Architectural design method and components based on diffusion model generation-optimization-relearning

Abstract

The present invention provides a building design method and component based on diffusion model generation-optimization-relearning, the method comprising: determining an initial building structure design data set, so as to generate and train a diffusion model according to the initial building structure design data set; generating a structure layout generation scheme based on a generative building structure design network of the diffusion model; optimizing the structure layout generation scheme based on a preset optimization algorithm to obtain an optimized structure layout scheme, and determining a relearned building structure design data set according to the optimized structure layout scheme; relearning and training the diffusion model according to the relearned building structure design data set, and repeatedly iterating to form a diffusion model generation-optimization-relearning cycle of the building structure design scheme, thereby continuously improving the total amount of learning data and the design quality of the diffusion model, and improving the capture and generalization performance of the generative building structure design network based on the diffusion model on underlying mechanics and design experience.

Description

Technical Field

[0001] The present invention relates to the field of building structure design and deep learning technology, and in particular to a building design method and component based on diffusion model generation-optimization-relearning. Background Art

[0002] As a labor-intensive industry, the traditional construction industry has high data barriers and low automation in drawing design, making it difficult to fully utilize and combine drawings and materials from other projects. In addition, since engineers need to optimize the scheme as much as possible while meeting the design specifications, the huge number of optimization alternatives means that engineers can only make judgments based on their own design experience, which makes it difficult to obtain the optimal design scheme. Therefore, intelligent design methods for building structures will be a research direction with broad prospects. The number of building structure design data sets available in the existing intelligent design of building structures is limited, which seriously affects the effect of data-driven generative intelligent design. It is still necessary to further expand high-quality data sets that conform to the rules of structural mechanics characteristics and design experience rules. Summary of the invention

[0003] The present invention provides a method and component for building design based on diffusion model generation-optimization-relearning, which is used to solve the defect that the number of available building structure design data sets in the prior art is limited, which seriously affects the effect of data-driven generative intelligent design. The present invention repeatedly iterates the rapid generation, optimization and relearning of building structure design schemes based on a generative building structure design network based on a diffusion model, a preset optimization algorithm and a relearned building structure design data set, thereby continuously improving the total amount of learning data and design quality of the diffusion model, and improving the capture and generalization performance of the generative building structure design network based on the diffusion model on the underlying mechanics and design experience.

[0004] The present invention provides a building design method based on diffusion model generation-optimization-relearning, comprising: determining an initial building structure design data set, so as to generate and train a diffusion model according to the initial building structure design data set; the initial building structure design data set comprises initial building design data and initial structural design data; generating a structure layout generation scheme based on a generative building structure design network of the diffusion model; optimizing the structure layout generation scheme based on a preset optimization algorithm to obtain an optimized structure layout scheme, so as to determine a relearned building structure design data set according to the optimized structure layout scheme; the relearned building structure design data set is a collection of the optimized structure layout scheme and corresponding building design data; relearning and training the diffusion model according to the relearned building structure design data set, and repeatedly and iteratively executing the generation step, the optimization step and the relearning step according to the diffusion model after the relearning training, until the relearned building structure design data set meets the preset data set requirements.

[0005] According to a building design method based on diffusion model generation-optimization-relearning provided by the present invention, the initial building design data includes the initial building design tensor and the initial mask tensor; the initial structural design data is an initial shear wall tensor; the initial building structure design data set is determined, and the diffusion model is generated and trained according to the initial building structure design data set, including: mapping the initial building drawings, initial structural drawings and initial building parameters to the feature space to obtain the initial shear wall tensor, the initial building design tensor and the initial mask tensor, so as to construct the initial building structure design data set according to the initial shear wall tensor, the initial building design tensor and the initial mask tensor; the initial building design tensor and the initial mask tensor are input into the diffusion model, and the generation training of the diffusion model is guided according to the initial shear wall tensor.

[0006] According to a building design method based on diffusion model generation-optimization-relearning provided by the present invention, the initial building drawings, initial structural drawings and initial building parameters are mapped to the feature space to obtain the initial shear wall tensor, the initial building design tensor and the initial mask tensor, including: extracting features from the initial building drawings to obtain an initial component position feature tensor; extracting features from the initial structural drawings to obtain the initial shear wall tensor; determining the initial building parameter feature tensor according to the initial building parameters; splicing the initial component position feature tensor and the initial building parameter feature tensor to obtain the initial building design tensor; and determining the initial mask tensor according to the initial component position feature tensor.

