CN114741766A

CN114741766A - Building structure design method and device based on joint structure optimization neural network

Info

Publication number: CN114741766A
Application number: CN202210444464.6A
Authority: CN
Inventors: 陆新征; 费一凡; 廖文杰
Original assignee: Tsinghua University
Current assignee: Tsinghua University
Priority date: 2022-04-25
Filing date: 2022-04-25
Publication date: 2022-07-12
Anticipated expiration: 2042-04-25
Also published as: CN114741766B

Abstract

The invention relates to a building structure design method and a device based on a combined structure optimization neural network, comprising the following steps: acquiring design conditions of a target building structure; substituting the design conditions into a pre-constructed building structure design scheme generation model to obtain a design scheme of a target building structure; the building structure design scheme generation model is constructed with the assistance of a building structure optimizer. The building structure optimizer is an optimization model which can be used for optimizing the design scheme of the building structure and outputting the optimized design scheme of the building structure and the corresponding specified target performance. On one hand, the scale and the quality of a neural network training set generated by a building structure design scheme are improved by using a building structure optimizer, and on the other hand, the neural network is optimized by using the specified target performance evaluation capability of the neural network on the characteristic tensor of the design scheme to generate the neural network; therefore, the intelligent design of the building structure is realized quickly and reliably.

Description

Building structure design method and device based on joint structure optimization neural network

Technical Field

The invention relates to the technical field of crossing of building structure design and artificial intelligence, in particular to a building structure design method and a building structure design device based on a joint structure optimization neural network.

Background

In the design stage of the scheme of the building structure, in order to ensure the compliance, the economy, the environmental protection and the like of the final structural design scheme, the preliminary scheme design of the plane arrangement and the section size of the structural members needs to be carried out quickly and reasonably under the constraint of structural design conditions.

However, the current preliminary scheme design mainly depends on personal experience of architects and structural engineers, which is time-consuming and labor-consuming and difficult to carry existing design experience.

The existing building structure design method based on artificial intelligence has a series of limitations, including that the performance of a design model for generating a building structure design scheme is influenced and restricted by the scale and quality of a data set, and the requirements in various aspects such as economy, environmental protection and the like are difficult to be fully considered.

Disclosure of Invention

The invention aims to provide a building structure design method and a building structure design device based on a joint structure optimization neural network, which are used for overcoming the defects that in the prior art, the performance of a design model is influenced and restricted by the scale and the quality of a data set, and various requirements such as economy, environmental protection and the like are difficult to be fully considered, so that the intelligent design of a building structure can be quickly and reliably completed.

In a first aspect, the present invention provides a method for designing an architectural structure based on a joint structure optimization neural network, the method comprising:

acquiring design conditions of a target building structure;

substituting the design conditions into a pre-constructed building structure design scheme generation model to obtain a design scheme of a target building structure;

the building structure design scheme generation model is constructed with the assistance of a building structure optimizer;

the building structure optimizer is generated based on a specified target, a building design empirical rule and an optimization algorithm, and can be used for optimizing a design scheme of a building structure and outputting an optimized design scheme of the building structure and an optimized model of the corresponding specified target performance.

According to the building structure design method based on the combined structure optimization neural network, when the design condition is composed of a structural layout drawing and natural environment factors, the design scheme is a component section size design scheme;

when the design condition is composed of a building design drawing and natural environment factors, the design scheme is a component arrangement design scheme;

wherein the natural environment factors include: seismic influence coefficient, wind influence coefficient, total structural height and standard layer characteristic height.

According to the building structure design method based on the combined structure optimization neural network provided by the invention, the building structure design scheme generation model comprises the following steps: generating a neural network by the pre-processor, the post-processor and the architectural structure design scheme; substituting the design conditions into a pre-constructed building structure design scheme generation model to obtain a design scheme of a target building structure, wherein the design scheme comprises the following steps:

processing the design condition into a design condition feature tensor based on a pre-processor;

inputting the design condition feature tensor to the architectural structure design scheme generation neural network to obtain a design scheme feature tensor output by the architectural structure design scheme generation neural network;

converting, based on a post-processor, the design feature tensor into a design, and taking the converted design as a design for the target building structure.

According to the building structure design method based on the joint structure optimization neural network provided by the invention, the building structure design scheme generation model further comprises the following steps: specifying a target performance evaluation neural network; the building process of generating the neural network by the architectural structure design scheme comprises the following steps:

step 1: collecting design conditions and design schemes of the building structure from the engineering cases, respectively converting the design conditions and the design schemes of the building structure into a design condition feature tensor and a design scheme feature tensor of the building structure, and constructing a basic data set taking the design condition feature tensor-the design scheme feature tensor of the building structure as a sample;

step 2: training the initial neural network by using a basic data set until a loss function of the initial neural network is stable by using a design condition characteristic tensor of the building structure as input of the initial neural network and using a design scheme characteristic tensor of the building structure as output of the initial neural network;

and step 3: acquiring a design condition feature tensor of the building structure input by the last k rounds of iteration before the initial neural network loss function is stabilized and a corresponding building structure design scheme feature tensor output by the initial neural network, and expanding a basic data set by using a design condition feature tensor-design scheme feature tensor of the building structure formed by the design condition feature tensor-design scheme feature tensor as a sample to obtain an expanded data set;

