CN114880739B

CN114880739B - Re-optimization design method and device for generative building structure design scheme

Info

Publication number: CN114880739B
Application number: CN202210444462.7A
Authority: CN
Inventors: 陆新征; 赵鹏举; 廖文杰; 费一凡
Original assignee: Tsinghua University
Current assignee: Tsinghua University
Priority date: 2022-04-25
Filing date: 2022-04-25
Publication date: 2023-03-24
Anticipated expiration: 2042-04-25
Also published as: CN114880739A

Abstract

The invention relates to a re-optimization design method and a device of a generative building structure design scheme, which comprises the following steps: acquiring a key feature mask of the target building structure according to the key features; combining the structural arrangement scheme image to be optimized of the target building structure, the building scheme image and the key characteristic mask to obtain a corresponding combination tensor; inputting the combined tensor into a prestored structure arrangement scheme optimization model to obtain an optimization result of the image of the structure arrangement scheme to be optimized; wherein, the structural arrangement scheme optimization model includes: the system comprises a neural network optimization module obtained by training a loss function defined by considering key features and based on a structural arrangement scheme optimization sample, and an evolution optimization module set based on an evolution optimization algorithm, mechanics and experience rules. According to the method, the key characteristics of the images of the structural arrangement scheme to be optimized are optimized, and the images are optimized based on mechanics and experience rules, so that a more accurate structural arrangement scheme of the building structure is obtained.

Description

Re-optimization design method and device for generative building structure design scheme

Technical Field

The invention relates to the crossing field of architectural structure design and machine learning, in particular to a re-optimization design method and a device of a generative building structure design scheme.

Background

In the design stage of the building structure scheme, in order to ensure the safety and the economy of the building structure design scheme, a machine learning method is often adopted to learn the existing design scheme so as to overcome the defects of the existing design scheme.

However, in the existing machine learning method for learning the existing design scheme, the arrangement of the key spatial position structure is easy to be not suitable for professional requirements, and the expected improvement effect is difficult to achieve.

Disclosure of Invention

The invention aims to provide a re-optimization design method and a re-optimization design device for a generated building structure design scheme, so as to solve the problem that the existing machine learning method for learning the existing design scheme is easy to cause the situation that the key spatial position structure arrangement is not matched with the professional requirement, and further efficiently and accurately optimize the design scheme of the building structure.

In a first aspect, the present invention provides a method for re-optimizing a design plan of a generative building structure, the method comprising:

acquiring a key feature mask of a target building structure according to predefined key features;

combining the key characteristic mask, the image of the structural arrangement scheme to be optimized of the target building structure and the image of the building scheme of the target building structure to obtain a combination tensor corresponding to the target building structure;

inputting the combined tensor into a prestored structure arrangement scheme optimization model to obtain an optimization result of the image of the structure arrangement scheme to be optimized;

wherein the structural arrangement scheme optimization model comprises: the device comprises a neural network optimization module and an evolution optimization module;

the neural network optimization module is obtained by optimizing a sample based on a structural arrangement scheme and training a loss function defined by considering the key features;

the evolution optimization module is set based on an evolution optimization algorithm, mechanics and experience rules.

According to the re-optimization design method of the generative building structure design scheme provided by the invention, the key characteristics comprise: a plurality of key spatial features and a plurality of key rule features;

the obtaining of the key feature mask of the target building structure according to the predefined key features includes:

for each key feature, representing the range of the target building structure under the corresponding key feature by using a non-zero number, and representing the part of the target building structure outside the range under the corresponding key feature by using zero to obtain a corresponding key feature mask of the target building structure;

and the corresponding key characteristic mask of the target building structure is a two-dimensional tensor consistent with the image resolution of the structural arrangement scheme to be optimized.

According to the re-optimization design method of the generative building structure design scheme provided by the invention, the dimension of the structural arrangement scheme image to be optimized is completely consistent with that of the building scheme image; the structural arrangement scheme image to be optimized and the building scheme image are characterized by different components by using different colors; the combining the key feature mask, the image of the structural arrangement scheme to be optimized of the target building structure and the image of the building scheme of the target building structure to obtain a combination tensor corresponding to the target building structure includes:

for each key feature mask of the target building structure, copying and splicing the corresponding key feature mask in the channel direction where the key feature mask is located to obtain a mask tensor which is identical to the image dimension of the arrangement scheme of the structure to be optimized;

carrying out tensor element one-to-one corresponding averaging on the mask tensor and the image of the structural arrangement scheme to be optimized to obtain an image of the structural arrangement scheme to be optimized, in which the corresponding key characteristic masks are embedded, and carrying out tensor element one-to-one corresponding averaging on the mask tensor and the image of the construction scheme to obtain an image of the construction scheme, in which the corresponding key characteristic masks are embedded;

tensor splicing is carried out on the image of the structural arrangement scheme to be optimized embedded in each key characteristic mask, the image of the building scheme embedded in each key characteristic mask, the image of the structural arrangement scheme to be optimized and the image of the building scheme in channel dimension, and a combined tensor corresponding to a target building structure is obtained.

According to the re-optimization design method of the generative architectural structure design scheme provided by the invention, the combination tensor is input into a pre-stored structure arrangement scheme optimization model to obtain an optimization result of the to-be-optimized structure arrangement scheme image, and the method comprises the following steps of:

inputting the combined tensor into the neural network optimization module so that the neural network optimization module optimizes the structural arrangement scheme image to be optimized based on the key features to obtain a structural arrangement scheme image meeting key feature constraints;

and based on the evolution optimization module, re-optimizing the structural arrangement scheme image meeting the key feature constraint on the aspect of mechanics and experience rules to obtain an optimization result of the structural arrangement scheme image to be optimized.

According to the re-optimization design method of the generative building structure design scheme provided by the invention, the structure arrangement scheme optimization sample comprises the following steps: the method comprises the following steps of combining a key characteristic mask of a building structure, a combined tensor formed by combining an image of a structural arrangement scheme to be optimized of the building structure and an image of the structural arrangement scheme to be optimized of the building structure, and an ideal optimization result of the image of the structural arrangement scheme to be optimized of the building structure;

and the total loss value of the neural network optimization module in the loss function defined by considering the key characteristics is equal to the weighted sum value of the network loss, the key space loss and the key rule loss of the neural network optimization module.

