Partition design method and device for section of concrete faced rockfill dam based on diffusion model

Abstract

The invention provides a face rockfill dam section partition design method and device based on a diffusion model, and the method comprises the steps: obtaining the section contour of a to-be-designed face rockfill dam and the volume parameters of each partition rockfill material of the to-be-designed face rockfill dam, generating a design constraint input tensor based on the profile contour and the volume parameters of the rockfill material in each partition; inputting the design constraint input tensor, the Gaussian noise input tensor and the feature mask tensor into a target diffusion model to obtain an output design tensor of each region of the profile; the target diffusion model is obtained by performing pre-training and fine tuning on the diffusion model based on a real data set and an artificial data set of a historical concrete faced rockfill dam profile design; determining arrangement designs of different partitions of the section according to the output design tensor of each area of the section, and performing vectorization correction and extraction on the arrangement designs of the different partitions of the section to obtain a target design result of each area of the section. Through the method provided by the invention, the section partition design precision and efficiency are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of deep learning, and particularly relates to a method and device for sectional zoning design of concrete face rockfill dams based on diffusion models. Background Art

[0002] Due to its superior safety, economy, and adaptability, the concrete face rockfill dam has become one of the mainstream dam types for hydropower facilities such as conventional hydropower stations and pumped storage power stations. However, traditional rockfill dam design relies on manual experience, and there is still room for improvement in design efficiency and accuracy. Moreover, existing design data has not been fully exploited.

[0003] In data-driven generative intelligent design methods such as deep learning, the scale of the dataset is crucial. A rich dataset can significantly improve the generalization ability and prediction accuracy of the model. However, due to factors such as intellectual property and confidentiality, the actual number of available concrete face rockfill dam design cases is limited, which greatly restricts the performance of intelligent design models.

[0004] How to improve the accuracy and efficiency of sectional zoning design of concrete face rockfill dams is a technical problem that needs to be solved currently. Summary of the Invention

[0005] The present invention provides a method and device for sectional zoning design of concrete face rockfill dams based on diffusion models to solve the defects existing in the prior art.

[0006] The present invention provides a method for sectional zoning design of concrete face rockfill dams based on diffusion models, including the following steps: Obtain the sectional profile of the concrete face rockfill dam to be designed and the volume parameters of the rockfill materials in each zone of the concrete face rockfill dam to be designed, and generate a design constraint input tensor based on the sectional profile and the volume parameters of the rockfill materials in each zone; Input the design constraint input tensor, Gaussian noise input tensor, and feature mask tensor into a target diffusion model to obtain an output design tensor for each section of the profile; wherein, the Gaussian noise input tensor and the feature mask tensor are constructed according to the sectional profile; the target diffusion model is obtained by pre-training and fine-tuning a diffusion model based on a real dataset and an artificial dataset of historical concrete face rockfill dam sectional designs; Determine the layout design of different sections of the profile according to the output design tensor of each section of the profile, and perform vectorization correction and extraction on the layout design of different sections of the profile to obtain the target design result of each section of the profile.

[0007] According to the method for sectional zoning design of concrete face rockfill dams based on diffusion models provided by the present invention, the generating a design constraint input tensor based on the sectional profile and the volume parameters of the rockfill materials in each zone includes: Construct a binary tensor based on the profile contour, and multiply the volume parameters of the embanked rockfill in each partition after normalization by the binary tensor of the profile contour to obtain a design parameter tensor; Stack the binary tensor of the profile contour and the design parameter tensor along the channel direction to generate the design constraint input tensor.

[0008] According to a method for sectional partition design of a concrete face rockfill dam based on a diffusion model provided by the present invention, the pre-training and fine-tuning process of the target diffusion model includes: Obtain the true design data of the historical sectional design of the concrete face rockfill dam, generate artificial design data of the historical sectional design of the concrete face rockfill dam through parametric generation based on the true design data, and convert both the true design data and the artificial design data of the historical sectional design of the concrete face rockfill dam into historical design constraint input tensors, historical Gaussian noise input tensors, historical feature mask tensors, and historical target design results of each section of the profile to obtain the true data set and the artificial data set of the historical sectional design of the concrete face rockfill dam; Pre-train the diffusion model based on the true data set and the artificial data set to obtain a basic diffusion model; Fine-tune the basic diffusion model based on the true data set through a low-rank adaptation method to obtain the target diffusion model.

[0009] According to a method for sectional partition design of a concrete face rockfill dam based on a diffusion model provided by the present invention, both the true design data and the artificial design data include: drawing data and text data; The artificial design data of the historical sectional design of the concrete face rockfill dam generated through parametric generation based on the true design data includes: Extract the key features of the drawing data of the historical sectional design of the concrete face rockfill dam to obtain the profile contour parameters of the historical concrete face rockfill dam and the layout parameters of the embankment material partitions, and determine the first distribution law of the profile contour parameters and the layout parameters of the embankment material partitions; Extract the key features of the text data of the historical sectional design of the concrete face rockfill dam to obtain the key design parameters in the text data, and determine the second distribution law of the key design parameters; Based on the first distribution law and the second distribution law, through an automatic parametric generation method, augment and expand the profile contour parameters, the layout parameters of the embankment material partitions, and the key design parameters to generate the artificial design data that meets the preset parameter range.

[0010] A method for sectional zoning design of concrete face rockfill dams based on diffusion models according to the present invention. Converting both the real design data and the artificial design data of the historical concrete face rockfill dam sectional design into historical design constraint input tensors, historical Gaussian noise input tensors, historical feature mask tensors, and historical target design results for each section of the section, to obtain the real dataset and the artificial dataset of the historical concrete face rockfill dam sectional design, includes: Extract features from the real design data and the artificial design data to obtain the historical sectional profile of the historical concrete face rockfill dam, the volume parameters of the historical rockfill materials in each zone, and the design positions and dimensions of each zone in the historical sectional profile; Construct a historical binary tensor based on the historical sectional profile, and multiply the normalized volume parameters of the historical rockfill materials in each zone by the historical binary tensor of the historical sectional profile to obtain a historical design parameter tensor; wherein, the historical design parameter tensor has the same size as the historical binary tensor of the historical sectional profile; Stack the historical binary tensor of the historical sectional profile and the historical design parameter tensor along the channel direction to generate a historical design constraint input tensor; Forward-diffuse the positions where the numbers inside the historical binary tensor are target numbers into Gaussian noise to obtain the historical Gaussian noise input tensor, and use the historical binary tensor as the historical feature mask tensor; Construct the design positions and dimensions of each zone in the historical sectional profile into sectional binary tensors respectively, and stack the sectional binary tensors of each zone along the corresponding channel direction to obtain the historical target design results for each section of the section; Based on the historical design constraint input tensor, the historical Gaussian noise input tensor, the historical feature mask tensor, and the historical target design results for each section of the section of the real design data and the artificial design data, respectively constitute the real dataset and the artificial dataset of the historical concrete face rockfill dam sectional design.

