Intelligent Design Method of Gravity Dam Based on Physical Mechanics Information Generative Adversarial Network

Abstract

The present invention belongs to the technical fields of gravity dam profile design and artificial intelligence. The present invention discloses an intelligent design method for gravity dams based on a physical mechanics information generative adversarial network, which includes the following steps: Step 1, collect and obtain the data characteristics of external loads, characteristic water levels, uplift pressure reduction coefficients, friction coefficients between concrete and rock foundations, and mechanical parameters of concrete and rock for the design of domestic concrete gravity dams, and normalize the data characteristics to obtain contour prediction input data; Step 2, construct and train a GAN model, input the contour prediction input data, and the GAN model outputs output data including the slope ratios of the predicted upstream and downstream dam slopes, the elevation of the upstream slope starting point, the horizontal width of the dam bottom, and the width of the dam crest. The present invention can solve the problems of difficult experience inheritance and bottleneck in efficiency improvement in the design of traditional concrete gravity dams, thereby promoting the advancement of the design of domestic concrete gravity dams towards the intelligent level.

Description

Technical Field

[0001] The present invention relates to the technical fields of gravity dam profile design and artificial intelligence. More specifically, the present invention relates to an intelligent design method for gravity dams based on a physical mechanics information generative adversarial network. Background Art

[0002] The design of traditional concrete gravity dams has long relied on a trial-and-error method dominated by manual experience. Designers need to complete the design by repeatedly adjusting the profile geometric parameters and cooperating with mechanical calculations based on limited geological survey data, hydraulic specifications, and historical cases. With the expansion of the scale of high dam and large reservoir projects, the increase in projects with complex geological conditions, and the continuous development of pumped-storage power stations and hydropower projects, digital and intelligent design has become an important means to improve design quality and efficiency and meet the requirements of intelligent construction.

[0003] However, the traditional mode of concrete gravity dam design relying on manual design faces challenges in terms of difficult inheritance of experience and difficult improvement of design efficiency.

[0004] In view of this, the present invention proposes an intelligent design method for gravity dams based on a physical mechanics information generative adversarial network to solve the above problems. Summary of the Invention

[0005] To overcome the above-mentioned defects of the prior art and to achieve the above purposes, the present invention provides the following technical solutions: An intelligent design method for gravity dams based on a physical mechanics information generative adversarial network, comprising the following steps:

[0006] Step 1, collect and obtain the data characteristics of external loads, characteristic water levels, uplift pressure reduction coefficients, friction coefficients between concrete and rock foundations, and mechanical parameters of concrete and rock for the design of domestic concrete gravity dams, and normalize the data characteristics to obtain contour prediction input data;

[0007] Step 2, construct and train a GAN model, input the contour prediction input data, the GAN model outputs output data including the slope ratios of the predicted upstream and downstream dam slopes, the elevation of the upstream starting slope point, the horizontal width at the dam bottom, and the width at the dam top, and combine the output data with the foundation surface parameters to obtain the dam body contour parameters;

[0008] Step 3, based on the GAN model and embedded physical mechanics constraints, to realize the intelligent generation and optimization of dam body design.

[0009] Further, the process of collecting and obtaining the data characteristics of external loads, characteristic water levels, uplift pressure reduction coefficients, friction coefficients between concrete and rock foundations, and mechanical parameters of concrete and rock for the design of domestic concrete gravity dams, and normalizing the data characteristics to obtain contour prediction input data includes;

[0010] Step 101: Collect the external loads, characteristic water levels, uplift pressure reduction coefficients, friction coefficients between concrete and the rock foundation surface, and mechanical parameters of concrete and the rock foundation for the concrete gravity dam project to be designed. Among them, the mechanical parameters of concrete and the rock foundation include density, elastic modulus, and Poisson's ratio.

[0011] Step 102: Process the data using the data feature extraction method, conduct correlation analysis, remove highly correlated features, reduce the dimension of the data through principal component analysis, simplify the complex redundant features into several key features, use the several key features as input data, and use the slope ratios of the upstream and downstream dam slopes, the elevation of the upstream starting slope point, the horizontal width at the dam bottom, and the dam top width parameters as output data.

[0012] Step 103: Characterize the input data and output data using vectors, and use floating-point numbers to characterize the input data and output data.

[0013] Step 104: Complete the feature expression of the input data and output data and construct the corresponding training and test data sets.