According to the re-optimization design method of the generative building structure design scheme provided by the invention, the mechanical and empirical rules comprise: mechanical property rules and experience-oriented rules; the process of setting the evolutionary optimization module based on the evolutionary optimization algorithm, mechanics and empirical rules comprises:

presetting a quantitative evaluation function of each mechanical characteristic rule and a quantitative evaluation function of each experience guide rule;

setting a quantitative target function based on the quantitative evaluation function of each mechanical characteristic rule and the quantitative evaluation function of each experience guide rule;

setting corresponding constraints for the quantified objective function;

and constructing the evolutionary optimization module based on a preselected optimization algorithm, the quantified objective function and the corresponding constraint.

According to the re-optimization design method of the design scheme of the generative building structure provided by the invention, the key space loss is determined according to the following formula:

in the above formula, n is the number of key spatial features, W _i Is a buildingWeight tensor, λ, corresponding to the ith key spatial mask of the structure _i The loss function weight coefficient corresponding to the ith key space mask of the building structure, which is an operation sign of Hada Ma Ji, D _gt For the ideal optimization of the structural arrangement plan image to be optimized of the building structure, D _out An optimization result, loss, of the structural arrangement plan image to be optimized of the building structure generated by the neural network optimization module _NN Computing a function for neural network Loss, loss _space Is a critical space loss;

the critical rule loss is determined according to the following formula:

in the above formula, m is the number of key rules, Y _j J' th key rule mask for building structure, F _j A function is calculated for quantification corresponding to the j key rule, and alpha is D _gt Beta is D _out Weight coefficient of (3), loss _rule Loss of key rules;

wherein, the quantitative calculation function corresponding to the j key rule is preset,

According to the re-optimization design method of the design scheme of the generative building structure provided by the invention, the quantified objective function is determined according to the following formula:

in the above formula, function _object Is the function value of the quantified objective function, I is the number of mechanical property rules, h is the number of experience oriented rules, μ _k Is the weight coefficient of the kth mechanical property rule, f _mk Normalized result of the quantitative evaluation function of the kth mechanical property rule, v _z Weight coefficient for the z-th empirically-directed rule, f _ez For quantitative evaluation functions of the z-th empirically-directed ruleNormalized result, D _out And optimizing the structural arrangement scheme image to be optimized of the building structure generated by the neural network optimization module.

According to the re-optimization design method of the generative building structure design scheme provided by the invention, the target building structure comprises: a frame structure, a shear wall structure and a frame shear wall structure;

the structural arrangement scheme image to be optimized comprises: an image of a structural arrangement scheme of a target building structure vertical force transfer component and an image of a structural arrangement scheme of a target building structure lateral force resistance component;

wherein, the mechanical property rule comprises: dynamic characteristic rules, torsional characteristic rules, and structural deformation rules.

The empirically guided rule comprising: material constraint rules, component space validity rules and personalized definition rules;

the vertical force transfer component comprising: a beam;

the lateral force resisting member includes: shear walls and structural columns.

In a second aspect, the present invention also provides a re-optimization design apparatus for a generative architectural structural design, the apparatus comprising:

the acquisition module is used for acquiring a key feature mask of the target building structure according to predefined key features;

the combination module is used for combining the key characteristic mask, the image of the structural arrangement scheme to be optimized of the target building structure and the image of the building scheme of the target building structure to obtain a combination tensor corresponding to the target building structure;

the optimization module is used for inputting the combined tensor into a prestored structure arrangement scheme optimization model to obtain an optimization result of the image of the structure arrangement scheme to be optimized;

In a third aspect, the present invention further provides an electronic device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the re-optimization design method of the generated architectural structure design solution as described in any one of the above when executing the program.

In a fourth aspect, the present invention also provides a non-transitory computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements a re-optimization design method of a generative architectural structure design as described in any one of the above.

The invention provides a re-optimization design method and a re-optimization design device of a generative building structure design scheme, wherein a structure layout scheme optimization model is constructed in advance, and comprises a neural network optimization module and an evolution optimization module; the neural network optimization module takes a combined tensor formed by combining a key feature mask of the building structure, the image of the structural arrangement scheme to be optimized and the image of the building scheme as input, takes the image of the structural arrangement scheme meeting key feature constraint as output, defines a network structure of a loss function according to key features, and has the capability of optimizing the image of the structural arrangement scheme to be optimized on the key feature level; the evolution optimization module is set based on an evolution optimization algorithm, mechanics and experience rules and has the capability of re-optimizing the structural arrangement scheme images meeting key feature constraints on the aspect of the mechanics and experience rules, namely the structural arrangement scheme optimization model can perform key feature-mechanics and experience rule double optimization on the structural arrangement scheme images to be optimized so as to improve the accuracy and the rationality of the structural arrangement scheme. In the application stage, acquiring a key feature mask of a target building structure according to predefined key features; combining the key characteristic mask, the image of the structural arrangement scheme to be optimized of the target building structure and the image of the building scheme of the target building structure to obtain a combination tensor corresponding to the target building structure; inputting the combination tensor into a prestored structure arrangement scheme optimization model to obtain an optimization result of the image of the structure arrangement scheme to be optimized; the method can efficiently optimize the structural arrangement scheme image to be optimized of the target building structure.

Drawings

In order to more clearly illustrate the technical solutions of the present invention or the prior art, the drawings needed to be used in the description of the embodiments or the prior art will be briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without creative efforts.

FIG. 1 is a flow chart of a re-optimization design method for a generative building structure design solution provided by the present invention;

FIG. 2 is a schematic diagram of a key spatial mask provided by the present invention;

FIG. 3 is a schematic diagram of a key rule mask provided by the present invention;

FIG. 4 is a diagram of the practical application of the re-optimization design method for the generated architectural structure design solution provided in the present invention;

FIG. 5 is a block diagram of a re-optimization design apparatus for a generative architectural design scheme provided by the present invention;

fig. 6 is a schematic structural diagram of an electronic device implementing a re-optimization design method of a generative building structure design scheme provided in the present invention.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings, and it is obvious that the described embodiments are some, but not all embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

The following describes a re-optimization design method and apparatus for a generative building structure design according to the present invention with reference to fig. 1 to 6.