[0007] According to a diffusion model-based generation-optimization-relearning architectural design method provided by the present invention, the generative architectural structure design network based on the diffusion model generates a structural layout generation scheme, including: modifying the initial architectural parameters to obtain input architectural parameters; mapping the initial architectural drawings and the input architectural parameters to a feature space to obtain an input architectural design tensor and an input mask tensor; inputting the input architectural design tensor and the input mask tensor into the generative architectural structure design network based on the diffusion model to obtain a structural layout feature tensor output by the network; and determining the structural layout generation scheme according to the structural layout feature tensor.

[0008] According to a diffusion model-based generation-optimization-relearning architectural design method provided by the present invention, the architectural design tensor and the mask tensor are input into the generative architectural structure design network based on the diffusion model to obtain the structural layout feature tensor output by the network, including: based on the mask tensor, forward diffusing the architectural design tensor position into Gaussian noise; and applying the network to perform reverse denoising to obtain the structural layout feature tensor.

[0009] According to a building design method based on diffusion model generation-optimization-relearning provided by the present invention, the structural layout generation scheme is optimized based on a preset optimization algorithm to obtain an optimized structural layout scheme, including: extracting component information from the structural layout generation scheme to obtain component position feature data; determining a structural analysis model based on the component position feature data; based on the structural analysis model, under the constraints of empirical rules and mechanical indicators, applying the preset optimization algorithm to optimize the structural layout generation scheme with the goal of minimizing material costs to obtain the optimized structural layout scheme; the preset optimization algorithms include gradient descent method, genetic algorithm, simulated annealing algorithm, particle swarm optimization algorithm and online optimization algorithm.

[0010] According to a diffusion model-based generation-optimization-relearning architectural design method provided by the present invention, the relearned architectural structure design data set is determined according to the optimized structural layout plan, including: determining the optimized structural drawings according to the optimized structural layout plan; mapping the optimized structural drawings and the corresponding architectural drawings and architectural parameters to the feature space to obtain a relearned shear wall tensor, a relearned architectural design tensor and a relearned mask tensor, so as to construct the relearned architectural structure design data set according to the relearned shear wall tensor, the relearned architectural design tensor and the relearned mask tensor.

[0011] The present invention also provides a building design system based on diffusion model generation-optimization-relearning, comprising: an initial data set determination module, used to determine an initial building structure design data set, so as to generate and train a diffusion model according to the initial building structure design data set; the initial building structure design data set includes initial building design data and initial structural design data; a generation scheme determination module, used to generate a structure layout generation scheme based on a generative building structure design network of a diffusion model; an optimization module, used to optimize the structure layout generation scheme based on a preset optimization algorithm to obtain an optimized structure layout scheme, so as to determine a relearned building structure design data set according to the optimized structure layout scheme; the relearned building structure design data set is a collection of the optimized structure layout scheme and corresponding building design data; a relearning module, used to relearn and train the diffusion model according to the relearned building structure design data set, and repeatedly iteratively execute the generation step, the optimization step and the relearning step according to the diffusion model after the relearning training, until the relearned building structure design data set meets the preset data set requirements.

[0012] The present invention also provides 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 program, the architectural design method based on diffusion model generation-optimization-relearning as described in any one of the above-mentioned methods is implemented.

[0013] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the above-described diffusion model-based generation-optimization-relearning architectural design methods.

[0014] The present invention provides a method and component for building design based on diffusion model generation-optimization-relearning. The method comprises: determining an initial building structure design data set, so as to generate and train a diffusion model according to the initial building structure design data set; generating a structure layout generation scheme based on a generative building structure design network of the diffusion model; optimizing the structure layout generation scheme based on a preset optimization algorithm to obtain an optimized structure layout scheme, so as to determine a relearned building structure design data set according to the optimized structure layout scheme; relearning and training the diffusion model according to the relearned building structure design data set, and repeatedly iterating to form a diffusion model generation-optimization-relearning cycle of the building structure design scheme, thereby continuously improving the total amount of learning data and the design quality of the diffusion model, and improving the capture and generalization performance of the generative building structure design network based on the diffusion model on underlying mechanics and design experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0016] Figure 1 It is a flow chart of a building design method based on diffusion model generation-optimization-relearning provided by the present invention;

[0017] Figure 2 It is a schematic diagram of the principle of an architectural design method based on diffusion model generation-optimization-relearning provided by the present invention;

[0018] Figure 3 It is a schematic diagram of the principle of determining a building structure design data set provided by the present invention;

[0019] Figure 4 It is a schematic diagram of the principle of optimizing the structural arrangement generation scheme provided by the present invention;

[0020] Figure 5 It is a structural schematic diagram of an architectural design system based on diffusion model generation-optimization-relearning provided by the present invention;

[0021] Figure 6 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0022] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0023] Please refer to Figure 1 , Figure 1 A schematic flow chart of an architectural design method based on diffusion model generation-optimization-relearning provided by the present invention.

