Gravity dam intelligent design method based on generative adversarial network

Abstract

The invention relates to the technical field of gravity dam profile design and artificial intelligence, in particular to an intelligent gravity dam design method based on a generative adversarial network, and the method comprises the steps: determining the parameter boundary conditions of a gravity dam, and constructing a gravity dam parameterized finite element model; a topological optimization method is adopted, an optimization target is determined, and gravity dam partition section image data are generated in combination with parameter boundary conditions; preprocessing the design text information and the gravity dam partition profile image data to obtain high-dimensional tensor data, and constructing a complete data set based on the high-dimensional tensor data; taking the complete data set as a training sample, introducing a design constraint condition and a feature mask, and constructing a generative adversarial network; and inputting the to-be-designed feature data into the generative adversarial network, and generating a gravity dam partition design image meeting requirements. The invention aims to provide a gravity dam partition generation type design method based on topological optimization and a generative adversarial network so as to overcome the defects of a traditional gravity dam design method and improve the gravity dam design efficiency.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of gravity dam profile design and artificial intelligence, and particularly relates to a gravity dam intelligent design method based on a generative adversarial network. BACKGROUND

[0002] Gravity dam is one of the core structures in hydropower engineering construction. As a common dam type, it has the characteristics of strong stability and high compression resistance, and is widely used in the construction of water conservancy facilities such as reservoirs and hydropower stations. Gravity dam resists water pressure through its own weight, so the strength of the material and the stability of the structure are emphasized during the design and construction process. In China, gravity dam, as a traditional dam type, has rich design and construction experience, and the number of dam construction is also quite large. However, traditional gravity dam design relies on a large amount of engineer experience, and there are problems such as low design efficiency and difficulty in experience inheritance. With the development of artificial intelligence technology, the leading AI technology represented by generative AI is reshaping the engineering design industry, bringing efficient design tools, effectively reusing existing design data, and improving design quality. Generative AI has good ability to extract and learn the characteristics of existing data and generate new designs. The digital and intelligent design of gravity dam is an inevitable choice to realize high-quality and efficient engineering design and meet the development needs of intelligent construction. The intelligent transformation of design is needed to solve current problems.

[0003] The present application generates a large number of gravity dam profile partition image datasets based on topological optimization, fuses multi-modal data features of image-text data of the project to be designed, calls a generative adversarial network generated by partition design images, and then generates gravity dam profile partition design images to realize intelligent design of gravity dam profile scheme, which can greatly improve design efficiency and further improve the intelligent level of gravity dam design in China. SUMMARY

[0004] To achieve the above object, the present application provides the following technical scheme: According to the first aspect of the present application, the present application claims a gravity dam intelligent design method based on a generative adversarial network, comprising: determining the parameter boundary conditions of the gravity dam, and constructing a parameterized finite element model of the gravity dam; using a topological optimization method to determine an optimization target, and generating a gravity dam partition profile image data in combination with the parameter boundary conditions; preprocessing design text information and the gravity dam partition profile image data to obtain high-dimensional tensor data, and constructing a complete data set based on the high-dimensional tensor data; using the complete data set as a training sample, introducing design constraint conditions and feature masks, and constructing a generative adversarial network; Input the to-be-designed feature data into the generative adversarial network, and generate a required gravity dam partition design image.

[0005] Further, the determination of the parameter boundary condition of the gravity dam and the construction of the parameterized finite element model of the gravity dam further comprise: The parameterized finite element model of the gravity dam is established with self-weight load, water pressure, uplift pressure, sediment pressure and foundation constraint as boundary conditions.

[0006] Further, the determination of the optimization target by the topology optimization method and the generation of the gravity dam partition profile image data in combination with the parameter boundary condition further comprise: The topology optimization method is used to calculate the minimum overall flexibility of the gravity dam under different boundary conditions and load combinations to generate the gravity dam partition profile image data.