In a first aspect, the present invention provides a re-optimization design method for a generative building structure design solution, as shown in fig. 1, including:

s11, acquiring a key feature mask of the target building structure according to predefined key features;

s12, combining the key feature mask, the image of the structural arrangement scheme to be optimized of the target building structure and the image of the building scheme of the target building structure to obtain a combination tensor corresponding to the target building structure;

s13, inputting the combined tensor into a prestored structural arrangement scheme optimization model to obtain an optimization result of the image of the structural arrangement scheme to be optimized;

the neural network optimization module is obtained by training a loss function defined by considering the key characteristics and optimizing a sample based on a structural arrangement scheme;

The invention provides a re-optimization design method of a generating type building structure design scheme, which is characterized in that a structure arrangement scheme optimization model is constructed in advance, and the model comprises a neural network optimization module and an evolution optimization module; the neural network optimization module takes a combined tensor formed by combining a key feature mask of the building structure, the image of the structural arrangement scheme to be optimized and the image of the building scheme as input, takes the image of the structural arrangement scheme meeting key feature constraint as output, defines a network structure of a loss function according to key features, and has the capability of optimizing the image of the structural arrangement scheme to be optimized on the key feature level; the evolution optimization module is set based on an evolution optimization algorithm, mechanics and experience rules, and has the capability of re-optimizing the structural arrangement scheme images meeting key feature constraints on the aspect of the mechanics and experience rules, namely the structural arrangement scheme optimization model can perform key feature-mechanics and experience rule double optimization on the structural arrangement scheme images to be optimized so as to improve the accuracy of the structural arrangement scheme. In the application stage, acquiring a key feature mask of a target building structure according to predefined key features; combining the key characteristic mask, the image of the structural arrangement scheme to be optimized of the target building structure and the image of the building scheme of the target building structure to obtain a combination tensor corresponding to the target building structure; inputting the combined tensor into a prestored structure arrangement scheme optimization model to obtain an optimization result of the image of the structure arrangement scheme to be optimized; the method can efficiently optimize the structural arrangement scheme image to be optimized of the target building structure.

On the basis of the above embodiments, as an alternative embodiment, the key features include: a plurality of key spatial features and a plurality of key rule features;

The present embodiment is described by taking optimization of the shear wall arrangement of the shear wall structure as an example:

1): according to the field experience, the shear wall of the shear wall structure has obvious space characteristics at the elevator shaft and a balcony. Therefore, the key space features adopt elevator shaft space features and balcony space features, and the corresponding key space masks adopt elevator shaft space masks and balcony space masks.

In the embodiment, the elevator shaft space mask and the balcony space mask are both expressed by using two-dimensional tensors with the same dimension as the resolution of the image of the structural arrangement scheme to be optimized; the tensor dimension corresponding to the image of the structural arrangement scheme to be optimized is 512 multiplied by 1024 multiplied by 3, the 512 multiplied by 1024 is a resolution scale, 3 represents three channels of RGB of the color image, and then the elevator shaft space mask and the balcony space mask are represented by two-dimensional tensors of 512 multiplied by 1024.

In addition, the present embodiment numerically represents the range of the key spatial mask. Fig. 2 is a schematic diagram of a key space mask, as shown in fig. 2, where the range of elements in the elevator hoistway space mask is represented by an integer 255 and the rest are represented by an integer 0; the element range where the balcony is located in the balcony space mask is represented by an integer 255, and the rest are represented by an integer 0; for convenience of presentation, the extent of the hoistway space and the balcony space is indicated in white, and the remaining spaces are indicated in black.

It should be noted that the above examples are merely used for demonstration, and other spatial features may be used as consideration of the key spatial features according to the design requirements of the shear wall structure.

2): the key rule features adopt an integral symmetry rule and a local symmetry rule, and the key rule mask corresponding to the key rule features adopts an integral symmetry mask and a local symmetry mask.

Similar to the key spatial mask, in this embodiment, the key rule mask is also expressed by a two-dimensional tensor that has the same dimension as the resolution of the image of the structural arrangement scheme to be optimized; the range of the key rule mask is represented using a number.

Fig. 3 illustrates a key rule mask diagram, and as shown in fig. 3, the overall symmetry mask may cover the overall symmetric part with the same integer, and may use an integer 255. The local symmetry mask may represent different locally symmetric parts by using different integers, for example, an integer 125 represents a left symmetric part, an integer 255 represents a right symmetric part, and tensors except the symmetric part are represented by an integer 0. For convenience of display, the overall symmetrical part and the local symmetrical part in fig. 3 are displayed by using different gray scales.

In addition, although the global symmetry mask and the local symmetry mask as illustrated in fig. 3 have regular shapes and are full of the whole two-dimensional tensors of the masks, in actual applications, the global symmetry mask and the local symmetry mask are required to clearly represent symmetric regions and to represent different symmetric regions by different numbers, and tensors other than symmetric portions are represented by integers 0, and may have irregular shapes.

In addition, in the process of representing the whole symmetry mask and the local symmetry mask, the symmetry axis coordinate corresponding to each symmetry mask needs to be recorded, so that the loss can be calculated conveniently in the following process.

It should be noted that the overall symmetry rule and the local symmetry rule of the present embodiment are merely demonstrated as examples, and other rules may also be used as key rule features according to the design requirements of the shear wall structure and the actual building conditions.

It is conceivable that, for the frame structure, the key spatial feature may select a space such as a staircase or a living room, which may have a large influence on the arrangement of the frame structure, and the key regular feature may select an overall symmetry rule, a local symmetry rule, a uniformity rule of arrangement of peripheral frame columns, and the like.

For a frame-shear wall structure, the key spatial features and key rule features may be a combination of the two structures, frame structure and shear wall structure.

