WO2021238113A1 - 基于对抗生成网络的剪力墙结构布置方法和装置 - Google Patents

基于对抗生成网络的剪力墙结构布置方法和装置 Download PDF

Info

Publication number
WO2021238113A1
WO2021238113A1 PCT/CN2020/131065 CN2020131065W WO2021238113A1 WO 2021238113 A1 WO2021238113 A1 WO 2021238113A1 CN 2020131065 W CN2020131065 W CN 2020131065W WO 2021238113 A1 WO2021238113 A1 WO 2021238113A1
Authority
WO
WIPO (PCT)
Prior art keywords
structural
image
shear wall
trained
architectural
Prior art date
Application number
PCT/CN2020/131065
Other languages
English (en)
French (fr)
Inventor
陆新征
廖文杰
郑哲
Original Assignee
清华大学
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 清华大学 filed Critical 清华大学
Publication of WO2021238113A1 publication Critical patent/WO2021238113A1/zh
Priority to US17/697,660 priority Critical patent/US20220207205A1/en

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/10—Geometric CAD
    • G06F30/13—Architectural design, e.g. computer-aided architectural design [CAAD] related to design of buildings, bridges, landscapes, production plants or roads
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0475—Generative networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/094—Adversarial learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T11/00—2D [Two Dimensional] image generation
    • G06T11/40—Filling a planar surface by adding surface attributes, e.g. colour or texture
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
    • G06F2119/02—Reliability analysis or reliability optimisation; Failure analysis, e.g. worst case scenario performance, failure mode and effects analysis [FMEA]

