Intelligent design method for section of concrete faced rockfill dam based on topological optimization and diffusion model

Abstract

The invention relates to the technical field of face rockfill dam profile design and artificial intelligence, in particular to a face rockfill dam profile intelligent design method based on topological optimization and a diffusion model, and the method comprises the steps: obtaining key design parameters and structural features of a face rockfill dam profile as a data source; performing data preprocessing to construct a high-dimensional feature space format suitable for the diffusion model, and generating a data set of the concrete faced rockfill dam; performing multi-stage training on a diffusion model based on the input data set to generate a noisy sample; calling a diffusion model, and inputting contour prediction data to obtain a concrete faced rockfill dam section design initial scheme of the concrete faced rockfill dam meeting engineering requirements; and evaluating and screening the initial design scheme of the section of the concrete faced rockfill dam according to the design target and the constraint condition to obtain an optimized design scheme of the section of the concrete faced rockfill dam. According to the method, the advantages of topological optimization and generative design technologies can be effectively fused, the intellectualization and innovativeness of the design are remarkably improved, and the method is suitable for the generative design of the earth and rockfill dam.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of panel stack dam profile design and artificial intelligence technology, in particular to a panel stack dam profile intelligent design method based on topology optimization and diffusion model. BACKGROUND

[0002] The traditional concrete panel stack dam design mode excessively relies on the experience of engineers, and faces the problems of difficult experience inheritance and limited design efficiency improvement. With the promotion of pumped storage power stations and hydropower projects in recent years, the digital and intelligent design of panel dams has gradually become an inevitable trend to realize high-quality and efficient design and match the demand for intelligent construction. At present, the design data is large and difficult to obtain, and some key data is missing.

[0003] Through the topology optimization method, the diffusion model is used to optimize and generate the input data set in multiple stages, and gradually extract the profile design scheme that meets the engineering requirements from random noise. At the same time, using the diffusion model generated by the partition design image, the partition design image of the panel stack dam can be effectively generated. This process not only significantly improves the design efficiency, but also further promotes the intelligent development of panel dam design in China. SUMMARY

[0004] To achieve the above purpose, the present application provides the following technical scheme: According to the first aspect of the present application, a panel stack dam profile intelligent design method based on topology optimization and diffusion model is claimed, comprising: Generating the panel stack dam using the topology optimization method, obtaining the key design parameters and structural characteristics of the panel stack dam profile as the data source; Data preprocessing is performed on the data source, a high-dimensional feature space format suitable for the diffusion model is constructed, and a data set is formed; Using forward diffusion and reverse denoising methods, introducing design parameter tensors, training the diffusion model based on the input data set in multiple stages, and generating noisy samples; Calling the diffusion model, inputting the contour prediction data, combining the noisy samples to obtain the initial scheme of the panel stack dam profile design of the panel stack dam that meets the engineering requirements; According to the design target and constraint condition, the initial scheme of the panel stack dam profile design is evaluated and screened to obtain the optimized panel stack dam profile design scheme.

[0005] Further, the generating the panel stack dam using the topology optimization method, obtaining the key design parameters and structural characteristics of the panel stack dam profile as the data source further comprises: Determine the design domain and boundary conditions, set the objective function and constraint conditions, and use the nonlinear finite element method to iteratively solve to generate the optimization results; Obtain the profile shape of the face slab rockfill dam and the rockfill parameters of each partition, and use the topology optimization technology to construct the data source of the face slab rockfill dam structure.

[0006] Further, the determination of the design domain and the boundary conditions, the setting of the objective function and the constraint conditions, and the generation of the optimization results by using the nonlinear finite element method to iteratively solve also include: Based on the physical and mechanical parameters of the rockfill material and its source, the self-weight load, the water load, and the foundation constraint parameters of different face slab rockfill dams, a corresponding face slab rockfill dam topology optimization model is established; By using APDL secondary development of ANSYS software, the Duncan-Zhang EB constitutive model of the face slab rockfill dam is realized. Based on the Duncan-Zhang EB constitutive model, the topology optimization algorithm of the homogenization method is used to solve the profile shape of the face slab rockfill dam under different material parameters and load combinations, and a training data set is constructed.