[0011] A method for sectional zoning design of concrete face rockfill dams based on diffusion models according to the present invention. Pre-training the diffusion model based on the real dataset and the artificial dataset to obtain a basic diffusion model, includes: Under the constraints of the historical design constraint input tensor and the historical feature mask tensor, train the diffusion model to predict the noise distribution at each time step, and denoise the historical Gaussian noise input tensor to obtain the historical output design tensors for each section of the section; Based on the historical target design results for each section of the section, by minimizing the difference between the model-predicted noise distribution and the real noise distribution within the range of the historical feature mask tensor at each time step, obtain the basic diffusion model.

[0012] A method for sectional zoning design of a concrete face rockfill dam based on a diffusion model. The method fine-tunes the basic diffusion model through a low-rank adaptation method based on the real dataset to obtain the target diffusion model, including: While freezing all the weights of the basic diffusion model, fine-tune and train the low-rank part of the low-rank adaptation method insertion layer based on the real dataset; wherein, the low-rank part of the low-rank adaptation method insertion layer is: the low-rank matrix corresponding to each insertion layer; After the fine-tuning training is completed, add the weights of the basic diffusion model and the low-rank matrices corresponding to each insertion layer obtained by the fine-tuning training to obtain the target diffusion model.

[0013] A method for sectional zoning design of a concrete face rockfill dam based on a diffusion model. The method determines the layout design of different sections of the section according to the output design tensor of each section of the section, and performs vectorization correction and extraction on the layout design of different sections of the section to obtain the target design result of each section of the section, including: Correspond the actual zoning at the target position within the section profile to the section with the largest predicted value at the target position in the output design tensor of each section of the section to obtain the layout design of different sections of the section; Perform vectorization correction and extraction on the layout design of different sections of the section according to the vectorization correction and extraction principle to obtain the target design result of each section of the section; wherein, the vectorization correction and extraction principle is: the upper and lower limits of the secondary rockfill and the increment model are horizontal lines, the upper limit of the drainage area is a horizontal line, the downstream slope of the secondary rockfill area is consistent with the downstream slope of the slope surface, and the top of the increment model area is consistent with the dam crest.

[0014] The present invention also provides a sectional zoning design device for a concrete face rockfill dam based on a diffusion model, including the following modules: An acquisition module, configured to acquire the section profile of the concrete face rockfill dam to be designed and the volume parameters of the rockfill materials in each zoning of the concrete face rockfill dam to be designed, and generate a design constraint input tensor based on the section profile and the volume parameters of the rockfill materials in each zoning; An output module, configured to input the design constraint input tensor, the Gaussian noise input tensor, and the feature mask tensor into the target diffusion model to obtain the output design tensor of each section of the section; wherein, the Gaussian noise input tensor and the feature mask tensor are: constructed according to the section profile; the target diffusion model is: pre-trained and fine-tuned based on the real dataset and the artificial dataset of the historical sectional design of the concrete face rockfill dam; A design module, configured to determine the layout design of different partitions of the profile according to the design tensors output by each area of the profile, and perform vectorization correction and extraction on the layout designs of different partitions of the profile to obtain the target design results of each area of the profile.

[0015] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, it implements the method for sectional partition design of a concrete face rockfill dam based on a diffusion model as described in any one of the above.

[0016] The present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the method for sectional partition design of a concrete face rockfill dam based on a diffusion model as described in any one of the above.

[0017] The present invention also provides a computer program product, including a computer program. When the computer program is executed by a processor, it implements the method for sectional partition design of a concrete face rockfill dam based on a diffusion model as described in any one of the above.

[0018] A method and device for sectional partition design of a concrete face rockfill dam based on a diffusion model provided by the present invention. By obtaining the profile contour of the concrete face rockfill dam to be designed and the volume parameters of the rockfill materials in each partition of the concrete face rockfill dam to be designed, and generating a design constraint input tensor based on the profile contour and the volume parameters of the rockfill materials in each partition; inputting the design constraint input tensor, a Gaussian noise input tensor, and a feature mask tensor into a target diffusion model to obtain design tensors output by each area of the profile; wherein, the Gaussian noise input tensor and the feature mask tensor are constructed according to the profile contour; the target diffusion model is obtained by pre-training and fine-tuning a diffusion model based on a real dataset and an artificial dataset of historical sectional designs of concrete face rockfill dams; determining the layout design of different partitions of the profile according to the design tensors output by each area of the profile, and performing vectorization correction and extraction on the layout designs of different partitions of the profile to obtain the target design results of each area of the profile. It can be seen from this that the present invention pre-trains and fine-tunes a diffusion model based on a real dataset and an artificial dataset of historical sectional designs of concrete face rockfill dams to obtain a target diffusion model. The full excavation of data makes the model have high accuracy, and the intelligent generation of the sectional partition design of the concrete face rockfill dam is realized based on the target diffusion model, improving the efficiency of the sectional partition design. Description of the Drawings

[0019] To more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0020] Figure 1 It is a schematic flow chart of the method for sectional zoning design of a concrete face rockfill dam based on a diffusion model provided by the present invention.

[0021] Figure 2 It is a schematic diagram of the training and application of the target diffusion model provided by the present invention.

[0022] Figure 3 It is a schematic flow chart of the construction process of the design constraint input tensor provided by the present invention.

[0023] Figure 4 It is a schematic diagram of the parameters defining the sectional design image of a concrete face rockfill dam provided by the present invention.

[0024] Figure 5 It is a specific architecture diagram of the target diffusion model provided by the present invention.

[0025] Figure 6 It is a schematic flow chart of the correction process of the design result of the target diffusion model provided by the present invention.

[0026] Figure 7 It is a schematic structural diagram of the device for sectional zoning design of a concrete face rockfill dam based on a diffusion model provided by the present invention.

[0027] Figure 8 It is a schematic structural diagram of the electronic device provided by the present invention. Detailed implementation manners

[0028] To make the objectives, technical solutions, and advantages of the present invention clearer, the following will clearly and completely describe the technical solutions in the present invention in conjunction with the drawings in the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art without creative efforts based on the embodiments in the present invention belong to the scope of protection of the present invention.

[0029] The following will describe Figures 1 - 8 a method and device for sectional zoning design of a concrete face rockfill dam based on a diffusion model of the present invention.

[0030] Figure 1 It is a schematic flow chart of the method for sectional zoning design of a concrete face rockfill dam based on a diffusion model provided by the present invention, as shown in Figure 1As shown, the method includes the following: Step 100: Obtain the profile contour of the panel rockfill dam to be designed and the volume parameters of the rockfill materials in each zone of the panel rockfill dam to be designed, and generate a design constraint input tensor based on the profile contour and the volume parameters of the rockfill materials in each zone.