Definitions

  • This application relates to the technical field of construction and civil structure engineering, and in particular to a method and device for arranging a shear wall structure based on a confrontation generation network.
  • a good preliminary design plan of the structure can assist the later deepening design of the building plan and the structure plan.
  • the current structural design methods that rely on manual experience are time-consuming and low in efficiency, which tends to reduce the efficiency of interaction design between construction engineers and structural engineers, and manual design relies on experience, which leads to certain differences in the design results of different designers, and it is difficult to make full use of the existing
  • the variability of design results is relatively large.
  • the existing computer-aided structural design optimization method consumes large computational resources, takes a long time, and is difficult to effectively apply the existing mature design results, which makes it difficult to meet the rapid design requirements in the preliminary design stage of the structure.
  • This application aims to solve one of the technical problems in the related technology at least to a certain extent.
  • one purpose of this application is to propose a shear wall structure layout method based on the confrontation generation network, which can quickly output the corresponding shear wall structure design according to the standard floor plan in the architectural design, and realize the corresponding architectural design Quick design of shear wall structure.
  • Another purpose of the present application is to provide a shear wall structure arrangement device based on a confrontation generation network.
  • the embodiment of the first aspect of the present application proposes a shear wall structure arrangement method based on a confrontation generation network, including:
  • the embodiment of the second aspect of the present application proposes a shear wall structure arrangement device based on a confrontation generation network, including:
  • the first acquisition module is used to acquire the architectural design drawings to be processed
  • the first extraction module is used to extract the key elements in the architectural design drawings, and fill the key elements with different colors to generate the image features to be input;
  • the generating module is used for inputting the features of the image to be input into the pre-trained structural confrontation generating network model for processing, and generating structural design drawings.
  • An embodiment of the third aspect of the present application proposes an electronic device, including: a memory, a processor, and a computer program stored on the memory and capable of running on the processor, and when the processor executes the computer program , To implement the shear wall structure arrangement method based on the confrontation generation network described in the embodiment of the first aspect of the present application.
  • the embodiment of the fourth aspect of the present application proposes a computer-readable storage medium on which a computer program is stored.
  • the computer program When the computer program is executed by a processor, it realizes the scissor based on the confrontation generation network described in the embodiment of the first aspect of the present application.
  • the layout method of the force wall structure When the computer program is executed by a processor, it realizes the scissor based on the confrontation generation network described in the embodiment of the first aspect of the present application.
  • FIG. 1 is a schematic flowchart of a method for arranging a shear wall structure based on a confrontation generation network provided by an embodiment of the application;
  • FIG. 2 is a detailed frame diagram of a shear wall structure arrangement method based on a confrontation generation network provided by an embodiment of the application;
  • FIG. 3 is a set of typical training-test image sets provided by an embodiment of this application.
  • FIG. 4 is a training process diagram of the confrontation generation network algorithm provided by an embodiment of the application.
  • FIG. 5 is a diagram of the results of extracting key elements in the generated image and the target image provided by an embodiment of the application;
  • FIG. 6 is a diagram of the evaluation method of the shear wall area intersection ratio (SIoU) provided by the embodiments of the application;
  • FIG. 7 is a plan design diagram of a standard floor of a shear wall structure designed by a structural engineer provided by an embodiment of the application;
  • FIG. 8 is a design diagram of a shear wall output by StructGAN (Structural Adversarial Generation Network Model) provided by an embodiment of the application and a comparison diagram with the result designed by an engineer;
  • StructGAN Structuretural Adversarial Generation Network Model
  • FIG. 9 is a comparison diagram of calculated interlayer displacement angles between the StructGAN (Structural Adversarial Generation Network Model) design provided by an embodiment of the application and the structural model designed by the engineer;
  • StructGAN Structuretural Adversarial Generation Network Model
  • FIG. 10 is a schematic structural diagram of a device for arranging a shear wall structure based on a confrontation generation network provided by an embodiment of the application.
  • FIG. 1 is a schematic flowchart of a method for arranging a shear wall structure based on a confrontation generation network provided by an embodiment of the application.
  • the method includes the following steps:
  • Step 101 Acquire architectural design drawings to be processed.
  • Step 102 Extract key elements in the architectural design drawings, and perform different color filling processing on the key elements to generate image features to be input.
  • the key elements in the architectural design drawings are extracted, and the key elements are filled with different colors to generate the image features to be input, including: extracting shear walls, ordinary infill walls, and indoor doors and windows in the architectural design drawings
  • Four key elements openings and outdoor door openings; different colors are used to fill the key elements. Among them, red represents shear walls, gray represents ordinary infill walls, green represents indoor door and window openings, and blue represents outdoor door openings.
  • Step 103 Input the features of the image to be input into the pre-trained structural confrontation generation network model for processing, and generate a structural design drawing.
  • the method before inputting the features of the input image into the pre-trained structural confrontation generation network model for processing and generating the structural design drawings, the method further includes: obtaining a sample of the architectural design drawing and its supporting structural design drawing sample; Extract the architectural sample elements of the architectural design drawing sample, and fill the architectural sample elements with different colors to generate the architectural image features to be trained; extract the structural sample elements of the structural design drawing sample, and fill the structural sample elements with different colors to generate
  • the image features of the structure to be trained; the image features of the building to be trained and the image features of the structure to be trained are divided into a training set and a test set; training is performed according to the training set input to the confrontation generation network, and the structural confrontation generation network model is obtained after the training is completed.
  • the image features of the building to be trained and the image features of the structure to be trained are divided into a training set and a test set, including: dividing the image features of the building to be trained and the image features of the structure to be trained into a training set according to the height of the structure and the seismic design of the structure And test set.
  • test set tests and evaluates the structural confrontation generative network model.
  • the test and evaluation of the structural confrontation generation network model according to the test set includes: classifying each pixel of the generated image and the target image according to the color range; classifying each pixel in the generated image and the target image; and classifying the classified pixels
  • the points are made into a confusion matrix, and the first evaluation index is calculated based on the confusion matrix.
  • the weighting coefficient is determined according to the first evaluation index and the second evaluation index; the evaluation is performed based on the weighted comprehensive index.
  • Fig. 2 is a detailed frame diagram of a method for arranging a shear wall structure based on a confrontation generation network provided by an embodiment of the present application.
  • the division and arrangement of the data set is a typical training-test image set after collection and semantics.
  • the design results are divided into two groups: Level 1 (height lower than 50m) and Level 2 (height higher than 50m, lower than 140m) according to the height, and the drawings are further divided into 7 according to the seismic design and waterproof criteria.
  • the structural design drawings are finally divided into three categories as a training set, namely Level 1-7 degree fortification (L1-7), Level 2-7 degree fortification (L2-7), and 8 degree fortification (L1&2-8).
  • L1-7 Level 1-7 degree fortification
  • L2--7 Level 2-7 degree fortification
  • L1&2-8 8 degree fortification
  • the selection and training of the GAN algorithm is the training process diagram of the confrontation generative network algorithm.
  • Applicable GAN algorithms include two types, pix2pix and pix2pixHD.
  • the key parameters of the pix2pix algorithm are ⁇ GAN and ⁇ L1.
  • the local and overall effects of the image are determined by the relative values of the two parameters. The larger the relative value of ⁇ GAN, the better the local effect of the generated image. The larger the relative value of ⁇ L1, the generated image The overall effect is better.
  • the key parameter of the pix2pixHD algorithm is ⁇ FM.
  • the ⁇ FM parameter is used to adjust the proportion of the loss of the generated image feature matching in the overall loss, which in turn affects the overall quality of the generated image.
  • the ⁇ FM parameter is used to adjust the proportion of the loss of the generated image feature matching in the overall loss, which in turn affects the overall quality of the generated image.
  • the ⁇ FM parameter is used to adjust the proportion of the loss of the generated image feature matching in the overall loss, which in turn affects the overall quality of the generated image.