[0007] Further, the high-dimensional tensor data obtained by preprocessing the design text information and the gravity dam partition profile image data and constructing a complete data set based on the high-dimensional tensor data further comprise: The design text information and the gravity dam partition profile image data generated by the topology optimization are subjected to data cleaning and standardization and are fused and converted into high-dimensional tensor data suitable for the generative adversarial network; The high-dimensional tensor data is used as the input of the generative adversarial network, and the output is the dam height, the profile material partition parameter, and the material partition area proportion parameter.

[0008] Further, the complete data set is used as a training sample, design constraints and feature masks are introduced, and the generative adversarial network is constructed, which further comprises: The high-dimensional tensor data is used as a training sample to enable the generative adversarial network to learn the gravity dam partition design; Design constraints and feature masks are introduced to guide the optimization of the generation process; In the generative adversarial network, the generator continuously adjusts the generated image to gradually approach the design target; The discriminator helps the generator optimize the generation effect by judging the difference between the generated image and the real image to generate gravity dam partition profile data that meets the design target.

[0009] Further, the input of the to-be-designed feature data into the generative adversarial network and the generation of the required gravity dam partition design image further comprise: The generative adversarial network for partition design image generation is called to input the to-be-designed feature data and generate a required gravity dam partition design image; The generative adversarial network for partition design image generation is trained and tested before having the prediction and generation capabilities.

[0010] Furthermore, the method of using topology optimization to minimize the overall flexibility of the gravity dam, performing calculations under different boundary conditions and load combinations to generate gravity dam sectional profile image data, also includes: With the goal of minimizing the overall flexibility of gravity dams, the material density distribution under different load combinations is calculated based on the variable density method, generating a gravity dam image dataset containing material distribution, stress state, and stability coefficient. A topology optimization model based on the variable density method is established with the objective function of minimizing the cross-sectional flexibility of the gravity dam.

[0011] In the formula, The compliance matrix of the gravity dam cross-section optimization region; Indicates the first Units; To optimize the total number of units within the region; design variables For the first The relative density of each unit cell. and These are the upper and lower limits; This represents the upper limit of the cross-sectional area of ​​the gravity dam. For the first The area of ​​each unit; , and These are the stiffness matrix, displacement matrix, and external force vector, respectively. and These are the element matrix and the element stiffness matrix, respectively.

[0012] Furthermore, the method also includes: The key features are refined using vectors, with floating-point numbers used to represent key parameters and one-hot vectors used to represent key categories. Collect characteristic water levels, seismic requirements, foundation constraints and material physical and mechanical parameters of gravity dam projects, corresponding dam height, profile material zoning parameters, and material zoning area ratio parameters; Data feature extraction methods are used for data feature processing, and correlation analysis is performed on text data features to clean up features with excessive correlation. Principal component analysis is performed to reduce the dimensionality of features, refining complex and redundant features into key features as input, and outputting dam height, profile material zoning parameters, and material zoning area ratio parameters. The refined key text features and the established gravity dam finite element image features are fused together, and multi-channel superposition fusion is used to convert them into a high-order tensor form suitable for generative adversarial networks. The feature expressions of input and output are completed, and the corresponding training and test data sets are constructed.

[0013] The application relates to the technical field of gravity dam profile design and artificial intelligence, in particular to a gravity dam intelligent design method based on a generative adversarial network. BRIEF DESCRIPTION OF DRAWINGS

[0014] Figure 1 A work flow diagram of the gravity dam intelligent design method based on the generative adversarial network claimed by the embodiment of the application; Figure 2 A generative adversarial network diagram of the gravity dam intelligent design method based on the generative adversarial network claimed by the embodiment of the application; Figure 3 A generative adversarial network diagram of the gravity dam intelligent design method based on the generative adversarial network claimed by the embodiment of the application; Figure 4 A typical gravity dam design profile output result diagram of the gravity dam intelligent design method based on the generative adversarial network claimed by the embodiment of the application. DETAILED DESCRIPTION

[0015] The technical solutions in the embodiments of the application will be clearly and completely described below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor fall within the protection scope of the application.