[0024] Please refer to Figure 2 , Figure 2 A schematic diagram of the principle of an architectural design method based on diffusion model generation-optimization-relearning provided by the present invention.

[0025] Please refer to Figure 3 , Figure 3 A schematic diagram of the principle of determining a building structure design data set provided by the present invention.

[0026] The present invention provides a building design method based on diffusion model generation-optimization-relearning, comprising:

[0027] 101: Determine an initial building structure design data set to generate training for a diffusion model according to the initial building structure design data set; the initial building structure design data set includes initial building design data and initial structure design data.

[0028] As a preferred embodiment, the initial building design data includes an initial building design tensor and an initial mask tensor; the initial structural design data is an initial shear wall tensor; an initial building structure design data set is determined, and a diffusion model is generated and trained according to the initial building structure design data set, including: mapping initial building drawings, initial structural drawings and initial building parameters to a feature space to obtain an initial shear wall tensor, an initial building design tensor and an initial mask tensor, so as to construct an initial building structure design data set according to the initial shear wall tensor, the initial building design tensor and the initial mask tensor; the initial building design tensor and the initial mask tensor are input into the diffusion model, and the generation training of the diffusion model is guided according to the initial shear wall tensor.

[0029] As a preferred embodiment, initial architectural drawings, initial structural drawings and initial architectural parameters are mapped to feature space to obtain an initial shear wall tensor, an initial architectural design tensor and an initial mask tensor, including: extracting features from the initial architectural drawings to obtain an initial component position feature tensor; extracting features from the initial structural drawings to obtain an initial shear wall tensor; determining an initial architectural parameter feature tensor based on the initial architectural parameters; concatenating the initial component position feature tensor and the initial architectural parameter feature tensor to obtain an initial architectural design tensor; and determining an initial mask tensor based on the initial component position feature tensor.

[0030] Considering the limited number of available building structure design data sets in existing building structure intelligent design, which seriously affects the effect of data-driven generative intelligent design, it is still necessary to further expand high-quality data sets that conform to the structural mechanics characteristics and design experience rules. In addition, in existing structural intelligent design, the data preprocessing method is usually to convert architectural drawings into RGB images. However, this method actually introduces erroneous prior information that different components have different distances in RGB space. And when relying solely on deep learning networks for structural intelligent design, there are still defects in the design of some key locations.

[0031] In order to solve the technical problems existing in the prior art, in this embodiment, when comparing the network model for generating the building structure layout, the diffusion model is introduced in combination with the mathematical model and the architectural design principle, so that the generative neural network for building structure design can quickly generate high-quality structural design solutions while considering the building parameters. In the construction stage of the generative building structure design network based on the diffusion model, an encoder is constructed to map the initial building drawings, initial structural drawings and initial building parameters to the feature space to obtain the initial shear wall tensor, the initial building design tensor and the initial mask tensor. The initial building design tensor and the initial mask tensor are input into the generative building structure design network based on the diffusion model, and the generation training of the diffusion model is guided according to the initial shear wall tensor (true label), and a decoder is constructed to reversely map the network output to the color space to obtain the structural layout solution.

[0032] In this embodiment, the objects of the building structure design method mainly include: frame structure, shear wall structure, frame shear wall structure, and the structural design objects mainly include: shear wall, beam, column and supporting member layout design. The shear wall structure has excellent lateral resistance and overall rigidity, and performs well against earthquakes and wind loads.

[0033] In this embodiment, the building parameters are design conditions that characterize the influence on the building structure design. The building parameters may include, but are not limited to, building structure appearance parameters, building structure seismic parameters, and building wind resistance design parameters.

[0034] In this embodiment, the building structure drawings are drawings that represent the building structure design, and the building structure drawings mainly include location information of building walls, doors and windows, shear walls, etc.

[0035] The specific process of mapping the initial architectural drawings, initial structural drawings and initial architectural parameters into the feature space to obtain the initial shear wall tensor, initial architectural design tensor and initial mask tensor is as follows:

[0036] First, the encoder extracts the position coordinates of the building components and structural components in the building CAD drawings and stores them as component position feature data. In this embodiment, the building components are building walls, doors and windows; the structural components are shear walls.