  • Evaluation method 1 Evaluation method based on semantic classification of image pixels. Judge the color range of each pixel of the generated image, determine the key element it represents according to the color, combine the pixel classification result of the generated image and the pixel classification result of the target image into a confusion matrix for evaluation .
  • the specific evaluation process is as follows:
  • (k+1) is the total category of pixels (category 0 is background, category 1 is shear wall, category 2 is infill wall, category 3 is indoor door and window opening, category 4 is outdoor door opening).
  • p ij is the number of pixels of category i that are judged to be category j, that is, p ii represents the number of pixels that are generated correctly, and p ij and p ji are the number of pixels that generate errors.
  • a swall is the area of the shear wall, and A inwall is the area of the infill wall.
  • Evaluation method 2 Evaluation method based on target detection and combination of shear wall graphics. By detecting the edge of the shear wall in the generated image and the edge of the shear wall in the target image, and then solving the intersection area of the contour edge and the total union area, the evaluation of the generation of the shear wall is obtained. If the generated shear wall completely matches the target shear wall, the result is 1, and if it does not match at all, the result is 0. As shown in Figure 6, the specific steps are as follows:
  • a inter is the intersection area of the generated shear wall and the target shear wall
  • a union is the union area of the generated shear wall and the target shear wall
  • a union A tar + A out- A inter
  • a tar is The area of the shear wall of the target image
  • a out is the area of the shear wall of the generated image.
  • Evaluation method three weighted comprehensive evaluation based on multiple indicators, fusing evaluation method one and evaluation method two to obtain a comprehensive evaluation index, and the IoU comprehensive score is calculated by formula (4).
  • Score IoU ( ⁇ SWratio ⁇ ( ⁇ SIoU ⁇ SIoU+ ⁇ WIoU ⁇ WIoU)) (4)
  • ⁇ SWratio 1-
  • /SWratio tar , SWratio out and SWratio tar are the SWratio of the generated image and the target image respectively;
  • ⁇ SIoU is the SIoU value coefficient, taking the value 0.5;
  • ⁇ SIoU is the WIoU value The coefficient is 0.5.
  • the trained StructGAN model is tested using the corresponding test data set, and the evaluation method proposed by the present invention is used to carry out the evaluation.
  • the final evaluation result is shown in Table 3. Among them, when the weighted comprehensive evaluation index of multiple indexes exceeds 0.5, the generation result can be considered to be good, and the StructGAN model can be applied.
  • the newly designed architectural drawings are semantically input into the qualified StructGAN model to generate the corresponding shear wall design.
  • the new design adopts the engineer’s design and the StructGAN-based rapid design method of shear wall structure layout proposed by the present invention, and then compares the dynamic characteristics and design results of the two design results to verify the safety and rationality of the design results based on this method sex.
  • the structure is a 38-story shear wall residential building, with a total height of 103m, of which the standard floors are 5-36 floors, and the floor heights are all 2.9m.
  • the second group of seismic fortification group, fortification intensity Intensity 7 (0.15g), site category II, shear wall seismic grade 2, and the design result of the standard floor of the structure is shown in Figure 7.
  • the method proposed by the present invention is used to generate the shear wall design results of the standard floor plan, as shown in Fig. 8.
  • the design results obtained by StructGAN are evaluated, and the evaluation results are shown in Table 4.
  • the structural engineer design model is adjusted accordingly based on the shear wall structure design drawing generated by StructGAN.
  • the process of converting the semantic drawing into the structural model is as follows:
  • Model adjustment follows the following principles: The shear wall generated by StructGAN has some pixels missing. In this case, it is considered that as long as there are shear wall pixels within this length, the shear wall is arranged; if the length of the shear wall is shorter than the wall thickness (200mm) , Then discard the wall; only adjust the wall length design of the original structure without changing its thickness and material and other attributes; the beam after the wall is shortened is lengthened accordingly.
  • Figure 8 shows the detailed comparison between the results of the engineer design and the shear wall layout designed by StructGAN. It can be seen that: 1) The shear wall designed by StructGAN has a high degree of overlap with the shear wall designed by the structural engineer; 2) The shear wall generated by StructGAN The layout of the force wall is relatively discrete, and there are some short-leg shear walls. 3) From the perspective of the number and distribution of walls as a whole, there is no obvious error in the generated results of StructGAN, which can basically meet the needs of the design. Subsequently, the engineer design model and the StructGAN design model were calculated using PKPM software, and the actual calculation results were compared.
  • the comparison results of the overall indicators of the PKPM calculation results are shown in Table 5 and Figure 9. From Table 5, it can be seen that the model designed by StructGAN is basically the same as the model designed by the engineer in terms of overall quality or power characteristics.
  • the key indicator of structural lateral force design is the envelope value of the displacement angle between the design layers.
  • the difference between the StructGAN model and the engineer's design model is basically about 10%. Therefore, the StructGAN model is only used as the initial design, and the difference between the final perfect design result is only about 10%, and the acceptable difference range indicates the reliability of the design result of this method.
  • this application also proposes a shear wall structure arrangement device based on a confrontation generation network.
  • FIG. 10 is a schematic structural diagram of a shear wall structure arrangement device based on a confrontation generation network provided by an embodiment of the application.
  • the device includes: a first acquisition module 1001, a first extraction module 1002, and a generation module 1003.
  • the first obtaining module 1001 is used to obtain the architectural design drawings to be processed
  • the first extraction module 1002 is used to extract key elements in the architectural design drawings, and perform different color filling processing on the key elements to generate image features to be input;
  • the generating module 1003 is used for inputting the features of the image to be input into the pre-trained structural confrontation generating network model for processing, and generating structural design drawings.
  • the device further includes: a second acquisition module for acquiring a sample of architectural design drawings and a sample of structural design drawings matching the sample of architectural design drawings; and a second extraction module , Used to extract the architectural sample elements of the architectural design drawing sample, and perform different color filling processing on the architectural sample elements to generate the architectural image features to be trained; the third extraction module is used to extract the structural design drawing sample Structural sample elements, and perform different color filling processing on the structural sample elements to generate structural image features to be trained; a division module for dividing the architectural image features to be trained and the structural image features to be trained into training sets and Test set; training module, used to train according to the training set input confrontation generation network, and the structured confrontation generation network model is obtained after training; evaluation module, used to evaluate the design effect of the generated structure confrontation generation network model , Models that are tested and qualified based on the three evaluation methods can be put into application.
  • a second acquisition module for acquiring a sample of architectural design drawings and a sample of structural design drawings matching the sample of architectural design drawings
  • the to-be-processed architectural design drawing is obtained; the key elements in the architectural design drawing are extracted, and the key elements are filled with different colors to generate the image to be input
  • an embodiment of the present application proposes an electronic device, including: a memory, a processor, and a computer program stored in the memory and running on the processor.
  • the processor executes the program, The method for arranging a shear wall structure based on a confrontation generation network as described in the foregoing terminal device execution method embodiment.
  • the embodiment of the present application proposes a computer-readable storage medium on which a computer program is stored.
  • the program When the program is executed by a processor, it realizes the shear force based on the confrontation generation network described in the foregoing method embodiment.
  • Wall structure layout method When the program is executed by a processor, it realizes the shear force based on the confrontation generation network described in the foregoing method embodiment.
  • first and second are only used for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features. Therefore, the features defined with “first” and “second” may explicitly or implicitly include at least one of the features. In the description of the present application, "a plurality of” means at least two, such as two, three, etc., unless specifically defined otherwise.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Geometry (AREA)
  • General Engineering & Computer Science (AREA)
  • Computer Hardware Design (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • Computing Systems (AREA)
  • Molecular Biology (AREA)
  • General Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Data Mining & Analysis (AREA)
  • Mathematical Physics (AREA)
  • Computational Linguistics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Biophysics (AREA)
  • Health & Medical Sciences (AREA)
  • Architecture (AREA)
  • Civil Engineering (AREA)
  • Structural Engineering (AREA)
  • Computational Mathematics (AREA)
  • Mathematical Analysis (AREA)
  • Mathematical Optimization (AREA)
  • Pure & Applied Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Image Analysis (AREA)