[0014] Furthermore, the process of constructing the GAN model includes:

[0015] Step 201: Based on the constructed data set and GAN model, conduct model training and testing. If the prediction accuracy meets the requirements, it can be used for design. Among them, the GAN model includes a generator, a discriminator, and a connection layer.

[0016] Step 202: Use the U-Net structure to construct the generator of the GAN model, input the contour prediction input data, and the output of the generator is the vectorized image of the gravity dam contour.

[0017] Step 203: Use a dual-channel discriminator architecture to create the discriminator of the GAN model. The discriminator includes an image discrimination branch and a mechanical stability discrimination branch. The image discrimination branch uses a traditional convolutional network to evaluate the geometric similarity between the generated profile contour and the real project. The mechanical stability discrimination branch is constructed based on ResNet-101, predicts the anti-sliding stability safety factor, anti-overturning safety factor, and maximum basal stress of the generated profile, outputs a design rationality score, and integrates the specification constraint loss function and the mechanical performance loss function at the same time.

[0018] Step 204: Innovate the connection layer: Add a mask enhancement channel between the generator and the discriminator, and guide the GAN model to focus on detail optimization through local masks.

[0019] Furthermore, the process of conducting model training and testing based on the constructed data set and GAN model includes:

[0020] Step 2011: Collect real data: Collect CAD drawings of 50 gravity dams, extract key parameters, and generate parametric data;

[0021] Step 2012: Generate the overall contour in the first stage, and optimize the local accuracy through gradient penalty in the second stage. The weight ratio of the regularization constraint loss to the mechanical loss gradually transitions from 3:1 to 1:2 to balance compliance and safety;

[0022] Step 2013: Input typical working conditions and compare the generated results with the manual design drawings of gravity dams;

[0023] Step 2014: Use geometric accuracy to generate the matching degree between the contour vertex sequence and the manual design coordinates;

[0024] Step 2015: Conduct mechanical performance verification on the GAN model, that is, complete the training and testing of the GAN model;

[0025] Among them, during the training and verification of the GAN model, k-fold cross-validation is adopted, and the root mean square error MSE is used as the error metric standard for accuracy evaluation.

[0026] Furthermore, the process of obtaining the dam body contour parameters includes:

[0027] Step 205: Use the contour prediction input data of the project to be designed as the input data, and the output data composed of the slope ratios of the upstream and downstream slopes, the elevation of the upstream starting slope point, the horizontal width of the dam bottom, and the width of the dam top can be predicted;

[0028] Step 206: Combine the output data with the foundation surface parameters to obtain the dam body contour parameters;

[0029] Among them, the foundation surface parameters include position parameters, geometric parameters, and mechanical parameters. The position parameters include the elevation of the foundation surface, the geometric parameters include the horizontal projection length of the foundation surface and the coordinates of the control points of the dam foundation rock body profile, and the mechanical parameters are the mechanical parameters of the bedrock.

[0030] Furthermore, the process of combining the output data with the foundation surface parameters includes:

[0031] Step 2061: Compare the generated horizontal width of the dam bottom with the horizontal projection length of the foundation surface. If they match, it is directly used as the basis of the dam body contour; if not, parameters such as the dam bottom width and the upstream and downstream slope ratios in the output data need to be adjusted according to the bedrock topography; among them, the elevation of the upstream starting slope point needs to be coordinated with the elevation of the foundation surface to ensure that the starting slope point of the dam body is located in the stable area of the bedrock;

[0032] Step 2062: Integrate the optimized upstream and downstream slope ratios, the width of the dam top, the width of the dam bottom, the elevation of the starting slope point, and the geometric and mechanical parameters of the foundation surface verified by mechanics into the dam body contour parameters.

[0033] Further, the process of realizing the intelligent generation and optimization of dam design based on the GAN model and embedding physical mechanics constraints includes:

[0034] Step 301: Determine the requirements for mechanical stability and stress constraints. The mechanical stability includes anti-sliding stability and anti-overturning stability, and the stress constraint requirements include uniform stress distribution, compressive strength, tensile strength, and fatigue effect;

[0035] Step 302: Parametrically generate a multi-source heterogeneous data set, which includes input data construction and output data construction;

[0036] Step 303: The generative adversarial network model for image generation consists of an image encoding feature extraction module and an image decoding generation module, and generates a corresponding optimized profile of the gravity dam by inputting and fusing multi-modal features;

[0037] Step 304: Based on the augmented training and test data sets, train and test the generative adversarial network model for image synthesis. If the generated quality meets the requirements, it can be used for design;

[0038] Step 305: Use the multi-modal data features of the project to be designed as the input to the trained and tested generative adversarial network model, and then the profile image of the concrete gravity dam can be generated, which is the generation result.

