessed data. The deterministic coefficient and root mean square error are used as evaluation indicators of the model fit goodness, and an acceptable hyperparameter combination is determined to establish a good regression mapping relationship between the multi-field coupling model parameters and the dynamic system output (such as permeability, porosity, solute concentration and mineral volume fraction). The R of the alternative model corresponding to all response variables is finally obtained. 2 The index is greater than 0.98, such as Figure 2 shown.

[0049] (4) Global sensitivity analysis: Sobol sampling (53,248 instances) was used to perform sensitivity analysis to quantify the Sobol index of the 24 parameters that control the multi-field coupling model to the output of each grid in the scCO2-EGS model. In this process, the output was efficiently predicted with the help of alternative models. The sensitivity analysis was divided into three parts: (i) Parameter sensitivity analysis was quantified by the domain-averaged first-order Sobol index, which represents the average contribution of a single input parameter to the model output over all spatial grids, to identify key parameters that have a greater impact on the model output, such as Figure 3(ii) The sensitivity contribution of each physical field to the model output is quantified by accumulating the S1 of all parameters in the field to clarify the importance of different physical fields in the system, and accordingly reasonably simplify the coupling effects between physical fields; (iii) Spatial sensitivity is characterized by plotting spatially resolved S1 to intuitively show the spatial heterogeneity of the contribution of parameters to the model output in the simulation area, providing a basis for optimizing the design of the monitoring network. Figure 4 shown.

[0050] An electronic device, characterized in that the device includes: a processor and a memory storing computer program instructions; when the processor executes the computer program instructions, it implements the sensitivity analysis method for accelerating the identification of parameters, space and physical fields in a multi-field coupling system based on deep learning.

[0051] A computer storage medium, characterized in that computer program instructions are stored on the computer storage medium, and when the computer program instructions are executed by a processor, the sensitivity analysis method of parameters, space and physical fields in a multi-field coupling system based on deep learning acceleration is implemented.

[0052] Alternative solution 1: In terms of alternative model construction, other methods can be tried to construct alternative models, such as polynomial model, radial basis function (RBF), Kriging model, DACE, etc.

[0053] Alternative Option 2: In terms of sensitivity analysis methods, other global sensitivity analysis methods can be considered to replace the Sobol index method, such as Fourier amplitude sensitivity test (FAST).

[0054] Alternative solution three: In terms of sampling, for simpler models, orthogonal experimental design can be used instead of Latin hypercube sampling (LHS) to generate training data sets while ensuring accuracy. This can effectively reduce the number of samples and thus improve computational efficiency.

[0055] The present invention is applicable to various underground engineering systems involving multi-field coupling processes, including but not limited to geothermal energy development, carbon dioxide geological storage, underground nuclear waste disposal, groundwater resource management and other fields. It is used to analyze the impact of the uncertainty of parameters, physical fields and spatial domains in the system on system performance and results.

[0056] The present invention discloses a sensitivity analysis method for accelerating the identification of parameters, space and physical fields in a multi-field coupling system based on deep learning. By constructing an integrated analysis framework based on deep learning, the efficiency and accuracy of uncertainty quantification in a multi-field coupling system are significantly improved, and a clear decision-making basis is provided for engineering optimization. First, in response to the pain point that traditional numerical models take a long time to calculate, in the sensitivity analysis process based on Monte Carlo simulation, the present invention uses a deep learning proxy model based on an improved ResNet-18 framework to replace the traditional numerical model, thereby increasing the computational efficiency by 4 orders of magnitude. At the same time, the deterministic coefficient R2>0.98 is guaranteed to effectively capture the high-order coupling effects between parameters and the nonlinear mapping of parameters to responses. Furthermore, the present invention innovatively proposes a three-level sensitivity analysis system: Parameter sensitivity is quantified by the global average first-order Sobol index, identifying key parameters such as injection rate and fracture spacing; the multi-field coupling mechanism is decoupled through the cumulative contribution value of the physical field. For CO2-fluid response, the full coupling can be simplified to a seepage-stress coupling model, and for mineral reactions, a seepage-stress-chemistry coupling model is adopted to avoid the error accumulation caused by traditional simplified methods; finally, the monitoring network layout is optimized through spatial sensitivity distribution maps. For example, in supercritical carbon dioxide geothermal systems, pressure sensors are deployed near production wells based on sensitive area identification. This technology achieves full-chain optimization from efficient calculation to precise decision-making, providing a standardized solution for the safe regulation of multi-field coupling systems in underground engineering.

[0057] The above are merely preferred embodiments of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A method for analyzing the sensitivity of parameters, space and physical fields in a multi-field coupled system based on deep learning acceleration, characterized in that: include: Constructing multi-field coupled numerical models: Building field-scale multi-physics coupled models based on geological data; Generate training data sets based on Monte Carlo simulation: Generate multiple sets of parameter instances through Latin hypercube sampling, perform numerical simulations, and store output results; Training deep learning agent models: using the improved ResNet-18 architecture, with R 2 and RMSE as indicators to optimize model parameters and establish regression mapping between parameters and outputs; Analyze three-level sensitivity: calculate the global average first-order Sobol index to quantify parameter sensitivity; accumulate the contribution values ​​of physical field parameters, distinguish the dominant physical fields and formulate decoupling strategies; Draw a spatial sensitivity distribution map and optimize the monitoring network layout.

2. The sensitivity analysis method for parameters, space and physical fields in a multi-field coupling system based on deep learning acceleration according to claim 1, characterized in that: The multi-field coupling model is based on TOUGHREACT-FLAC 3D Software construction and parameter calibration are based on the geostatistical characteristics of the target reservoir.

3. The sensitivity analysis method for parameters, space and physical fields in a multi-field coupling system based on deep learning acceleration as claimed in claim 1, characterized in that: The proposed deep learning agent model consists of 4 stages and 6 residual units, capturing spatial heterogeneity through convolutional layers.

4. The method for analyzing the sensitivity of parameters, space, and physical fields in a multi-field coupled system based on deep learning acceleration according to claim 1, wherein: The decoupling strategy includes: simplifying the CO2-fluid response to a seepage-stress coupling model; and adopting a seepage-stress-chemistry coupling model for the mineral response.

5. The sensitivity analysis method for parameters, space and physical fields in a multi-field coupling system based on deep learning acceleration identification according to claim 1, characterized in that: The optimizing the monitoring network layout includes optimizing the monitoring point layout according to the spatial sensitivity distribution.

6. The method for analyzing the sensitivity of parameters, space, and physical fields in a multi-field coupled system based on deep learning acceleration according to claim 1, wherein: The deep learning model training includes: 2 >0.98 and root mean square error RMSE are the optimization goals.

7. An electronic device, characterized in that: The device includes: a processor and a memory storing computer program instructions; when the processor executes the computer program instructions, it implements the sensitivity analysis method for accelerating identification of parameters, space and physical fields in a multi-field coupling system based on deep learning as described in any one of claims 1 to 6.

8. A computer storage medium, characterized in that The computer storage medium stores computer program instructions, which, when executed by a processor, implement the sensitivity analysis method for accelerating identification of parameters, space, and physical fields in a multi-field coupling system based on deep learning as described in any one of claims 1 to 6.

Images