[0007] Further, the data preprocessing of the data source and the construction of a high-dimensional feature space format suitable for the diffusion model to form a data set also include: After extracting the design information from the topology optimization results, processing and standardization are performed, and the high-dimensional tensor form suitable for the model is converted as the input of the diffusion model, and the output is the dam height, the slope ratio of the upstream and downstream dam slopes, and a complete data set is constructed.

[0008] Further, the extraction of design information from the data source, the processing and standardization, and the conversion into a high-dimensional tensor form suitable for the model as the input of the diffusion model also include: The geometric parameters, material parameters, and environmental conditions are normalized before input; After the parameter standardization is completed, the input tensor of each parameter is constructed according to the input structure of the diffusion model; The input tensors of all parameters are spliced to form a global input tensor of a fixed dimension, which is used as the input of the conditional diffusion model; The global input tensor and the random noise tensor are jointly input into the diffusion model to generate a dam profile structure vector that meets the target performance indicators and constraint conditions.

[0009] Further, the use of the forward diffusion and reverse denoising method, the introduction of the design parameter tensor, and the multi-stage training of the diffusion model based on the input data set to generate a noisy sample also include: In the forward diffusion stage, the original dam profile data is gradually added to the Gaussian noise to generate a noisy sample; a reverse denoising stage, training a neural network to predict noise and gradually restore to the original data at each step; During training, a design parameter tensor is introduced as conditional information, which is input into the diffusion model together with the noisy sample and time step, guiding the network to learn the restoration path under the condition; By minimizing the mean square error between the model predicted noise and the true noise, conditional diffusion model training is performed.

[0010] Further, the diffusion model is called to input the contour prediction data and combine the noisy sample to obtain an initial scheme of the face slab rockfill dam profile design of the face slab rockfill dam that meets the engineering requirements, and further includes: The diffusion model samples an initial noise image from a standard Gaussian distribution, combines a given design parameter tensor as a conditional input, and gradually denoises according to the reverse diffusion path learned in the training process, and then iterates; In each iteration, the diffusion model predicts noise residuals based on the current noise state, time step, and conditional tensor, and updates the image until the entire denoising process is completed, outputting a dam profile graph or its structural expression corresponding to the input parameters, and performing conditional generation from random noise to structural morphology.

[0011] Further, the initial scheme of the face slab rockfill dam profile design is evaluated and screened according to the design target and constraint condition to obtain an optimized face slab rockfill dam profile design scheme, and further includes: Before partitioning the diffusion model, construct the corresponding training and test data sets for model training and testing; Based on topological optimization, the profile design data of the face slab rockfill dam is obtained, including drawings and design information texts, and the profile design information of the existing face slab rockfill dam is extracted from the drawings and text data key features, the face slab rockfill dam profile contour parameters and partition arrangement parameters in the image are counted, and the characteristic water level, seismic requirement, and face slab rockfill dam rockfill physical and mechanical parameters in the text are counted to obtain the distribution law of different parameters; Convert to design data through parameterization method, process it into multiple input tensors, including design constraint input tensor, Gaussian noise input tensor, feature mask tensor, and target arrangement design tensor; The tensors correspond to the constraint conditions of the design, the initial noise of the diffusion model, the regional features that need to be paid attention to, and the arrangement targets of each zone of the face slab rockfill dam profile; Through the generated profile data, the position and range of each functional partition are determined, and corresponding vectorization processing is performed to convert the geometric shape and position of each partition into vector data, obtaining a complete vectorization layout scheme of the face slab rockfill dam profile partition.

[0012] Based on the training and test data sets, the diffusion model is trained and tested to generate a quality that meets the design requirements.