[0037] Then, the encoder characterizes the position information of each building component and structural component in the component position feature data as a component position feature tensor, such as a second-order tensor (two-dimensional abstract feature tensor). In this embodiment, the second-order tensor is used to characterize whether a corresponding component is arranged at a certain position in the layout diagram. If the corresponding component is arranged, the value of the second-order tensor at the position is 1, otherwise it is 0.

[0038] Then, the encoder processes the building parameters into a first-order tensor, performs a copy and dimension expansion operation after normalization, generates a third-order tensor, and forms a building parameter feature tensor. In this embodiment, the category data in the building parameters is converted into a first-order tensor through unique hot encoding, and the numerical data is converted into a zero-order tensor. Subsequently, all types of data are integrated into a first-order tensor, and a copy and dimension expansion operation is performed after normalization to generate a third-order tensor to form a building parameter feature tensor.

[0039] Finally, the encoder represents the second-order tensors (component position feature tensors) of building walls, doors, and windows and the building parameter feature tensors as a stacked third-order tensor, that is, the feature tensors are concatenated in the channel dimension to obtain the input feature tensor of the building structure generation network; the second-order tensor of the union of building walls and shear walls is represented as a third-order tensor, that is, the spatial position information of the key target area is extracted and mapped to a multidimensional abstract feature tensor with the same dimension as the feature tensor resolution to obtain a mask tensor; the second-order tensor of the shear wall is expanded to a third-order tensor to obtain the true label of the building structure generation network to guide model training.

[0040] The specific process of inputting the initial building design tensor and the initial mask tensor into the diffusion model and guiding the generation and training of the diffusion model according to the initial shear wall tensor is as follows:

[0041] First, based on the initial mask tensor, the position of the initial building design tensor is forward diffused into Gaussian noise through a Markov process (based on the mask tensor, the corresponding position of the initial building design tensor is treated as Gaussian noise).

[0042] In this embodiment, the initial building design tensor x 0 ~q(x) performs T times of small-amplitude Gaussian noise accumulation to form a noise sequence x 1 , x 2 , …, x T :

[0043]

[0044] Among them, β t To control the size of the added noise, a series of hyperparameters characterizing the variance of the Gaussian distribution, β t The larger the value, the more noise is added, and the generated samples are more random, which helps the model to explore the data distribution more comprehensively. t Linear interpolation from 0.02 to 0.0001. represents a Gaussian distribution, where z is a random variable and μ θ is the mean, represents the variance. q(x t |x t-1 ) means that at a given past time x t-1Under the condition that the current time x t The conditional probability distribution of .

[0045] By using the reparameterization technique and the additivity of independent Gaussian distributions, we can get x at any time t At the same time, in order to concentrate the diffusion process features in the key area, the mask tensor is introduced to improve the feature density:

[0046]

[0047] Among them, x mask is the input mask tensor, Represents a multidimensional Gaussian distribution with mean 0 and covariance matrix equal to the identity matrix.

[0048] Then, the generative building structure design network based on the diffusion model is trained to learn the reverse denoising process (gradually eliminating noise) to obtain the denoised sample sequence, that is, the structural layout feature tensor. The model is trained by calculating the loss value of the generated tensor and the target tensor (initial shear wall tensor), so that the diffusion process features of the generative building structure design network based on the diffusion model are concentrated in the key target area. The feature concentration in the key target area is achieved by adding a mask.

[0049] The reversed distribution can be obtained by Bayes' formula:

[0050]

[0051] In this embodiment, the deep learning network (generative architectural structure design based on diffusion model) adopts the structure of U-Net fusion self-attention mechanism to predict Gaussian noise, and trains the network by maximizing the log-likelihood of the model prediction distribution:

[0052]

[0053] Optimize the neural network parameters M based on the above loss function θ , and a generative building structure design network based on the diffusion model is obtained. Through this network, noise prediction can be achieved, thereby obtaining the structural layout feature tensor.

[0054] 102: Generative building structure design network based on diffusion model to generate structural layout generation schemes.

[0055] As a preferred embodiment, a generative building structure design network based on a diffusion model generates a structural layout generation plan, including: modifying initial building parameters to obtain input building parameters; mapping the initial building drawings and the input building parameters to a feature space to obtain an input building design tensor and an input mask tensor; inputting the input building design tensor and the input mask tensor into a generative building structure design network based on a diffusion model to obtain a structural layout feature tensor output by the network; and determining a structural layout generation plan according to the structural layout feature tensor.