[0016] The terms "first", "second", "third", etc. in the present application are only for descriptive purpose and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the technical features indicated. Therefore, the features defined with "first", "second", "third" can explicitly or implicitly include at least one of the features. In the description of the present application, the meaning of "a plurality of" is at least two, such as two, three, etc., unless otherwise explicitly and specifically limited. All directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present application are only used to explain the relative position relationship, movement condition, etc. between components in a certain posture (as shown in the drawings), and if the certain posture changes, the directional indications also change accordingly. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device including a series of steps or units is not limited to the listed steps or units, but can optionally include steps or units not listed, or can optionally include other steps or units inherent to the process, method, product or device.

[0017] Reference herein to "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment can be included in at least one embodiment of the application. The appearance of the phrase in various places in the specification does not necessarily all refer to the same embodiment, nor is it necessarily mutually exclusive of other embodiments. It is explicitly and implicitly understood that the embodiments described herein can be combined with other embodiments.

[0018] According to the first embodiment of the present application, with reference to Figure 1 The present application claims a gravity dam intelligent design method based on a generative adversarial network, comprising: determining parameter boundary conditions of the gravity dam, and constructing a parameterized finite element model of the gravity dam; adopting a topology optimization method to determine an optimization target, and generating a gravity dam partition profile image data in combination with the parameter boundary conditions; preprocessing design text information and the gravity dam partition profile image data to obtain high-dimensional tensor data, and constructing a complete data set based on the high-dimensional tensor data; taking the complete data set as a training sample, introducing design constraint conditions and feature masks, and constructing a generative adversarial network; inputting to-be-designed feature data into the generative adversarial network to generate a gravity dam partition design image meeting the requirements.

[0019] Further, the determination of the parameter boundary conditions of the gravity dam and the construction of the parameterized finite element model of the gravity dam further comprise: This embodiment takes a concrete gravity dam with a height of 80.0 m as the research object, the dam body has a bottom width of 60.0 m, a top width of 8.0 m, and a vertical height of 10.0 m on the downstream side of the dam site, and a two-dimensional plane stress finite element model is used for modeling: In this embodiment, the gravity load (concrete density kg / m 3 , elastic modulus of 25.0 GPa, and Poisson's ratio of 0.167, W1=17700 kN), water pressure (upstream water depth of 70 m, W2=24034.5 kN / m, downstream water depth of 5 m, w2=122.625 kN / m), uplift pressure (U=22072.5 kN / m), sediment pressure (P s =100 kN / m), and foundation restraint (density of bedrock is 2600 kg / m 3 , elastic modulus of 29.0 GPa, and Poisson's ratio of 0.2) are used as boundary conditions to establish a parameterized finite element model of the gravity dam.

[0020] Further, the topological optimization method is used to determine the optimization target, and the gravity dam partition profile image data is generated in combination with the parameter boundary conditions, which further includes: In this embodiment, the topological optimization method is used to minimize the overall flexibility of the gravity dam, and the calculation is performed under different boundary conditions (upstream water level is 70 m, 80 m and 90 m) and load combinations (water pressure, sediment pressure, etc. are different when the upstream water level is different) to generate gravity dam partition profile image data.

[0021] The variable density method is used for topological optimization, and the overall flexibility of the dam under the given load combination is used as the optimization objective function, the material penalty factor P=3; the minimum density threshold: min =10 -3 ; maximum iteration number: 150; filter radius: 1.0 m.