[0013] The application relates to the field of panel rockfill dam profile design and artificial intelligence technology, in particular to a panel rockfill dam profile intelligent design method based on topology optimization and a diffusion model, key design parameters and structural characteristics of a panel rockfill dam profile are obtained as a data source; data preprocessing is performed to construct a high-dimensional feature space format suitable for a diffusion model, and a data set of the panel rockfill dam is generated; the diffusion model is trained in multiple stages based on the input data set, and a noisy sample is generated; the diffusion model is called, contour prediction data is input, and an initial scheme of the panel rockfill dam profile design of the panel rockfill dam meeting engineering requirements is obtained; the initial scheme of the panel rockfill dam profile design is evaluated and screened according to a design target and a constraint condition, and an optimized panel rockfill dam profile design scheme is obtained. The application can effectively integrate the advantages of topology optimization and generative design technology, significantly improve the intelligence and innovation of design, and is suitable for generative design of earth and rockfill dams. BRIEF DESCRIPTION OF DRAWINGS

[0014] Figure 1 A workflow diagram of a panel rockfill dam profile intelligent design method based on topology optimization and a diffusion model claimed by the embodiments of the application; Figure 2 A specific parameter definition schematic diagram of dam body contour parameterization generation of a panel rockfill dam profile intelligent design method based on topology optimization and a diffusion model claimed by the embodiments of the application; Figure 3 A diffusion model prediction schematic diagram of a panel rockfill dam profile intelligent design method based on topology optimization and a diffusion model claimed by the embodiments of the application; Figure 4 A typical panel rockfill dam design profile output result schematic diagram of a panel rockfill dam profile intelligent design method based on topology optimization and a diffusion model claimed by the embodiments of the application; Figure 5 A topology optimization calculation result presentation diagram of a panel rockfill dam profile intelligent design method based on topology optimization and a diffusion model claimed by the embodiments 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 "multiple" is at least two, such as two, three, etc., unless otherwise explicitly specified. 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 a first embodiment of the present application, with reference to Figure 1 , the present application claims a panel rockfill dam profile intelligent design method based on topology optimization and diffusion model, comprising: Generating the panel rockfill dam by using the topology optimization method, obtaining the key design parameters and structural characteristics of the panel rockfill dam profile as a data source; Data preprocessing is performed on the data source to construct a high-dimensional feature space format suitable for the diffusion model, forming a data set; Using forward diffusion and reverse denoising methods, introducing a design parameter tensor, and based on the input data set, the diffusion model is trained in multiple stages to generate noisy samples; Calling the diffusion model, inputting the profile prediction data, and combining the noisy samples to obtain the initial scheme of the panel rockfill dam profile design of the panel rockfill dam that meets the engineering requirements; According to the design target and constraint condition, the initial scheme of the panel rockfill dam profile design is evaluated and screened to obtain an optimized panel rockfill dam profile design scheme.

[0019] Further, the face slab rockfill dam is generated by using the topology optimization method, key design parameters and structural characteristics of a profile of the face slab rockfill dam are obtained as a data source, and the method further comprises the following steps: The design domain and boundary conditions are determined, the objective function and constraint conditions are set, and the nonlinear finite element method is used to iteratively solve to generate an optimization result. The profile shape of the face slab rockfill dam and the rockfill parameters of each subzone are obtained, and a data source of the face slab rockfill dam structure is constructed by using a topology optimization technique.

[0020] Further, the design domain and boundary conditions are determined, the objective function and constraint conditions are set, and the nonlinear finite element method is used to iteratively solve to generate an optimization result, and the method further comprises the following steps: Based on the physical and mechanical parameters of the rockfill and its source, the self-weight load, the water load, and the foundation constraint parameters of different face slab rockfill dams, a corresponding topology optimization model of the face slab rockfill dam is established. The Duncan-Chang EB constitutive model of the face slab rockfill dam is realized by using APDL secondary development of the ANSYS software. Based on the Duncan-Chang EB constitutive model, a topology optimization algorithm of the homogenization method is used to solve the profile shape of the face slab rockfill dam under different material parameters and load combinations, and a training data set is constructed.