[0056] As a preferred embodiment, the architectural design tensor and the mask tensor are input into a generative architectural structure design network based on a diffusion model to obtain a structural layout feature tensor output by the network, including: based on the mask tensor, forward diffusing the architectural design tensor position into Gaussian noise; and applying the network to perform reverse denoising to obtain a structural layout feature tensor.

[0057] The generative building structure design network based on the diffusion model is a convolutional neural network built on the theoretical basis of the diffusion model. It uses the mathematical diffusion process and its inverse process to train a noise prediction network, and realizes the generation of building structure layout design for the input building plan through forward diffusion noise and reverse denoising process.

[0058] The existing initial architectural drawings are combined with different architectural parameters (for example, changing the design parameters of the building structure's earthquake resistance and wind resistance), and the encoder is used to map the initial architectural drawings and the input architectural parameters to the feature space to obtain different input architectural design tensors and input mask tensors.

[0059] In this embodiment, the building structure appearance parameters, building structure earthquake resistance parameters and building wind resistance design parameters corresponding to the building structure drawings can be changed to form different input data.

[0060] The input architectural design tensor and the input mask tensor are input into the generative architectural structure design network based on the diffusion model to obtain the structural layout feature tensor output by the network. The output structural layout feature tensor is then reversely mapped to the RGB color space by the decoder to obtain the structural layout generation scheme.

[0061] Please refer to Figure 4 , Figure 4 A schematic diagram of the principle of optimizing the structural arrangement generation scheme provided by the present invention.

[0062] 103: Optimize the structure layout generation scheme based on a preset optimization algorithm to obtain an optimized structure layout scheme, and determine a re-learned building structure design data set according to the optimized structure layout scheme; the re-learned building structure design data set is a collection of the optimized structure layout scheme and the corresponding building design data.

[0063] As a preferred embodiment, the structure layout generation scheme is optimized based on a preset optimization algorithm to obtain an optimized structure layout scheme, including: extracting component information from the structure layout generation scheme to obtain component position feature data; determining a structure analysis model based on the component position feature data; based on the structure analysis model, under the constraints of empirical rules and mechanical indicators, applying a preset optimization algorithm to optimize the structure layout generation scheme with the goal of minimizing material costs to obtain an optimized structure layout scheme; the preset optimization algorithms include gradient descent method, genetic algorithm, simulated annealing algorithm, particle swarm optimization algorithm and online optimization algorithm.

[0064] In this embodiment, the specific process of optimizing the structure arrangement generation scheme based on the preset optimization algorithm to obtain the optimized structure arrangement scheme is as follows:

[0065] First, the position information of each building component and structural component in the structural layout generation scheme is extracted and stored as component position feature data. The component position feature data is converted into a structural analysis model using the application program interface (API) of the structural analysis software. In this embodiment, the component position feature data is stored in a vector data file; the structural analysis software is PKPM software.

[0066] Then, based on the structural analysis software and structural analysis model, under the constraints of structural design experience rules and quantitative indicators of mechanical properties, the optimization algorithm is applied to optimize the generated structural layout plan with the goal of minimizing material cost, and the optimized structural layout plan is obtained.

[0067] In this embodiment, the structural design empirical rules may be, but are not limited to, structural topology rules, component pattern rules, and component size rules.

[0068] Structural topology rules include: shear walls need to be arranged at the corners of the building; shear walls cannot be set outside the open balcony space; shear walls are distributed all around the elevator; when the shear wall is too long (the threshold in this study is set to 8m, as required by the Technical Code for Concrete Structures of High-rise Buildings JGJ 3-2010), it should be interrupted at a suitable position in the middle.

[0069] Component pattern rules, for example: avoid the straight shear wall pattern; give priority to the shear wall structure layout patterns based on L-shape and T-shape, as well as C-shape, Z-shape and other patterns composed of multiple L-shapes or T-shapes; when the straight shear wall pattern cannot be avoided, an end column structure should be added at the end.

[0070] Component size rules, for example: avoid short-leg shear walls, that is, the cross-sectional aspect ratio should be greater than 4 (required by "Technical Code for Concrete Structures of High-rise Buildings JGJ 3-2010"). Usually the shear wall thickness is 200mm, so the shear wall length needs to be greater than 800mm.

