A dam multi-physics field coupling design method, device, equipment and medium

Abstract

The present application relates to the technical field of water conservancy and hydropower engineering design, and discloses a dam multi-physical field coupling design method, device, equipment and medium, the present application constructs a heterogeneous hybrid solution architecture of embedding a machine learning model and a numerical solution model which is constrained by a physical control equation, and differentiates model selection based on the difference in the rate of change of each physical field in space and time, fundamentally solving the core contradiction between the low efficiency of traditional pure numerical simulation and the lack of physical constraints in pure data-driven machine learning models which are prone to non-physical solutions; at the same time, through mutual feedback type two-way coupling iteration of the response results of each physical field, the solution efficiency and solution accuracy are simultaneously considered; on this basis, combined with the adaptive screening mechanism of the multi-working-condition differentiated response weight strategy, the design scheme can dynamically adapt to the core needs of different service working conditions, significantly improving the engineering practicability of the design scheme.

Description

Technical Field

[0001] This invention relates to the field of water conservancy and hydropower engineering design technology, specifically to a multi-physics field coupling design method, device, equipment, and medium for dams. Background Technology

[0002] Hydropower dams are typical complex multiphysics coupled structures, simultaneously subjected to the superposition of multiple fields, including seepage, stress-strain, temperature, and dynamic-fluid-structure interaction, throughout their service life. Strong nonlinear interactions exist between these physical fields, making multiphysics coupled design a core and critical factor determining the dam's safety performance and overall benefits. Currently, the mainstream technical approaches for multiphysics coupled design of hydropower dams fall into two main categories: One type is the pure physics solution technique based on numerical simulation methods such as the finite element method (FEM), scaled boundary finite element method (SBFEM), and finite volume method (FVM). This method constructs a refined mesh model of the dam body, foundation, and surrounding environment, substitutes the corresponding physical control equations, and uses iterative algorithms to solve for the multi-physics coupling distribution. Based on the solution results, design parameters such as dam cross-sectional dimensions and material mix proportions are then adjusted. This type of method offers high computational accuracy, but the computational load increases exponentially with mesh density and coupling complexity. Solving a single design scheme often takes several days or even weeks, making it difficult to meet the needs of rapid iteration and real-time comparison of multiple schemes during the design phase.

[0003] Another type is the data-driven design-aided technology based on neural networks (CNN, RNN), deep learning (DNN), and other data-driven models. These technologies rely on a large amount of historical engineering data and on-site monitoring data to train predictive models and quickly output multi-physics response results, assisting designers in optimizing solutions. However, this type of method depends entirely on fitting the probability distribution of data samples and lacks intrinsic constraints on core physical mechanisms such as hydraulics and seepage mechanics. When faced with insufficient training data or skewed data distribution, it is highly prone to generating non-physical solutions that violate the fundamental laws of engineering physics, resulting in design schemes with significant safety hazards.

[0004] In addition, existing design methods often focus on optimizing a single performance index, making it difficult to take into account multiple dimensions such as safety performance, economic cost, construction period and long-term durability. Moreover, the modeling process is static and fixed, unable to dynamically respond to sudden changes in geological conditions during construction or to make differentiated adaptations for different service conditions, resulting in a serious disconnect between the design scheme and the dynamic feedback of the actual project. Summary of the Invention

[0005] This invention provides a multi-physics coupling design method, device, equipment, and medium for dams, which solves the problems in the prior art where it is difficult to balance the physical solution efficiency and solution accuracy in multi-physics coupling scenarios, and the design process cannot be dynamically and adaptively adjusted according to changes in working conditions.

[0006] In a first aspect, the present invention provides a multi-physics field coupled design method for dams, comprising: acquiring data of each physical field of the dam, inputting them into a solution model selected based on the differences in the spatiotemporal change rates of each physical field for solution, and obtaining the response results of each physical field; wherein the solution model includes at least a numerical solution model and a machine learning model, and the machine learning model is embedded with physical control equation constraints; performing a bidirectional coupling iteration of the response results of each physical field in a mutual feedback manner to obtain a coupling result; based on the coupling result, obtaining multiple feasible design schemes by adjusting the design variables of the dam; and selecting the optimal design scheme of the dam from the multiple feasible design schemes based on a preset differentiated response weight strategy for different service conditions.

[0007] This invention fundamentally solves the core contradiction between the low computational efficiency of traditional pure numerical simulation and the lack of physical constraints in pure data-driven machine learning models, which are prone to non-physical solutions, by constructing a heterogeneous hybrid solution architecture that integrates machine learning models with embedded physical control equation constraints and numerical solution models. Furthermore, it achieves a balance between solution efficiency and accuracy through mutual feedback and bidirectional coupling iteration of the response results of each physical field. Based on this, an adaptive selection mechanism using a multi-condition differentiated response weight strategy enables the design scheme to dynamically adapt to the core requirements of different service conditions, significantly improving the engineering practicality of the design scheme.

[0008] In one alternative implementation, the numerical solution model is a scaled boundary finite element model. Using a scaled boundary finite element model can significantly reduce the mesh degrees of freedom and computational scale, while ensuring the high-precision solution requirements for seismic dynamic response and other aspects, and further improve the overall computational efficiency of multi-physics coupled solution.