Abstract

一种基于对抗生成网络的剪力墙布置方法和装置，其中，方法包括：获取待处理的建筑设计图纸（S101）；提取建筑设计图纸中的关键元素，并对关键元素进行不同颜色填充处理，生成待输入图像特征（S102）；将待输入图像特征输入预先训练的结构对抗生成网络模型进行处理，生成结构设计图纸（S103）。由此，能够根据建筑设计中的标准层平面图纸，快速输出对应的剪力墙结构设计，实现建筑设计对应的剪力墙结构快速设计。

Description

基于对抗生成网络的剪力墙结构布置方法和装置
相关申请的交叉引用
本申请要求清华大学于2020年05月25日提交的、发明名称为“基于对抗生成网络的剪力墙结构布置方法和装置”的、中国专利申请号“CN202010446468.9”的优先权。
技术领域
本申请涉及建筑与土木结构工程技术领域，特别涉及一种基于对抗生成网络的剪力墙结构布置方法和装置。
背景技术
在高层剪力墙住宅建筑方案设计时、以及结构初始设计时，为保证最终设计结果的安全性与合理性，需在建筑平面图纸的基础上进行快速合理的剪力墙结构构件初步设计。
良好的结构初步设计方案，可辅助建筑方案与结构方案的后期深化设计。但目前依赖人工经验的结构设计方法时间长、效率低，易降低建筑工程师与结构工程师之间的交互设计效率，并且人工设计依赖经验导致不同设计人员的设计结果存在一定差异，难以充分利用既有的资源，设计成果变异性较大。同时，现有基于计算机辅助的结构设计优化方法计算资源消耗大，耗时长，且难以有效应用既有的成熟设计结果，导致结构初步设计阶段的快速设计需求难以满足。
发明内容
本申请旨在至少在一定程度上解决相关技术中的技术问题之一。
为此，本申请的一个目的在于提出一种基于对抗生成网络的剪力墙结构布置方法,能根据建筑设计中的标准层平面图纸，快速输出对应的剪力墙结构设计，实现建筑设计对应的剪力墙结构快速设计。
本申请的另一个目的在于提出一种基于对抗生成网络的剪力墙结构布置装置。
为达到上述目的，本申请第一方面实施例提出了一种基于对抗生成网络的剪力墙结构布置方法，包括：
获取待处理的建筑设计图纸；
提取建筑设计图纸中的关键元素，并对关键元素进行不同颜色填充处理，生成待输入图像特征；
将待输入图像特征输入预先训练的结构对抗生成网络模型进行处理，生成结构设计图纸。
本申请第二方面实施例提出了一种基于对抗生成网络的剪力墙结构布置装置，包括：
第一获取模块，用于获取待处理的建筑设计图纸；
第一提取模块，用于提取建筑设计图纸中的关键元素，并对关键元素进行不同颜色填充处理，生成待输入图像特征；
生成模块，用于将待输入图像特征输入预先训练的结构对抗生成网络模型进行处理，生成结构设计图纸。
本申请第三方面实施例提出了一种电子设备，包括：存储器、处理器及存储在所述存储器上并可在所述处理器上运行的计算机程序，所述处理器执行所述计算机程序时，实现本申请第一方面实施例所述的基于对抗生成网络的剪力墙结构布置方法。
本申请第四方面实施例提出了一种计算机可读存储介质，其上存储有计算机程序，所述计算机程序被处理器执行时实现本申请第一方面实施例所述的基于对抗生成网络的剪力墙结构布置方法。
本申请实施例所提供的技术方案可以包含如下的有益效果：
通过获取待处理的建筑设计图纸；提取建筑设计图纸中的关键元素，并对关键元素进行不同颜色填充处理，生成待输入建筑图像特征；将待输入建筑图像特征输入预先训练的结构对抗生成网络模型进行处理，生成对应结构设计图纸。由此，能够根据建筑设计中的标准层平面图纸，快速输出对应的剪力墙结构设计，实现建筑设计对应的剪力墙结构快速设计。
附图说明
本申请上述的和/或附加的方面和优点从下面结合附图对实施例的描述中将变得明显和容易理解，其中：
图1为本申请实施例所提供的一种基于对抗生成网络的剪力墙结构布置方法的流程示意图；
图2为本申请实施例所提供的基于对抗生成网络的剪力墙结构布置方法的详细框架图；
图3为本申请实施例所提供的一组典型训练-测试图像集；
图4为本申请实施例所提供的对抗生成网络算法的训练过程图；
图5为本申请实施例所提供的生成图像与目标图像中关键元素分别提取的结果图；
图6为本申请实施例所提供的剪力墙面积交并比(SIoU)的评价方法图；
图7为本申请实施例所提供的结构工程师设计的剪力墙结构标准层平面设计图；
图8为本申请实施例所提供的StructGAN(结构对抗生成网络模型)输出的剪力墙设计图以及与工程师设计的结果对比图；
图9为本申请实施例所提供的StructGAN(结构对抗生成网络模型)设计与工程师设计的结构模型的计算层间位移角对比图；
图10为本申请实施例所提供的一种基于对抗生成网络的剪力墙结构布置装置的结构示意图。
具体实施方式
下面详细描述本申请的实施例，所述实施例的示例在附图中示出，其中自始至终相同或类似的标号表示相同或类似的元件或具有相同或类似功能的元件。下面通过参考附图描述的实施例是示例性的，旨在用于解释本申请，而不能理解为对本申请的限制。
下面参考附图描述本申请实施例的基于对抗生成网络的剪力墙结构布置方法和装置。
图1为本申请实施例所提供的一种基于对抗生成网络的剪力墙结构布置方法的流程示意图。
如图1所示，该方法包括以下步骤：
步骤101，获取待处理的建筑设计图纸。
步骤102，提取建筑设计图纸中的关键元素，并对关键元素进行不同颜色填充处理，生成待输入图像特征。
在本申请实施例中，提取建筑设计图纸中的关键元素，并对关键元素进行不同颜色填充处理，生成待输入图像特征，包括：提取建筑设计图纸中的剪力墙、普通填充墙、室内门窗洞口和室外门洞四个关键元素；采用不同颜色对关键元素进行填充，其中，红色代表剪力墙，灰色代表普通填充墙，绿色代表室内门窗洞口，蓝色代表室外门洞。
步骤103，将待输入图像特征输入预先训练的结构对抗生成网络模型进行处理，生成结构设计图纸。
在本申请的一个实施例中，在将待输入图像特征输入预先训练的结构对抗生成网络模型进行处理，生成结构设计图纸之前，还包括：获取建筑设计图纸样本及其配套的结构设计图纸样本；提取建筑设计图纸样本的建筑样本元素，并对建筑样本元素进行不同颜色填充处理，生成待训练建筑图像特征；提取结构设计图纸样本的结构样本元素，并对结构样本元素进行不同颜色填充处理，生成待训练结构图像特征；将待训练建筑图像特征和待训练结构图像特征划分为训练集和测试集；根据训练集输入对抗生成网络进行训练，训练完成得到结构对抗生成网络模型。
进一步地,将待训练建筑图像特征和待训练结构图像特征划分为训练集和测试集，包括：根据按照结构高度和结构抗震设防水准将待训练建筑图像特征和待训练结构图像特征划分为训练集和测试集。
其中，测试集对结构对抗生成网络模型进行测试评估。
具体地，根据测试集对结构对抗生成网络模型进行测试评估，包括：将生成图像与目标图像的各像素点按照颜色范围将生成图像与目标图像中的各像素点进行分类；将分类后的像素点制作成混淆矩阵，基于混淆矩阵计算第一评价指标。将生成图像与目标图像划分为多个子图；提取每个子图的剪力墙，并获取剪力墙的轮廓坐标；计算生成图像与目标图像中的每片剪力墙的交集面积和并集面积，并根据交集面积和并集面积计算第二评价指标。