[0071] In this embodiment, the quantitative indexes of mechanical properties may include but are not limited to the displacement angle between structural layers and the natural vibration period of the structure; the minimization of material cost may include but is not limited to the minimization of the amount of concrete and steel used; the optimization object of the structural layout generation scheme may include but is not limited to the size and layout position of the shear wall.

[0072] The preset optimization algorithm may be, but is not limited to, a variety of optimization algorithms such as a gradient descent method, a genetic algorithm, a simulated annealing algorithm, a particle swarm optimization algorithm, and an online optimization algorithm.

[0073] 104: Relearn and train the diffusion model according to the relearned building structure design data set, and iteratively execute the generation step, the optimization step and the relearning step according to the relearned diffusion model until the relearned building structure design data set meets the preset data set requirements.

[0074] As a preferred embodiment, a re-learned building structure design data set is determined according to the optimized structural layout plan, including: determining the optimized structural drawings according to the optimized structural layout plan; mapping the optimized structural drawings and the corresponding architectural drawings and architectural parameters to the feature space to obtain a re-learned shear wall tensor, a re-learned building design tensor and a re-learned mask tensor, so as to construct a re-learned building structure design data set according to the re-learned shear wall tensor, the re-learned building design tensor and the re-learned mask tensor.

[0075] In this embodiment, an encoder is used to encode the optimized structural layout scheme into a relearned building design tensor, a relearned shear wall tensor and a relearned mask tensor, and the relearned mask tensor is input as new training data into a generative building structure design network based on a diffusion model. The network is further trained and iterated to form a diffusion model generation-optimization-relearning cycle of the building structure design scheme until the relearned building structure design data set meets the preset data set requirements, thereby enriching the available data for model learning noise prediction, continuously improving the total amount of learning data and design quality of the diffusion model, and improving the capture and generalization performance of the generative building structure design network based on the diffusion model on the underlying mechanics and design experience.

[0076] The following is a description of the architectural design system based on diffusion model generation-optimization-relearning provided by the present invention. The architectural design system based on diffusion model generation-optimization-relearning described below and the architectural design method based on diffusion model generation-optimization-relearning described above can be referenced to each other.

[0077] Please refer to Figure 5 , Figure 5 A schematic diagram of the structure of an architectural design system based on diffusion model generation-optimization-relearning provided by the present invention.

[0078] The present invention also provides a building design system based on diffusion model generation-optimization-relearning, comprising: an initial data set determination module 1, used to determine an initial building structure design data set, so as to generate and train a diffusion model according to the initial building structure design data set; the initial building structure design data set includes initial building design data and initial structural design data; a generation scheme determination module 2, used to generate a structure layout generation scheme based on a generative building structure design network of the diffusion model; an optimization module 3, used to optimize the structure layout generation scheme based on a preset optimization algorithm to obtain an optimized structure layout scheme, so as to determine a relearned building structure design data set according to the optimized structure layout scheme; the relearned building structure design data set is a collection of the optimized structure layout scheme and the corresponding building design data; a relearning module 4, used to relearn and train the diffusion model according to the relearned building structure design data set, and repeatedly iteratively execute the generation step, the optimization step and the relearning step according to the diffusion model after the relearning training, until the relearned building structure design data set meets the preset data set requirements.

[0079] Figure 6 An example of a structural diagram of an electronic device is shown in FIG. Figure 6As shown, the electronic device may include: a processor (processor) 601, a communication interface (Communications Interface) 602, a memory (memory) 603 and a communication bus 604, wherein the processor 601, the communication interface 602, and the memory 603 communicate with each other through the communication bus 604. The processor 601 can call the logic instructions in the memory 603 to execute the architectural design method based on the diffusion model generation-optimization-relearning, which includes: determining an initial architectural structure design data set to generate and train the diffusion model according to the initial architectural structure design data set; the initial architectural structure design data set includes initial architectural design data and initial structural design data; generating a structural layout generation scheme based on a generative architectural structure design network of the diffusion model; optimizing the structural layout generation scheme based on a preset optimization algorithm to obtain an optimized structural layout scheme, and determining a relearned architectural structure design data set according to the optimized structural layout scheme; the relearned architectural structure design data set is a set of the optimized structural layout scheme and the corresponding architectural design data; relearning and training the diffusion model according to the relearned architectural structure design data set, and repeatedly iteratively executing the generation step, the optimization step and the relearning step according to the diffusion model after the relearning training, until the relearned architectural structure design data set meets the preset data set requirements.