[0039] Further, the process of determining the requirements for mechanical stability and stress constraints includes:

[0040] Step 3011: For anti-sliding stability, it is mandatory to meet the anti-sliding safety factor standard. For anti-overturning stability, it is mandatory to meet the anti-overturning safety factor standard. For stress constraints, it is required to limit the tensile stress;

[0041] Step 3012: The cross-entropy loss between the discriminator output and the authenticity label forms the total loss. The generator and the discriminator are alternately optimized. After the generator generates a contour, the anti-sliding, overturning coefficients, and maximum tensile stress are predicted in real time through a surrogate model, and the constraint loss is calculated. Then, the weights are adjusted by backpropagation;

[0042] Step 3013: By analyzing the loss gradient in the initial stage of training, dynamically adjust λ1 and λ2 to balance geometric generation and mechanical constraints, and realize the closed-loop feedback of physical constraints of the GAN.

[0043] The technical effects and advantages of the intelligent design method of gravity dam based on physical mechanics information generative adversarial network of the present invention:

[0044] By integrating existing design data and experience of concrete gravity dams, combining with the generative adversarial network model to generate dam body profile images, and embedding the training method of mechanical stability and stress constraints, the design efficiency and safety can be significantly improved, further promoting the intelligent development of gravity dam design in China. Description of the Drawings

[0045] Figure 1 It is a schematic flow chart of the intelligent design method of gravity dam based on physical and mechanical information generative adversarial network of the present invention;

[0046] Figure 2 It is a schematic diagram of specific parameter definitions for parametric generation of dam body contours of the present invention;

[0047] Figure 3 It is a schematic diagram of physical and mechanical constraint generative adversarial network of the present invention;

[0048] Figure 4 It is a schematic diagram of the output result of the design profile of a typical concrete gravity dam of the present invention. Detailed Embodiment

[0049] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative work shall fall within the protection scope of the present invention.

[0050] Embodiment 1

[0051] Please refer to Figures 1 to 4 As shown, the intelligent design method of gravity dam based on physical and mechanical information generative adversarial network in this embodiment includes:

[0052] Step 1: Collect and obtain the data characteristics of external loads, characteristic water levels, uplift pressure reduction coefficients, friction coefficients between concrete and rock foundation surfaces, and mechanical parameters of concrete and rock foundation required for the design of domestic concrete gravity dams, and normalize the data characteristics to obtain contour prediction input data;

[0053] Step 2: Construct and train a GAN model, input the contour prediction input data, and the GAN model outputs output data including the slope ratios of the predicted upstream and downstream dam slopes, the elevation of the upstream starting slope point, the horizontal width of the dam bottom, and the width of the dam top. Combine the output data with the foundation surface parameters to obtain the dam body contour parameters;

[0054] Step 3: Based on the GAN model and embedded physical and mechanical constraints, realize the intelligent generation and optimization of dam body design;

[0055] Among them, the GAN model is based on a large amount of engineering data. The generator is used to generate design schemes, and the discriminator screens the results that meet the mechanical requirements. By inputting constraint conditions such as external loads and material parameters, the GAN model can automatically adjust design parameters (such as the slope ratio of the dam slope, the width of the dam, etc.) to ensure that the generated scheme meets the anti-sliding, anti-overturning stability, and stress uniformity.

[0056] Specifically, before the GAN model has the ability of prediction and generation, it needs to be trained and tested to ensure its excellent prediction and generation ability.

[0057] Furthermore, the process of collecting and obtaining the data characteristics of the external loads, characteristic water levels, uplift pressure reduction coefficients, friction coefficients between concrete and bedrock surfaces, and mechanical parameters of concrete and bedrock required for the design of domestic concrete gravity dams, and normalizing the data characteristics to obtain the input data for contour prediction includes:

[0058] Step 101: Collect the external loads, characteristic water levels, uplift pressure reduction coefficients, friction coefficients between concrete and bedrock surfaces, and mechanical parameters of concrete and bedrock for the concrete gravity dam project to be designed. Among them, the mechanical parameters of concrete and bedrock include density, elastic modulus, and Poisson's ratio, etc.