[0009] In one optional implementation, the machine learning model is a physically informed neural network, constructed by: building a deep network structure for the machine learning model, including an input layer, hidden layers, and an output layer; embedding the physical control equations into the loss function of the machine learning model to form a double loss function containing both data loss and physical loss terms; training the machine learning model using pre-collected historical dam data, and adjusting the network parameters to bring the double loss function to converge. By constructing the machine learning model as a physically informed neural network and embedding the physical control equations into the loss function to form a dual constraint of data loss and physical loss, the model can both fit the patterns of historical data and follow the laws of engineering mechanics during training, eliminating the generation of non-physical interpretations from the source, and significantly improving the physical interpretability and predictive reliability of the model.

[0010] In one optional implementation, the solution model is selected based on the differences in the spatiotemporal change rates of each physical field. This includes: dividing each physical field into a first type of physical field and a second type of physical field according to the differences in their spatiotemporal change rates, wherein the spatiotemporal change rate of the first type of physical field is lower than that of the second type of physical field; for the first type of physical field, a machine learning model is used for solution; for the second type of physical field, a numerical solution model is used for solution. By dividing each physical field into a first type of physical field and a second type of physical field according to the differences in their spatiotemporal change rates, and using machine learning models and numerical solution models respectively for differentiated solutions, a heterogeneous hybrid solution architecture of fast prediction of slow-changing fields and fine calculation of fast-changing fields is realized. This effectively avoids the waste of computational resources caused by the traditional method of solving all physical fields in a one-size-fits-all manner, and significantly improves the overall computational efficiency while ensuring the accuracy of the solution.

[0011] In one optional implementation, the response results of each physical field are subjected to a mutual feedback bidirectional coupling iteration, including: using the response result of the first type of physical field as the boundary condition for solving the second type of physical field; and feeding back the response result of the second type of physical field to the solution model of the first type of physical field to correct the parameters of the solution model of the first type of physical field. By establishing a mutual feedback bidirectional coupling iteration mechanism between the first and second types of physical fields, the response result of the slowly varying field is used as the boundary condition input for solving the rapidly varying field, while the response result of the rapidly varying field is fed back to correct the parameters of the solution model of the slowly varying field. This achieves co-evolution and dynamic calibration between the two heterogeneous solution models, effectively eliminating the error accumulation caused by unidirectional data transmission, and significantly improving the overall accuracy and convergence stability of multi-physics coupling solution.

[0012] In one optional implementation, multiple feasible design schemes are obtained by adjusting the dam's design variables. This includes using the dam's geometric and material parameters as design variables, and pre-defined specifications, geological conditions, and construction techniques as constraints. Based on the coupling results, multi-objective optimization is performed on the design variables to obtain multiple feasible design schemes. By using the dam's geometric and material parameters as design variables and performing multi-objective optimization under multiple constraints of pre-defined specifications, geological conditions, and construction techniques, multiple design requirements such as structural safety, economic cost, construction feasibility, and long-term durability can be considered simultaneously. This effectively avoids the shortcomings of traditional single-objective optimization methods that tend to overlook certain aspects, providing designers with a diverse set of feasible schemes and significantly improving the overall engineering benefits and decision-making flexibility.

[0013] In one optional implementation, for different service conditions, an optimal design scheme for the dam is selected from multiple feasible design schemes based on a preset differentiated response weight strategy. This includes: determining the response weights of each physical field according to the current service condition type; and selecting the optimal design scheme from multiple feasible design schemes based on the response weights. By dynamically adjusting the response weights of each physical field according to different service condition types and selecting the design scheme that best matches the current service condition from multiple feasible design schemes, adaptive response of the same model to different service scenarios is achieved. This effectively overcomes the shortcomings of traditional static design methods that cannot take into account the differentiated needs of multiple service conditions, and significantly improves the engineering practicality of the design scheme in complex service environments.

[0014] Secondly, the present invention provides a multi-physics coupling design device for dams, comprising: a response solving module, used to acquire data of each physical field of the dam, inputting them into a solution model selected based on the differences in the spatiotemporal change rates of each physical field for solving, and obtaining the response results of each physical field; wherein the solution model includes at least a numerical solution model and a machine learning model, and the machine learning model is embedded with physical control equation constraints; a bidirectional coupling module, used to perform bidirectional coupling iteration of the response results of each physical field in a mutual feedback manner to obtain coupling results; a scheme design module, used to obtain multiple feasible design schemes by adjusting the design variables of the dam based on the coupling results; and a scheme screening module, used to select the optimal design scheme of the dam from multiple feasible design schemes based on a preset differentiated response weight strategy for different service conditions.

[0015] Thirdly, the present invention provides an electronic device, comprising: a memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the dam multiphysics coupling design method described in the first aspect or any corresponding embodiment thereof.

[0016] Fourthly, the present invention provides a computer-readable storage medium storing computer instructions for causing a computer to execute the dam multiphysics coupling design method described in the first aspect or any corresponding embodiment thereof. Attached Figure Description

[0017] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0018] Figure 1This is a flowchart illustrating the multiphysics coupling design method for dams according to an embodiment of the present invention; Figure 2 This is a structural block diagram of a dam multiphysics coupling design device according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, 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.

[0020] It is understood that before using the technical solutions disclosed in the various embodiments of the present invention, users should be informed of the types, scope of use, and usage scenarios of the personal information involved in the present invention and their authorization should be obtained in accordance with relevant laws and regulations through appropriate means.