进一步地，根据第一评价指标和第二评价指标确定加权系数；基于加权后的综合指标进行评价。
为了本领域人员更加清楚上述实施例，下面结合图2-图9进行详细说明。
图2本申请实施例所提供的基于对抗生成网络的剪力墙结构布置方法的详细框架图。
如图2所示，收集高层剪力墙住宅建筑的建筑-结构配套CAD图纸数据集，以及CAD图纸中关键元素的语义化清洗；数据集整理划分与StructGAN模型(结构对抗生成网络模型)训练；StructGAN训练后进行模型评估，采用测试集图纸输入StructGAN中并生成结构设计图纸，将生成结构图纸与目标结构图纸进行差异评估，评估合格的StructGAN投入应用；将全新设计的建筑图纸语义化后输入合格的StructGAN模型中，生成对应剪力墙设计；全新设计分别采用工程师设计以及本发明提出的基于StructGAN的剪力墙结构布置快速设计方法，随后将二者设计结果进行动力特性和设计结果的对比，验证基于该方法的设计结果的安全性与合理性。
其中，收集高层剪力墙住宅建筑的建筑-结构配套CAD图纸数据集，以及CAD图纸中关键元素的语义化清洗。一共收集了187份结构设计图纸，包含了不同高度，不同抗震设防水准信息。对所有的图纸进行图纸内部元素清洗，仅保留图纸中的墙体、门、窗洞口元素，去除掉常规CAD图纸中存在的不必要信息，包括轴网、标注、家具、文字信息；对所有收集的CAD图纸元素的进行语义化前处理，采用4种颜色在图纸中对关键元素进行填充，红色代表剪力墙，灰色代表普通填充墙，绿色代表室内门窗洞口，蓝色代表室外门洞。
其中，数据集的划分与整理，如图3所示，为收集并语义化后的典型训练-测试图像集。根据收集的图纸信息，将设计结果按照高度分为Level 1(高度低于50m)和Level 2(高度高于50m，低于140m)两个组别，再进一步将图纸按照抗震设防水准分为7度设防与8度设防共四个组别，并对组别编号。需指出，当抗震设防烈度为8度时，Level 1和Level 2组别的剪力墙设计结果接近，表明当抗震设计需求较高时， 抗震需求将主导剪力墙的设计结果，而结构高度的影响相对较小。因此，最终将结构设计图纸作为训练集划分为三类，即Level 1-7度设防(L1-7)，Level 2-7度设防(L2-7)，以及8度设防(L1&2-8)。其中，L1-7的训练集55份设计图纸，测试集6份图纸；L2-7的训练集55份，测试集6份；L1&2-8的训练集57份，测试集6份。
其中，GAN算法的选取与训练，如图4所示，为对抗生成网络算法的训练过程图。可应用的GAN算法包括两类，pix2pix与pix2pixHD。pix2pix算法的关键参数为γGAN与γL1，对图像局部效果和整体效果的由二者参数的相对值确定，其中γGAN相对值越大，生成图像的局部效果更好，γL1相对值越大，生成图像的整体效果越好。经过分析表明，对与剪力墙生成问题而言，γGAN＝1，γL1＝100时的pix2pix算法效果较好。pix2pixHD算法的关键参数为γFM，γFM参数用于调整生成图像特征匹配的损失在整个损失中的占比，进而影响生成图像的整体质量，经过分析表明，γFM＝10时pix2pixHD算法较好。pix2pix与pix2pixHD两种算法均可以得到较好的设计结果，总体而言pix2pixHD效果更佳。但pix2pixHD的训练硬件条件的需求更高，训练2048×1024分辨率的图像需要24G显存的NVIDIA显卡，而pix2pix算法在6G显存的NVIDIA显卡中即可开展应用。因此，应用时需根据硬件条件选用相应算法，本实施例选用1024×512分辨率的pix2pixHD算法。选取算法后，将分组的训练集输入StructGAN模型中开展训练，直到训练的损失达到稳定即可停止训练。
进一步地，StructGAN训练后模型评估有3种评价方法。
评价方法一：基于图像像素点语义分类的评估方法。对生成图像的每个像素点所处的颜色范围进行判断，根据颜色确定其所代表的关键元素，将生成图像的像素点分类结果与目标图像的像素点分类结果组合成混淆矩阵，进而开展评估。具体评估流程如下：
1)根据图像中每个像素HSV值，直接判定像素点类别并分离。由于图像的不同颜色RGB值范围离散，不利于应用，因此采用OpenCV进行图像处理将其转化为HSV空间，各颜色范围如表1所示，元素分离后的结果如图5所示。
2)图像像素点分类后，可得到每个图像每个像素点的类别矩阵，将矩阵重组变为向量，将合成图像与目标图像的向量输入“sklearn.metrics.confusion_matrix(y_true,y_pred)”API中，得到混淆矩如表2所示。
3)基于该混淆矩阵或提取的像素结果，计算相应的评价指标，包括加权交并比(WIoU，公式(1))，剪力墙占比率(SWratio，公式(2))。
表1 应用的5种颜色的HSV范围
Figure PCTCN2020131065-appb-000001
Figure PCTCN2020131065-appb-000002
表2 混淆矩阵
Figure PCTCN2020131065-appb-000003
Figure PCTCN2020131065-appb-000004
Figure PCTCN2020131065-appb-000005
式中，(k+1)是像素点总共的类别(类别0是背景，类别1是剪力墙，类别2是填充墙，类别3是室内门窗洞口，类别4是室外门洞)。p ij是类别i的像素点被判断为类别j的数量，即p ii代表生成正确的像素点的数量，而p ij和p ji则是生成错误的像素点的数量。w 0＝0，w 1＝0.4，w 2＝0.4，w 3＝0.1，w 4＝0.1，代表5个类别的像素权重。A swall为剪力墙面积，A inwall为填充墙面积。
评价方法二：基于目标检测剪力墙图形交并比评价方法。通过检测生成图像中的剪力墙边缘与目标图像中剪力墙的边缘，再求解轮廓边缘的交集面积以及总的并集面积，进而得到剪力墙生成情况的评价。若生成的剪力墙与目标剪力墙完全匹配，则结果为1，若完全不匹配，则结果为0。如图6所示，具体步骤如下：
1)将合成图像与目标图像划分为多个子图。
2)基于表1的HSV提取每个子图的剪力墙，随后基于OpenCV.findContours(image)的API获取每个子图中每片剪力墙的轮廓坐标。
3)基于shapely.geometry.Polygon(coordinates)的API计算合成图像中与目标图像中，每片剪力墙的交集面积，并集面积，采用公式(3)计算剪力墙交并比(SIoU)。
Figure PCTCN2020131065-appb-000006
式中，A inter是生成剪力墙与目标剪力墙交集面积，A union是生成剪力墙与目标剪力墙的并集面积，A union＝A tar+A out-A inter，A tar为目标图像剪力墙面积，A out为生成图像剪力墙面积。
评价方法三：基于多指标的加权综合评价，融合评价方法一与评价方法二，得到综合评价指标，IoU综合得分由公式(4)计算。