[0021] In this embodiment, the method further comprises the following steps:

[0022] In the formula, the first formula of the constraint is a structural equilibrium equation expressed in a variational form. - design variables; - the "parameterized" density field for the i-th periodic microstructure, a and b represent the geometric size of the microstructure, θ represents the orientation angle, L represents the periodic unit size, and i represents different microstructure types or design unit numbers; - the "macroscopic" continuous density field over the entire domain, used for global volume constraints and material distribution; - the structural compliance functional; - node displacement; - the node equivalent boundary load to which the structure is subjected; - the node equivalent body force to which the structure is subjected; - the strain energy of the structure; - the set of allowable structural stiffness tensors; - the initial volume of the structure; - the specified removed volume. - the linear type of external load work; - the trial function space that satisfies the displacement boundary conditions; The geometric domain of the structure.

[0023] The mathematical model corresponds to the core link of "solving the optimal section shape under the condition of given material parameters and load combination" in the topology optimization process. First, the equivalent elastic parameters are calculated and the macro finite element model is constructed by giving the microstructure parameters Then, under the premise of satisfying the balance equation and volume constraint, the density distribution and microstructure combination that can minimize the material volume are found through iterative optimization, so as to obtain the optimal dam section structure shape. This process is executed multiple times to generate a large number of representative input-output pairs (material parameters, load combination to optimal section shape) for constructing the training data set to support the subsequent deep learning-based fast dam generation model training.

[0024] Further, the data source is preprocessed to construct a high-dimensional feature space format suitable for the diffusion model to form a data set, which also includes: After extracting the design information from the data source, it is processed and standardized to convert it into a high-dimensional tensor form suitable for the model as input, and the output is the dam height, upstream and downstream dam slope ratio, to construct a complete data set.

[0025] Referring to Figure 2 , specific parameter definitions are generated for dam profile parameterization.

[0026] Further, the design information extracted from the data source is processed and standardized to convert it into a high-dimensional tensor form suitable for the model as input, which also includes: The geometric parameters, material parameters and environmental conditions are normalized before input; After parameter standardization, input tensors of each parameter are constructed according to the input structure of the diffusion model; All parameter input tensors are spliced to form a global input tensor of fixed dimension, which is used as the input of the conditional diffusion model; The global input tensor and random noise tensor are jointly input into the diffusion model to generate a dam section structure vector that meets the target performance indicators and constraint conditions.

[0027] In this embodiment, to enhance the training stability and generation accuracy of the model, the geometric parameters, material parameters and environmental conditions need to be normalized before input. Numerical parameters are uniformly normalized by linear normalization method; categorical parameters (such as rockfill type) can be embedded in the form of one-hot encoding; After the parameter standardization is completed, a unified input tensor is constructed according to the input structure of the diffusion model. Specifically, the parameters are encoded into a one-dimensional tensor in a structured vector form, and the tensor content includes: dam body geometric parameters, material parameters of each partition (spliced in region order), and environmental working condition parameters; All parameters are finally spliced into an input tensor with a fixed dimension, which is used as the input of the conditional diffusion model. The input tensor can be input into the diffusion model together with a random noise tensor to generate a dam profile structure vector that meets the target performance indicators and constraint conditions.

[0028] Further, the method of using forward diffusion and reverse denoising, introducing a design parameter tensor, training the diffusion model in multiple stages based on the input data set, and generating noisy samples further comprises: The forward diffusion stage gradually adds Gaussian noise to the original dam profile data to generate noisy samples; In the reverse denoising stage, a neural network is trained to predict noise and gradually restore the original data at each step; During training, a design parameter tensor is introduced as conditional information, which is input into the diffusion model together with the noisy sample and time step, guiding the network to learn the restoration path under the condition; The conditional diffusion model training is performed by minimizing the mean square error between the model predicted noise and the real noise.

[0029] In this embodiment, with reference to Figure 3 , for the diffusion model, the forward diffusion process: the real dam profile image is regarded as the input original face slab rockfill dam profile image x0, a series of time series samples Each step follows the following Gaussian perturbation process:

[0030] In the formula: is a preset noise scheduling parameter at time step T is the total number of diffusion steps. Finally, after T steps, x T approaches the standard normal distribution. , is, the dimension is height H, width W, and channel number C; The noisy image after the t-th step diffusion process is ; wherein the noise scheduling parameter satisfies ; is a multi-source Gaussian distribution, the mean is , and the covariance matrix is ; , is an identity matrix, indicating a diagonal covariance matrix (isotropic noise).