[0031] It should be noted that in data-driven generative intelligent design methods such as deep learning, the scale of the dataset is crucial. A rich dataset can significantly improve the generalization ability and prediction accuracy of the model. However, the number of actual available panel rockfill dam design cases is limited, which greatly restricts the performance of the intelligent design model. Therefore, further expanding more datasets that meet the design specifications is the key to improving the effect of generative intelligent design.

[0032] Specifically, the design objects of the panel rockfill dam profile zoning mainly include the design positions and dimensions of the main rockfill, secondary rockfill, drainage, and additional model zones in the panel rockfill dam profile contour.

[0033] Generating the design constraint input tensor based on the profile contour and the volume parameters of the rockfill materials in each zone in Step 100 includes: Step 110: Construct a binary tensor based on the profile contour, multiply the normalized volume parameters of the rockfill materials in each zone with the binary tensor of the profile contour respectively to obtain a design parameter tensor.

[0034] Step 120: Stack the binary tensor of the profile contour and the design parameter tensor along the channel direction to generate the design constraint input tensor.

[0035] Figure 2 It is a schematic diagram of the training and application of the target diffusion model provided by the present invention. As Figure 2 shown, obtain the profile contour of the panel rockfill dam to be designed and the volume parameters of the rockfill materials in each zone; construct the profile contour into a binary tensor; then multiply the normalized volume parameters of the rockfill materials in each zone with the binary tensor of the profile contour respectively to obtain a design parameter tensor; stack the contour tensor and the design parameter tensor along the channel direction to form a design constraint input tensor.

[0036] In one embodiment, Figure 3 It is a schematic diagram of the construction process of the design constraint input tensor provided by the present invention. The following combines Figure 3 , and describes the construction process of the design constraint input tensor provided in this embodiment.

[0037] 1. Extract the profile contour of the real CAD panel rockfill dam and the Excel statistical data of the rockfill material parameters based on the Python library, and store them as profile contour vectorized data and the volume parameters of the rockfill materials in each zone.

[0038] In one embodiment, the volume parameters of the rockfill materials in each zone mainly include the volumes of the rockfill materials in the main rockfill zone, secondary rockfill zone, additional model zone, and drainage zone.

[0039] 2. Characterize the profile contour information in the profile contour vectorized data as a second-order tensor.

[0040] In one embodiment, the profile contour vectorized data is scaled to a profile contour matrix of size 512×256×1 at the image scaling ratio. Here, the image scaling ratio is the ratio of the actual size of the rockfill dam scaled to the pixel map size.

[0041] In yet another embodiment, the second-order tensor is used to characterize whether a certain position in the profile contour matrix is inside the contour. If it is inside the contour, the value of the second-order tensor at this position is 1, otherwise it is 0, as shown in the following formula: In the formula, M cnt is the contour mask matrix, m ij is the value of each point in the matrix.

[0042] 3. Divide the volume of the rockfill materials in each zone by the total volume of the rockfill materials respectively for parameter normalization, so as to obtain the characteristic scalars of each zone; multiply the characteristic scalars of each zone by the profile contour binary tensor respectively, and then the design parameter tensors of each zone with the same size as the profile contour binary tensor can be obtained.

[0043] In one embodiment, directly multiply the characteristic scalars of each zone by the profile contour binary tensor, and then the design parameter tensors of each zone with the same size as the profile contour binary tensor can be obtained. The construction method is shown in the following formula: In the formula, M k are different characteristic matrices, which are the main rockfill zone ( M mra ), secondary rockfill zone ( M srz ), drainage zone ( M drz ), additional model zone ( M miz ), and the scaling ratio ( M sr ), are the normalized scalar values of different characteristics.

[0044] 4. Stack the contour tensor and the design parameter tensor along the channel direction to form a design constraint input tensor.

[0045] Step 200: Input the design constraint input tensor, Gaussian noise input tensor, and feature mask tensor into the target diffusion model to obtain the output design tensor for each area of the profile. Among them, the Gaussian noise input tensor and the feature mask tensor are constructed according to the profile contour. The target diffusion model is obtained by pre-training and fine-tuning the diffusion model based on the real dataset and artificial dataset of historical panel rockfill dam profile designs.

[0046] In this embodiment, the positions where the internal numbers of the profile contour binary tensor are 1 are accumulated with small Gaussian noises T times through a Markov process, and the forward diffusion is Gaussian noise: In the formula, is a set of hyperparameters used to control the intensity of the added noise, representing the variance of the Gaussian distribution. The larger it is, the greater the added noise, and the more random the generated samples are, which helps the model explore the data distribution more comprehensively. In the embodiment, is obtained by linear interpolation and gradually decreases from 0.02 to 0.0001. represents the Gaussian distribution, where z is a random variable. is the mean value. represents the variance. represents the conditional probability distribution of the current moment given the past moment .

[0047] Through the reparameterization trick and the additivity of independent Gaussian distributions, the at any time can be directly derived. At the same time, to make the characteristics of the diffusion process concentrated in the key areas, a mask tensor is introduced to enhance the feature density: In the formula, is the input mask tensor, which is numerically equivalent to the profile contour binary tensor, that is, ; . represents a multi-dimensional Gaussian distribution with a mean of 0 and a covariance matrix of the identity matrix.

[0048] Continue to refer to Figure 2 to illustrate the pre-training and fine-tuning process of the target diffusion model provided in this embodiment.

[0049] In this embodiment, the training of the target diffusion model is a two-stage training mode, specifically including: In the first stage, training data is generated parametrically, and a basic diffusion model is jointly pre-trained using a real dataset and a large amount of artificially generated data generated parametrically; in the second stage, low-rank adaptation technology is introduced, and the basic diffusion model is fine-tuned using a real rockfill dam profile design dataset. Finally, the hyperparameters of the diffusion model are tuned by calculating the intersection and union ratio of the output design results of each section of the test dataset and the target design results of each section of the concrete face rockfill dam profile, so as to obtain the optimal target diffusion model.

[0050] Specifically, the pre-training and fine-tuning process of the target diffusion model includes: Step 210: Obtain the real design data of the historical concrete face rockfill dam profile design. Based on the real design data, artificially generated design data of the historical concrete face rockfill dam profile is generated parametrically. Both the real design data and the artificially generated design data of the historical concrete face rockfill dam profile are converted into historical design constraint input tensors, historical Gaussian noise input tensors, historical feature mask tensors, and historical target design results of each section of the profile, so as to obtain the real dataset and the artificial dataset of the historical concrete face rockfill dam profile.

[0051] It should be noted that both the real design data and the artificially generated design data include: drawing data and text data.

[0052] The artificially generated design data of the historical concrete face rockfill dam profile generated parametrically based on the real design data in step 210 includes: Step 211: Extract the key features of the drawing data of the historical concrete face rockfill dam profile design to obtain the profile contour parameters of the historical concrete face rockfill dam and the layout parameters of the rockfill zones, and determine the first distribution law of the profile contour parameters and the layout parameters of the rockfill zones.