Score IoU＝(η SWratio×(η SIoU×SIoU+η WIoU×WIoU))            (4)
式中，η SWratio＝1-|SWratio out-SWratio tar|/SWratio tar，SWratio out和SWratio tar分别为生成图像和目标图像的SWratio；η SIoU为SIoU值系数，取值0.5；η SIoU为WIoU值系数，取值0.5。
将训练完成的StructGAN模型，采用相应的测试数据集进行测试，并采用本发明提出的评价方法开展评价，最终的评价结果如表3所示。其中，当多指标的加权综合评价指标超过0.5时，可认为该生成结果良好，StructGAN模型可以应用。
表3 训练完成模型测试结果评价
Figure PCTCN2020131065-appb-000007
进一步地，将全新设计的建筑图纸语义化后输入评估合格的StructGAN模型中， 生成对应剪力墙设计。全新设计分别采用工程师设计以及本发明提出的基于StructGAN的剪力墙结构布置快速设计方法，随后将二者设计结果进行动力特性和设计结果的对比，验证基于该方法的设计结果的安全性与合理性。
首先由结构工程师完成结构的设计，结构为38层剪力墙住宅建筑，结构主体总高103m，其中标准层为5-36层，层高均为2.9m，设防地震分组第二组，设防烈度7度(0.15g)，场地类别Ⅱ类，剪力墙抗震等级2级，结构的标准层的设计结果如图7所示。随后利用本发明提出的方法生成标准层平面的剪力墙设计结果,如图8所示。对StructGAN得到的设计结果进行评估，评估结果见表4。
表4 定量评价StructGAN剪力墙设计结果
Figure PCTCN2020131065-appb-000008
随后基于StructGAN生成的剪力墙结构设计图相应调整结构工程师设计模型，其中语义化图纸转化为结构模型的过程为：
1)采用AutoCAD附图功能，将像素化图纸附着于原结构的AutoCAD图纸中。
2)采用AutoCAD标注功能，获取StructGAN生成设计结果的剪力墙坐标及其墙体长度。
3)根据步骤2)获取的剪力墙坐标在PKPM软件中调整工程师设计结构模型。
4)模型调整遵循以下原则：StructGAN生成的剪力墙存在部分像素缺失，此情况认为该长度内只要存在剪力墙像素即布置剪力墙；若剪力墙长度短于墙体厚度(200mm)，则舍弃该墙体；仅调整原结构的墙体长度设计，不更改其厚度和材料等其他属性；墙体缩短后的梁体相应加长。
图8展示了工程师设计的结果与StructGAN设计的剪力墙布置详细对比情况，可以看到：1)StructGAN设计的剪力墙与结构工程师设计的剪力墙重合度高；2)StructGAN生成的剪力墙布置则相对离散，且存在部分短肢剪力墙。3)从墙体的数量和分布整体来看，StructGAN的生成结果并无明显错误，基本能满足设计的需求。随后，将工程师设计模型与StructGAN设计模型采用PKPM软件开展计算，对比实际计算结果。
PKPM计算结果的整体指标的对比结果如表5和图9的结果所示，从表5中可以看到，无论是整体质量还是动力特性，StructGAN设计的模型与工程师设计的模型基本一致。此外，结构抗侧力设计的关键指标为设计层间位移角包络值，StructGAN模型与工程师设计模型的差异基本在10％左右。因此，StructGAN模型仅作为初始设计，与最终的完善设计结果相差仅10％左右，在可接受差异范围内，表明了该方法设计结果 的可靠性。
表5 结构设计结果整体指标对比
Figure PCTCN2020131065-appb-000009
为了实现上述实施例，本申请还提出一种基于对抗生成网络的剪力墙结构布置装置。
图10为本申请实施例提供的一种基于对抗生成网络的剪力墙结构布置装置的结构示意图。
如图10所示，该装置包括：第一获取模块1001、第一提取模块1002、和生成模块1003。
第一获取模块1001，用于获取待处理的建筑设计图纸；
第一提取模块1002，用于提取建筑设计图纸中的关键元素，并对关键元素进行不同颜色填充处理，生成待输入图像特征；
生成模块1003，用于将待输入图像特征输入预先训练的结构对抗生成网络模型进行处理，生成结构设计图纸。
进一步地，在本申请的一个实施例中，所述装置还包括：第二获取模块，用于获取建筑设计图纸样本，以及与所述建筑设计图纸样本配套的结构设计图纸样本；第二提取模块，用于提取所述建筑设计图纸样本的建筑样本元素，并对所述建筑样本元素进行不同颜色填充处理，生成待训练建筑图像特征；第三提取模块，用于提取所述结构设计图纸样本的结构样本元素，并对所述结构样本元素进行不同颜色填充处理，生成待训练结构图像特征；划分模块，用于将所述待训练建筑图像特征和所述待训练结构图像特征划分为训练集和测试集；训练模块，用于根据所述训练集输入对抗生成网络进行训练，训练完成得到所述结构对抗生成网络模型；评估模块，用于对所述生成的结 构对抗生成网络模型进行设计效果评估，基于三种评估方法测试合格的模型可投入应用。
需要说明的是，前述对方法实施例的解释说明也适用于该实施例的装置，此处不再赘述。
本申请实施例的基于对抗生成网络的剪力墙结构布置装置中，通过获取待处理的建筑设计图纸；提取建筑设计图纸中的关键元素，并对关键元素进行不同颜色填充处理，生成待输入图像特征；将待输入图像特征输入预先训练的结构对抗生成网络模型进行处理，生成结构设计图纸。由此，能够根据建筑设计中的标准层平面图纸，快速输出对应的剪力墙结构设计，实现建筑设计对应的剪力墙结构快速设计。
为了实现上述实施例，本申请实施例提出了一种电子设备，包括：存储器、处理器及存储在存储器上并可在处理器上运行的计算机程序，所述处理器执行所述程序时，实现如前述终端设备执行方法实施例所述的基于对抗生成网络的剪力墙结构布置方法。
为了实现上述实施例，本申请实施例提出了一种计算机可读存储介质，其上存储有计算机程序，该程序被处理器执行时，实现前述方法实施例所述的基于对抗生成网络的剪力墙结构布置方法。
此外，术语“第一”、“第二”仅用于描述目的，而不能理解为指示或暗示相对重要性或者隐含指明所指示的技术特征的数量。由此，限定有“第一”、“第二”的特征可以明示或者隐含地包括至少一个该特征。在本申请的描述中，“多个”的含义是至少两个，例如两个，三个等，除非另有明确具体的限定。
在本说明书的描述中，参考术语“一个实施例”、“一些实施例”、“示例”、“具体示例”、或“一些示例”等的描述意指结合该实施例或示例描述的具体特征、结构、材料或者特点包含于本申请的至少一个实施例或示例中。在本说明书中，对上述术语的示意性表述不必须针对的是相同的实施例或示例。而且，描述的具体特征、结构、材料或者特点可以在任一个或多个实施例或示例中以合适的方式结合。此外，在不相互矛盾的情况下，本领域的技术人员可以将本说明书中描述的不同实施例或示例以及不同实施例或示例的特征进行结合和组合。
尽管上面已经示出和描述了本申请的实施例，可以理解的是，上述实施例是示例性的，不能理解为对本申请的限制，本领域的普通技术人员在本申请的范围内可以对上述实施例进行变化、修改、替换和变型。