[0021] The existing mainstream technical approaches for multiphysics coupling design of the two types of water conservancy and hydropower dams are still insufficient to meet the core requirements of high dam and large reservoir projects for coupling design in terms of "high precision, high efficiency, multi-objective, and strong adaptability," and have four significant technical defects: First, there is an imbalance between computational efficiency and solution accuracy. Pure numerical simulation methods require fine meshing for complex dam structures and geological conditions. The multiphysics coupling iteration process requires processing massive amounts of data. The coupling analysis of a single design scheme often takes several days or even weeks, resulting in extremely low computational efficiency. This makes it impossible to support rapid iterative comparison of multiple design schemes, leading to a lengthy design cycle and difficulty in meeting project schedule requirements.

[0022] Secondly, the model lacks generalization ability and physical interpretability. Traditional data-driven models rely entirely on data fitting ability and lack constraints from core physical mechanisms such as hydraulics and seepage mechanics. When the amount of training data is insufficient or the data distribution is uneven, it is easy to generate non-physical interpretations that "conform to the statistical laws of data but violate the basic laws of engineering physics." For example, the predicted seepage field distribution may contradict Darcy's law, which may lead to potential safety hazards in the optimized design scheme and fail to meet the dam safety design specifications.

[0023] Third, the optimization objectives are too singular. Existing design methods often focus on optimizing a single objective, such as minimizing dam stress safety or project cost. This makes it difficult to consider multiple design requirements, such as safety performance, economic cost, construction period, and long-term durability. As a result, problems such as "safety standards are met but costs far exceed the budget" or "construction period is shortened but dam carbonization rate exceeds the standard" often occur, which fail to maximize the overall benefits of the project.

[0024] Fourth, the design process is static and fixed. The modeling and design process is static and fixed. Once the initial design parameters are determined, the model cannot dynamically adapt to actual engineering variables such as sudden changes in geological conditions and load conditions during the construction period. Moreover, the same set of optimization strategies cannot be specifically adapted to the core requirements of different working conditions such as normal operation, earthquakes, floods, and ice jams. This results in a serious disconnect between the design scheme and the actual situation on the engineering site, with extremely poor adaptability and flexibility.

[0025] In recent years, Physically Informed Neural Networks (PINNs) have become a core carrier of machine learning technology. Their core advantage lies in their ability to embed physical control equations into the loss function of the neural network, achieving a deep integration of physical mechanism constraints and data-driven modeling. This not only improves prediction efficiency by relying on data but also ensures the rationality of results through physical laws, providing a new technical path to solve the technical pain points of multi-physics coupled design.

[0026] However, the application of machine learning technology in the design of water conservancy and hydropower dams is still in the exploratory stage. Existing research is mostly limited to modeling single physical fields and has not yet formed a complete and systematic optimization design method adapted to the multi-physics coupling characteristics of dams. It cannot effectively balance practical engineering problems such as "physical constraints and solution efficiency", "multi-objective collaborative optimization" and "dynamic operating condition adaptation". Therefore, it is urgent to construct a multi-physics coupling optimization design method for dams based on machine learning to fill the existing technological gap and meet the high-quality design requirements of modern water conservancy and hydropower projects.

[0027] Based on this, the present invention provides a multi-physics field coupling design method, device, equipment and medium for dams, to solve the problems in the prior art where it is difficult to balance the physical solution efficiency and solution accuracy in multi-physics field coupling scenarios, and the design process cannot be dynamically and adaptively adjusted with changes in working conditions. It is applicable to multi-physics field collaborative design of various dam types and can be widely used in the preliminary design, construction drawing design and dynamic optimization of the construction process of water conservancy and hydropower projects.

[0028] According to an embodiment of the present invention, a multiphysics coupling design method for dams is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0029] This embodiment provides a multiphysics coupling design method for dams. Figure 1 This is a flowchart of a multiphysics coupling design method for dams according to an embodiment of the present invention, such as... Figure 1 As shown, the process includes the following steps: Step S101: Obtain data of each physical field of the dam, input the solution model selected based on the difference in the spatiotemporal change rate of each physical field, and obtain the response results of each physical field; wherein, the solution model includes at least a numerical solution model and a machine learning model, and the machine learning model is embedded with physical control equation constraints.

[0030] In one alternative implementation, the numerical solution model preferably employs a scaled boundary finite element model.

[0031] In one optional implementation, the machine learning model preferably employs a Physics-Informed Neural Network (PINN). PINN exhibits good adaptability to multi-physics coupling problems, directly embedding arbitrary forms of physical control equations into the loss function without simplification. It is suitable for complex physical control equations such as those related to dam seepage, stress, temperature, and dynamic-fluid-structure interaction. Furthermore, the model has a simple structure, high training efficiency, and is suitable for practical engineering applications. It should be noted that other machine learning models with physical constraint embedding capabilities are also applicable to this method; this embodiment uses PINN only as an example. The construction method of PINN includes: Step a1: Construct the deep network structure of the machine learning model, which includes an input layer, hidden layers, and an output layer.

[0032] In this embodiment, the deep network structure is used to establish a nonlinear mapping relationship between the core feature parameters of the dam and key physical field parameters. Specifically, the input layer corresponds to the core feature parameters extracted from multi-source basic data; the output layer corresponds to key physical field parameters such as seepage pressure, stress, strain, and temperature.

[0033] The process of collecting multi-source basic data is as follows: Comprehensive collection of basic data required throughout the entire dam design process, covering: geological survey data (such as rock permeability coefficient, elastic modulus, Poisson's ratio, and rock integrity coefficient); hydrological and water resource data (such as long-term flood processes in the basin, dynamic changes in groundwater levels, and spatiotemporal distribution of precipitation); engineering material parameters (such as concrete strength grade, elastic modulus, compaction degree of fill material, and permeability coefficient of anti-seepage materials); load data (such as static loads such as self-weight and water pressure, dynamic loads such as seismic acceleration and blasting vibration, and temperature loads such as environmental temperature changes and heat of hydration); and regulatory constraint data (such as the minimum anti-sliding stability safety factor of the dam body, seepage flow limit, crack width control standards, and other current industry standard requirements). Simultaneously, measured seepage pressure, displacement, and temperature data from 3 to 5 similar dam projects are collected, along with real-time monitoring data from sensors pre-embedded during the construction phase of this project.