[0053] Step 212: Extract the key features of the text data of the historical concrete face rockfill dam profile design to obtain the key design parameters in the text data, and determine the second distribution law of the key design parameters.

[0054] Specifically, key features of the drawing and text data of the real design data of the historical concrete face rockfill dam profile design are extracted, the profile contour parameters of the historical concrete face rockfill dam and the layout parameters of the rockfill zones in the image are counted, and the key design parameters in the text are counted; the distribution laws of different parameters are obtained.

[0055] Step 213: Based on the first distribution law and the second distribution law, through an automatic parametric generation method, the profile contour parameters, the layout parameters of the rockfill zones, and the key design parameters are augmented and expanded to generate the artificially generated design data that meets the preset parameter range.

[0056] In this embodiment, by statistically analyzing and summarizing a large amount of real data, the parameters that make up the historical rockfill dam profile design image are determined. Then, through the statistical analysis of existing design data and the investigation of engineers, the ranges of relevant parameters are clarified. Figure 4 It is a schematic diagram of the parameters that define the profile design image of the concrete face rockfill dam provided by the present invention. The defined parameters are as Figure 4 shown, and the corresponding relevant parameter symbols and ranges are shown in Table 1.

[0057] Table 1

[0058] Finally, based on the statistical distribution law of the parameters, an automatic parametric generation method is adopted to augment and expand the profile contour parameters, the layout parameters of the fill zones, and the key design parameters of the historical concrete face rockfill dam, and a large amount of artificial design data that meet the parameter ranges is generated.

[0059] Furthermore, in step 210, both the real design data and the artificial design data of the historical concrete face rockfill dam profile design are converted into a historical design constraint input tensor, a historical Gaussian noise input tensor, a historical feature mask tensor, and the historical target design results of each zone of the profile, so as to obtain the real data set and the artificial data set of the historical concrete face rockfill dam profile design, including: Step 214: Extract features from the real design data and the artificial design data to obtain the historical profile contour of the historical concrete face rockfill dam, the volume parameters of the historical fill materials in each zone, and the design positions and dimensions of each zone in the historical profile contour.

[0060] Specifically, extract features from the real design data and the artificial design data to obtain the historical profile contour of the historical concrete face rockfill dam, the volume parameters of the historical fill materials in each zone, and the design positions and dimensions of each zone in the historical profile contour.

[0061] Step 215: Construct a historical binary tensor based on the historical profile contour, and multiply the normalized volume parameters of the historical fill materials in each zone by the historical binary tensor of the historical profile contour to obtain a historical design parameter tensor; wherein, the historical design parameter tensor has the same size as the historical binary tensor of the historical profile contour.

[0062] Specifically, a historical binary tensor is constructed based on the historical profile contour. In this tensor, the numbers inside the contour are 1, and the numbers outside the contour are 0. The historical volumes of rockfill in each partition are divided by the total volume of rockfill respectively for parameter normalization, so as to obtain the characteristic scalars of each partition. The characteristic scalars of each partition are multiplied by the historical binary tensor of the historical profile contour respectively, and then the historical design parameter tensors of each partition with the same size as the historical binary tensor of the historical profile contour can be obtained. The historical design parameter tensors of each partition are stacked along the channel direction to obtain the historical design parameter tensor.

[0063] Step 216: Stack the historical binary tensor of the historical profile contour and the historical design parameter tensor along the channel direction to generate a historical design constraint input tensor.

[0064] Step 217: Forward-diffuse the positions where the numbers inside the historical binary tensor are the target numbers into Gaussian noise to obtain the historical Gaussian noise input tensor, and use the historical binary tensor as the historical feature mask tensor.

[0065] Specifically, forward-diffuse the positions where the numbers inside the historical binary tensor of the historical profile contour are 1 into Gaussian noise to obtain the historical Gaussian noise input tensor; the historical feature mask tensor is numerically equivalent to the historical binary tensor of the historical profile contour.

[0066] Step 218: Construct the design positions and sizes of each partition in the historical profile contour into partition binary tensors respectively, and stack the partition binary tensors of each partition along the corresponding channel direction to obtain the historical target design results of each section of the profile.

[0067] Specifically, construct the design positions and sizes of each partition in the historical profile contour into partition binary tensors respectively. Specifically, in the binary tensor of a certain partition, the number at the design position of this partition is 1, and the rest of the numbers are 0. Stack the binary tensors of all partitions along their channel direction to obtain the historical target design results of each section of the profile.

[0068] In actual application, the design data is converted into the above-mentioned Gaussian noise input tensor, feature mask tensor and design constraint input tensor and jointly input into the trained target diffusion model.

[0069] Step 219: Based on the historical design constraint input tensor, the historical Gaussian noise input tensor, the historical feature mask tensor and the historical target design results of each section of the profile of the real design data and the artificial design data, respectively constitute the real dataset and the artificial dataset for the historical panel rockfill dam profile design.

[0070] Specifically, the historical design constraint input tensors, historical Gaussian noise input tensors, historical feature mask tensors, and historical target design results of each section of the profile of the real design data and the artificial design data respectively constitute the real dataset and the artificial dataset for the panel rock-fill dam profile design.

[0071] Step 220: Pre-train the diffusion model based on the real dataset and the artificial dataset to obtain a basic diffusion model.

[0072] Step 220 specifically includes: Step 221: Under the constraints of the historical design constraint input tensor and the historical feature mask tensor, train the diffusion model to predict the noise distribution at each time step, and denoise the historical Gaussian noise input tensor to obtain the historical output design tensors of each section of the profile.

[0073] Specifically, under the constraints of the historical design constraint input tensor and the historical feature mask tensor, train the diffusion model to predict the noise distribution at each time step, so as to gradually denoise the historical Gaussian noise input tensor in the dataset, and finally obtain the historical output design tensors of each section of the profile.

[0074] Step 222: Based on the historical target design results of each section of the profile, obtain the basic diffusion model by minimizing the difference between the model-predicted noise distribution and the real noise distribution within the range of the historical feature mask tensor at each time step.

[0075] Specifically, based on the historical target design results of each section of the profile, guide the optimization of the network parameters by minimizing the difference between the model-predicted noise distribution and the real noise distribution within the range of the historical feature mask tensor at each time step, so as to pre-train and obtain the basic diffusion model.

[0076] Step 230: Fine-tune the basic diffusion model based on the real dataset by using the low-rank adaptation method to obtain the target diffusion model.

[0077] Step 230 specifically includes: Step 231: With all the weights of the basic diffusion model frozen, fine-tune and train the low-rank part of the insertion layer of the low-rank adaptation method based on the real dataset; where the low-rank part of the insertion layer of the low-rank adaptation method is: the low-rank matrices corresponding to each insertion layer.

[0078] Specifically, freeze all the weights of the pre-trained basic diffusion model; based on the real dataset, fine-tune and train the low-rank part of the insertion layer of the low-rank adaptation technique according to the aforementioned pre-training method to adapt to the data designed by real engineers.