[0059] Step 102: Process the data using a data characteristic extraction method, and conduct a correlation analysis to remove highly correlated characteristics. Through principal component analysis (PCA), reduce the dimension of the data, simplify the complex redundant characteristics into several key characteristics, use the several key characteristics (refined key characteristics) as input data, and use the slope ratios of the upstream and downstream dam slopes, the elevation of the upstream starting slope point, the horizontal width of the dam bottom, and the dam top width parameters as output data.

[0060] Step 103: Use vectors to represent the input data (refined key characteristics) and output data, and use floating-point numbers to represent the key parameters of the input data and output data.

[0061] Step 104: Complete the feature expression of the input data and output data and construct the corresponding training and test data sets.

[0062] It should be noted that the external loads, characteristic water levels, reduction coefficients of uplift pressure, friction coefficients between concrete and rock foundation surfaces, mechanical parameters of concrete and rock foundation, as well as the slope ratios of the upstream and downstream dam slopes, elevation of the upstream slope starting point, horizontal width at the dam bottom, and dam top width parameters of existing design projects are collected; data is processed using data feature extraction methods to ensure data integrity and consistency; correlation analysis is carried out to remove highly correlated features to eliminate redundant information; through principal component analysis (PCA), data dimensionality reduction is performed to simplify complex redundant features into several key features, thereby reducing the dimensionality of model input; the refined key features are used as input, and the slope ratios of the upstream and downstream dam slopes, elevation of the upstream slope starting point, horizontal width at the dam bottom, and dam top width parameters are used as output; the feature expressions of input and output are completed and the corresponding training and test data sets are constructed.

[0063] Furthermore, the process of constructing the GAN model includes:

[0064] Step 201, based on the constructed data set and GAN model, conduct model training and testing. If the prediction accuracy meets the requirements, it can be used for design. Among them, the GAN model includes a generator, a discriminator, and a connection layer;

[0065] Step 202, construct the generator of the GAN model using the U-Net structure, input the contour prediction input data (data features of external loads, characteristic water levels, reduction coefficients of uplift pressure, friction coefficients between concrete and rock foundation surfaces, physical and mechanical parameters of concrete and rock foundation), and the output of the generator is a vectorized image of the gravity dam contour (which includes key coordinates such as upstream and downstream slope break points, dam top width, and horizontal width at the dam bottom);

[0066] Step 203, create the discriminator of the GAN model using a two-channel discriminator architecture. The discriminator includes an image discrimination branch and a mechanical stability discrimination branch. The image discrimination branch uses a traditional convolutional network to evaluate the geometric similarity between the generated profile contour and the real project; the mechanical stability discrimination branch is constructed based on ResNet-101, predicts the anti-sliding stability safety factor, anti-overturning safety factor, and maximum basal stress of the generated profile, outputs a design rationality score, and simultaneously integrates a specification constraint loss function and a mechanical performance loss function;

[0067] Step 204, innovate the connection layer: add a mask enhancement channel between the generator and the discriminator, and guide the GAN model to focus on detail optimization through local masks (such as the dam toe).

[0068] Furthermore, the process of conducting model training and testing based on the constructed data set and GAN model includes:

[0069] Step 2011: Collect real data: Collect CAD drawings of 50 gravity dams, extract key parameters (i.e., key features, slope ratios of upstream and downstream slopes, elevation of the starting point of the upstream slope, horizontal width at the dam bottom, and width at the dam top), and generate parametric data: Based on Python scripts, batch generate 300,000 sets of contour data covering extreme working conditions (such as seismic intensity of 9 degrees and deep silt pressure);

[0070] Step 2012: In the first stage, generate the overall contour (dominated by the MSE loss function), and in the second stage, optimize the local accuracy through gradient penalty. The weight ratio of the regularization constraint loss (errors in the width at the dam top and slope ratios of upstream and downstream slopes) to the mechanical loss (anti-sliding stability coefficient, anti-overturning safety factor, and maximum tensile stress error) gradually transitions from 3:1 to 1:2 to balance compliance and safety;