[0034] The collected data underwent systematic preprocessing: First, the 3σ criterion was used to remove abnormal peak values ​​from seismic load and monitoring data; second, missing data was supplemented using linear interpolation; then, the Z-score standardization method was used to normalize data of different dimensions to the same interval; finally, principal component analysis (PCA) was used to extract core feature parameters such as seepage coefficient, elastic modulus, and peak flow rate, forming a dimensionality-reduced set of core feature parameters. In this way, a standardized and structured multi-physics coupled design database for dams was constructed, providing data support for the model's input layer.

[0035] Step a2 involves embedding the physical control equations into the loss function of the machine learning model, forming a dual loss function that includes both data loss terms and physical loss terms.

[0036] For the various physical fields faced by the dam, the following core physical control equations are embedded into the physical loss term of the loss function: For the seepage field, Darcy's law is used, and its governing equation is: ,in, Here, is the Hamiltonian operator, k is the permeability coefficient of the porous medium in the dam body or foundation, and h is the seepage head.

[0037] For stress-strain fields, the elastoplastic constitutive equations are adopted, and the governing equations include: static equilibrium equations (when there are no volume forces). Geometric equations (strain-displacement relationship) Elastic-plastic incremental constitutive relations .in, Let f be the stress tensor and f be the volume force vector. For strain tensor, For displacement components, The constitutive matrix is ​​elastoplastic. , These are the stress increment and strain increment, respectively.

[0038] For the temperature field, the heat conduction equation is used, and its governing equation is: Where ρ is the density of concrete or rock mass, c is the specific heat capacity of the material, T is the temperature, t is the time, λ is the thermal conductivity coefficient of the material, and Q(t) is the heat generation rate of concrete hydration.

[0039] For dynamic fluid-structure interaction fields, the Navier-Stokes equations are adopted, and their governing equations are: ,in, Let ρ be the reservoir water velocity vector, ρf be the reservoir water density, p be the reservoir water dynamic pressure, and μ be the reservoir water dynamic viscosity coefficient. Here, g is the Laplace operator, and g is the gravitational acceleration vector.

[0040] By incorporating the aforementioned governing equations as constraints into the loss function, a dual-loss function structure is formed: "data loss (fitting measured data) + physical loss (satisfying physical laws)". During training, the weight coefficients are adjusted; for example, the data loss weight is set to 0.4 and the physical loss weight to 0.6, to balance the accuracy of data fitting with physical rationality. This ensures that the model output not only conforms to statistical data patterns but also strictly adheres to the fundamental laws of hydraulic mechanics and seepage mechanics.

[0041] In this embodiment, to further enhance the model's ability to capture the spatiotemporal distribution features of multi-physics fields, Fourier feature decomposition technology is introduced to optimize the input layer before training. Specifically: First, the core feature parameters are normalized to adapt to the numerical requirements of Fourier feature mapping; second, a Fourier feature mapping function is constructed for the spatiotemporal features, introducing a random frequency matrix and phase offset, and mapping the low-dimensional spatiotemporal features to a high-dimensional high-frequency feature space through sine and cosine transforms, generating multiple sets of high-frequency Fourier features to enhance the linear representation ability of the spatiotemporal distribution of multi-physics fields; finally, the generated high-frequency Fourier features are concatenated with the normalized non-spatiotemporal original core features to form a combined feature set that integrates high-frequency spatiotemporal features, which serves as the final input layer feature of the physical knowledge neural network.

[0042] Step a3: Train the machine learning model using pre-collected historical dam data, and adjust the network parameters to make the double loss function converge.

[0043] The database constructed in step a1 was used as the source of historical dam data to train the model. During training, the model hyperparameters were adjusted using a grid search method, with the optimal settings being 3 hidden layers, 64 neurons per layer, 5000 iterations, and a learning rate of 0.001. Mean squared error (MSE) and relative error (RE) were used as accuracy evaluation metrics. Iterative training continued until the double loss function converged, ensuring that the model's prediction accuracy met preset requirements, such as MSE ≤ 0.02 and RE ≤ 3%. The trained physical-informed neural network, as a machine learning model with embedded physical control equation constraints, was used for the rapid solution of the first type of physical field.

[0044] During training, the hyperparameters of the machine learning model are tuned to obtain optimal model performance. Taking the determination of the number of hidden layers as an example, the optimization process is as follows: First, multiple candidate values ​​for hidden layers are preset, for example, including layer 1, layer 2, layer 3 and layer 4; Secondly, with the number of neurons, learning rate and other hyperparameters fixed, each candidate model is trained separately using the dam coupling design database. Finally, the candidate models were validated and screened based on both prediction accuracy and computational efficiency. The candidate model with the highest prediction accuracy and computational efficiency was determined as the optimal parameter for the number of hidden layers. In this embodiment, after the above optimization process, the final number of hidden layers was determined to be 3.

[0045] In addition to the number of hidden layers, other hyperparameters such as the number of neurons, learning rate, and number of iterations can also be jointly tuned using similar methods or grid search to ensure that the double loss function converges to the preset accuracy requirements.