[0080] In addition, the logic instructions in the above-mentioned memory 603 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0081] On the other hand, the present invention also provides a computer program product, which includes a computer program. The computer program 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 architectural design method based on the diffusion model generation-optimization-relearning provided by the above methods, the method including: determining an initial architectural structure design data set to generate and train the diffusion model according to the initial architectural structure design data set; the initial architectural structure design data set includes initial architectural design data and initial structural design data; generating a structural layout generation scheme based on a generative architectural structure design network of the diffusion model; optimizing the structural layout generation scheme based on a preset optimization algorithm to obtain an optimized structural layout scheme, and determining a relearned architectural structure design data set according to the optimized structural layout scheme; the relearned architectural structure design data set is a set of optimized structural layout schemes and corresponding architectural design data; relearning and training the diffusion model according to the relearned architectural structure design data set, and repeatedly iteratively executing the generation step, the optimization step and the relearning step according to the diffusion model after the relearning training until the relearned architectural structure design data set meets the preset data set requirements.

[0082] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the architectural design method based on diffusion model generation-optimization-relearning provided by the above-mentioned methods, the method comprising: determining an initial architectural structure design data set to generate and train a diffusion model according to the initial architectural structure design data set; the initial architectural structure design data set comprises initial architectural design data and initial structural design data; generating a structural layout generation scheme based on a generative architectural structure design network of the diffusion model; optimizing the structural layout generation scheme based on a preset optimization algorithm to obtain an optimized structural layout scheme, and determining a relearned architectural structure design data set according to the optimized structural layout scheme; the relearned architectural structure design data set is a set of optimized structural layout schemes and corresponding architectural design data; relearning and training the diffusion model according to the relearned architectural structure design data set, and repeatedly and iteratively executing the generation step, the optimization step and the relearning step according to the diffusion model after the relearning training, until the relearned architectural structure design data set meets the preset data set requirements.

[0083] The device embodiments described above are merely illustrative, wherein 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, i.e., they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Those of ordinary skill in the art may understand and implement it without creative effort.

[0084] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method 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 solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0085] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for architectural design based on diffusion model generation-optimization-relearning, characterized in that: include: Determining an initial building structure design data set to generate training for a diffusion model based on the initial building structure design data set; The initial building structure design data set includes initial building design data and initial structure design data; A generative building structure design network based on a diffusion model to generate structural layout solutions; Optimizing the structure arrangement generation scheme based on a preset optimization algorithm to obtain an optimized structure arrangement scheme, and determining a re-learned building structure design data set according to the optimized structure arrangement scheme; The relearned building structure design data set is a collection of the optimized structural layout scheme and the corresponding building design data; Retraining the diffusion model according to the relearned building structure design data set, and iteratively executing the generating step, the optimizing step and the relearning step according to the relearned diffusion model until the relearned building structure design data set meets the preset data set requirement; The initial building design data includes an initial building design tensor and an initial mask tensor; The initial structural design data is the initial shear wall tensor; The step of determining an initial building structure design data set and generating and training a diffusion model according to the initial building structure design data set includes: Mapping the initial building drawings, initial structural drawings and initial building parameters to the feature space to obtain the initial shear wall tensor, the initial building design tensor and the initial mask tensor, so as to construct the initial building structure design data set according to the initial shear wall tensor, the initial building design tensor and the initial mask tensor; inputting the initial building design tensor and the initial mask tensor into the diffusion model, and guiding the generation training of the diffusion model according to the initial shear wall tensor; The generative building structure design network based on the diffusion model generates a structural layout generation scheme, including: Modifying the initial building parameters to obtain input building parameters; Mapping the initial architectural drawing and the input architectural parameters to a feature space to obtain an input architectural design tensor and an input mask tensor; Inputting the input building design tensor and the input mask tensor into the generative building structure design network based on the diffusion model to obtain a structural arrangement feature tensor output by the network; Determining the structural arrangement generation scheme according to the structural arrangement feature tensor; The step of optimizing the structure arrangement generation scheme based on a preset optimization algorithm to obtain an optimized structure arrangement scheme includes: Extracting component information from the structural arrangement generation scheme to obtain component position feature data; Determining a structural analysis model according to the component position characteristic data; Based on the structural analysis model, under the constraints of empirical rules and mechanical indicators, the preset optimization algorithm is applied to optimize the structural layout generation plan with the goal of minimizing material cost to obtain the optimized structural layout plan; the preset optimization algorithm includes gradient descent method, genetic algorithm, simulated annealing algorithm, particle swarm optimization algorithm and online optimization algorithm.