The invention provides key space masks and key rule masks, wherein each key space mask is a separate mask tensor, and each key rule mask is also a separate mask tensor. Based on the independence of the mask tensors, the numbers of different mask tensors can be controlled to represent the space represented by the different mask tensors or the specificity of the rules, so that convenience is provided for the subsequent key feature constraint optimization.

On the basis of the above embodiments, as an optional embodiment, the dimension of the to-be-optimized structural arrangement scheme image is completely consistent with that of the building scheme image; the structural arrangement scheme image to be optimized and the building scheme image both represent different components by using different colors; the combining the key feature mask, the image of the structural arrangement scheme to be optimized of the target building structure and the image of the building scheme of the target building structure to obtain a combination tensor corresponding to the target building structure includes:

carrying out one-to-one corresponding tensor element averaging on the mask tensor and the image of the structural arrangement scheme to be optimized to obtain an image of the structural arrangement scheme to be optimized in which the corresponding key feature masks are embedded, and carrying out one-to-one corresponding tensor element averaging on the mask tensor and the image of the building scheme to obtain an image of the building scheme in which the corresponding key feature masks are embedded;

The embodiment is described by taking as an example that the key space mask includes 2 512 × 1024 two-dimensional tensors, which are respectively an elevator shaft space mask and a balcony space mask, the key rule mask includes 2 512 × 1024 two-dimensional tensors, which are respectively an overall symmetry mask and a local symmetry mask, and the to-be-optimized structure arrangement scheme image and the building scheme image are both 512 × 1024 × 3 three-dimensional tensors:

for 1 512 × 1024 elevator shaft space masks, firstly, copying and splicing the elevator shaft space masks in a channel dimension to obtain a 512 × 1024 × 3 three-dimensional tensor; carrying out tensor element one-to-one corresponding averaging on the 512 multiplied by 1024 multiplied by 3 three-dimensional tensor and the image of the arrangement scheme of the structure to be optimized to obtain an image of the arrangement scheme of the structure to be optimized embedded in the elevator shaft space mask; similar to the structural arrangement scheme image to be optimized, carrying out tensor element one-to-one corresponding averaging on the three-dimensional tensor of 512 multiplied by 1024 multiplied by 3 and the building scheme image to obtain the building scheme image embedded with the elevator shaft space mask; for example: assuming that the element of the elevator shaft space mask at the (1,2,1) position of the three-dimensional tensor is 255 and the element of the to-be-optimized structure arrangement scheme image at the (1,2,1) position of the three-dimensional tensor is 125, after the elements are correspondingly averaged in a one-to-one correspondence manner, the element of the to-be-optimized structure arrangement scheme image embedded by the obtained elevator shaft space mask at the (1,2,1) position of the three-dimensional tensor is 190, that is, (125 + 255)/2 =190.

Secondly, similarly, the balcony space mask, the overall symmetry mask and the local symmetry mask can obtain a corresponding mask-embedded image of the structural arrangement scheme to be optimized and a mask-embedded image of the building scheme;

finally, tensor splicing is carried out on the image of the structure arrangement scheme to be optimized embedded in each mask, the image of the building scheme embedded in each mask, the image of the structure arrangement scheme to be optimized and the image of the building scheme to be optimized originally in the dimensionality of the channel, and a combined tensor can be obtained.

Specifically, 10 plan images of 512 × 1024 × 3 in total are subjected to tensor splicing in the dimension of a channel to obtain a three-dimensional tensor of 512 × 1024 × 30, which is a combined tensor.

Besides the method adopted by this embodiment, a method of directly performing tensor splicing in the channel dimension to obtain a combined tensor can be adopted:

namely: and directly splicing the mask tensor, the image of the structural arrangement scheme to be optimized and the image of the building scheme in a channel direction to obtain a combined tensor corresponding to the target building structure.

Specifically, 4 512 × 1024 two-dimensional tensors, namely an elevator shaft space mask, a balcony space mask, an overall symmetry mask and a local symmetry mask, and 2 512 × 1024 × 3 three-dimensional tensors, namely an image of a structural arrangement scheme to be optimized and an image of a building scheme, are spliced in a channel dimension to obtain the 512 × 1024 × 10 three-dimensional tensor, namely the combined tensor.

According to the method, the key feature mask of the target building structure, the image of the structural arrangement scheme to be optimized and the features of the image of the building scheme are combined in the same tensor, so that the neural network optimization module optimizes the key feature level of the image of the structural arrangement scheme to be optimized, and the image of the structural arrangement scheme meeting the optimization of the key features is obtained.

On the basis of the foregoing embodiments, as an optional embodiment, the inputting the combination tensor into a prestored structural arrangement scheme optimization model to obtain an optimization result of the to-be-optimized structural arrangement scheme image includes:

inputting the combined tensor into the neural network optimization module so that the neural network optimization module optimizes the image of the structural arrangement scheme to be optimized based on the key features to obtain an image of the structural arrangement scheme meeting key feature constraints;

Fig. 4 illustrates a diagram of an actual application of a re-optimization design method of a generated architectural structural design scheme, as shown in fig. 4, a structural arrangement scheme optimization model includes two parts, namely, a neural network optimization module and an evolutionary optimization algorithm, so that the combination tensor is input into a pre-stored structural arrangement scheme optimization model, and an optimization result of obtaining an image of the structural arrangement scheme to be optimized also involves two steps:

the first step is as follows: inputting a combination tensor formed by combining the key feature mask, the image of the structural arrangement scheme to be optimized and the image of the building scheme into a neural network optimization module, wherein the neural network optimization module realizes the key feature optimization of the image of the structural arrangement scheme to be optimized by adopting a key feature constraint method to obtain an image of the structural arrangement scheme meeting the key feature constraint;

the second step: and the evolution optimization module generates the structural arrangement scheme image meeting the key characteristic constraint, then performs constraint through mechanics and experience rules, and optimizes the generated structural arrangement scheme image meeting the key characteristic constraint by using an evolution optimization algorithm.

It should be noted that, in the construction stage, the neural network optimization module defines a loss function according to the key features, and optimizes the network parameters with the loss function; the trained neural network optimization module can be used after passing the evaluation of the test set.

The method can perform double optimization of key characteristics, mechanics and experience rules on the image of the structural arrangement scheme to be optimized so as to improve the accuracy of the structural arrangement scheme.