[0071] Step 2013: Input typical working conditions and compare the generated results with the manual design drawings of gravity dams;

[0072] Step 2014: Use geometric accuracy (Hausdorff distance) to generate the matching degree (threshold < 1.5 pixels) between the contour vertex sequence and the manual design coordinates;

[0073] Step 2015: Conduct mechanical performance verification on the GAN model, that is, complete the training and testing of the GAN model: Predict the anti-sliding stability coefficient, anti-overturning safety factor, and maximum tensile stress of the dam body through a pre-trained ResNet-101 proxy model (based on 500,000 sets of finite element data), and compare the error rate with the calculation results of the finite element software ≤ 8%;

[0074] Among them, during the training and verification of the GAN model, k-fold cross-validation is adopted to solve the problem of insufficient intelligent prediction accuracy with a small amount of data, and the root mean square error MSE is used as the error metric standard for accuracy evaluation.

[0075] Furthermore, the process of obtaining the dam body contour parameters includes:

[0076] Step 205: Use the contour prediction input data of the project to be designed as the input data to predict the output data composed of the slope ratios of the upstream and downstream slopes, the elevation of the starting point of the upstream slope, the horizontal width at the dam bottom, and the width at the dam top;

[0077] Step 206: Combine the output data with the foundation surface parameters to obtain the dam body contour parameters;

[0078] Among them, the foundation surface parameters include position parameters, geometric parameters, and mechanical parameters. The position parameters include the elevation of the foundation surface, the geometric parameters include the horizontal projection length of the foundation surface and the coordinates of the control points of the dam foundation rock body profile, and the mechanical parameters are the mechanical parameters of the bedrock (the mechanical parameters have been used as inputs in the GAN);

[0079] It should be noted that the parameters of the foundation surface refer to the geometric characteristics and mechanical property parameters of the contact surface between the gravity dam and the foundation, which mainly include: position parameters (elevation of the foundation surface), geometric parameters (horizontal projection length of the foundation surface, coordinates of the control points of the foundation rock body profile), and mechanical parameters.

[0080] Furthermore, the process of combining the output data with the parameters of the foundation surface includes:

[0081] Step 2061, integration of geometric parameters: Compare the generated bottom width of the dam (one of the output data) with the horizontal projection length of the foundation surface. If they match, it is directly used as the basis for the dam body contour; if not, parameters such as the bottom width of the dam and the upstream and downstream slope ratios in the output data need to be adjusted according to the bedrock topography. Among them, the elevation of the upstream slope starting point needs to be coordinated with the elevation of the foundation surface to ensure that the slope starting point of the dam body is located in the stable area of the bedrock.

[0082] Step 2062, integration into dam body contour parameters: The combined parameters include: optimized upstream and downstream slope ratios, top width of the dam, bottom width of the dam, elevation of the slope starting point, and geometric characteristics (length, position) and mechanical parameters of the foundation surface verified by mechanics.

[0083] Furthermore, the process of realizing the intelligent generation and optimization of the dam body design based on the GAN model and embedding physical and mechanical constraints includes:

[0084] Step 301, determine the requirements for mechanical stability and stress constraints. The mechanical stability includes anti-sliding stability and anti-overturning stability, and the stress constraint requirements include uniform stress distribution, compressive strength, tensile strength, and fatigue effect;

[0085] Step 302: Parametrically generate a multi-source heterogeneous dataset, which includes input data construction and output data construction. It should be noted that constructing the input data includes: drawing the input image according to the dam profile parameters, combining the external load, characteristic water level, uplift pressure reduction coefficient, friction coefficient between concrete and rock foundation surface, and mechanical parameter text data of concrete and rock foundation, extracting multi-modal features and performing data normalization and feature size alignment. Introduce a working condition sensitive scaling factor in the feature normalization stage, construct a parameter coupling matrix (PCM) for gravity dams, and convert the discrete engineering parameters (such as friction coefficient, anti-shear parameters) in the text modality into a tensor structure matching the pixel scale of the image through a numerical mapping function to complete the feature fusion of multi-modal input data. Introduce a weighted fusion mechanism to further improve the effect of data fusion. By setting the weights of different modal data, according to their correlation and contribution to the model prediction performance, perform weighted fusion on each modality to optimize the combination of multi-modal features; connect the mechanical parameter mode of typical dam instability cases (such as the friction coefficient threshold of sliding failure) in the fusion layer to endow the model with the ability of feature selection guided by engineering experience; and construct a mask with the dam profile parameters to process the image matrix and text vector parameters with mask technology.