Claims (12)

  1. 一种基于对抗生成网络的剪力墙结构布置方法，其特征在于，包括:
    获取待处理的建筑设计图纸；
    提取所述建筑设计图纸中的关键元素，并对所述关键元素进行不同颜色填充处理，生成待输入图像特征；
    将所述待输入图像特征输入预先训练的结构对抗生成网络模型进行处理，生成结构设计图纸。
  2. 如权利要求1所述的方法，其特征在于，在将所述待输入图像特征输入预先训练的结构对抗生成网络模型进行处理，生成结构设计图纸之前，还包括：
    获取建筑设计图纸样本，以及与所述建筑设计图纸样本配套的结构设计图纸样本；
    提取所述建筑设计图纸样本的建筑样本元素，并对所述建筑样本元素进行不同颜色填充处理，生成待训练建筑图像特征；
    提取所述结构设计图纸样本的结构样本元素，并对所述结构样本元素进行不同颜色填充处理，生成待训练结构图像特征；
    将所述待训练建筑图像特征和所述待训练结构图像特征划分为训练集和测试集；
    根据所述训练集输入对抗生成网络进行训练，训练完成得到所述结构对抗生成网络模型。
  3. 如权利要求2所述的方法，其特征在于，还包括：
    根据所述测试集对所述结构对抗生成网络模型进行测试评估。
  4. 如权利要求1至3中任一项所述的方法，其特征在于，所述提取所述建筑设计图纸中的关键元素，并对所述关键元素进行不同颜色填充处理，生成待输入图像特征，包括：
    提取所述建筑设计图纸中的剪力墙、普通填充墙、室内门窗洞口和室外门洞四个关键元素；
    采用不同颜色对所述关键元素进行填充，其中，红色代表所述剪力墙，灰色代表所述普通填充墙，绿色代表所述室内门窗洞口，蓝色代表所述室外门洞。
  5. 如权利要求2至4中任一项所述的方法，其特征在于，所述将所述待训练建筑图像特征和所述待训练结构图像特征划分为训练集和测试集，包括：
    根据按照结构高度和结构抗震设防水准将所述待训练建筑图像特征和所述待训练结构图像特征划分为训练集和测试集。
  6. 如权利要求3至5中任一项所述的方法，其特征在于，所述根据所述测试集对所述结构对抗生成网络模型进行测试评估，包括：
    将生成图像与目标图像的各像素点按照颜色范围将所述生成图像与目标图像中的各像素点进行分类；
    将分类后的像素点制作成混淆矩阵，基于所述混淆矩阵计算第一评价指标。
  7. 如权利要求3至5中任一项所述的方法，其特征在于，所述根据所述测试集对所述结构对抗生成网络模型进行测试评估，包括：
    将生成图像与目标图像划分为多个子图；
    提取每个子图的剪力墙，并获取所述剪力墙的轮廓坐标；
    计算所述生成图像与目标图像中的每片剪力墙的交集面积和并集面积，并根据所述交集面积和所述并集面积计算第二评价指标。
  8. 如权利要求3所述的方法，其特征在于，所述根据所述测试集对所述结构对抗生成网络模型进行测试评估，包括：
    将生成图像与目标图像的各像素点按照颜色范围将所述生成图像与目标图像中的各像素点进行分类；
    将分类后的像素点制作成混淆矩阵，基于所述混淆矩阵计算第一评价指标；
    将生成图像与目标图像划分为多个子图；
    提取每个子图的剪力墙，并获取所述剪力墙的轮廓坐标；
    计算所述生成图像与目标图像中的每片剪力墙的交集面积和并集面积，并根据所述交集面积和所述并集面积计算第二评价指标；
    其中，所述方法还包括：
    根据所述第一评价指标和所述第二评价指标确定加权系数；
    基于加权后的综合指标进行评价。
  9. 一种基于对抗生成网络的剪力墙结构布置装置，其特征在于，包括:
    第一获取模块，用于获取待处理的建筑设计图纸；
    第一提取模块，用于提取所述建筑设计图纸中的关键元素，并对所述关键元素进行不同颜色填充处理，生成待输入图像特征；
    生成模块，用于将所述待输入图像特征输入预先训练的结构对抗生成网络模型进行处理，生成结构设计图纸。
  10. 如权利要求9所述的装置，其特征在于，还包括：
    第二获取模块，用于获取建筑设计图纸样本，以及与所述建筑设计图纸样本配套的结构设计图纸样本；
    第二提取模块，用于提取所述建筑设计图纸样本的建筑样本元素，并对所述建筑样本元素进行不同颜色填充处理，生成待训练建筑图像特征；
    第三提取模块，用于提取所述结构设计图纸样本的结构样本元素，并对所述结构样本元素进行不同颜色填充处理，生成待训练结构图像特征；
    划分模块，用于将所述待训练建筑图像特征和所述待训练结构图像特征划分为训练集和测试集；
    训练模块，用于根据所述训练集输入对抗生成网络进行训练，训练完成得到所述结构对抗生成网络模型；
    评估模块，用于对所述结构对抗生成网络模型进行设计效果评估，基于三种评估方法测试合格的模型可投入应用。
  11. 一种电子设备，其特征在于，包括：存储器、处理器及存储在所述存储器上并可在所述处理器上运行的计算机程序，所述处理器执行所述程序时，实现如权利要求1至8中任一项所述的基于对抗生成网络的剪力墙结构布置方法。
  12. 一种计算机可读存储介质，其上存储有计算机程序，其特征在于，所述计算机程序被处理器执行时，实现如权利要求1至8中任一项所述的基于对抗生成网络的剪力墙结构布置方法。
PCT/CN2020/131065 2020-05-25 2020-11-24 基于对抗生成网络的剪力墙结构布置方法和装置 WO2021238113A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US17/697,660 US20220207205A1 (en) 2020-05-25 2022-03-17 Method and device for scheme design of shear wall structure based on generative adversarial network