and 4, step 4: constructing a structure optimization data set and a structure evaluation data set based on the augmentation data set and the building structure optimizer; the structure optimization data set is based on a design condition feature tensor-optimized design scheme feature tensor of the building structure as a sample, and the structure evaluation data set is based on a specified target performance feature tensor corresponding to the design scheme feature tensor-optimized design scheme feature tensor of the building structure as a sample; the optimized design scheme feature tensor is a result of optimizing the design scheme feature tensor by the building structure optimizer;

and 5: training the designated target performance evaluation neural network by using a structure evaluation data set, retraining the initial neural network by using the structure optimization data set, and inputting a design scheme characteristic tensor of the building structure output by the initial neural network in a retraining stage into the designated target performance evaluation neural network to obtain a corresponding designated target performance characteristic tensor;

step 6: if the optimization effect of the initial neural network compared with the previous iteration is judged to be smaller than the expected effect based on the appointed target performance characteristic tensor, finishing training and taking the obtained initial neural network as the building structure design scheme to generate the neural network; otherwise, optimizing the parameters of the initial neural network by using the weighted sum of the loss of the initial neural network and the designated target performance characteristic tensor until the loss function of the initial neural network is stable, and returning to the step 3.

According to the building structure design method based on the combined structure optimization neural network provided by the invention, the construction process of the building structure optimizer comprises the following steps:

setting an optimization object, an optimization output, an optimization target and an optimization algorithm of the building structure optimizer;

grouping the components in the optimized object according to building design experience rules, and setting the constraint which needs to be met by each group of components;

constructing the building structure optimizer based on the set optimization object, the optimization output, the optimization objective, the optimization algorithm and the constraint;

the optimization object is a structural analysis model corresponding to the characteristic tensor of the design scheme of the building structure;

the optimization target is the specified target;

the optimized output is an optimized design scheme of the building structure and a corresponding specified target performance.

According to the building structure design method based on the joint structure optimization neural network provided by the invention, the specified target comprises the following steps: a material cost minimization objective and a carbon emission minimization objective;

the optimization algorithm comprises the following steps: genetic algorithm, tabu search algorithm, simulated annealing algorithm and particle swarm algorithm;

the building design experience rules comprise: symmetry rules and constructability rules;

the constraint, comprising: plane arrangement regularity constraint, vertical arrangement regularity constraint, bearing capacity constraint, rigidity constraint and stability constraint;

wherein the floorplanning regularity constraint comprises: torsion period ratio constraint, displacement ratio constraint under the action of accidental eccentricity earthquake and interlayer displacement ratio constraint under the action of accidental eccentricity earthquake;

the vertical arrangement regularity constraint comprising: floor mass ratio constraint, floor lateral stiffness ratio constraint and floor shear-bearing capacity ratio constraint;

the load bearing restraint includes: constraint of the shear-weight ratio;

the stiffness constraint comprising: restricting the interlayer displacement angle under the action of horizontal force;

the stability constraint comprising: and (4) constraining the rigidity-to-weight ratio.

According to the building structure design method based on the joint structure optimization neural network provided by the invention, the step 4 comprises the following steps:

for each sample of the augmented data set, determining a structural analysis model corresponding to a design scheme feature tensor of the building structure in the sample;

substituting the structural analysis model into the building structure optimizer to correspondingly obtain an optimized design scheme of the building structure and a corresponding specified target performance;

obtaining an optimized design scheme characteristic tensor of a building structure and a corresponding specified target performance characteristic tensor based on an optimized design scheme of the building structure and a corresponding specified target performance, a pre-processor and a post-processor;

taking a design condition feature tensor of an architectural structure and an optimized design scheme feature tensor of the architectural structure in a sample as a sample in the structure optimization data set;

taking the optimized design scheme feature tensor of the building structure and a specified target performance feature tensor corresponding to the optimized design scheme feature tensor of the building structure as a sample in the structure evaluation data set;

and traversing the augmented data set to obtain the structure optimization data set and the structure evaluation data set.

According to the building structure design method based on the joint structure optimization neural network provided by the invention, the determination process of the structural analysis model corresponding to the characteristic tensor of the design scheme of the building structure comprises the following steps:

storing a design solution feature tensor for the architectural structure as a pixel picture format;

converting the design scheme characteristic tensor of the building structure in the pixel picture format into a vector format based on a computer vision algorithm;

converting the design scheme feature tensor of the building structure in a vector format into structured modeling data by adopting an application program interface of structural analysis software;

and processing the structured modeling data by adopting structural analysis software to obtain a structural analysis model corresponding to the design scheme feature tensor of the building structure.

In a second aspect, the present invention further provides an architectural structure design apparatus based on a joint structure optimization neural network, the apparatus including:

the acquisition module is used for acquiring the design conditions of the target building structure;

the substitution module is used for substituting the design conditions into a pre-constructed building structure design scheme generation model to obtain a design scheme of a target building structure;

In a third aspect, the present invention also discloses an electronic device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and when the processor executes the program, the building structure design method based on the joint structure optimization neural network according to the first aspect is implemented.