[0086] The output data construction includes: the image of the dam optimization design parameters, and the text data of the mechanical performance and stability index parameters. Extract multi-modal features and perform data normalization and feature size alignment, stack different-sized features along the data channels to complete the feature fusion of multi-modal output data.

[0087] Specifically, by collecting the existing concrete gravity dam profile design data, including drawings and design information text, extract the key features of the drawings and text data of the existing concrete gravity dam profile design, count the profile parameters of the concrete gravity dam section and the layout parameters of materials in the image, and count the external load, characteristic water level, uplift pressure reduction coefficient, friction coefficient between concrete and rock foundation surface, and mechanical parameters of concrete and rock foundation in the text. Use parametric adjustment technology to expand the geometric and physical and mechanical parameters such as the dam slope and section materials, so as to construct a multi-source heterogeneous training and testing dataset.

[0088] Step 303: The generative adversarial network model for image generation consists of an image encoding feature extraction module and an image decoding generation module, and generates the corresponding optimized gravity dam profile by inputting the fused multi-modal features (refer to Figure 3 )

[0089] Step 304: Based on the augmented training and testing dataset, train and test the generative adversarial network model for image synthesis. If the generated quality meets the requirements, it can be used for design.

[0090] Step 305: Use the multi-modal data features of the project to be designed as the input to train and test the completed generative adversarial network model, and then the cross-section image of the concrete gravity dam can be generated. The generation result (reference Figure 4 ).

[0091] Furthermore, the process of determining the mechanical stability and stress constraint requirements includes:

[0092] Step 3011: For anti-sliding stability, force the anti-sliding safety factor to meet the standard; for anti-overturning stability, force the anti-overturning safety factor to meet the standard; for stress constraint, limit the tensile stress.

[0093] Step 3012: The cross-entropy loss between the discriminator output and the authenticity label forms the total loss. The generator and the discriminator are alternately optimized. After the generator generates the contour, the anti-sliding and overturning coefficients and the maximum tensile stress are predicted in real time through the surrogate model, and the constraint loss is calculated. Then, backpropagation is used to adjust the weights.

[0094] Step 3013: By analyzing the loss gradient at the initial stage of training, dynamically adjust λ1 and λ2 to balance geometric generation and mechanical constraints, and realize the closed-loop feedback of physical constraints of GAN.

[0095] It should be noted that the anti-sliding stability requirement means that the dam body does not slide under the action of external forces, which is usually evaluated by calculating the anti-sliding safety factor; the anti-overturning stability ensures that the dam body does not overturn under the action of loads, which is usually judged by the anti-overturning safety factor. Stress constraint focuses on the stress distribution inside the dam body, requiring uniform stress distribution, avoiding local stress concentration or exceeding the ultimate strength of the material, and ensuring the long-term safety of the dam body. In addition, the design also needs to consider the compressive and tensile strengths of the material and the fatigue effect to ensure that the dam body does not crack or break under multiple loadings. These constraint requirements ensure the stability and safety of the dam body in actual projects through reasonable design and calculation; the discriminator predicts the anti-sliding stability safety factor, anti-overturning stability safety factor and maximum tensile stress of the generated cross-section, and at the same time integrates the mechanical performance loss function.

[0096] Specifically, the GAN model is based on a large amount of engineering data. The generator receives the input constraint conditions (external load, characteristic water level, uplift pressure reduction coefficient, friction coefficient between concrete and bedrock surface, mechanical parameters of concrete and bedrock) and generates corresponding design schemes. The discriminator judges whether the generated design scheme meets the mechanical requirements, such as anti-sliding stability, anti-overturning stability and stress uniformity. The discriminator evaluates the rationality of the design scheme by calculating the mechanical performance of the design scheme; through training, the generator can automatically adjust the design parameters so that the generated design scheme can meet the mechanical requirements such as anti-sliding, anti-overturning stability and stress uniformity.