[0022] Further, the high-dimensional tensor data is obtained by preprocessing the design text information and the gravity dam partition profile image data, and the complete data set is constructed based on the high-dimensional tensor data, which further includes: The design text information (dam height, self-weight load, water pressure, sediment pressure and uplift pressure) and the gravity dam partition profile image data generated by topology optimization are subjected to data cleaning and standardization. The design text information extracts five structural design parameters, which are combined into a vector after standardization. The design parameter tensor size is [N, 5], N is the number of design text information of the gravity dam under different working conditions, and 5 is the five structural design parameters (dam height, self-weight load, water pressure, uplift pressure and sediment pressure). The topology optimization image is converted into tensor data, and the normalized gray value is obtained. The image tensor size is [N, 1, 120, 160], N is the number of generated sample images under different working conditions, 1 is the number of channels, 120 is the number of pixels in the vertical direction of the image, and 160 is the number of pixels in the horizontal direction of the image. Finally, the fusion is converted into high-dimensional tensor data suitable for generating a generative adversarial network, and the final combination is a high-dimensional tensor: complete input tensor: [N, 6, 120, 160], N is the number of data sets, and 6 is the fusion of five text design parameters and one image data into six channels. The high-dimensional tensor data is used as the input of the generative adversarial network, and the output is the dam height, the profile material partition parameter, and the material partition area ratio parameter.

[0023] Further, with reference to Figure 2 , the complete data set is used as a training sample, design constraints and feature masks are introduced, and a generative adversarial network is constructed. The high-dimensional tensor data is used as a training sample, and the generative adversarial network learns the gravity dam partition design. Design constraints and feature masks are introduced to guide the optimization of the generation process. In the generative adversarial network, the generator continuously adjusts the generated image to gradually approach the design target. The discriminator helps the generator optimize the generation effect by judging the difference between the generated image and the real image, and generates the gravity dam partition profile data that meets the design target.

[0024] Further, the feature data to be designed is input into the generative adversarial network to generate a gravity dam partition design image that meets the requirements. The generative adversarial network for partition design image generation is called, and the feature data to be designed (dam height, self-weight load, water pressure, uplift pressure and sediment pressure) is input to generate a gravity dam partition design image that meets the requirements. The generative adversarial network for partition design image generation has the ability to predict and generate before the model is trained and tested.

[0025] Further, the method employs topology optimization method to minimize the overall flexibility of the gravity dam, and calculates under different boundary conditions and load combinations to generate the gravity dam partition profile image data, which further includes: Referring to Figure 3 The method minimizes the overall flexibility of the gravity dam, and calculates the material density distribution under different load combinations (upstream water level of 70m, 80m, 90m) based on the variable density method to generate the gravity dam image dataset containing material distribution, stress state and stability coefficient. Different load combinations are the load combinations under different water level combinations mentioned above, and then the variable density method mentioned above is used to generate the image dataset, which contains material distribution, stress distribution and stability coefficient. The stress state and stability coefficient are used to evaluate whether the stability requirement under the load condition can be met, and if the requirement is met, the image is retained.

[0026] The variable density method topology optimization model is established with the minimum flexibility of the gravity dam section as the objective function;

[0027] In the formula, is the flexibility matrix of the optimized area of the gravity dam section; represents the first unit; is the total number of units in the optimized area; the design variable is the relative density of the first unit, and are the upper and lower limits; is the upper limit of the gravity dam section area; is the area of the first unit; , and are the stiffness matrix, displacement matrix and external force vector respectively; and are the element matrix and element stiffness matrix respectively.

[0028] Further, the method further includes: The key features after refinement are characterized by vectors, the key parameters are characterized by floating point numbers, and the key categories are characterized by one-hot vectors; Collect the characteristic water level, seismic requirements, foundation constraints and material physical and mechanical parameters of the gravity dam project, the corresponding dam height, profile material partition parameters, and material partition area proportion parameters; Data feature extraction method is used for data feature processing, and correlation analysis is performed on text data features, and features with high correlation are cleaned; ​Perform principal component analysis for feature dimension reduction, refine complex redundant features into key features as input, and output dam height, profile material partition parameters, and material partition area ratio parameters; Fuse the refined key text features and the established gravity dam finite element image features, and convert them into high-order tensor form suitable for the generative adversarial network through multi-channel superposition fusion; Complete the feature expression of input and output and build the corresponding training and test data sets.