On the basis of the above embodiments, as an optional embodiment, the structural arrangement scheme optimization sample includes: the method comprises the following steps of combining a key characteristic mask of the building structure, a combination tensor formed by combining an image of a structural arrangement scheme to be optimized of the building structure and an image of the structural arrangement scheme to be optimized of the building structure, and an ideal optimization result of the image of the structural arrangement scheme to be optimized of the building structure;

the structural arrangement scheme optimization sample comprises structural arrangement scheme images to be optimized of the building structure, and the structural arrangement scheme images are constructed manually; the ideal optimization result of the structural arrangement scheme image to be optimized of the building structure is the result of engineer design obtained by collecting drawings.

Taking a shear wall structure as an example, the manual construction method for the image of the structural arrangement scheme to be optimized of the building structure in the structural arrangement scheme optimization sample may be:

partition wall elements in the architectural plan image of the architectural structure are randomly filled with shear wall elements in a certain proportion (exemplarily, for example, 40%) as the structural plan image to be optimized.

In this embodiment, the neural network structure selected by the neural network optimization module may adopt pix2pixHD as a generation antagonistic neural network, or may adopt other generation antagonistic neural networks, such as pix2pix, or other convolutional neural networks, such as U-Net.

Assuming that the key space mask includes 2 512 × 1024 two-dimensional tensors, which are an elevator shaft space mask and a balcony space mask, respectively, the key rule mask includes 2 512 × 1024 two-dimensional tensors, which are an overall symmetry mask and a local symmetry mask, respectively, the image of the to-be-optimized structure layout scheme and the image of the building scheme are both 512 × 1024 × 3 three-dimensional tensors, the neural network optimization module needs to be implemented, a combination tensor (512 × 1024 × 30 three-dimensional tensor) formed by combining the image of the to-be-optimized structure layout scheme, the key feature mask and the image of the building scheme is input, and an optimization design result (512 × 1024 × 3 three-dimensional tensor) of the image of the to-be-optimized structure layout scheme is output.

When the neural network optimization module is constructed, besides the key space mask and the key rule mask, the loss function needs to be perfected. Assuming that the neural network optimization module is a pix2pixHD architecture, in addition to the pix2pixHD generating the loss against the neural network itself, the critical spatial loss and the critical rule loss need to be defined, and the critical spatial loss and the critical rule loss are added to the pix2pixHD generating the loss against the neural network itself to obtain the total loss.

According to the method, the loss function is defined according to the key features, and the parameters of the neural network model are optimized by the loss function, so that the trained neural network model can perform key feature constraint optimization on the structural arrangement scheme image.

On the basis of the above embodiments, as an alternative embodiment, the mechanical and empirical rules include: mechanical property rules and experience-oriented rules; the process of setting the evolutionary optimization module based on the evolutionary optimization algorithm, mechanics and empirical rules comprises:

setting corresponding constraints for the quantified objective function;

and constructing the evolution optimization module based on a preselected optimization algorithm, the quantified objective function and the corresponding constraint.

In this embodiment, the evolutionary optimization algorithm employs a genetic algorithm. The method expresses the mechanical property rules and the experience guiding rules as quantitative objective functions, further optimizes the building structure arrangement scheme meeting the key feature constraints by using the progress optimization algorithm on the basis, and further improves the rationality of the building structure arrangement scheme.

The invention sets the quantitative objective function by prescribing the quantitative evaluation function of each mechanical characteristic rule and the quantitative evaluation function of each experience guide rule in advance, so that the evolution optimization model has the capabilities of mechanical characteristic rule constraint optimization and experience guide rule constraint optimization.

On the basis of the above embodiments, as an alternative embodiment, the critical space loss is determined according to the following formula:

/>

in the above formula, n is the number of key spatial features, W _i Weight tensor, lambda, corresponding to ith key space mask of building structure _i The loss function weight coefficient corresponding to the ith key space mask of the building structure, which is an operation sign of Hada Ma Ji, D _gt For the ideal optimization of the structural arrangement plan image to be optimized of the building structure, D _out Optimization node of structural arrangement scheme image to be optimized of building structure generated by neural network optimization moduleFruit, loss _NN Computing function for neural network Loss, loss _space Is a critical space loss;

in this embodiment, the loss function weight coefficient λ corresponding to each key spatial feature _i All take 1,loss _NN And taking the loss function as a smooth L1 loss function, wherein the smooth L1 loss function is a function of loss calculation commonly used by a neural network.

The critical rule loss is determined according to the following formula:

in the above formula, m is the number of key rules, Y _j J' th key rule mask for building structure, F _j A quantitative calculation function corresponding to the jth key rule, wherein alpha is D _gt Beta is D _out Weight coefficient of (D) _gt For the ideal optimization of the structural arrangement plan image to be optimized of the building structure, D _out An optimization result, loss, of the structural arrangement plan image to be optimized of the building structure generated by the neural network optimization module _rule Loss of key rules;

and the quantitative calculation function corresponding to the jth key rule is preset.

The embodiment is explained by taking the example that the key space characteristics comprise balcony space characteristics and elevator shaft space characteristics, and the key rule characteristics comprise the shear wall arrangement optimization of the whole symmetry rule and the local symmetry rule:

if the range of balcony/elevator shaft in balcony/elevator shaft space mask is represented by integer 1 and the rest by integer 0, then W may be expressed as _i Directly taking the code as the ith key space mask; if the range of balconies/elevator shafts in the balcony/elevator shaft space mask is represented by the remaining integer number, for example 255, and the remaining integer number 0, then W can be expressed as _i The resulting tensor is normalized by 255, taken as the ith key spatial mask. This means that the embodiment only takes into account the critical spatial losses when calculating themThe difference between the optimization result and the corresponding truth value at the key space.

In addition, considering that the symmetry of the optimization result is irrelevant to the true value of the shear wall design, α is 0, and β is not 1.