[0046] Specifically, step S101 includes: Step S1011: Divide each physical field into a first type of physical field and a second type of physical field according to the difference in spatiotemporal change rate, wherein the spatiotemporal change rate of the first type of physical field is lower than that of the second type of physical field.

[0047] Based on the spatiotemporal variation characteristics of the multi-physics field of the dam, a differentiated coupled solution architecture is established to achieve synergistic optimization of solution efficiency and solution accuracy. The first type of physical field consists of slowly varying fields with spatiotemporal scales measured in days or months and relatively gradual change rates, exemplified by seepage fields and temperature fields. The second type of physical field consists of rapidly varying fields with millisecond-level changes and extremely high requirements for solution accuracy, exemplified by dynamic response fields and instantaneous stress fields under seismic conditions.

[0048] Step S1012: For the first type of physical field, a machine learning model is used to solve it; for the second type of physical field, a numerical solution model is used to solve it.

[0049] For the first type of physical field, a trained physical-informed neural network model is used for fast solution. Since the physical-informed neural network model does not require complex mesh partitioning, it directly maps the input-output relationship through the trained network parameters, which can significantly shorten the calculation time for solving slowly varying fields.

[0050] The core feature parameters corresponding to the first type of physical field are input into the trained physical-information neural network model. Examples of core feature parameters include: geological parameters (such as rock permeability coefficient), material parameters (such as concrete thermal conductivity coefficient), and load parameters (such as static load at normal water level). The physical-information neural network model performs forward propagation calculations and directly outputs the spatial distribution results of the first type of physical field, achieving rapid solution without the need for complex mesh generation.

[0051] For example, when the first type of physical field is the seepage field and the temperature field, the input parameters include the rock permeability coefficient, the concrete thermal conductivity coefficient, the static load of the normal water level, etc., and the output results include the three-dimensional spatial distribution data and key characteristic values ​​of the dam, such as the seepage pressure and seepage head distribution of the seepage field and the temperature value distribution of the temperature field.

[0052] For the second type of physical field, a scaled boundary finite element model is used for refined numerical solutions. The scaled boundary finite element model only requires mesh discretization of the computational domain boundary, significantly reducing the mesh degrees of freedom. This effectively reduces the complexity of numerical computation while maintaining the accuracy of solutions for rapidly changing fields such as seismic dynamic responses. The solution process includes the following steps: The first step is to construct a detailed analysis model of the rapidly changing field of the dam body and foundation, and to divide the mesh according to the proportional boundary finite element method rules to adapt to the millisecond-level dynamic response. Unlike the finite element method, which requires meshing within the entire computational domain, the proportional boundary finite element model only needs to discretize the mesh at the boundary of the computational domain, while maintaining analytical solutions for directions within the domain. This significantly reduces the mesh degrees of freedom and improves the solution accuracy for complex structures.

[0053] The second step involves substituting the response results of the first type of physical field obtained from the physical-informed neural network model in the previous steps as boundary conditions into the rapid-changing field fine analysis model. For example, the seepage field distribution results obtained from the slow-changing field solution are transformed into stress boundary conditions or displacement boundary conditions and applied to the scaled boundary finite element model.

[0054] The third step involves combining the seismic load parameters and the dynamic balance control equations, and then using numerical iteration calculations to solve for the spatiotemporal distribution and millisecond-level response data of rapidly changing fields such as instantaneous stress, dynamic displacement, and hydrodynamic pressure under seismic conditions.

[0055] It is worth noting that the data for each physical field of the dam in step S101 refer to the current design input parameters of the dam. For the first type of physical field, the design input parameters include, for example, permeability coefficient, thermal conductivity coefficient, and static load at normal water level; for the second type of physical field, the design input parameters include, for example, seismic acceleration, flood level elevation, and material elastic modulus. These design input parameters differ from the historical dam data used in the model training phase; they are the specific engineering parameters of the dam currently being designed.

[0056] Through the above steps, a high-precision and detailed solution for the second type of physical field was achieved, providing initial response results for the rapidly changing field for subsequent bidirectional coupled iterations.

[0057] Step S102: Perform a bidirectional coupling iteration on the response results of each physical field in a mutual feedback manner to obtain the coupling result.

[0058] Specifically, the response results of the first type of physical field are used as the boundary conditions for solving the second type of physical field, and the initial parameters of the second type of physical field are modified. The response results of the second type of physical field are fed back to the solution model of the first type of physical field, and the weight parameters of the solution model of the first type of physical field are modified. Through the above bidirectional coupling iteration, the accuracy and efficiency of multi-physics coupling solution are balanced, and the coupling result after the mutual coordination of each physical field is finally obtained.

[0059] Step S103: Based on the coupling results, multiple feasible design schemes are obtained by adjusting the design variables of the dam.

[0060] Specifically, using the dam's geometric and material parameters as design variables, and pre-defined specifications, geological conditions, and construction techniques as constraints, a multi-objective optimization is performed on the design variables based on the coupling results to obtain multiple feasible design schemes. Among them, the geometric parameters include at least one of the upstream and downstream slope ratio of the dam body, the thickness of the concrete pouring layer, and the depth of the anti-seepage wall, and the material parameters include material mix parameters. The multi-objective optimization uses a multi-objective genetic algorithm to solve for the Pareto optimal solution set.

[0061] For example, a four-dimensional coupled optimization objective function covering the core requirements of dam design can be established, including safety objectives, economic objectives, schedule objectives, and durability objectives, to achieve multi-dimensional collaborative optimization.