[0029] In this embodiment, existing gravity dam profile design data, including drawings (dwg format) and design information text, are collected. The key features of the existing face gravity dam profile design information are extracted from the drawings (dam height 80m, upstream and downstream slope ratio, material partition area ratio (area of various strength concrete materials)) and text data (upstream water level, foundation elastic model 29.0GPa, foundation Poisson's ratio 0.2). The gravity dam profile contour parameters and partition geometric coordinate parameters in the image, as well as the profile partition area ratio parameters, are counted. The characteristic water level, seismic requirement parameters, foundation constraints and their material physical and mechanical parameters in the text are counted. The distribution of different parameters is obtained, and a multi-source heterogeneous training and test data set is constructed; Fuse the data features of the gravity dam partition profile image data and the text design information, and construct the corresponding fusion data input and output feature data sets. At the same time, the gravity dam profile contour image is constructed as a mask, i.e. the number in the contour is 1 and the number outside the contour is 0. The mask data is used in conjunction with the input data;

[0030]

[0031] In the formula, M cnt is the contour mask matrix, m ij is the value of each point in the matrix; M k is a different feature matrix, which is different materials partition of gravity dam and scaling ratio, is the normalized scalar value of different features.

[0032] Based on the constructed data set and the generative adversarial network, the model is trained and tested, and the prediction accuracy meets the requirements for design, generating gravity dam partition profile data that meets the design target.

[0033] Before the model has partition design capability, the training and test data sets of the corresponding parallel integrated learning model need to be constructed for model training and testing; The generative adversarial network mainly consists of two parts, a generator and a discriminator. The generator receives an initial random noise input and gradually generates a gravity dam profile image. The generated image is guided by design constraints to ensure that the generated image meets engineering specifications. The discriminator is responsible for determining whether the generated image is real and conducting adversarial training with the generator to continuously optimize the output of the generator. To ensure that the generated data is strictly limited within the effective profile of the dam body, a profile mask of the gravity dam is introduced in the decoder part of the generator. The mask filters the output features of the generator by multiplying them channel by channel, ensuring that the generated gravity dam profile data strictly meets the design constraints. During training, the generator gradually optimizes the generated gravity dam image through adversarial training, and the discriminator continuously guides the generator to improve the authenticity and design requirements of the image. Referring to Figure 4 Through multiple optimization iterations, the output of the generator is evaluated and fed back by the discriminator each time, and finally the gravity dam profile partition image that meets the design target is generated.

[0034] In several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the above-described device embodiments are only illustrative, and for example, the division of units is only a logical function division, and actual implementation can have another division manner, for example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the displayed or discussed units can be indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0035] In addition, the functional units in each embodiment of the present application can be integrated in one processing unit, or each unit can exist physically, or two or more units can be integrated in one unit. The above integrated unit can be realized in the form of hardware or in the form of a software functional unit. The above is only an embodiment of the present application, and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation using the contents of the specification and drawings, or direct or indirect application in other related technical fields, are also included in the patent protection scope of the present application.

[0036] The specific embodiments of the application are described in detail above, but they are only examples. The present application is not limited to the above-described specific embodiments. Any equivalent modification or substitution of the present application by those skilled in the art is also within the scope of the present application, and therefore, any equivalent transformation and modification, improvement, etc. made without departing from the spirit and principle range of the present application should be included in the scope of the present application.