F _j Is a function of calculating the loss of symmetry and can be defined as:

in the formula, T is the number of symmetric parts corresponding to the j-th key rule, and in this example, the overall symmetry T =1 and the local symmetry T =2.dim _ x and dim _ y are resolution dimensions of tensors, 512 and 1024 respectively in the example. D _out Shear wall structure layout scheme image optimization results D 'generated for neural network models and meeting key feature constraints' _out Is D _out The resulting tensor flipped around the d-th axis of symmetry. W is a group of _d The weighting tensor is the d-th symmetric part, and the dimension of the weighting tensor is consistent with the tensor mask corresponding to each key rule, and the weighting tensor is a 512 × 1024 two-dimensional tensor in this example, but only focusing on the d-th symmetric part, the mask area of the d-th symmetric feature is the integer 1, and the rest is the integer 0.α and β represent the position of values in the resolution dimension of the two tensors dim _ x and dim _ y, respectively, in D _outαβ For example, if α is 100 and β is 200, then D _outαβ Is shown by D _out The value at this position.

For example: in the present example, T =2 is the local symmetry, so d has two values, 1 and 2, and when d takes 1, it represents the left local symmetric region, and corresponds to W ₁ The tensor part takes 1, the rest takes 0, when d takes 2, the right local symmetrical area is shown, and the W corresponding to the area ₂ The tensor part takes 1 and the rest takes 0.

signF is a sign function defined in this embodiment, and takes a value of 0 when the argument is 0, otherwise takes a value of 1.

Experiments show that in the areas of an elevator shaft and a balcony and the symmetry of a structure, compared with the structural arrangement scheme without the key feature optimization, the structural arrangement scheme with the optimized key features has obvious effect improvement, for example, a shear wall with better continuity is arranged near the elevator shaft, the wall ratio is improved, and fewer shear walls are arranged near the balcony.

The invention represents the weight of the key characteristic mask in the loss function by controlling the number of the key characteristic mask so as to reflect the importance degree of the space or the rule corresponding to different key characteristic masks, so that the trained neural network model has good optimization performance.

On the basis of the above embodiments, as an optional embodiment, the objective function of quantification is determined according to the following formula:

in the above formula, function _object Is the function value of the quantified objective function, I is the number of mechanical property rules, h is the number of experience oriented rules, μ _k Is the weight coefficient of the kth mechanical property rule, f _mk Normalized result of the quantitative evaluation function of the kth mechanical property rule, v _z Weight coefficient for the z-th empirically-directed rule, f _ez Normalization of the quantitative evaluation function for the z-th empirically-directed rule, D _out And optimizing the structural arrangement scheme image to be optimized of the building structure generated by the neural network optimization module.

Taking the optimization of the arrangement of the shear wall structure as an example, 1 mechanical property rule and 1 empirical guidance rule are exemplarily used.

The mechanical property rule is a torsion property rule, the difference between the centroid and the rigid center of the shear wall design result is calculated through a quantitative evaluation function, and the difference is defined as:

in the formula (x) _m ，y _m ) And (x) _s ，y _s ) Respectively represent D _out Center of mass and rigidity of the shear wall, f _m Normalized to have a value range of [0,1 ], when the centroid and centroid coincide, the value is 0, otherwise, it tends to 1, so f _m The smaller, the better the result.

The experience guidance rule is a component space validity rule, the difference between the shear wall radiation area and the floor area is calculated through a quantitative evaluation function, the shear wall radiation area is the area covered by the range of the peripheral distance S of the shear wall in the floor, S is defined according to professional experience, and in the embodiment, 2m is selected, namely the area covered by the range of 2m away from the shear wall. Then, the quantitative merit function of the empirically directed rule is defined as:

in the formula, A _s Is the shear wall radiation area, A _f Is floor area, f _e Normalized, its value range is [0,1 ], when the shear wall radiation area covers the entire floor area, the value is 0, otherwise, 1 is favored. It is desirable that the shear wall affect the entire floor as much as possible, so f _e The smaller the better the result.

In this example, μ _k And v _z If 1 is not taken, i.e. the mechanical property rule and the experience-oriented rule are considered to have the same weight, then the optimization goal of this embodiment is to minimize the Function _object Namely:

D _out the constraint condition of the shear wall is that the shear wall can be increased only on the partition wall and cannot be arranged on doors and windows; and at each iteration, the changed pixels that are increased and decreased need to be next to the existing shear wall elements, namely:

Pixel _action ＝Neighbor(Pixel _shear )

in the formula, pixel _action For variably changeable pixels in iteration, pixels _shear Is D _out The Neighbor function is a function for finding the pixels in the immediate vicinity.

The principle of the Neighbor function is shown for a single shear wall pixel as an example. If the pixel (i, j) is a shear wall pixel, the pixel existing in the four adjacent pixels (i-1,j), (i, j-1), (i, j + 1), (i +1,j) is the adjacent pixel corresponding to the Neighbor function. The existence of the four pixel points adjacent to the pixel points means that the pixel points do not exceed the pixel boundary of the image, are still on the partition wall and are not above the blank and the door and window pixels.

And Function _object In this embodiment, the following are specific:

wherein (x) _m ，y _m ) And (x) _s ，y _s ) Respectively represent D _out Center of mass and rigidity of shear wall, A _s And A _f Are respectively D _out The corresponding radiation area of the shear wall and the floor area.

The invention can quantitatively set the objective function aiming at the mechanical rule and the empirical guidance rule, so that the building structure arrangement scheme meeting the key characteristic constraint can be further optimized in the aspects of the mechanical rule and the empirical guidance rule, and the reasonability of the building structure arrangement is improved.

On the basis of the above embodiments, as an alternative embodiment, the target building structure includes: a frame structure, a shear wall structure and a frame shear wall structure;

the image of the structural arrangement scheme to be optimized comprises the following steps: an image of a structural arrangement scheme of a target building structure vertical force transfer component and an image of a structural arrangement scheme of a target building structure lateral force resistance component;

wherein the mechanical property rule comprises: dynamic characteristic rules, torsional characteristic rules, and structural deformation rules.

the vertical force transfer component comprising: a beam;

In this embodiment, the dynamic characteristics rule is used to constrain the dynamic characteristics of the building structure. The natural vibration frequency of the building structure can be calculated by adopting a multi-degree-of-freedom model or other structural finite element models, and the upper limit and the lower limit of the natural vibration frequency are restricted not to exceed a defined range by defining a function. The dynamic characteristics rules may also be defined in other forms, as long as the dynamic characteristics of the building structure need to be constrained. The dynamic characteristics of the building structure may be considered to include: natural vibration frequency, natural vibration period, vibration mode.