[0031] Inverse denoising process: By training a neural network model (U-Net structure), learn the mapping function from any time step of noisy samples to the original sample . The network is denoted as:

[0032] where, is the neural network for predicting noise, the parameters are , the input includes the current noisy sample , the time step and the conditional information tensor c, i.e. the extracted dam design parameter tensor.

[0033] Conditional training mechanism: To achieve the responsiveness of the generated results to the dam design parameters, the conditional tensor c is introduced in the model training. The tensor contains normalized geometric parameters and material parameter information, corresponding to each training sample. The training goal is to minimize the mean square error (MSE) between the predicted noise and the real noise, and the loss function is defined as follows:

[0034] where, is the real noise added to x0, x t is the noisy sample generated according to the diffusion process.

[0035] The training data is composed of the original dam profile image (i.e. the target output) and the corresponding structural design parameter tensor (i.e. the conditional input) noise disturbance step number T is set to 1000 steps, which supports detailed denoising learning.

[0036] The diffusion model backbone can adopt a time-aware U-Net architecture, which includes multi-layer convolution downsampling, residual connection, time embedding module and conditional attention module. The conditional tensor is processed through the fully connected network embedding module and fused with the intermediate feature map to guide the generation results with parameter constraint information. Through repeated training and optimization, the diffusion model can finally learn to generate dam profile images that meet the mechanical laws and geometric characteristics of the dam profile images starting from any structural design parameters.

[0037] Further, the diffusion model is called to input the profile prediction data, and the noisy sample is combined to obtain the initial scheme of the face slab rockfill dam profile design of the face slab rockfill dam that meets the engineering requirements, which further includes: The diffusion model samples the initial noise image from the standard Gaussian distribution, combines the given design parameter tensor as the conditional input, and iterates after gradually denoising according to the inverse diffusion path learned in the training process; In each step of iteration, the diffusion model predicts the noise residual according to the current noise state, time step and conditional tensor, and updates the image until the entire denoising process is completed, outputting the dam profile or its structural expression corresponding to the input parameters, and performing conditional generation from random noise to structural morphology.

[0038] In this embodiment, in the prediction stage, the diffusion model first samples an initial noise image from a standard Gaussian distribution, and combines the given design parameter tensor as a conditional input, and gradually denoises according to the reverse diffusion path learned in the training process. In each step of iteration, the model predicts the noise residual according to the current noise state, time step and conditional tensor, and updates the image accordingly, until the entire denoising process is completed at step 0, and finally outputs the dam profile or its structural expression corresponding to the input parameters, realizing conditional generation from random noise to structural morphology.

[0039] Further, the evaluation and screening of the initial scheme of the face slab rockfill dam profile design according to the design target and constraint condition to obtain the optimized face slab rockfill dam profile design scheme further comprises: Before partitioning the diffusion model, the corresponding training and test data sets are constructed for model training and testing; Based on the topological optimization, the profile design data of the face slab rockfill dam is obtained, including drawings and design information texts, and the profile design information of the existing face slab rockfill dam is extracted for key features of drawing and text data, the face slab rockfill dam profile contour parameters and partition arrangement parameters in the image are counted, and the characteristic water level, seismic requirement, face slab rockfill dam stockpile physical and mechanical parameters in the text are counted to obtain the distribution law of different parameters; The design data is converted into a plurality of input tensors by a parameterization method, including a design constraint input tensor, a Gaussian noise input tensor, a feature mask tensor and a target arrangement design tensor; The tensors correspond to the constraint conditions of the design, the initial noise of the diffusion model, the regional features that need to be paid attention to, and the arrangement targets of each zone of the face slab rockfill dam profile; Referring to Figure 4 By generating the profile data, the positions and ranges of each functional partition are determined, and corresponding vectorization processing is performed to convert the geometric shape and position of each partition into vector data to obtain a complete vectorization layout scheme of the face slab rockfill dam profile partition.