[0097] In this embodiment, an intelligent and automated dam optimization design is achieved by using a generative adversarial network, ensuring that the stability requirements of anti-sliding and anti-overturning are met, and the rationality of material and stress distribution is satisfied; a dataset for training is constructed by means of a parametric method and a mechanical analysis method, a generative adversarial network architecture capable of fusing image and text features is designed, and the model is gradually optimized and trained based on the constructed dataset; through the model training method embedding physical and mechanical stability and stress constraint information, the generated design can effectively avoid problems such as sliding, overturning, and insufficient foundation bearing capacity, and significantly improve the safety of dam projects.

[0098] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed in the present invention can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.

[0099] In several embodiments provided by the present invention, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is only one example, and there can be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections to each other can be through some interfaces, and the indirect couplings or communication connections of the devices or units can be in electrical, mechanical, or other forms.

[0100] As mentioned above, the above is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of changes or substitutions, which should all be covered by the protection scope of the present invention.

[0101] Finally: The above is only the preferred embodiment of the present invention and is not used to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should all be included in the protection scope of the present invention.

Claims

1. An intelligent design method for gravity dams based on a physical mechanics information generation adversarial network, characterized in that The method includes the following steps: Step 1: Collect and obtain the data characteristics of the external loads, characteristic water levels, uplift pressure reduction coefficients, friction coefficients between concrete and rock foundations, and mechanical parameters of concrete and rock for the design of domestic concrete gravity dams, and normalize the data characteristics to obtain the input data for profile prediction; Step 2: Construct and train a GAN model, input the input data for profile prediction, and the GAN model outputs the output data including the slope ratios of the predicted upstream and downstream dam slopes, the elevation of the upstream starting slope point, the horizontal width at the dam bottom, and the width at the dam top. Combine the output data with the foundation surface parameters to obtain the dam body profile parameters; Step 3: Based on the GAN model and embedded physical and mechanical constraints, realize the intelligent generation and optimization of dam design; The specific content of Step 3 includes: Step 301: Determine the requirements for mechanical stability and stress constraints. The mechanical stability includes anti-sliding stability and anti-overturning stability, and the stress constraint requirements include uniform stress distribution, compressive strength, tensile strength, and fatigue effects; Step 302: Parametrically generate a multi-source heterogeneous data set, which includes input data construction and output data construction; Step 303: The generative adversarial network model for image generation consists of an image encoding feature extraction module and an image decoding generation module, and generates the corresponding optimized profile of the gravity dam by inputting the fused multi-modal features; Step 304: Based on the augmented training and test data sets, train and test the generative adversarial network model for image synthesis. If the generated quality meets the requirements, it can be used for design; Step 305: Use the multi-modal data characteristics of the project to be designed as the input to the trained and tested generative adversarial network model, and the profile image of the concrete gravity dam can be generated, and the generation result can be obtained.

2. The intelligent design method of gravity dam based on physical mechanics information generating adversarial network according to claim 1, characterized in that The process of collecting and obtaining the data characteristics of the external loads, characteristic water levels, uplift pressure reduction coefficients, friction coefficients between concrete and rock foundations, and mechanical parameters of concrete and rock for the design of domestic concrete gravity dams, and normalizing the data characteristics to obtain the input data for profile prediction includes: Step 101: Collect the external loads, characteristic water levels, uplift pressure reduction coefficients, friction coefficients between concrete and rock foundations, and mechanical parameters of the concrete gravity dam project to be designed. Among them, the mechanical parameters of concrete and rock include density, elastic modulus, and Poisson's ratio; Step 102: Process the data using the data feature extraction method, and perform correlation analysis to remove highly correlated features. Reduce the dimension of the data through principal component analysis, simplify the complex redundant features into several key features, use the several key features as input data, and use the slope ratios of the upstream and downstream dam slopes, the elevation of the upstream starting slope point, the horizontal width at the dam bottom, and the width at the dam top parameters as output data; Step 103: Use vectors to represent the input data and output data, and use floating-point numbers to represent the input data and output data; Step 104: Complete the feature expression of the input data and output data and construct the corresponding training and test data sets.