[0079] It should be noted that in this embodiment, a basic diffusion model is constructed based on the U-Net architecture and the low-rank adaptation technology is introduced. By predicting the actual noise as accurately as possible at each time step, the Gaussian noise input tensor is gradually denoised to obtain the output design tensor for each section of the profile. The low-rank adaptation technology refers to decomposing the attention layer in the basic diffusion model into low-rank matrices and changing the weights of the model by only training the low-rank part. Figure 5 is the specific architecture diagram of the target diffusion model provided by the present invention. For the specific architecture of the target diffusion model, reference can be made to Figure 5 , where the training method of the basic diffusion model is as follows: According to the Bayes formula, the distribution of reverse denoising of the Gaussian noise input tensor based on the design constraint input tensor can be obtained: In this embodiment, under the constraints of the design constraint input tensors and feature mask tensors of each real data and artificial data, a structure that combines the U-Net with the self-attention mechanism is used to predict Gaussian noise, so as to gradually denoise the Gaussian noise input tensor in the data set, and finally obtain the output design tensor for each section of the profile.

[0080] At the same time, based on the target design results of each section of the profile, by maximizing the log-likelihood of the model prediction distribution, the network parameters are optimized, so as to pre-train the basic diffusion model.

[0081] Specifically, the U-Net architecture includes: linear layer 1, convolutional layer 1, convolutional layer 2, convolutional layer 3, convolutional layer 4, convolutional layer 5, transposed convolutional layer 1, transposed convolutional layer 2, transposed convolutional layer 3, transposed convolutional layer 4, transposed convolutional layer 5, and linear layer 2. Among them, the resolution remains unchanged when passing through convolutional layer 1, convolutional layer 5, transposed convolutional layer 1, and transposed convolutional layer 5; the resolution is reduced by 1 / 2 when passing through other convolutional layers and transposed convolutional layers; among them, convolutional layer 4, convolutional layer 5, transposed convolutional layer 1, and transposed convolutional layer 2 include self-attention mechanisms.

[0082] The low-rank adaptation technology refers to inserting trainable layers in linear layer 1, convolutional layer 4, convolutional layer 5, transposed convolutional layer 1, transposed convolutional layer 2, and linear layer 2, that is, decomposing the weight matrix W update amount of each inserted layer into the product of two low-rank matrices A and B; the low-rank matrices A and B are trainable.

[0083] After splitting, the network parameters of the low-rank part of the inserted layer are . During the second-stage training, all the weights of the basic diffusion model pre-trained in the first stage are frozen Based on the real dataset of the panel rockfill dam profile design, fine-tune the training according to the aforementioned training method , to adapt to the data designed by real engineers. It should be noted that during the second-stage training, the parameters of the network are .

[0084] Step 232. When the fine-tuning training is completed, add the weights of the basic diffusion model to the low-rank matrices corresponding to each inserted layer obtained by the fine-tuning training to obtain the target diffusion model.

[0085] Specifically, after the fine-tuning training is completed, add the weights of the trained basic diffusion model to the low-rank matrices A and B corresponding to each inserted layer obtained by the fine-tuning, and the final target diffusion model can be obtained.

[0086] Through model test evaluation, the optimal target diffusion model is obtained, including: 1) Obtain real design cases that do not overlap with the real dataset to form a test dataset.

[0087] 2) Process the cases in the test dataset into Gaussian noise input tensors, feature mask tensors, design constraint input tensors, and target design results for each section of the profile according to the aforementioned dataset construction method.

[0088] 3) Input the Gaussian noise input tensor, feature mask tensor, and design constraint input tensor in the test set into diffusion models with different hyperparameters, output the corresponding design tensors for each section, and correct them according to the aforementioned method to obtain the output design results for each section of the profile.

[0089] 4) Calculate the intersection over union of the output design results for each section of the profile and the target design results for each section of the profile, and select the diffusion model composed of the hyperparameter combination with the highest intersection over union in the test set as the optimal target diffusion model.

[0090] In this embodiment, the intersection over union measures the overlapping degree of the contour generated by the diffusion model and the contour designed by the engineer by calculating the overlap rate between the output design results for each section of the profile and the target design results for each section of the profile (i.e., the ratio of their intersection to their union). The higher the intersection over union value, the better the model performance. Specifically, compare the coordinates of the secondary embankment, drainage, and additional model areas in the output design results for each section of the profile with the target design results for each section of the profile to calculate the corresponding intersection over union, and take the weighted average intersection over union as the evaluation index. It should be noted that since the main embankment is equal to the outer contour minus the secondary embankment, drainage, and additional model areas, the intersection over union of the main embankment is no longer considered in the calculation of the weighted intersection over union: In the formula, is the weighted average intersection over union, is the weight of the i-th element, is the i Intersection over Union (IoU) of the -th element, and i are the -th element contours generated by the diffusion model and designed by the engineer respectively, and

[0091] Step 300: Determine the layout design of different sections of the profile according to the output design tensor of each section of the profile, and perform vectorization correction and extraction on the layout design of different sections of the profile to obtain the target design results of each section of the profile.

[0092] Step 300 specifically includes: Step 310: Corresponding the actual section of the target position within the profile contour to the section with the largest predicted value at the target position of the output design tensor of each section of the profile to obtain the layout design of different sections of the profile.

[0093] Step 320: Perform vectorization correction and extraction on the layout design of different sections of the profile according to the vectorization correction and extraction principle to obtain the target design results of each section of the profile; wherein, the vectorization correction and extraction principle is: the upper and lower boundaries of the secondary embankment and the increasing module are horizontal lines, the upper boundary of the drainage area is a horizontal line, the downstream slope of the secondary embankment area is consistent with the downstream slope of the slope surface, and the top of the increasing module area is consistent with the dam crest.

[0094] Specifically, the actual section at a certain position within the profile contour corresponds to the section with the largest predicted value of the output design tensors of each section, and the layout design results of different sections of the profile are obtained after extraction and processing according to this principle.

[0095] Use the contour detection algorithm to outline the effective contours of different section layouts. By setting the convex hull morphology, ensure that the extracted contours are as convex as possible and the parameter points are relatively sparse. During this process, perform morphological operations on the local contour mask through the cv2.getStructuringElement and cv2.morphologyEx functions to remove noise, and call the cv2.findContours and cv2.contourArea functions to find the contour with the largest area as the local contour. Subsequently, find the maximum and minimum y coordinates of the contour as the calibration basis.

[0096] Finally, according to the following correction principles, the constraint points intersecting the outer contour are searched for the secondary heap, the incremental model, and the drainage area in sequence, and the design result of the diffusion model is corrected. Finally, the vectorized layout design result of the rockfill dam profile zoning is obtained: 1) The upper and lower boundaries of the secondary heap, the incremental model, and the drainage area need to be horizontal lines; 2) The downstream slope of the secondary heap area is consistent with the downstream slope of the slope; 3) The top of the incremental model area is consistent with the dam crest. Figure 6 It is a schematic diagram of the correction process of the target diffusion model design result provided by the present invention. The specific process can be referred to Figure 6 .