The invention provides a building structure design method and a building structure design device based on a combined structure optimization neural network, wherein a building structure design scheme generation model is constructed in advance, and the model is constructed with the assistance of a building structure optimizer; the building structure optimizer is generated based on a specified target, a building design empirical rule and an optimization algorithm, and can be used for optimizing a design scheme of a building structure and outputting an optimized design scheme of the building structure and an optimized model of the corresponding specified target performance. According to the method, on one hand, the scale and the quality of a data set of a neural network generated by a building structure design scheme are improved through a building structure optimizer of a specified target in consideration of experience rules, on the other hand, the data set of the neural network for performance evaluation of the specified target is obtained by using the building structure optimizer, the training of the neural network for performance evaluation of the target is realized, then the performance evaluation capability of the specified target on the characteristic tensor of the design scheme is evaluated by using the specified target performance evaluation neural network, and the building structure design scheme is optimized to generate the neural network. In the application stage, the design condition of a target building structure is obtained; substituting the design conditions into a pre-constructed building structure design scheme generation model to obtain a design scheme of a target building structure; and then the intelligent design of the building structure is realized quickly and reliably, and the design result can reach the specified target.

Drawings

In order to more clearly illustrate the technical solutions of the present invention or the prior art, the drawings needed for 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 those skilled in the art can also obtain other drawings according to the drawings without creative efforts.

FIG. 1 is a schematic flow diagram of a method of designing a building structure provided by the present invention;

FIG. 2 is a schematic diagram of a building structure design scenario generation model construction process provided by the present invention;

FIG. 3 is a schematic structural diagram of a design generation model of a building structure provided by the present invention;

FIG. 4 is a schematic diagram of an application of the building structure optimizer provided by the present invention;

FIG. 5 is a schematic diagram of a structural analysis model and corresponding design conditions provided by the present invention;

FIG. 6 is a schematic structural view of a design device for a building structure provided by the present invention;

fig. 7 is a schematic structural diagram of an electronic device implementing a design method of a building structure according to the present invention.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention more apparent, 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 building structure design method and device based on the joint structure optimization neural network of the present invention are described below with reference to fig. 1 to 7.

In a first aspect, the present invention provides a method for designing an architectural structure based on a joint structure optimization neural network, as shown in fig. 1, the method includes:

s11, obtaining the design condition of the target building structure;

s12, substituting the design conditions into a pre-constructed building structure design scheme generation model to obtain a design scheme of a target building structure;

the invention can also process the design scheme of the target building structure into a structure analysis model so as to facilitate the modification and adjustment of the design by a user.

The invention provides a building structure design method based on a combined structure optimization neural network, which is characterized in that a building structure design scheme generation model is constructed in advance, and the model is constructed with the assistance of a building structure optimizer; the building structure optimizer is generated based on a specified target, a building design empirical rule and an optimization algorithm, and can be used for optimizing a design scheme of a building structure and outputting an optimized design scheme of the building structure and an optimized model of the corresponding specified target performance. According to the method, on one hand, the scale and the quality of a data set of a neural network generated by a building structure design scheme are improved through a building structure optimizer of a specified target in consideration of experience rules, on the other hand, the data set of the neural network for performance evaluation of the specified target is obtained by using the building structure optimizer, the training of the neural network for performance evaluation of the target is realized, then the performance evaluation capability of the specified target on the characteristic tensor of the design scheme is evaluated by using the specified target performance evaluation neural network, and the building structure design scheme is optimized to generate the neural network. In the application stage, the design condition of a target building structure is obtained; substituting the design conditions into a pre-constructed building structure design scheme generation model to obtain a design scheme of a target building structure; and then the intelligent design of the building structure is realized quickly and reliably, and the design result can reach the specified target.

On the basis of the above embodiments, as an alternative embodiment, when the design condition is composed of a structural layout and a natural environment factor, the design scheme is a component section size design scheme;

wherein the natural environment factors include: seismic influence coefficient, wind influence coefficient, total structural height and standard layer feature height.

In the technical field of the present invention, design conditions and design solutions (design results) are generally determined according to design tasks, for example: the design task is the design of the cross section size of a building structural member, so that the corresponding design conditions are a structural arrangement diagram and natural environment factors (earthquake resistance, wind resistance and the like), and the design scheme is a member size diagram; similarly, the design task is the arrangement design of building structural components, and the corresponding design conditions are the building design drawing and the natural environment factors, and the design scheme is the component arrangement drawing.

The present embodiment exemplifies two design cases to better explain the design conditions and the design solutions; it will be appreciated that other forms of design in the construction field are equally applicable to the method provided by the present invention.

On the basis of the above embodiments, as an alternative embodiment, the generating a model of the architectural structure design scheme includes: generating a neural network by the pre-processor, the post-processor and the architectural structure design scheme; substituting the design conditions into a pre-constructed building structure design scheme generation model to obtain a design scheme of a target building structure, wherein the design scheme comprises the following steps:

converting, based on a post-processor, the design feature tensor to a design, and taking the converted design as a design for the target architectural structure.

The invention can realize rapid and reliable intelligent design of the building structure, and the design result can reach the specified target.