[0062] Among them, the safety objective is to ensure that the minimum anti-sliding stability safety factor of the dam body is not lower than the standard limit, while controlling the maximum principal stress of the dam body to not exceed the material design strength; the economic objective is to minimize the project cost ratio, covering key cost items such as concrete usage, seepage prevention works, and construction machinery input; the construction period objective is to minimize the dam body filling or pouring period, combined with optimization of construction procedures; and the durability objective is to minimize the concrete carbonation rate and freeze-thaw damage rate, and control the long-term service performance degradation.

[0063] In the optimization process, an optimization surrogate model is first constructed. This surrogate model is a fast response model built around the coupling results, used to replace traditional time-consuming numerical simulations, and can quickly output the response results of each physical field of the dam. The physical field response results output by the optimization surrogate model serve as the quantitative calculation basis for the aforementioned four-dimensional coupled optimization objective function.

[0064] Secondly, using the dam's core design parameters as optimization variables, the response results of each physical field are obtained by substituting them into the optimization surrogate model, and then the four-dimensional objective function value is calculated. For example, the core design parameters include the upstream and downstream slope ratio of the dam body, the thickness of the concrete pouring layer, the depth or thickness of the anti-seepage wall, and the proportion of engineering materials. Pre-set specifications, geological conditions, and construction technology are used as constraints.

[0065] Then, the NSGA-Ⅲ multi-objective genetic algorithm is used for multi-generation iterative optimization. In each generation of evolution, new combinations of design parameters are generated through selection, crossover, and mutation operations. These parameters are then substituted into the optimization surrogate model to evaluate the corresponding physical field response and the four-dimensional objective function value. Non-dominated solutions are screened through non-dominated sorting, and a selection mechanism based on reference points is introduced to maintain the distribution of solutions.

[0066] During the iteration process, sensitivity analysis is used to eliminate low-sensitivity parameters with an impact of less than 5% on the optimization objective, reducing the dimensionality of optimization variables and improving the efficiency of the optimization iteration. After the iteration converges, a set of Pareto optimal solutions that satisfy all constraints is finally obtained, i.e., multiple feasible design schemes.

[0067] Based on this, once the construction phase begins, as the actual measured data accumulates, the established machine learning model needs to be dynamically revised to maintain consistency between the model's predictions and the actual state of the project. This includes the following steps: I. Acquisition and Preprocessing of Real-time Monitoring Data Real-time monitoring data collected by sensors embedded during construction is transmitted in real time to the machine learning model via an IoT module. The real-time monitoring data includes measured values ​​of physical fields such as seepage pressure, displacement, and temperature. The real-time monitoring data is preprocessed to adapt it to the input format requirements of the machine learning model.

[0068] II. Dynamic Adjustment of Model Parameters Based on Incremental Learning Incremental learning algorithms are used to dynamically correct the parameters of machine learning models without retraining the entire model; only the subset of parameters that deviate significantly from the measured data is updated.

[0069] In one alternative implementation, preprocessed real-time monitoring data is input into a machine learning model, and the deviation between the model's predicted values ​​and the measured values ​​is calculated. Based on this deviation, backpropagation is used to update only the weight parameters of the model's output layer and the last hidden layer, while keeping the parameters of the remaining network layers unchanged.

[0070] After the correction is completed, the response results of each physical field and the optimization objective function are updated synchronously to ensure that the optimized design scheme is consistent with the actual engineering state.

[0071] III. Model Accuracy Feedback and Retraining Trigger Mechanism Establish a model accuracy feedback mechanism with a preset error threshold. If the deviation between the corrected model prediction and the measured data exceeds the preset error threshold, the model retraining process is automatically triggered, returning to step a2 to readjust the network parameters and physical control equation constraints until the model prediction accuracy meets the preset requirements. If the deviation between the corrected model prediction and the measured data is within the preset error threshold, the corrected model is used for subsequent solutions.

[0072] The above-mentioned dynamic correction mechanism ensures that the optimized design scheme can adapt to dynamic factors such as sudden changes in geological conditions and load changes during the construction period.

[0073] After dynamic correction, the model's output coupling results and optimization objective function are updated synchronously. Based on the updated Pareto optimal solution set, the model proceeds to the screening stage in step S104.

[0074] Step S104: For different service conditions, based on a preset differentiated response weight strategy, select the optimal design scheme for the dam from multiple feasible design schemes.

[0075] Specifically, the response weights of each physical field are determined based on the current service conditions; and the optimal design scheme is selected from multiple feasible design schemes based on the response weights.

[0076] This embodiment distinguishes between two stages: preliminary design and construction drawing design, covering four typical operating conditions: normal operation, earthquake, flood, and ice jam. It should be noted that the "weight" adjustment here refers to the response weights of each physical field in the multi-field coupling solution, rather than the weight coefficients in the model loss function. By differentially presetting the coupling solution weights of each physical field under different operating conditions, the optimization and screening process prioritizes ensuring the core safety requirements of the current operating condition.

[0077] I. Operating Condition Identification and Weight Strategy Preset First, the system automatically determines the current working condition type based on preset working condition identification rules. The system receives input parameters of the dam's current design stage and actual service load characteristics, such as flood level elevation and extreme ambient temperature, and matches them with preset threshold values ​​for four types of working conditions to identify the corresponding working condition type.