The torsional properties rules are used to constrain the torsional properties of the building structure. The deviation between the centroid and the rigid center can be used for evaluation, and other forms can also be used for definition.

The structural deformation rules are used to constrain the deformation characteristics of the building structure. The deformation condition of the structure can be obtained by adopting a multi-degree-of-freedom model or other structure finite element models and adopting static analysis or dynamic analysis, and the deformation condition of the structure is realized by restraining the deformation of the building structure. Generally, the structural variations that can be adopted are: vertical maximum displacement under the action of gravity, the displacement angle between layers of the structure, the maximum lateral displacement of the structure and the like.

The material constraint rules are used to constrain the amount of structural material, and can generally be implemented by limiting the upper and lower limits of the amount of structural material.

The space validity rule of the structural member can be realized by calculating the difference between the radiation area of the structural member and the floor area, so as to quantify whether the structural member can effectively conduct the load of the structure.

The individualized definition rules are used for restricting individualized experience rules of some experts, for a shear wall structure, for example, the length of a wall section of a constrained shear wall is greater than 0.5m, and for a frame structure, for example, the column pitch of a constrained frame must be between 3.0m and 8.0 m.

The invention exemplifies common building structures, structural arrangement scheme images, mechanical characteristic rules and experience guiding rules, but a person skilled in the art should clearly understand that the building structures, structural arrangement scheme images, mechanical characteristic rules and experience guiding rules which are in accordance with the situation all belong to the protection scope of the invention.

The present invention further defines, by way of example, architectural structures, structural layout plan images, mechanical property rules, and empirically directed rules to better explain the present invention.

In a second aspect, the present invention provides a re-optimization design apparatus for a generative architectural structure design scheme, as shown in fig. 5, comprising:

the acquiring module 21 is configured to acquire a key feature mask of the target building structure according to a predefined key feature;

the combination module 22 is configured to combine the key feature mask, the image of the to-be-optimized structure arrangement scheme of the target building structure, and the image of the building scheme of the target building structure, so as to obtain a combination tensor corresponding to the target building structure;

the optimization module 23 is configured to input the combination tensor into a prestored structure arrangement scheme optimization model to obtain an optimization result of the image of the structure arrangement scheme to be optimized;

wherein, the structural arrangement scheme optimization model comprises: the device comprises a neural network optimization module and an evolution optimization module;

The building structure component size design device with embedded domain knowledge provided in the embodiments of the present invention specifically executes the process of the building structure component size design method with embedded domain knowledge, and please refer to the content of the building structure component size design method with embedded domain knowledge in detail, which is not described herein again.

The invention provides a re-optimization design device of a generation type building structure design scheme, which is characterized in that a structure arrangement scheme optimization model is constructed in advance, and the model comprises a neural network optimization module and an evolution optimization module; the neural network optimization module takes a key feature mask of the building structure, a combined tensor formed by combining the image of the structural arrangement scheme to be optimized and the image of the building scheme as input, takes the image of the structural arrangement scheme meeting key feature constraint as output, defines a network structure of a loss function according to key features, and has the capability of optimizing the image of the structural arrangement scheme to be optimized on the key feature level; the evolution optimization module is set based on an evolution optimization algorithm, mechanics and experience rules, and has the capability of re-optimizing the structural arrangement scheme images meeting key feature constraints on the aspect of the mechanics and experience rules, namely the structural arrangement scheme optimization model can perform key feature-mechanics and experience rule double optimization on the structural arrangement scheme images to be optimized so as to improve the accuracy of the structural arrangement scheme. In the application stage, acquiring a key feature mask of a target building structure according to predefined key features; combining the key characteristic mask, the image of the structural arrangement scheme to be optimized of the target building structure and the image of the building scheme of the target building structure to obtain a combination tensor corresponding to the target building structure; inputting the combined tensor into a prestored structure arrangement scheme optimization model to obtain an optimization result of the image of the structure arrangement scheme to be optimized; it is possible to efficiently optimize the structural arrangement plan image to be optimized of the target building structure.

In a third aspect, fig. 6 illustrates a schematic physical structure diagram of an electronic device, and as shown in fig. 6, the electronic device may include: a processor (processor) 610, a communication Interface (Communications Interface) 620, a memory (memory) 630 and a communication bus 640, wherein the processor 610, the communication Interface 620 and the memory 630 communicate with each other via the communication bus 640. Processor 610 may invoke logic instructions in memory 630 to perform a method of re-optimizing a design of a generative architectural structure design, the method comprising: acquiring a key feature mask of a target building structure according to predefined key features; combining the key characteristic mask, the image of the structural arrangement scheme to be optimized of the target building structure and the image of the building scheme of the target building structure to obtain a combination tensor corresponding to the target building structure; inputting the combined tensor into a prestored structure arrangement scheme optimization model to obtain an optimization result of the image of the structure arrangement scheme to be optimized; wherein the structural arrangement scheme optimization model comprises: the device comprises a neural network optimization module and an evolution optimization module; the neural network optimization module is obtained by optimizing a sample based on a structural arrangement scheme and training a loss function defined by considering the key features; the evolution optimization module is set based on an evolution optimization algorithm, mechanics and experience rules.

In addition, the logic instructions in the memory 630 may be implemented in software functional units and stored in a computer readable storage medium when the logic instructions are sold or used as independent products. Based on such understanding, the technical solution of the present invention may be embodied in the form of a software product, which is stored in a storage medium and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, and other various media capable of storing program codes.