[0097] The above is the step description of the panel rockfill dam profile zoning design method based on the diffusion model provided by the present invention. It can be seen from the description of the above steps that according to the panel rockfill dam profile zoning design method based on the diffusion model provided by the present invention, by obtaining the profile contour of the panel rockfill dam to be designed and the volume parameters of the rockfill materials in each zoning of the panel rockfill dam to be designed, and generating a design constraint input tensor based on the profile contour and the volume parameters of the rockfill materials in each zoning; inputting the design constraint input tensor, the Gaussian noise input tensor, and the feature mask tensor into the target diffusion model to obtain the output design tensors of each section of the profile; wherein, the Gaussian noise input tensor and the feature mask tensor are: constructed according to the profile contour; the target diffusion model is: obtained by pre-training and fine-tuning the diffusion model based on the real dataset and the artificial dataset of the historical panel rockfill dam profile design; determining the layout design of different sections of the profile according to the output design tensors of each section of the profile, and performing vectorization correction and extraction on the layout design of different sections of the profile to obtain the target design results of each section of the profile. It can be seen that the present invention pre-trains and fine-tunes the diffusion model based on the real dataset and the artificial dataset of the historical panel rockfill dam profile design to obtain the target diffusion model. The full mining of data makes the model have high accuracy. Based on the target diffusion model, the intelligent generation of the panel rockfill dam profile zoning design is realized, and the profile zoning design efficiency is improved.

[0098] Next, the panel rockfill dam profile zoning design device based on the diffusion model provided by the present invention will be described. The panel rockfill dam profile zoning design device described below can be mutually referred to the panel rockfill dam profile zoning design method described above.

[0099] Figure 7 It is a schematic diagram of the structure of the panel rockfill dam profile zoning design device based on the diffusion model provided by the present invention. As Figure 7 shown, the panel rockfill dam profile zoning design device based on the diffusion model provided by the present invention includes: An acquisition module 701, configured to acquire the profile contour of the panel rockfill dam to be designed and the volume parameters of the rockfill materials in each partition of the panel rockfill dam to be designed, and generate a design constraint input tensor based on the profile contour and the volume parameters of the rockfill materials in each partition; An output module 702, configured to input the design constraint input tensor, the Gaussian noise input tensor, and the feature mask tensor into a target diffusion model to obtain an output design tensor for each section of the profile; wherein, the Gaussian noise input tensor and the feature mask tensor are: constructed according to the profile contour; the target diffusion model is: pre-trained and fine-tuned based on the real dataset and the artificial dataset of the historical panel rockfill dam profile design; A design module 703, configured to determine the layout design of different partitions of the profile according to the output design tensor of each section of the profile, and perform vector quantization correction and extraction on the layout design of different partitions of the profile to obtain the target design result of each section of the profile.

[0100] The device for sectional partition design of a panel rockfill dam based on a diffusion model provided by the present invention generates a design constraint input tensor by acquiring the profile contour of the panel rockfill dam to be designed and the volume parameters of the rockfill materials in each partition of the panel rockfill dam to be designed, and based on the profile contour and the volume parameters of the rockfill materials in each partition; inputs the design constraint input tensor, the Gaussian noise input tensor, and the feature mask tensor into a target diffusion model to obtain an output design tensor for each section of the profile; wherein, the Gaussian noise input tensor and the feature mask tensor are: constructed according to the profile contour; the target diffusion model is: pre-trained and fine-tuned based on the real dataset and the artificial dataset of the historical panel rockfill dam profile design; determines the layout design of different partitions of the profile according to the output design tensor of each section of the profile, and performs vector quantization correction and extraction on the layout design of different partitions of the profile to obtain the target design result of each section of the profile. It can be seen that the present invention pre-trains and fine-tunes a diffusion model based on the real dataset and the artificial dataset of the historical panel rockfill dam profile design to obtain a target diffusion model. The full mining of data makes the model have high accuracy, and the intelligent generation of the sectional partition design of the panel rockfill dam is realized based on the target diffusion model, improving the efficiency of the sectional partition design.

[0101] Based on the above embodiments, in this embodiment, the device further includes a generation module, specifically configured to: Construct a binary tensor based on the profile contour, and multiply the normalized volume parameters of the rockfill materials in each partition by the binary tensor of the profile contour to obtain a design parameter tensor; Stack the binary tensor of the profile contour and the design parameter tensor along the channel direction to generate the design constraint input tensor.

[0102] Based on the above embodiments, in this embodiment, the device further includes a training module, specifically for: Obtain the true design data of the historical panel rockfill dam profile design, generate the artificial design data of the historical panel rockfill dam profile design through parametric generation based on the true design data, and convert both the true design data and the artificial design data of the historical panel rockfill dam profile design into historical design constraint input tensors, historical Gaussian noise input tensors, historical feature mask tensors, and historical target design results for each section area, so as to obtain the true dataset and the artificial dataset of the historical panel rockfill dam profile design; Pre-train the diffusion model based on the true dataset and the artificial dataset to obtain a basic diffusion model; Fine-tune the basic diffusion model based on the true dataset through the low-rank adaptation method to obtain the target diffusion model.

[0103] Based on the above embodiments, in this embodiment, both the true design data and the artificial design data include: drawing data and text data; The training module is specifically for: Extract the key features of the drawing data of the historical panel rockfill dam profile design to obtain the profile contour parameters of the historical panel rockfill dam and the layout parameters of the filling material partitions, and determine the first distribution law of the profile contour parameters and the layout parameters of the filling material partitions; Extract the key features of the text data of the historical panel rockfill dam profile design to obtain the key design parameters in the text data, and determine the second distribution law of the key design parameters; Based on the first distribution law and the second distribution law, through an automatic parametric generation method, augment and expand the profile contour parameters, the layout parameters of the filling material partitions, and the key design parameters to generate the artificial design data that meets the preset parameter range.

[0104] Based on the above embodiments, in this embodiment, the training module is specifically for: Extract features from the true design data and the artificial design data to obtain the historical profile contour of the historical panel rockfill dam, the volume parameters of the historical filling materials in each partition, and the design positions and sizes of each partition in the historical profile contour; Construct a historical binary tensor based on the historical profile contour, and multiply the normalized volume parameters of the historical filling materials in each partition by the historical binary tensor of the historical profile contour to obtain a historical design parameter tensor; wherein, the historical design parameter tensor has the same size as the historical binary tensor of the historical profile contour; Stack the historical binary tensor of the historical profile contour and the historical design parameter tensor along the channel direction to generate a historical design constraint input tensor; Forward diffuse the positions where the numbers inside the historical binary tensor are the target numbers into Gaussian noise to obtain the historical Gaussian noise input tensor, and use the historical binary tensor as the historical feature mask tensor; Construct the design positions and sizes of each partition in the historical profile contour into partition binary tensors respectively, and stack the partition binary tensors of each partition along the corresponding channel direction to obtain the historical target design results of each section of the profile; Based on the historical design constraint input tensor, the historical Gaussian noise input tensor, the historical feature mask tensor, and the historical target design results of each section of the profile of the real design data and the artificial design data, respectively constitute the real dataset and the artificial dataset for the historical panel rockfill dam profile design.