[0040] Based on the training and test data sets, the diffusion model is trained and tested to generate a quality that meets the design requirements.

[0041] wherein, in this embodiment, the dam crest settlement after completion is not more than 1% of the dam height. The minimum safety factor of the rockfill dam slope anti-sliding stability is met. When the profile generated by the diffusion model meets or is better than the corresponding industry specifications and design standards in terms of earth-rock dam deformation, stress and stability, it is considered to "meet the design requirements" and can be included in the final design scheme or as part of the training data set.

[0042] The specific implementation content and example data of each link are described below.

[0043] I. Obtaining of data source The design domain of the face rockfill dam and its boundary conditions are determined. The design domain is a two-dimensional profile of the dam body, 240 m high, with fixed constraints at the bottom of the dam body, normal water storage pressure applied at the top, symmetric boundary conditions at the sides, and the foundation simulated by an elastic constraint model with an elastic modulus of 200 MPa and a Poisson's ratio of 0.3. This boundary condition accurately reflects the interaction of the dam body, foundation and water body, ensuring the authenticity and reliability of the finite element analysis. Then, the objective function and constraint conditions of the topology optimization are set. The objective function is to minimize the total volume of the dam body, and the constraint conditions include that the stress of each part of the dam body should not exceed the allowable limit of the material, and the total volume of the rockfill material is limited to not exceed the upper limit of the design volume, ensuring the safety of the structure and the economy of the material.

[0044] On this basis, the topology optimization model of the face rockfill dam is constructed. Combined with the physical and mechanical parameters of different rockfill materials and their sources (the main rockfill material has a density of 2050 kg / m 3 , an elastic modulus of 35 GPa and an internal friction angle of 42°; the transition material has a density of 2150 kg / m 3 , an elastic modulus of 50 GPa and an internal friction angle of 38°), as well as water pressure, the APDL secondary development based on ANSYS software is carried out to accurately realize the Duncan-Zhang EB constitutive model, and the specific parameters are shown in Table 1. Through iterative solution by the nonlinear finite element method, the model aims to minimize the volume and meet the constraint conditions, and finally obtains the structure optimization result, with the iteration convergence condition set to the target function change rate being less than 1%.

[0045] Table 1 Duncan-Zhang (E-B) model parameters

[0046] The topology optimization calculation result presents a reasonable profile shape of the face rockfill dam, as shown in Figure 5 . The typical structure includes a main rockfill area with a thickness of 74 to 368 meters, a transition layer with a horizontal width of 3 meters, and a cushion layer with a horizontal width of 3-6 meters. The material parameters of each area are reasonably divided according to the functional requirements to ensure the mechanical properties and safety requirements of the structure.

[0047] Based on different material parameter perturbations and diversified load combinations (different water storage levels), about 500 groups of optimized profile data are batch-generated. Each group of data includes profile contour vector diagrams, partition material parameter tables, and text descriptions of loads and boundary conditions, constituting a high-quality data source.

[0048] II. Drawing and text feature extraction For CAD vector diagrams, we use OpenCV to assist with OCR and semantic segmentation algorithms to automatically identify and extract embankment profile contour line coordinates (vertices, toe points), interface positions of each partition, and area width and inclination information. For design specification texts, we use natural language processing techniques to analyze normal and design flood levels, dry density, elastic modulus, and other physical and mechanical parameters of different dam materials. For example, the main rockfill material: dry density 2050 kg / m³, internal friction angle 42°, elastic modulus 35 MPa.

[0049] III. Design parameter conversion to tensor After standardizing the extracted numerical parameters in a predetermined order, we construct a “design constraint input tensor” [dam height, upstream and downstream slope ratio, design flood level, foundation stiffness, dry density of rockfill material, internal friction angle, cohesion, Poisson's ratio, elastic modulus]. We generate a “Gaussian noise tensor” through random sampling to enhance model diversity. We generate a “feature mask tensor” based on the partition area of interest, with elements of 0 / 1. Finally, we convert the pixel-level partition labels of the standard profile diagram into a “target design tensor”, with class labels corresponding to rockfill, transition, and impervious zones. The tensor data composed of design constraint tensor, Gaussian noise, feature mask, and target profile layout (vector) form a dataset, which is divided into a training set and a test set. The training set accounts for 80%, and the test set accounts for 20%.