On the basis of the foregoing embodiments, as an optional embodiment, the generating a model of the architectural structure design scheme further includes: specifying a target performance evaluation neural network; the building structure in this embodiment may be a frame core tube structure. The frame core tube structure is a form of a frame shear wall structure, and is a building structure with various component types and high design difficulty. The frame core tube structure is characterized in that a central core tube is formed by enclosing an elevator shaft, a stair, a ventilation shaft, a cable shaft, a public washroom and partial equipment rooms in the central part of a building, an outer frame inner tube structure is formed by the central core tube and a peripheral frame, and the central core tube structure is poured by reinforced concrete. The structure is very beneficial to structural stress, has excellent shock resistance and is a mainstream structural form widely adopted by super high-rise buildings.

The invention relates to a training process of a building structure design scheme generation model, which mainly comprises the following steps: a first collection stage and a first training stage of the building structure design scheme generation neural network aiming at a basic data set, a second collection stage and a second training stage of the building structure design scheme generation neural network aiming at a structure optimization data set, and an agent collection stage and an agent training stage of the designated target performance evaluation neural network aiming at a structure evaluation data set; the initial architecture for generating the neural network by the architectural structure design scheme in this embodiment may be U-net, which is a network architecture widely used in the field of image segmentation. Its input and output dimensions are defined as 256 × 256 × 5 and 256 × 256 × 1, respectively. The initial architecture of the neural network architecture that specifies the target performance evaluation neural network in this embodiment may be U-net, with input and output dimensions defined as 256 × 256 × 1 and 1 × 1 × 1, respectively. The building structure design scheme generating neural network construction process, as shown in fig. 2, includes:

the following description will be given taking a frame core tube structure as an example:

a first collection phase: first, the present embodiment collects the design data of 52 frame core tube structures, including 276 standard layers. The structures are distributed in different partitions, have different earthquake-resistant and wind-resistant design conditions, and have the structural height ranging from 56.7m to 213.7m and the width ranging from 16.0m to 63.0 m. Then, corresponding design conditions and design schemes are extracted from design data of 52 frame core tube structures, finally, the design conditions and the design schemes are respectively processed into a design condition feature tensor (256 × 256 × 5) and a design scheme feature tensor (256 × 256 × 1), and the tensors are combined into input-label data pairs to construct a basic data set (52 samples). According to the following steps of 3: a scale of 1 divides the data set into a training set (39 samples) and a test set (13 samples).

It should be noted that, in the present embodiment, the basic data may be integrated and multiplied by a data augmentation (rotation and mirroring) mode, if necessary.

a first training stage: carrying out supervised training on the initial neural network by adopting a training set of a basic data set; observing the Loss function of the neural network generated by the architectural structure design scheme, and when the architectural structure design scheme generates the Loss function Loss of the neural network_genAnd stopping training when the training is stable. The stability of the loss function of the neural network generated by the architectural structure design scheme is determined according to working conditions, for example: loss in 10 consecutive rounds of training_genAre all less than 1%, then stable;

and 4, step 4: constructing a structure optimization data set and a structure evaluation data set based on the augmentation data set and the building structure optimizer; the structure evaluation data set is obtained by taking a design condition feature tensor of the building structure and an optimized design scheme feature tensor as samples, and the structure evaluation data set is obtained by taking a specified target performance feature tensor corresponding to the design scheme feature tensor of the building structure and the optimized design scheme feature tensor as samples; the optimized design scheme feature tensor is a result of optimizing the design scheme feature tensor by the building structure optimizer;

second collection phase and agent collection phase: the k of the invention can be freely set, for example, 10; firstly: the tensor of architectural design solution features (256 × 256 × 1, 390 samples) output by the initial neural network in the last 10 rounds of training are collected. And forming an input-label data pair (390 samples) by using the characteristic tensor of the design condition and the characteristic tensor of the architectural structure design scheme output by the initial neural network, and adding the input-label data pair into a basic data set to obtain an augmented data set (429 samples).

And then, constructing a structure optimization data set and a structure evaluation data set by utilizing the augmentation data set and the building structure optimizer.

an agent training stage: the structure assessment data set was as follows 3: a scale of 1 is divided into a training set (965 samples) and a testing set (322 samples). And performing supervised training on the specified target performance evaluation neural network by adopting a training set of a structure evaluation data set. Observing Loss function Loss of the designated target performance evaluation neural network_evalWhen the Loss function Loss of the specified target performance evaluation neural network_evalAnd stopping training when the training is stable. The stability of the loss function of the specified target performance evaluation neural network is determined according to working conditions, such as: loss during 10 consecutive rounds of training_evalThe decrease is less than 1 percent, and then the stability is ensured; and testing the training result of the specified target performance evaluation neural network by adopting a test set of the structure evaluation data set, wherein the average relative error is 8.7 percent and is less than 10 percent of the required error, and the test set can be put into use.

And a second training stage: keeping the parameters of the initial neural network in the first training stage unchanged, and continuing training by adopting a structure optimization data set; in the training process, the initial output data is input into the designated target performance evaluation neural network which is put into use to obtain a designated target performance characteristic tensor, and the designated target performance characteristic tensor and the Loss function of the initial neural network are subjected to weighted addition to obtain a comprehensive Loss function Loss_all；

Loss_all＝Loss_gen+λP_trg

Therein, Loss_genAs a loss function of said initial neural network, P_trgTo specify the target performance feature tensor, λ is the weight.