Claims

1. A method for intelligent design of gravity dam based on generative adversarial network, characterized in that, The application relates to a method for generating a gravity dam design image, and belongs to the technical field of gravity dam design. The method comprises the following steps: determining parameter boundary conditions of a gravity dam, and constructing a parameterized finite element model of the gravity dam; adopting a topological optimization method to determine an optimization target, and generating gravity dam partition profile image data in combination with the parameter boundary conditions; preprocessing design text information and the gravity dam partition profile image data to obtain high-dimensional tensor data, and constructing a complete data set based on the high-dimensional tensor data; taking the complete data set as a training sample, introducing design constraint conditions and a feature mask, and constructing a generative adversarial network; 2.The method of claim 1, wherein, inputting to-be-designed feature data into the generative adversarial network to generate a gravity dam partition design image meeting requirements. The method further comprises the following steps for determining the parameter boundary conditions of the gravity dam and constructing the parameterized finite element model of the gravity dam: 3.The method of claim 1, wherein, establishing the parameterized finite element model of the gravity dam by taking dead weight, water pressure, uplift pressure, silt pressure and foundation constraints as boundary conditions. The method further comprises the following steps for adopting the topological optimization method to determine the optimization target and generating the gravity dam partition profile image data in combination with the parameter boundary conditions: 4.The method of claim 1, wherein, applying the topological optimization method to calculate under different boundary conditions and load combinations with the minimum overall flexibility of the gravity dam as the target, and generating the gravity dam partition profile image data. The method further comprises the following steps for preprocessing the design text information and the gravity dam partition profile image data to obtain the high-dimensional tensor data and constructing the complete data set based on the high-dimensional tensor data: performing data cleaning processing and standardization on the design text information and the gravity dam partition profile image data generated by the topological optimization, and fusing and converting the data into high-dimensional tensor data suitable for the generative adversarial network; 5. The method of claim 1, wherein, taking the high-dimensional tensor data as input of the generative adversarial network, and taking dam height, profile material partition parameters and material partition area proportion parameters as output to construct the complete data set. The method further comprises the following steps for taking the complete data set as the training sample, introducing the design constraint conditions and the feature mask, and constructing the generative adversarial network: taking the high-dimensional tensor data as the training sample to enable the generative adversarial network to learn gravity dam partition design; introducing the design constraint conditions and the feature mask to guide optimization of the generation process; in the generative adversarial network, the generator continuously adjusts generated images to gradually approach a design target; 6. The method of claim 1, wherein the method is based on a generative adversarial network. the discriminator helps the generator to optimize generation effect by judging differences between the generated images and real images, and generates gravity dam partition profile data meeting the design target. The method further comprises the following steps for inputting the to-be-designed feature data into the generative adversarial network to generate the gravity dam partition design image meeting the requirements: calling the generative adversarial network for partition design image generation, inputting the to-be-designed feature data, and generating the gravity dam partition design image meeting the requirements; 7. The method of claim 3, wherein the method is based on a generative adversarial network. the generative adversarial network for partition design image generation is trained and tested before having the prediction and generation capabilities. The method further comprises the following steps for applying the topological optimization method to calculate under different boundary conditions and load combinations with the minimum overall flexibility of the gravity dam as the target, and generating the gravity dam partition profile image data: taking minimization of the overall flexibility of the gravity dam as the optimization target, calculating material density distribution under different load combinations based on a variable density method, and generating a gravity dam image data set containing material distribution, stress state and stability coefficient; Topology optimization model of variable density method is established with the minimum flexibility of gravity dam section as the objective function; wherein, is the flexibility matrix of the optimization region of the gravity dam section; represents the i-th element; represents the i-th element; is the total number of elements in the optimization region; design variable is the i-th element; is the relative density of the i-th element, and are the upper and lower limit values; is the upper limit of the gravity dam section area; is the area of the i-th element; , , and are the stiffness matrix, displacement matrix and external force vector, respectively; and are the element matrix and element stiffness matrix, respectively.

8. The method of claim 4, wherein the method is based on a generative adversarial network. Also includes: After the key feature representation is refined by using vectors, the key parameters are represented by floating-point numbers, and the key categories are represented by one-hot vectors. Collect the characteristic water level, seismic requirements, foundation constraints, and material physical and mechanical parameters of gravity dam projects, as well as the corresponding dam height, cross-section material partition parameters, and material partition area proportion parameters. Perform data feature processing using data feature extraction methods, and analyze the correlation of text data features to clean features with high correlation. Perform principal component analysis for feature dimension reduction, refine complex and redundant features into key features as input, and output dam height, cross-section material partition parameters, and material partition area proportion parameters. Fuse the refined key text features with the established gravity dam finite element image features, and convert them into high-order tensor form suitable for generative adversarial networks through multi-channel superposition fusion. Complete the feature representation of input and output and build the corresponding training and test data sets.

Images