[0078] Based on the identified operating condition type, a preset differentiated response weighting strategy is invoked. For example, the physics coupling weights for each operating condition are configured as follows: Earthquake condition: Increase the response weight of the dynamic fluid-structure interaction field to 40% to prioritize the anti-sliding stability and seismic performance of the dam body; Flood conditions: Increase the response weight of the seepage-stress coupling field to 45% to strictly control seepage flow and dam displacement deformation; Ice-related conditions: Increase the coupling weight of temperature field-stress field-seepage field to 50%, prioritize the protection of dam body against ice pressure, frost heave damage and seepage blockage, and strictly control the stress on the ice-facing surface of the dam body, seepage pressure in the freeze-thaw zone and temperature gradient; Normal operating conditions: The weights of each physical field are balanced. For example, the seepage field is 30%, the stress field is 30%, the temperature field is 20%, and the dynamic field is 20%, taking into account both safety and economic benefits.

[0079] II. Normalized Weighted Screening Based on the Pareto optimal solution set obtained in step S103, and combined with the current design stage and the identified working condition type, the optimal design scheme is adaptively selected using the weighted summation method.

[0080] Since each solution in the Pareto optimal solution set contains objective function values ​​for four dimensions—safety, economy, time, and durability—and the dimensions have different units (e.g., safety factor is a dimensionless ratio, project cost is in monetary units, time is in time units, and durability is in rate units), it is necessary to normalize the four-dimensional optimization objective indicators corresponding to each solution before weighted summation to eliminate dimensional differences. After normalization, each indicator is mapped to the same dimensionless interval, making it additive and comparable.

[0081] Then, based on the priority of the objectives under the current working condition, differentiated weights are assigned to the normalized four-dimensional objective indicators, and a weighted sum is obtained to obtain the comprehensive score for each design scheme. For example, the safety objective has the highest weight under the seismic condition, while the safety and economic objectives related to seepage control have priority under the flood condition. The design scheme with the highest comprehensive score is selected as the optimal design scheme under the current working condition, ranked from highest to lowest.

[0082] III. Scheme Verification and Output Finally, the optimal design scheme was verified using a pure finite element numerical simulation method. The results of the coupled solution method were compared with the results of the pure finite element numerical simulation to verify their consistency. If the deviation was within the preset allowable range, the scheme was confirmed to meet the specification requirements, and the final output included dam cross-sectional parameters, material proportions, construction techniques, and other design results that could be directly implemented.

[0083] The multiphysics coupling design method for dams provided in this embodiment has the following beneficial effects: I. Significantly improved solution efficiency By selecting a solution model based on the differences in the spatiotemporal change rates of various physical fields, the first type of physical field is solved quickly by a machine learning model, and the second type of physical field is solved in detail by a numerical solution model. The efficiency of multi-physics coupled computation is improved by more than 80% compared with the traditional pure numerical simulation method. The optimization iteration time of multiple design schemes is shortened from several weeks to several hours, which significantly shortens the dam design cycle and reduces the cost of computational resources invested in the design phase.

[0084] II. Enhanced Design Safety and Reliability By embedding the physical control equations into the loss function of the machine learning model, a dual loss function consisting of data loss terms and physical loss terms is formed, eliminating the generation of non-physical solutions at the source. Combined with dynamic model correction driven by engineering measurement data, the fit between the design scheme and the actual engineering situation is improved by more than 90%, effectively avoiding safety risks such as seepage failure, stress exceeding limits, and crack propagation, meeting the stringent safety design requirements of high dams and large reservoirs, and improving the stability of the dam throughout its entire service life.

[0085] III. Achieving Optimal Multi-Objective Collaboration Breaking through the limitations of traditional single-objective optimization, a four-dimensional coupled optimization objective function covering safety, economy, schedule, and durability objectives is established. Through Pareto optimal solution set screening and sensitivity analysis, the project cost ratio is reduced by 5% to 15% and the construction period is shortened by 10% to 20% while meeting the constraints of the specifications. At the same time, the concrete carbonation rate and freeze-thaw damage rate are effectively controlled, the long-term service durability of the dam body is improved, and the comprehensive benefits of the project are maximized.

[0086] IV. Strong adaptability and versatility This method is compatible with various mainstream dam types, including concrete gravity dams, face-faced rockfill dams, and arch dams. It can be flexibly adapted to different design stages and various service conditions such as normal operation, earthquakes, floods, and ice jams. By pre-setting differentiated response weight strategies for different service conditions, it adaptively selects and optimizes design schemes, making it easy to promote and apply in various water conservancy and hydropower dam projects. It has broad engineering value and market prospects.

[0087] This embodiment also provides a dam multiphysics coupling design device, which is used to implement the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that performs a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0088] This embodiment provides a multiphysics coupling design device for dams, such as... Figure 2 As shown, it includes: The response solving module 201 is used to acquire data of various physical fields of the dam, input the solution model selected based on the difference in the spatiotemporal change rate of each physical field, and solve the response results of each physical field. The solution model includes at least a numerical solution model and a machine learning model, and the machine learning model is embedded with physical control equation constraints. The bidirectional coupling module 202 is used to perform bidirectional coupling iteration of the response results of each physical field in a mutual feedback manner to obtain the coupling result; The scheme design module 203 is used to obtain multiple feasible design schemes by adjusting the design variables of the dam based on the coupling results. The scheme selection module 204 is used to select the optimal design scheme of the dam from multiple feasible design schemes based on a preset differentiated response weight strategy for different service conditions.

[0089] The dam multiphysics coupling design device provided in this embodiment of the invention can execute the dam multiphysics coupling design method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the method. Further functional descriptions of the above modules and units are the same as in the corresponding embodiments described above, and will not be repeated here.