When the synthetic Loss function of the initial neural network is stable (in 10 consecutive training rounds, Loss)_allAll drops are less than 1%), the training is stopped; the design feature tensor output by the initial neural network in the last 10 rounds of training (256 × 256 × 1, 12870 samples) is collected. And testing the initial neural network by adopting a test set of the basic data set. And evaluating the last test result and the current test result by adopting the designated target performance evaluation neural network, wherein the average designated target performance characteristic tensor of the current test result is lower than the last time (reduced by 3.6%), so that the second collection stage, the agent training stage and the second training stage are repeated. And the second cycle is ended, the average specified target performance characteristic tension of the test result is higher than that of the last time (improved by 0.05%), and therefore the training is ended.

A schematic structural diagram of the generated model of the architectural structural design plan based on the above steps is shown in fig. 3.

According to the method, on one hand, the scale and the quality of a data set of a building structure design scheme generating neural network are improved through a building structure optimizer of a specified target considering experience rules, on the other hand, the data set of the specified target performance evaluation neural network is obtained through the building structure optimizer, the training of the target performance evaluation neural network is realized, then the specified target performance evaluation capability of the specified target performance evaluation neural network on the design scheme feature tensor is utilized, the building structure design scheme is optimized, the neural network is generated, and therefore the building structure design scheme considering the specified target is trained and generated into the neural network.

On the basis of the above embodiments, as an alternative embodiment, the building structure optimizer building process includes:

the optimization target is the specified target;

the optimized output is an optimized design scheme of the building structure and a corresponding specified target performance;

the invention integrates the empirical rules into the building structure optimizer in a mode of grouping the components according to the empirical rules and respectively setting constraints for all groups of components. It can be said that the building structure optimizer takes into account the empirical rules and achieves the optimization of the design scheme based on the specified objective to obtain the optimized design scheme and the corresponding specified objective performance. The invention utilizes the optimization design scheme and the corresponding designated target performance to improve the quality of the neural network data set generated by the building structure design scheme and construct the data set of the designated target performance evaluation neural network, thereby laying a foundation for the optimization of the neural network generated by the building structure design scheme. Generally speaking, the quality of the neural network generated by the architectural structure design scheme is improved in a mode of efficiently matching the optimization algorithm with the neural network training.

On the basis of the foregoing embodiments, as an alternative embodiment, the specifying the target includes: a material cost minimization target and a carbon emission minimization target;

the building design experience rule comprises the following steps: symmetry rules and constructability rules;

the load bearing restraint includes: constraint of the shear-weight ratio;

Assuming that the cross-sectional dimensions of the shear wall and the frame columns are optimized with the aim of minimizing the material cost, the shear wall and the frame columns are grouped according to the empirical rule (the symmetrically arranged frame columns are in the same group, the frame columns are arranged in a collinear manner, and the shear walls which can share the template during construction are in the same group), and the constraints which need to be met by the shear wall and the frame columns respectively are set; then selecting an optimization algorithm to construct a corresponding building structure optimizer; fig. 4 illustrates an application diagram of the building structure optimizer, specifically: and substituting the structural analysis model corresponding to the design scheme corresponding to the optimization of the section sizes of the shear wall and the frame column into a building structure optimizer to obtain a corresponding optimization design result and the material cost thereof. It should be noted that the cross-sectional dimensions of the same set of components are consistent during the optimization process. The material cost in this embodiment may be the sum of the concrete and steel material costs.

The optimization algorithm in this embodiment may be a genetic algorithm, and the basic idea thereof is: the solution to the structural optimization problem is called an individual, which represents a sequence of building block cross-sectional dimensions, called a chromosome; based on the existing structure analysis model, a series of individuals are randomly generated by an algorithm, the fitness of the individuals is evaluated according to the material cost and the compliance, and the individual fitness with compliance and low material cost is high; selecting according to the fitness of the individual, wherein the higher the fitness is, the higher the probability that the individual is selected is; breeding two selected individuals, wherein on one hand, each of the two individuals provides a part of a chromosome, which is called mating, and on the other hand, the chromosomes are randomly changed with a small probability, which is called mutation; through a series of selections, matings, and mutations, new generation individuals are generated, adjusted from generation to generation, and eventually converge to the target solution.

The method provides powerful support for realizing the method by specifically setting the optimization algorithm type, the constraint type and the building design experience rule type.

On the basis of the foregoing embodiments, as an optional embodiment, the step 4 includes:

In the second collection stage of this embodiment, the construction of the structure optimization data set and the structure assessment data set is implemented by using the augmentation data set and the building structure optimizer, which specifically includes:

structural analysis models (429 samples) corresponding to the tag data in the augmented dataset are determined.

Then, substituting the structural analysis model into a pre-constructed building structure optimizer to obtain an optimized design scheme corresponding to the structural analysis model and a corresponding specified target performance (if the specified target is that the material cost is minimum, the specified target performance is the material cost, and if the specified target is that the carbon emission is minimum, the specified target performance is the carbon emission); for each structural analysis model, S design achievements which meet constraint conditions and have the lowest specified target performance are saved (when S takes 3, 1287 samples are total).