2. The architectural design method based on diffusion model generation-optimization-relearning according to claim 1 is characterized in that: The mapping of the initial architectural drawings, the initial structural drawings and the initial architectural parameters to the feature space to obtain the initial shear wall tensor, the initial architectural design tensor and the initial mask tensor includes: Extracting features from the initial architectural drawing to obtain an initial component position feature tensor; Performing feature extraction on the initial structural drawings to obtain the initial shear wall tensor; Determine an initial building parameter feature tensor according to the initial building parameters; splicing the initial component position feature tensor and the initial building parameter feature tensor to obtain the initial building design tensor; The initial mask tensor is determined according to the initial component position feature tensor.

3. The architectural design method based on diffusion model generation-optimization-relearning according to claim 1 is characterized in that: The step of inputting the input architectural design tensor and the input mask tensor into the generative architectural structure design network based on the diffusion model to obtain the structural arrangement feature tensor output by the network comprises: Based on the input mask tensor, forward diffusing the input architectural design tensor position into Gaussian noise; The network is applied to perform reverse denoising to obtain the structural arrangement feature tensor.

4. The architectural design method based on diffusion model generation-optimization-relearning according to any one of claims 1 to 3, characterized in that: Determining the building structure design data set to be relearned according to the optimized structural layout scheme includes: Determine an optimized structural drawing according to the optimized structural layout plan; The optimized structural drawings and the corresponding architectural drawings and architectural parameters are mapped to the feature space to obtain a relearned shear wall tensor, a relearned architectural design tensor and a relearned mask tensor, so as to construct the relearned architectural structure design data set according to the relearned shear wall tensor, the relearned architectural design tensor and the relearned mask tensor.

5. An architectural design system based on diffusion model generation-optimization-relearning, characterized in that: include: An initial data set determination module, used to determine an initial building structure design data set, so as to generate and train a diffusion model according to the initial building structure design data set; The initial building structure design data set includes initial building design data and initial structure design data; A generation scheme determination module is used for generating a structural layout generation scheme based on a generative building structure design network of a diffusion model; an optimization module, configured to optimize the structural arrangement generation scheme based on a preset optimization algorithm to obtain an optimized structural arrangement scheme, and to determine a relearned building structure design data set according to the optimized structural arrangement scheme; the relearned building structure design data set is a collection of the optimized structural arrangement scheme and corresponding building design data; a relearning module, configured to relearn and train the diffusion model according to the relearned building structure design data set, and iteratively execute the generating step, the optimizing step and the relearning step according to the relearned diffusion model until the relearned building structure design data set meets the preset data set requirement; The initial building design data includes an initial building design tensor and an initial mask tensor; The initial structural design data is the initial shear wall tensor; The step of determining an initial building structure design data set and generating and training a diffusion model according to the initial building structure design data set includes: Mapping the initial building drawings, initial structural drawings and initial building parameters to the feature space to obtain the initial shear wall tensor, the initial building design tensor and the initial mask tensor, so as to construct the initial building structure design data set according to the initial shear wall tensor, the initial building design tensor and the initial mask tensor; inputting the initial building design tensor and the initial mask tensor into the diffusion model, and guiding the generation training of the diffusion model according to the initial shear wall tensor; The generative building structure design network based on the diffusion model generates a structural layout generation scheme, including: Modifying the initial building parameters to obtain input building parameters; Mapping the initial architectural drawing and the input architectural parameters to a feature space to obtain an input architectural design tensor and an input mask tensor; Inputting the input building design tensor and the input mask tensor into the generative building structure design network based on the diffusion model to obtain a structural arrangement feature tensor output by the network; Determining the structural arrangement generation scheme according to the structural arrangement feature tensor; The step of optimizing the structure arrangement generation scheme based on a preset optimization algorithm to obtain an optimized structure arrangement scheme includes: Extracting component information from the structural arrangement generation scheme to obtain component position feature data; Determining a structural analysis model according to the component position characteristic data; Based on the structural analysis model, under the constraints of empirical rules and mechanical indicators, the preset optimization algorithm is applied to optimize the structural layout generation plan with the goal of minimizing material cost to obtain the optimized structural layout plan; the preset optimization algorithm includes gradient descent method, genetic algorithm, simulated annealing algorithm, particle swarm optimization algorithm and online optimization algorithm.

6. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the architectural design method based on diffusion model generation-optimization-relearning as described in any one of claims 1 to 4 is implemented.

7. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the architectural design method based on diffusion model generation-optimization-relearning as claimed in any one of claims 1 to 4 is implemented.

Images