[0090] Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention.

[0091] The following is a detailed reference. Figure 3 This diagram illustrates a structural schematic suitable for implementing an electronic device according to embodiments of the present invention. The electronic device may include a processor (e.g., a central processing unit, graphics processor, etc.) 301, which can perform various appropriate actions and processes based on a program stored in read-only memory (ROM) 302 or a program loaded from memory 308 into random access memory (RAM) 303. The RAM 303 also stores various programs and data required for the operation of the electronic device. The processor 301, ROM 302, and RAM 303 are interconnected via a bus 304. An input / output (I / O) interface 305 is also connected to the bus 304.

[0092] Typically, the following devices can be connected to I / O interface 305: input devices 306 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 307 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; memory devices 308 including, for example, magnetic tapes, hard disks, etc.; and communication devices 309. Communication device 309 allows electronic devices to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 3 Electronic devices with various devices are shown, but it should be understood that it is not required to implement or have all of the devices shown, and more or fewer devices may be implemented or have instead.

[0093] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 309, or installed from a memory 308, or installed from a ROM 302. When the computer program is executed by the processor 301, it performs the functions defined in the dam multiphysics coupling design method of the embodiments of the present invention.

[0094] Figure 3 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0095] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code. When the software or computer code is accessed and executed by the computer, processor, or hardware, the dam multiphysics coupling design method shown in the above embodiments is implemented.

[0096] A portion of this invention can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to the invention through the operation of the computer. Those skilled in the art will understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executing the instructions, or the computer compiling the instructions and then executing the corresponding compiled program, or the computer reading and executing the instructions, or the computer reading and installing the instructions and then executing the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to a computer.

[0097] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A multiphysics coupling design method for dams, characterized in that, The method includes: Data on various physical fields of the dam are acquired, and the solutions are obtained by inputting them into a solution model selected based on the differences in the spatiotemporal change rates of each physical field. The response results of each physical field are then obtained. The solution model includes at least a numerical solution model and a machine learning model, and the machine learning model is embedded with physical control equations. The response results of each physical field are subjected to a bidirectional coupling iteration with mutual feedback to obtain the coupling result; Based on the coupling results, multiple feasible design schemes can be obtained by adjusting the design variables of the dam. For different service conditions, based on a preset differentiated response weight strategy, the optimal design scheme of the dam is selected from the multiple feasible design schemes.

2. The multiphysics coupling design method for dams according to claim 1, characterized in that, The numerical solution model is a scaled boundary finite element model.

3. The multiphysics coupling design method for dams according to claim 1, characterized in that, The machine learning model is a physical information neural network, and its construction method includes: Construct the deep network structure of the machine learning model, the deep network structure including an input layer, hidden layers and an output layer; The physical control equations are embedded into the loss function of the machine learning model to form a double loss function that includes data loss terms and physical loss terms. The machine learning model is trained using pre-collected historical data of the dam, and the network parameters are adjusted to make the double loss function converge.

4. The multiphysics coupling design method for dams according to claim 1, characterized in that, The solution model is selected based on the differences in the spatiotemporal change rates of various physical fields, including: The physical fields are divided into a first type of physical field and a second type of physical field according to the difference in their spatiotemporal change rates, wherein the spatiotemporal change rate of the first type of physical field is lower than that of the second type of physical field. For the first type of physical field, the machine learning model is used to solve it; For the second type of physical field, the numerical solution model described above is used for solution.

5. The multiphysics coupling design method for dams according to claim 4, characterized in that, The step of performing bidirectional coupling iteration with mutual feedback on the response results of each physical field includes: The response results of the first type of physical field are used as the boundary conditions for solving the second type of physical field; The response results of the second type of physical field are fed back to the solution model of the first type of physical field to correct the parameters of the solution model of the first type of physical field.

6. The multiphysics coupling design method for dams according to claim 1, characterized in that, The process of adjusting the dam's design variables yields multiple feasible design schemes, including: Using the dam's geometric and material parameters as design variables, and pre-defined specifications, geological conditions, and construction techniques as constraints, the design variables are optimized through multi-objective search based on the coupling results to obtain multiple feasible design schemes.

7. The multiphysics coupling design method for dams according to claim 1, characterized in that, The process of selecting an optimal dam design scheme from multiple feasible design schemes based on a preset differentiated response weighting strategy for different service conditions includes: Determine the response weights of each physical field based on the current service condition type; Based on the response weights, the optimal design scheme is selected from the multiple feasible design schemes.

8. A multiphysics coupling design device for dams, characterized in that, The device includes: The response solving module is used to acquire data of various physical fields of the dam, input the solution model selected based on the difference in the spatiotemporal change rate of each physical field, and solve the problem to obtain the response results of each physical field; wherein, the solution model includes at least a numerical solution model and a machine learning model, and the machine learning model is embedded with physical control equation constraints; The bidirectional coupling module is used to perform bidirectional coupling iteration of the response results of each physical field in a feedback manner to obtain the coupling result; The scheme design module is used to obtain multiple feasible design schemes by adjusting the design variables of the dam based on the coupling results. The scheme selection module is used to select the optimal design scheme of the dam from the multiple feasible design schemes based on a preset differentiated response weight strategy for different service conditions.

9. An electronic device, characterized in that, include: A memory and a processor are interconnected, the memory stores computer instructions, and the processor executes the computer instructions to perform the dam multiphysics coupling design method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing the computer to execute the dam multiphysics coupling design method according to any one of claims 1 to 7.

Images