Finally, the optimized design scheme and the designated target performance are respectively coded into an optimized design scheme feature tensor (256 multiplied by 1) and a designated target performance feature scalar (1 multiplied by 1); forming an input-label data pair by using the characteristic tensor of the design condition and the characteristic tensor of the optimized design scheme to obtain a structure optimization data set (1287 samples); and (4) forming an input-label data pair by the optimized design scheme characteristic tensor and the designated target performance characteristic scalar to obtain a structure evaluation data set (1287 samples).

The invention utilizes the optimization design scheme and the corresponding designated target performance to improve the quality of the neural network data set generated by the building structure design scheme and construct the data set of the designated target performance evaluation neural network, thereby laying a foundation for the optimization of the neural network generated by the building structure design scheme.

On the basis of the foregoing embodiments, as an optional embodiment, the determining process of the structural analysis model corresponding to the design solution feature tensor of the architectural structure includes:

converting the design scheme feature tensor of the architectural structure in the vector format into structured modeling data by adopting an application program interface of structural analysis software;

Namely: the method comprises the steps of storing label data in an augmentation data set into a pixel picture format (PNG format), converting the label data in the PNG format into a vector format (CSV format) by adopting a computer vision algorithm, converting the label data in the CSV format into structured modeling data by adopting an Application Program Interface (API) of structural analysis software, and opening the modeling data by adopting the structural analysis software to obtain a structural analysis model (429 samples).

The computer vision algorithms in this embodiment may be OpenCV and PIL, which are widely used open source image processing libraries. The structure analysis software can be PKPM software, which is the mainstream software widely used in the field of structure design, and the corresponding modeling data formats are JWS format and BWS format.

The design scheme of the invention is generally stored in the form of drawings, and the final design result is expressed for subsequent construction; the design feature tensor is a centralized representation of the design features; the structural analysis model is used in the design process, is generated, read, modified, etc. by structural analysis software, and is variable, calculable, analyzable. Therefore, the design scheme feature tensor is converted into a structural analysis model, and data reading and processing of a building structure optimizer are facilitated.

The invention can use the trained building structure design scheme to generate the neural network to develop the design of the building structure.

First, the design conditions of the building structure to be designed are obtained. A structural analysis model of a 25-story office building and its design conditions are obtained from a real engineering project, as shown in fig. 5. The cross-sectional dimensions of the wall studs of the tower section (frame core tube structure) are planned to be redesigned, and the other sections are kept unchanged.

And then, processing the design conditions into a design condition feature tensor, inputting the design scheme of the building structure to generate a neural network, obtaining the feature tensor of the design scheme, converting the design scheme feature tensor into structural modeling data, and further obtaining a corresponding structural analysis model.

Then, the PKPM software is used for respectively carrying out structural analysis and material cost statistics on the two structural analysis models before and after redesign. The results show that the design of engineers and the design of intelligent building structures can meet all the constraint conditions. Taking the rigidity requirement as an example, compared with the design of engineers, the maximum interlayer displacement angle of the intelligent building structure design under the action of earthquake and wind is respectively larger by 4.63% and 5.43%, but is lower than the 1/800 limit value of the rigidity requirement, so that the constraint condition is met. In terms of material cost, compared with the design of engineers, the intelligent building structure design is 5.31% less in concrete cost, 3.74% less in steel cost and 4.19% less in total consumption of the concrete and the steel. Therefore, the intelligent building structure design can reduce the material cost on the premise of meeting the constraint condition. Moreover, a skilled engineer generally takes about 30 minutes to design the section size once, while the intelligent building structure design only takes 9 seconds to improve the efficiency by 200 times.

In a second aspect, the building structure design device based on the joint structure optimization neural network provided by the present invention is described, and the building structure design device based on the joint structure optimization neural network described below and the building structure design method based on the joint structure optimization neural network described above can be referred to correspondingly. Fig. 6 illustrates a schematic structural diagram of an architectural structural design apparatus based on a joint structural optimization neural network, as shown in fig. 6, the apparatus includes: an acquisition module 21 and a substitution module 22;

an obtaining module 21, configured to obtain design conditions of a target building structure;

a substituting module 22, configured to substitute the design condition into a pre-constructed building structure design scheme generation model to obtain a design scheme of a target building structure;

The invention provides a building structure design device based on a combined structure optimization neural network, which is characterized in that a building structure design scheme generation model is constructed in advance, and the model is constructed with the assistance of a building structure optimizer; the building structure optimizer is generated based on a specified target, a building design empirical rule and an optimization algorithm, and can be used for optimizing a design scheme of a building structure and outputting an optimized design scheme of the building structure and an optimized model of the corresponding specified target performance. According to the method, on one hand, the scale and the quality of a data set of a neural network generated by a building structure design scheme are improved through a building structure optimizer of a specified target considering experience rules, on the other hand, the data set of the neural network for performance evaluation of the specified target is obtained by using the building structure optimizer, the training of the neural network for performance evaluation of the target is realized, then, the performance evaluation capability of the specified target on the characteristic tensor of the design scheme is evaluated by using the specified target performance evaluation neural network, and the building structure design scheme is optimized to generate the neural network. In the application stage, the design condition of a target building structure is obtained; substituting the design conditions into a pre-constructed building structure design scheme generation model to obtain a design scheme of a target building structure; and then the intelligent design of the building structure is realized quickly and reliably, and the design result can reach the specified target.