[0105] Based on the above embodiments, in this embodiment, the training module is specifically used for: Under the constraints of the historical design constraint input tensor and the historical feature mask tensor, train the diffusion model to predict the noise distribution at each time step, denoise the historical Gaussian noise input tensor, and obtain the historical output design tensors of each section of the profile; Based on the historical target design results of each section of the profile, by minimizing the difference between the model-predicted noise distribution and the real noise distribution within the range of the historical feature mask tensor at each time step, obtain the basic diffusion model.

[0106] Based on the above embodiments, in this embodiment, the training module is specifically used for: With all the weights of the basic diffusion model frozen, fine-tune and train the low-rank part of the low-rank adaptation method insertion layer based on the real dataset; where the low-rank part of the low-rank adaptation method insertion layer is: the low-rank matrices corresponding to each insertion layer; After the fine-tuning training is completed, add the weights of the basic diffusion model and the low-rank matrices corresponding to each insertion layer obtained by the fine-tuning training to obtain the target diffusion model.

[0107] Based on the above embodiments, in this embodiment, the design module 703 is specifically used for: Correspond the actual partition at the target position in the profile contour to the partition with the largest predicted value at the target position in the output design tensors of each section of the profile to obtain the layout design of different partitions of the profile; Perform vectorization correction and extraction on the layout designs of different zones of the section according to the vectorization correction and extraction principle to obtain the target design results of each zone of the section; wherein, the vectorization correction and extraction principle is: the upper and lower limits of the secondary heap and the additional mold are horizontal lines, the upper limit of the drainage zone is a horizontal line, the downstream slope of the secondary heap zone is the same as the downstream slope of the slope surface, and the top of the additional mold zone is the same as the dam crest.

[0108] Figure 8 An example of a schematic diagram of the physical structure of an electronic device is shown as Figure 8 shown. The electronic device can be a robot or other electronic device. The electronic device can include: a processor 810, a communications interface 820, a memory 830, and a communication bus 840. Among them, the processor 810, the communications interface 820, and the memory 830 complete mutual communication through the communication bus 840. The processor 810 can call the logical instructions in the memory 830 to execute a method for designing the section partition of a concrete face rockfill dam based on a diffusion model, including: Obtain the section profile of the concrete face rockfill dam to be designed and the volume parameters of the rockfill materials in each zone of the concrete face rockfill dam to be designed, and generate a design constraint input tensor based on the section profile and the volume parameters of the rockfill materials in each zone; Input the design constraint input tensor, the Gaussian noise input tensor, and the feature mask tensor into the target diffusion model to obtain the output design tensors of each zone of the section; wherein, the Gaussian noise input tensor and the feature mask tensor are constructed according to the section profile; the target diffusion model is obtained by pre-training and fine-tuning the diffusion model based on the real dataset and the artificial dataset of the historical concrete face rockfill dam section design; Determine the layout designs of different zones of the section according to the output design tensors of each zone of the section, and perform vectorization correction and extraction on the layout designs of different zones of the section to obtain the target design results of each zone of the section.

[0109] In addition, when the logical instructions in the above-mentioned memory 830 are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0110] On the other hand, the present invention also provides a computer program product. The computer program product includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the panel rockfill dam profile zoning design method based on the diffusion model provided by the above-mentioned various methods, including: Obtain the profile contour of the panel rockfill dam to be designed and the volume parameters of the rockfill materials in each zone of the panel rockfill dam to be designed, and generate a design constraint input tensor based on the profile contour and the volume parameters of the rockfill materials in each zone; Input the design constraint input tensor, the Gaussian noise input tensor, and the feature mask tensor into the target diffusion model to obtain the output design tensors for each zone of the profile; wherein, the Gaussian noise input tensor and the feature mask tensor are: constructed according to the profile contour; the target diffusion model is: obtained by pre-training and fine-tuning the diffusion model based on the real dataset and the artificial dataset of the historical panel rockfill dam profile design; Determine the layout design of different zones of the profile according to the output design tensors for each zone of the profile, and perform vector quantization correction and extraction on the layout design of different zones of the profile to obtain the target design results for each zone of the profile.

[0111] On another aspect, the present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it is implemented to execute the panel rockfill dam profile zoning design method provided by the above-mentioned various methods, including: Obtain the profile contour of the panel rockfill dam to be designed and the volume parameters of the rockfill materials in each zone of the panel rockfill dam to be designed, and generate a design constraint input tensor based on the profile contour and the volume parameters of the rockfill materials in each zone; Input the design constraint input tensor, Gaussian noise input tensor, and feature mask tensor into the target diffusion model to obtain the output design tensor for each section of the profile; wherein, the Gaussian noise input tensor and the feature mask tensor are constructed according to the profile contour; the target diffusion model is obtained by pre-training and fine-tuning the diffusion model based on the real dataset and artificial dataset of the historical panel rockfill dam profile design; Determine the layout design of different sections of the profile according to the output design tensor of each section of the profile, and perform vectorization correction and extraction on the layout design of different sections of the profile to obtain the target design results of each section of the profile.

[0112] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative work.

[0113] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence, or the part that contributes to the prior art can be embodied in the form of a software product, which can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions for causing a computer device (which can be a personal computer, server, or network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0114] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for designing the section zoning of a concrete face rockfill dam based on a diffusion model, characterized in that: include: Acquire a cross-sectional profile of the face rockfill dam to be designed and a volume parameter of rockfill materials in each partition of the face rockfill dam to be designed, and generate a design constraint input tensor based on the cross-sectional profile and the volume parameter of rockfill materials in each partition; The design constraint input tensor, Gaussian noise input tensor and feature mask tensor are input into the target diffusion model to obtain the output design tensor of each section; wherein the Gaussian noise input tensor and the feature mask tensor are obtained by constructing the profile of the section; the target diffusion model is obtained by pre-training and fine-tuning the diffusion model based on the real data set and artificial data set of the historical panel rockfill dam profile design; The layout design of different sections of the section is determined according to the design tensors outputted from each section, and the layout design of different sections of the section is vectorized and corrected and extracted to obtain target design results of each section.

2. The method for designing the section zoning of a concrete face rockfill dam based on a diffusion model according to claim 1 is characterized in that: The generating of the design constraint input tensor based on the cross-sectional profile and the volume parameters of the rockfill materials in each partition comprises: A binary tensor is obtained based on the cross-sectional profile, and the normalized volume parameters of the rockfill materials in each partition are multiplied by the binary tensor of the cross-sectional profile to obtain a design parameter tensor; The binary tensor of the cross-sectional profile and the design parameter tensor are stacked along the channel direction to generate the design constraint input tensor.