[0050] IV. Profile image generation and vectorization layout After receiving the design constraint input tensor, the diffusion model undergoes multiple forward-backward diffusion iterations to generate a profile image with clear partition boundaries. Subsequently, we use boundary tracking and polygon fitting algorithms to convert image coordinates into vector data format. For example, the impervious zone can be accurately described by a sequence of vertices; the transition zone and rockfill zone can also be represented by corresponding polygons or polyline, respectively. The vectorized results can be seamlessly imported into CAD / BIM systems, facilitating subsequent design and construction integration.

[0051] V. Generation result verification and design requirement conformity Finally, we conduct strict engineering verification on the profile generated by the model. The settlement of the completed dam top should not exceed 1% of the dam height. It meets the minimum safety factor for rockfill dam slope stability against sliding.

[0052] In several embodiments provided in the present application, it should be understood that the disclosed system, device and method can be implemented in other manners. For example, the division of the units is only a logical function division, and there can be another division manner for the 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. In addition, the displayed or discussed mutual couplings or direct couplings or communication connections between different units, or the among different units, can be indirect couplings or communication connections through some interfaces, devices or units, and can be in electrical, mechanical or other forms.

[0053] In addition, each function unit in the various embodiments of the present application can be integrated into a processing unit, or each unit can exist alone physically, or two or more units can be integrated into one unit. The integrated unit can be implemented 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 content of the present application 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.

[0054] The specific embodiments of the application are described in detail above, but they are only examples. The present application is not limited to the specific embodiments described above. Any equivalent modification or substitution made by those skilled in the art to the present application is also within the scope of the present application. 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 panel rockfill dam profile intelligent design method based on topological optimization and diffusion model, characterized in that, The utility model relates to a kind of face slab rockfill dam profile design method based on diffusion model, including: Generate the face slab rockfill dam using topological optimization method, obtain the key design parameters and structural features of face slab rockfill dam profile as data source; Data preprocessing is carried out on the data source, and a high-dimensional feature space format suitable for diffusion model is constructed to form a data set; Using forward diffusion and reverse denoising method, introduce design parameter tensor, based on the input data set, the diffusion model is trained in multiple stages, and noisy samples are generated; Call the diffusion model, input profile prediction data, and obtain the initial scheme of face slab rockfill dam profile design of the face slab rockfill dam meeting engineering requirements in combination with the noisy samples; According to the design target and constraint condition, the initial scheme of face slab rockfill dam profile design is evaluated and screened to obtain the optimized face slab rockfill dam profile design scheme.

2. The intelligent design method for the profile of a face rockfill dam based on topological optimization and diffusion model according to claim 1, characterized in that, The utility model relates to a kind of face slab rockfill dam profile design method based on diffusion model, including: Determine the design domain and boundary condition, set the objective function and constraint condition, and use nonlinear finite element method to iteratively solve and generate optimization results; Obtain the profile shape of the face slab rockfill dam and the parameters of the rockfill material in each partition, and use topological optimization technology to construct the data source of the face slab rockfill dam structure.

3. The intelligent design method for the profile of a face rockfill dam based on topological optimization and diffusion model according to claim 2, characterized in that, The utility model relates to a kind of face slab rockfill dam profile design method based on diffusion model, including: Based on the physical and mechanical parameters of the rockfill material and its source of different face slab rockfill dams, self-weight load, water load, foundation constraint parameters, the corresponding topological optimization model of face slab rockfill dam is established; By using APDL secondary development on ANSYS software, the Duncan-Zhang EB constitutive model of face slab rockfill dam is realized; Based on the Duncan-Zhang EB constitutive model, the topological optimization algorithm of homogenization method is used to solve the profile shape of face slab rockfill dam under different material parameters and load combinations, and the training data set is constructed.