when the design condition is composed of a building design drawing and a natural environment factor, the design scheme is a component arrangement design scheme;

On the basis of the above embodiments, as an alternative embodiment, the generating a model of the architectural structure design scheme includes: generating a neural network by a pre-processor, a post-processor and a design scheme of the building structure; the substituting module comprises:

a design condition feature tensor generation unit for processing the design condition into a design condition feature tensor based on a pre-processor;

the design scheme feature tensor generating unit is used for inputting the design condition feature tensor to the building structure design scheme generation neural network to obtain a design scheme feature tensor output by the building structure design scheme generation neural network;

and the design scheme generation unit is used for converting the design scheme characteristic tensor into the design scheme based on the post processor and taking the converted design scheme as the design scheme of the target building structure.

On the basis of the foregoing embodiments, as an optional embodiment, the generating a model of the architectural structure design scheme further includes: specifying a target performance evaluation neural network; the device further comprises: a first building block for pre-building an architectural structure design to generate a neural network, the first building block comprising:

the basic data set generating unit is used for collecting the design conditions and the design scheme of the building structure from the engineering cases, respectively converting the design conditions and the design scheme of the building structure into a design condition feature tensor and a design scheme feature tensor of the building structure, and constructing a basic data set taking the design condition feature tensor-the design scheme feature tensor of the building structure as a sample;

the building structure design scheme generation neural network first-stage training unit is used for training the initial neural network by using a design condition feature tensor of a building structure as input of the initial neural network and using a design scheme feature tensor of the building structure as output of the initial neural network and by using basic data set until a loss function of the initial neural network is stable;

the augmented data set generating unit is used for acquiring a design condition feature tensor of the building structure input by the last k rounds of iteration before the initial neural network loss function is stabilized and a building structure design scheme feature tensor output by the initial neural network corresponding to the design condition feature tensor, and expanding a basic data set by taking the design condition feature tensor-design scheme feature tensor of the building structure formed by the design condition feature tensor and the design scheme feature tensor as a sample to obtain an augmented data set;

a structure optimization data set construction unit for constructing a structure optimization data set and a structure evaluation data set based on the augmentation data set and the building structure optimizer; the structure evaluation data set is obtained by taking a design condition feature tensor of the building structure and an optimized design scheme feature tensor as samples, and the structure evaluation data set is obtained by taking a specified target performance feature tensor corresponding to the design scheme feature tensor of the building structure and the optimized design scheme feature tensor as samples; the optimized design scheme feature tensor is a result of optimizing the design scheme feature tensor by the building structure optimizer;

a second-stage training unit of the building structure design scheme generation neural network, which is used for retraining the initial neural network by using the structure optimization data set and inputting the design scheme feature tensor of the building structure output by the initial neural network in the retraining stage into the specified target performance evaluation neural network to obtain a corresponding specified target performance feature tensor;

a building structure design scheme generation neural network generation unit, configured to, if it is determined that an optimization effect of the initial neural network compared to a previous iteration is smaller than an expected effect based on the specified target performance feature tensor, end training and use the obtained initial neural network as the building structure design scheme generation neural network; otherwise, optimizing the parameters of the initial neural network by using the weighted sum of the loss of the initial neural network and the designated target performance characteristic tensor until the loss function of the initial neural network is stable, and returning to the augmented data set generating unit.

On the basis of the foregoing embodiments, as an optional embodiment, the apparatus further includes: a second building module for pre-building a building structure optimizer, the second building module comprising:

the first setting unit is used for setting an optimization object, an optimization output, an optimization target and an optimization algorithm of the building structure optimizer;

the component grouping and constraint setting unit is used for grouping the components in the optimized object according to the building design experience rule and setting the constraint which needs to be met by each group of components;

a construction unit configured to construct the building structure optimizer based on the set optimization object, the optimization output, the optimization objective, the optimization algorithm, and the constraint;

the optimization target is the specified target;

the load bearing restraint includes: limiting the shear-weight ratio;

On the basis of the foregoing embodiments, as an optional embodiment, the structure optimization data set and structure evaluation data set construction unit includes:

the structural analysis model determining submodule is used for determining a structural analysis model corresponding to the design scheme feature tensor of the building structure in each sample of the augmented data set;

the optimization design scheme and the corresponding designated target performance determining submodule are used for substituting the structural analysis model into the building structure optimizer to correspondingly obtain the optimization design scheme and the corresponding designated target performance of the building structure;

the characteristic tensor generation submodule is used for obtaining an optimized design scheme characteristic tensor of the building structure and a corresponding specified target performance characteristic tensor based on an optimized design scheme of the building structure and a corresponding specified target performance, a pre-processor and a post-processor;

a first setting submodule, configured to use a design condition feature tensor of an architectural structure and an optimized design scheme feature tensor of the architectural structure in a sample as a sample in the structure optimization dataset;

a second setting submodule, configured to use the optimized design solution feature tensor of the architectural structure and a specified target performance feature tensor corresponding to the optimized design solution feature tensor of the architectural structure as a sample in the structure evaluation dataset;

and the traversal submodule is used for traversing the augmentation data set to obtain the structure optimization data set and the structure evaluation data set.