3. The method for designing the section zoning of a concrete face rockfill dam based on a diffusion model according to claim 1 is characterized in that: The pre-training and fine-tuning process of the target diffusion model includes: Acquire real design data of historical panel rockfill dam profile design, generate artificial design data of the historical panel rockfill dam profile design through parameterization based on the real design data, convert both the real design data and the artificial design data of the historical panel rockfill dam profile design into historical design constraint input tensors, historical Gaussian noise input tensors, historical feature mask tensors and historical target design results of each section of the profile, so as to obtain real data sets and artificial data sets of the historical panel rockfill dam profile design; Pre-training the diffusion model based on the real data set and the artificial data set to obtain a basic diffusion model; The basic diffusion model is fine-tuned based on the real data set through a low-rank adaptation method to obtain the target diffusion model.

4. The method for designing the section zoning of a concrete face rockfill dam based on a diffusion model according to claim 3 is characterized in that: The real design data and the artificial design data both include: drawing data and text data; The artificial design data of the historical panel rockfill dam profile design is obtained by parameterized generation based on the real design data, including: Extracting key features of the drawing data of the historical panel rockfill dam cross-section design, obtaining the cross-section profile parameters and the layout parameters of the stockpile partitions of the historical panel rockfill dam, and determining a first distribution law of the cross-section profile parameters and the layout parameters of the stockpile partitions; Extracting key features of the text data of the historical panel rockfill dam profile design to obtain key design parameters in the text data, and determining a second distribution law of the key design parameters; Based on the first distribution law and the second distribution law, the cross-section profile parameters, the layout parameters of the stockpile partitions and the key design parameters are augmented and expanded through an automatic parameter generation method to generate the artificial design data that meets the preset parameter range.

5. The method for designing the section zoning of a concrete face rockfill dam based on a diffusion model according to claim 3 is characterized in that: The real design data and the artificial design data of the historical panel rockfill dam profile design are converted into a historical design constraint input tensor, a historical Gaussian noise input tensor, a historical feature mask tensor and historical target design results of each section of the profile to obtain a real data set and an artificial data set of the historical panel rockfill dam profile design, including: Performing feature extraction on the real design data and the artificial design data to obtain the historical cross-sectional profile of the historical panel rockfill dam, the volume parameters of historical rockfill materials in each sub-area, and the design position and size of each sub-area in the historical cross-sectional profile; Based on the historical profile, a historical binary tensor is constructed, and normalized volume parameters of historical rockfill materials in each partition are multiplied by the historical binary tensor of the historical profile to obtain a historical design parameter tensor; wherein the historical design parameter tensor has the same size as the historical binary tensor of the historical profile; Stacking the historical binary tensor of the historical profile and the historical design parameter tensor along the channel direction to generate a historical design constraint input tensor; Forward diffuse the position where the internal number of the historical binary tensor is the target number into Gaussian noise to obtain the historical Gaussian noise input tensor, and use the historical binary tensor as the historical feature mask tensor; The design position and size of each partition in the historical profile are respectively constructed as partition binary tensors, and the partition binary tensors of each partition are stacked along the corresponding channel direction to obtain the historical target design results of each zone of the profile; The historical design constraint input tensor, the historical Gaussian noise input tensor, the historical feature mask tensor and the historical target design results of each section of the profile based on the real design data and the artificial design data respectively constitute the real data set and artificial data set of the historical panel rockfill dam profile design.

6. The method for designing the section zoning of a concrete face rockfill dam based on a diffusion model according to claim 3 is characterized in that: The pre-training of the diffusion model based on the real data set and the artificial data set to obtain a basic diffusion model includes: Under the constraints of the historical design constraint input tensor and the historical feature mask tensor, the diffusion model is trained to predict the noise distribution of each time step, and the historical Gaussian noise input tensor is denoised to obtain the historical output design tensor of each section area; Based on the historical target design results of each area of ​​the profile, the basic diffusion model is obtained by minimizing the difference between the model predicted noise distribution and the real noise distribution within the range of the historical feature mask tensor in each time step.

7. The method for designing the section zoning of a concrete face rockfill dam based on a diffusion model according to claim 3 is characterized in that: The step of fine-tuning the basic diffusion model by a low-rank adaptation method based on the real data set to obtain the target diffusion model includes: Under the condition of freezing all weights of the basic diffusion model, fine-tuning and training the low-rank part of the low-rank adaptation method insertion layer based on the real data set; wherein the low-rank part of the low-rank adaptation method insertion layer is: the low-rank matrix corresponding to each insertion layer; When the fine-tuning training is completed, the weight of the basic diffusion model is added to the low-rank matrix corresponding to each insertion layer obtained by the fine-tuning training to obtain the target diffusion model.

8. The method for designing the section zoning of a concrete face rockfill dam based on a diffusion model according to claim 1 is characterized in that: The step of determining the layout design of different sections of the section according to the design tensors outputted from each section, and performing vector correction and extraction on the layout design of different sections of the section to obtain target design results of each section of the section includes: The actual partition of the target position within the cross-section contour corresponds to the partition with the largest predicted value of the design tensor output at the target position in each zone of the cross-section, so as to obtain the layout design of different partitions of the cross-section; According to the vector correction and extraction principles, the layout design of different partitions of the section is vectorized and extracted to obtain the target design results of each area of ​​the section; wherein, the vector correction and extraction principles are: the upper and lower limits of the secondary pile and the increased mold are horizontal lines, the upper limit of the drainage area is the horizontal line, the downstream slope of the secondary pile area is consistent with the slope of the downstream slope, and the top of the increased mold area is consistent with the dam top.

9. A device for designing the section zoning of a concrete face rockfill dam based on a diffusion model, characterized in that: include: An acquisition module, used for acquiring a cross-sectional profile of the face rockfill dam to be designed and a volume parameter of rockfill materials in each partition of the face rockfill dam to be designed, and generating a design constraint input tensor based on the cross-sectional profile and the volume parameter of rockfill materials in each partition; An output module is used to input the design constraint input tensor, Gaussian noise input tensor and feature mask tensor into the target diffusion model to obtain the output design tensor of each section; wherein the Gaussian noise input tensor and the feature mask tensor are obtained by constructing the profile of the section; the target diffusion model is obtained by pre-training and fine-tuning the diffusion model based on the real data set and artificial data set of the historical panel rockfill dam profile design; The design module is used to determine the layout design of different sections of the section according to the design tensors output by each section area, and to perform vector correction and extraction on the layout design of different sections of the section to obtain the target design results of each section area.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the method for designing the section zoning of a concrete face rockfill dam based on a diffusion model as claimed in any one of claims 1 to 8 is implemented.

Images