4. The intelligent design method for the profile of a face rockfill dam based on topological optimization and diffusion model according to claim 1, characterized in that, The utility model relates to a kind of face slab rockfill dam profile design method based on diffusion model, including: After extracting design information from the data source, processing and standardization are carried out, and the high-dimensional tensor form suitable for the model is converted as the input of the diffusion model, and the output is the height of the dam body, the slope ratio of the upstream and downstream dam slopes, and a complete data set is constructed.

5. The intelligent design method for the profile of a face rockfill dam based on topological optimization and diffusion model according to claim 1, characterized in that, The utility model relates to a kind of face slab rockfill dam profile design method based on diffusion model, including: Geometric parameters, material parameters and environmental conditions are normalized before input; After parameter standardization is completed, input tensors of each parameter are constructed according to the input structure of the diffusion model; All parameter input tensors are spliced to form a fixed-dimension global input tensor, which is used as the input of the conditional diffusion model; The global input tensor and random noise tensor are jointly input into the diffusion model to generate a dam profile structure vector that meets the target performance index and constraint condition.

6. The intelligent design method for the profile of a face rockfill dam based on topological optimization and diffusion model according to claim 1, characterized in that, The forward diffusion and reverse denoising method introduces a design parameter tensor, and the diffusion model is trained in multiple stages based on the input data set to generate noisy samples, which also includes: The forward diffusion stage gradually adds Gaussian noise to the original dam profile data to generate noisy samples; In the reverse denoising stage, a neural network is trained to predict noise and gradually restore the original data at each step; Set the time step to control the evolution accuracy during profile generation, and at each step, the denoising result of the current state is predicted by the diffusion model, and the evolution is generated through multiple time steps to generate a dam partition profile that meets the design requirements; During training, introduce a design parameter tensor as conditional information, input the noisy sample and time step into the diffusion model, and guide the network to learn the restoration path under the condition; By minimizing the mean square error between the model predicted noise and the true noise, conditional diffusion model training is performed.

7. The intelligent design method for the profile of a face rockfill dam based on topological optimization and diffusion model according to claim 1, characterized in that, The diffusion model is called to input the contour prediction data, and the initial scheme of the face slab rockfill dam profile design of the face slab rockfill dam that meets the engineering requirements is obtained by combining the noisy sample, which also includes: The diffusion model samples the initial noise image from the standard Gaussian distribution, combines the given design parameter tensor as the conditional input, and gradually denoises according to the reverse diffusion path learned during the training process, and then iterates; In each iteration, the diffusion model predicts the noise residual based on the current noise state, time step and conditional tensor, and updates the image until the entire denoising process is completed, and outputs the dam profile image or its structural expression corresponding to the input parameters, and performs conditional generation from random noise to structural form.

8. The intelligent design method for the profile of a face rockfill dam based on topological optimization and diffusion model according to claim 1, characterized in that, The initial scheme of the face slab rockfill dam profile design is evaluated and screened according to the design target and constraint condition to obtain an optimized face slab rockfill dam profile design scheme, which also includes: Before partitioning the design of the diffusion model, construct the corresponding training and test data sets for model training and testing; Based on the topological optimization, the profile design data of the face slab rockfill dam is obtained, including drawings and design information texts, the profile design information of the face slab rockfill dam obtained by topological optimization is extracted, the face slab rockfill dam profile contour parameters and partition arrangement parameters in the image are counted, and the characteristic water level, seismic requirements, and face slab rockfill dam rockfill physical and mechanical parameters in the text are counted to obtain the distribution law of different parameters; Convert the design data into multiple input tensors through the parameterization method, including design constraint input tensor, Gaussian noise input tensor, feature mask tensor and target arrangement design tensor, to form a data set; The tensors correspond to the constraint conditions of the design, the initial noise of the diffusion model, the area features that need to be paid attention to, and the arrangement targets of each zone of the face slab rockfill dam profile; Through the generated profile data, the position and range of each functional partition are determined, and corresponding vectorization processing is performed to convert the geometric shape and position of each partition into vector data to obtain a complete vectorization layout scheme of the face slab rockfill dam profile partition, Based on the training and test data sets, the diffusion model is trained and tested to generate a quality that meets the design requirements.

Images