3. The intelligent design method of gravity dam based on physical mechanics information generation adversarial network according to claim 1, characterized in that The process of constructing the GAN model includes: Step 201: Based on the constructed dataset and the GAN model, conduct model training and testing. If the prediction accuracy meets the requirements, it can be used for design. Among them, the GAN model includes a generator, a discriminator, and a connection layer; Step 202: Use the U-Net structure to construct the generator of the GAN model. Input the contour prediction input data, and the output of the generator is the vectorized image of the gravity dam contour; Step 203: Create the discriminator of the GAN model using a dual-channel discriminator architecture. The discriminator includes an image discrimination branch and a mechanical stability discrimination branch. The image discrimination branch uses a traditional convolutional network to evaluate the geometric similarity between the generated profile contour and the real project. The mechanical stability discrimination branch is constructed based on ResNet-101, predicts the anti-sliding stability safety factor, anti-overturning safety factor, and maximum base stress of the generated profile, outputs the design rationality score, and integrates the specification constraint loss function and the mechanical performance loss function at the same time; Step 204: Innovate the connection layer: Add a mask enhancement channel between the generator and the discriminator, and guide the GAN model to focus on detail optimization through local masks.

4. The intelligent design method of gravity dam based on physical mechanics information generating adversarial network according to claim 3, characterized in that The process of conducting model training and testing based on the constructed dataset and the GAN model includes: Step 2011: Collect real data: Collect CAD drawings of 50 gravity dams, extract key parameters, and generate parametric data; Step 2012: Generate the overall contour in the first stage, and optimize the local accuracy through gradient penalty in the second stage. The weight ratio of the specification constraint loss to the mechanical loss gradually transitions from 3:1 to 1:2 to balance compliance and safety; Step 2013: Input typical working conditions, and compare the generated results with the manual design drawings of the gravity dam; Step 2014: Use geometric accuracy to generate the matching degree between the contour vertex sequence and the manual design coordinates; Step 2015: Conduct mechanical performance verification on the GAN model, that is, complete the training and testing of the GAN model; Among them, when training and validating the GAN model, k-fold cross-validation is adopted, and the root mean square error MSE is used as the error metric standard for accuracy evaluation.

5. The intelligent design method of gravity dam based on physical mechanics information generation adversarial network according to claim 1, characterized in that The process of obtaining the dam body contour parameters includes: Step 205: Use the contour prediction input data of the project to be designed as the input data, and the output data composed of the slope ratios of the upstream and downstream dam slopes, the elevation of the upstream starting slope point, the horizontal width of the dam bottom, and the width of the dam top can be predicted; Step 206: Combine the output data with the foundation surface parameters to obtain the dam body contour parameters; Among them, the foundation surface parameters include position parameters, geometric parameters, and mechanical parameters. The position parameters include the elevation of the foundation surface, the geometric parameters include the horizontal projection length of the foundation surface and the coordinates of the control points of the foundation rock body profile, and the mechanical parameters are the mechanical parameters of the bedrock.

6. The intelligent design method of gravity dam based on physical mechanics information generating adversarial network according to claim 5, characterized in that, The process of combining the output data with the foundation surface parameters includes: Step 2061: Compare the generated horizontal width of the dam bottom with the horizontal projection length of the foundation surface. If the two match, it is directly used as the basis of the dam body contour; if they do not match, the dam bottom width and the upstream and downstream slope ratio parameters in the output data need to be adjusted according to the bedrock topography. Among them, the elevation of the upstream starting slope point needs to be coordinated with the elevation of the foundation surface to ensure that the starting slope point of the dam body is located in the stable area of the bedrock; Step 2062: Integrate the optimized upstream and downstream slope ratios, crest width, bottom width, starting slope point elevation, and the geometric characteristics and mechanical parameters of the foundation surface verified by mechanics into the dam body profile parameters.

7. The intelligent design method of gravity dam based on physical mechanics information generation adversarial network according to claim 1, characterized in that The process of determining the requirements for mechanical stability and stress constraints includes: Step 3011: For anti-sliding stability, the anti-sliding safety factor is required to meet the standard; for anti-overturning stability, the anti-overturning safety factor is required to meet the standard; for stress constraints, the tensile stress is restricted. Step 3012: The cross-entropy loss between the discriminator output and the authenticity label forms the total loss. The generator and the discriminator are alternately optimized. After the generator generates the profile, the anti-sliding and overturning coefficients and the maximum tensile stress are predicted in real time through the surrogate model, and the constraint loss is calculated. The weights are adjusted by backpropagation. Step 3013: By analyzing the loss gradient at the initial stage of training, λ1 and λ2 are dynamically adjusted to balance geometric generation and mechanical constraints, realizing the closed-loop feedback of physical constraints of GAN.

Images