On the basis of the foregoing embodiments, as an optional embodiment, the structural analysis model determining sub-module includes:

a first format conversion subunit, configured to store a design solution feature tensor of the building structure as a pixel picture format;

the second format conversion subunit is used for converting the design scheme characteristic tensor of the building structure in the pixel picture format into a vector format based on a computer vision algorithm;

the modeling data determining subunit is used for converting the design scheme feature tensor of the building structure in the vector format into structured modeling data by adopting an application program interface of structural analysis software;

and the structural analysis model obtaining subunit is used for processing the structured modeling data by adopting structural analysis software to obtain a structural analysis model corresponding to the design scheme feature tensor of the building structure.

In a third aspect, fig. 7 illustrates a schematic physical structure diagram of an electronic device, and as shown in fig. 7, the electronic device may include: a processor (processor)710, a communication Interface (Communications Interface)720, a memory (memory)730, and a communication bus 740, wherein the processor 710, the communication Interface 720, and the memory 730 communicate with each other via the communication bus 740. Processor 710 may invoke logic instructions in memory 730 to perform a method for building structure design based on joint structure optimization neural networks, the method comprising: acquiring design conditions of a target building structure; substituting the design conditions into a pre-constructed building structure design scheme generation model to obtain a design scheme of a target building structure; the building structure design scheme generation model is constructed with the assistance of a building structure optimizer; the building structure optimizer is generated based on a specified target, a building design empirical rule and an optimization algorithm, and can be used for optimizing a design scheme of a building structure and outputting an optimized design scheme of the building structure and an optimized model of the corresponding specified target performance.

In addition, the logic instructions in the memory 730 can be implemented in the form of 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 or a part thereof which substantially contributes to the prior art may be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device) 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, storable on a non-transitory computer readable storage medium, which when executed by a processor performs a method for building structure design based on joint structure optimization neural network, the method comprising: acquiring design conditions of a target building structure; substituting the design conditions into a pre-constructed building structure design scheme generation model to obtain a design scheme of a target building structure; the building structure design scheme generation model is constructed with the assistance of a building structure optimizer; the building structure optimizer is generated based on a specified target, a building design empirical rule and an optimization algorithm, and can be used for optimizing a design scheme of a building structure and outputting an optimized design scheme of the building structure and an optimized model of the corresponding specified target performance.

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 for building structure design based on joint structure optimization neural network, the method comprising: acquiring design conditions of a target building structure; substituting the design conditions into a pre-constructed building structure design scheme generation model to obtain a design scheme of a target building structure; the building structure design scheme generation model is constructed with the assistance of a building structure optimizer; the building structure optimizer is generated based on a specified target, a building design empirical rule and an optimization algorithm, and can be used for optimizing a design scheme of a building structure and outputting an optimized design scheme of the building structure and an optimized model of the corresponding specified target performance.

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 for designing an architectural structure based on a joint structure optimization neural network, the method comprising:

acquiring design conditions of a target building structure;

2. The method for designing an architectural structure based on a joint structure optimization neural network according to claim 1, wherein when the design condition is composed of a structural layout drawing and a natural environment factor, the design scheme is a component section size design scheme;

3. The method for designing an architectural structure based on a joint structure optimization neural network according to claim 1 or 2, wherein the architectural structure design scheme generation model comprises: generating a neural network by the pre-processor, the post-processor and the architectural structure design scheme; substituting the design conditions into a pre-constructed building structure design scheme generation model to obtain a design scheme of a target building structure, wherein the design scheme comprises the following steps:

4. The method of claim 3, wherein the architectural structure design solution generates a model, further comprising: specifying a target performance evaluation neural network; the building structure design scheme generation neural network construction process comprises the following steps:

and step 3: acquiring a design condition feature tensor of the building structure input by the last k iterations before the initial neural network loss function is stabilized and a building structure design scheme feature tensor output by the initial neural network corresponding to the design condition feature tensor, and expanding a basic data set by using the design condition feature tensor-design scheme feature tensor of the building structure formed by the design condition feature tensor-design scheme feature tensor as a sample to obtain an expanded data set;

5. The method for designing an architectural structure based on a joint structure optimization neural network according to claim 4, wherein the building structure optimizer comprises the following steps:

the optimization target is the specified target;

6. The method of claim 5, wherein the assigning the target comprises: a material cost minimization target and a carbon emission minimization target;

the load bearing restraint includes: constraint of the shear-weight ratio;

the stiffness constraint comprising: interlayer displacement angle constraint under the action of horizontal force;

7. The method for designing an architectural structure based on a joint structure optimization neural network according to claim 4, wherein the step 4 comprises:

8. The method for designing an architectural structure based on a joint structure optimization neural network according to claim 7, wherein the process for determining the structural analysis model corresponding to the design solution feature tensor of the architectural structure comprises:

9. An apparatus for designing an architectural structure based on a joint structure optimization neural network, the apparatus comprising:

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 the method for architectural structure design based on joint structure optimization neural networks of any of claims 1 to 8.