In a fourth aspect, the present invention also provides a computer program product comprising a computer program, the computer program being storable on a non-transitory computer readable storage medium, the computer program, when executed by a processor, being operable to perform a method of re-optimization design of a generative architectural structural design, the method comprising: acquiring a key feature mask of a target building structure according to predefined key features; combining the key characteristic mask, the image of the structural arrangement scheme to be optimized of the target building structure and the image of the building scheme of the target building structure to obtain a combination tensor corresponding to the target building structure; inputting the combined tensor into a prestored structure arrangement scheme optimization model to obtain an optimization result of the image of the structure arrangement scheme to be optimized; wherein the structural arrangement scheme optimization model comprises: the device comprises a neural network optimization module and an evolution optimization module; the neural network optimization module is obtained by optimizing a sample based on a structural arrangement scheme and training a loss function defined by considering the key features; the evolution optimization module is set based on an evolution optimization algorithm, mechanics and experience rules.

In a fifth aspect, the present invention also provides a non-transitory computer readable storage medium having stored thereon a computer program to perform a method of re-optimizing a design of a generative architectural structure, the method comprising: acquiring a key feature mask of a target building structure according to predefined key features; combining the key characteristic mask, the image of the structural arrangement scheme to be optimized of the target building structure and the image of the building scheme of the target building structure to obtain a combination tensor corresponding to the target building structure; inputting the combined tensor into a prestored structure arrangement scheme optimization model to obtain an optimization result of the image of the structure arrangement scheme to be optimized; wherein the structural arrangement scheme optimization model comprises: the device comprises a neural network optimization module and an evolution optimization module; the neural network optimization module is obtained by optimizing a sample based on a structural arrangement scheme and training a loss function defined by considering the key features; the evolution optimization module is set based on an evolution optimization algorithm, mechanics and experience rules.

The above-described embodiments of the apparatus are merely illustrative, and the units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of the present embodiment. One of ordinary skill in the art can understand and implement it without inventive effort.

Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment may be implemented by software plus a necessary general hardware platform, and may also be implemented by hardware. With this understanding in mind, the above-described technical solutions may 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., and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the methods described in the embodiments or some parts of the embodiments.

Finally, it should be noted that: the above examples are only intended to illustrate the technical solution of the present invention, and not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.

Claims

1. A method of re-optimizing a design of a generative architectural structure, the method comprising:

inputting the combination tensor into a prestored structure arrangement scheme optimization model to obtain an optimization result of the image of the structure arrangement scheme to be optimized;

the evolution optimization module is set based on an evolution optimization algorithm, mechanics and experience rules;

the key features include: a plurality of key spatial features and a plurality of key rule features; the key rule features comprise an overall symmetry rule feature and a local symmetry rule feature;

2. The method of generating a re-optimized design of a architectural structural design solution according to claim 1, wherein the image of the structural layout solution to be optimized is identical in dimensions to the image of the architectural solution; the structural arrangement scheme image to be optimized and the building scheme image are characterized by different components by using different colors; the combining the key feature mask, the image of the structural arrangement scheme to be optimized of the target building structure and the image of the building scheme of the target building structure to obtain a combination tensor corresponding to the target building structure includes:

3. The method of claim 1, wherein inputting the combination tensor into a pre-stored optimization model of the structural arrangement scheme to obtain an optimization result of the image of the structural arrangement scheme to be optimized, comprises:

4. The method of claim 1, wherein the structural arrangement optimization paradigm comprises: the method comprises the following steps of combining a key characteristic mask of a building structure, a combined tensor formed by combining an image of a structural arrangement scheme to be optimized of the building structure and an image of the structural arrangement scheme to be optimized of the building structure, and an ideal optimization result of the image of the structural arrangement scheme to be optimized of the building structure;

5. The method of generating a re-optimized design of a structural design of a building according to claim 1, wherein the mechanical and empirical rules include: mechanical property rules and experience-oriented rules; the process of setting the evolutionary optimization module based on evolutionary optimization algorithm, mechanics and empirical rules comprises:

setting corresponding constraints for the quantified objective function;

6. The method of reoptimized design of a generative architectural structural design scheme according to claim 4 wherein the critical space loss is determined according to the following equation:

in the above formula, n is the number of key spatial features, W _i Weight tensor, lambda, corresponding to ith key space mask of building structure _i The loss function weight coefficient corresponding to the ith key space mask of the building structure, an operation symbol of [ "Hadar Ma Ji" ], D _gt For the ideal optimization of the structural arrangement plan image to be optimized of the building structure, D _out Building generated for the neural network optimization moduleOptimization of the structural arrangement scheme image of the structure to be optimized, loss _NN Computing a function for neural network Loss, loss _space Is a critical space loss;

the critical rule loss is determined according to the following formula:

in the above formula, m is the number of key rules, Y _j For the j-th key rule mask of the building structure, F _j A function is calculated for the quantification corresponding to the jth key rule, alpha is D _gt Beta is D _out Weight coefficient of (3), loss _rule Loss of key rules;

and the quantitative calculation function corresponding to the j key rule is preset.

7. The method of claim 5, wherein the quantified objective function is determined according to the following equation:

in the above formula, function _object For the function value of the quantified objective function, l is the number of mechanical property rules, h is the number of experience oriented rules, μ _k Is the weight coefficient of the kth mechanical property rule, f _mk Normalized result of the quantitative evaluation function of the kth mechanical property rule, v _z Weight coefficient for the z-th empirically-directed rule, f _ez Normalized result of quantitative evaluation function for the z-th empirically directed rule, D _out And optimizing the structural arrangement scheme image to be optimized of the building structure generated by the neural network optimization module.

8. The method of reoptimized design of a generative architectural structure design according to claim 5, wherein the target architectural structure comprises: a frame structure, a shear wall structure and a frame shear wall structure;

wherein the mechanical property rule comprises: a dynamic characteristic rule, a torsional characteristic rule and a structural deformation rule;

the vertical force transfer component comprising: a beam;

9. A generative architectural structural design solution re-optimization design apparatus, the apparatus comprising:

an acquisition module for acquiring, based on predefined key features, acquiring a key feature mask of a target building structure;

10. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor, when executing the program, implements a method of re-optimization design of a generative architectural structure design according to any one of claims 1 to 8.

11. A non-transitory computer readable storage medium having stored thereon a computer program, wherein the computer program, when executed by a processor, implements a method of re-optimization design of a generative architectural structure design according to any one of claims 1 to 8.