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202010446468.9A CN111666617B (zh) 2020-05-25 2020-05-25 基于对抗生成网络的剪力墙结构布置方法和装置
CN202010446468.9 2020-05-25

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US17/697,660 Continuation US20220207205A1 (en) 2020-05-25 2022-03-17 Method and device for scheme design of shear wall structure based on generative adversarial network

Publications (1)

Publication Number Publication Date
WO2021238113A1 true WO2021238113A1 (zh) 2021-12-02

Family

ID=72384438

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/131065 WO2021238113A1 (zh) 2020-05-25 2020-11-24 基于对抗生成网络的剪力墙结构布置方法和装置

Country Status (3)

Country Link
US (1) US20220207205A1 (zh)
CN (1) CN111666617B (zh)
WO (1) WO2021238113A1 (zh)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116011084A (zh) * 2023-02-20 2023-04-25 中国建筑西南设计研究院有限公司 结构平面布置整体生成方法、装置、电子设备及存储介质
CN116680778A (zh) * 2023-04-27 2023-09-01 清华大学 建筑结构布置生成方法、系统及生成模型的构建方法

Families Citing this family (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111666617B (zh) * 2020-05-25 2021-09-14 清华大学 基于对抗生成网络的剪力墙结构布置方法和装置
CN112966760B (zh) * 2021-03-15 2021-11-09 清华大学 融合文本和图像数据的神经网络及其建筑结构的设计方法
CN113779675B (zh) * 2021-09-02 2022-04-19 清华大学 物理-数据驱动的智能化剪力墙建筑结构设计方法和装置
CN113987637B (zh) * 2021-10-25 2022-07-01 清华大学 基于生成对抗网络的楼盖结构设计方法和装置
CN114880739B (zh) * 2022-04-25 2023-03-24 清华大学 生成式建筑结构设计方案的再优化设计方法和装置
CN114880741B (zh) * 2022-04-25 2023-01-31 清华大学 一种嵌入领域知识的建筑结构构件尺寸设计方法和装置
CN115563674B (zh) * 2022-09-29 2023-06-16 清华大学 初始平面布置生成方法和装置、电子设备和存储介质
CN116186825B (zh) * 2022-11-29 2023-10-31 清华大学 基于图节点分类图神经网络的剪力墙设计方法和装置
CN116150837B (zh) * 2022-11-29 2023-12-29 清华大学 基于边特征预测图神经网络的剪力墙设计方法和装置
CN115828396B (zh) * 2023-01-03 2023-04-28 哈尔滨工业大学(深圳)(哈尔滨工业大学深圳科技创新研究院) 基于混合约束条件下的民宿平面自动生成方法及系统
CN116150858B (zh) * 2023-04-19 2023-07-14 深圳市嘉闰州生态建工有限公司 建筑工程模板支架的自动排布系统及方法
CN117195626B (zh) * 2023-08-24 2024-04-19 东南大学 一种基于生成对抗网络的建筑自由曲面网格结构划分的设计方法

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109360232A (zh) * 2018-09-10 2019-02-19 南京邮电大学 基于条件生成对抗网络的室内场景布局估计方法和装置
CN109635511A (zh) * 2019-01-16 2019-04-16 哈尔滨工业大学 一种基于条件生成对抗网络的高层居住区强排方案生成设计方法
CN110633553A (zh) * 2019-10-09 2019-12-31 郑豪 一种住宅户型平面图自动生成方法和系统
CN110766786A (zh) * 2019-09-18 2020-02-07 湖南大学 基于生成对抗网络的草图到浅浮雕模型生成的方法
US20200057824A1 (en) * 2018-08-20 2020-02-20 Sri International Machine learning system for building renderings and building information modeling data
US20200151508A1 (en) * 2018-11-09 2020-05-14 Adobe Inc. Digital Image Layout Training using Wireframe Rendering within a Generative Adversarial Network (GAN) System
CN111666617A (zh) * 2020-05-25 2020-09-15 清华大学 基于对抗生成网络的剪力墙结构布置方法和装置

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110263192B (zh) * 2019-06-06 2021-04-27 西安交通大学 一种基于条件生成对抗网络的磨粒形貌数据库创建方法

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20200057824A1 (en) * 2018-08-20 2020-02-20 Sri International Machine learning system for building renderings and building information modeling data
CN109360232A (zh) * 2018-09-10 2019-02-19 南京邮电大学 基于条件生成对抗网络的室内场景布局估计方法和装置
US20200151508A1 (en) * 2018-11-09 2020-05-14 Adobe Inc. Digital Image Layout Training using Wireframe Rendering within a Generative Adversarial Network (GAN) System
CN109635511A (zh) * 2019-01-16 2019-04-16 哈尔滨工业大学 一种基于条件生成对抗网络的高层居住区强排方案生成设计方法
CN110766786A (zh) * 2019-09-18 2020-02-07 湖南大学 基于生成对抗网络的草图到浅浮雕模型生成的方法
CN110633553A (zh) * 2019-10-09 2019-12-31 郑豪 一种住宅户型平面图自动生成方法和系统
CN111666617A (zh) * 2020-05-25 2020-09-15 清华大学 基于对抗生成网络的剪力墙结构布置方法和装置

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
WANG JUN-BO;JIN QI;YANG GUO-DONG;ZHANG XU-QING;ZHAN GUO-QI;SHAO PENG: "A Building Extraction Method Based on Object-Oriented Technology and DSM Image", GLOBAL GEOLOGY, vol. 37, no. 4, 25 December 2018 (2018-12-25), pages 1258 - 1264, XP055872018, ISSN: 1004-5589, DOI: 10.3969/j.issn.1004-5589.2018.04.026 *

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116011084A (zh) * 2023-02-20 2023-04-25 中国建筑西南设计研究院有限公司 结构平面布置整体生成方法、装置、电子设备及存储介质
CN116011084B (zh) * 2023-02-20 2023-11-28 中国建筑西南设计研究院有限公司 结构平面布置整体生成方法、装置、电子设备及存储介质
CN116680778A (zh) * 2023-04-27 2023-09-01 清华大学 建筑结构布置生成方法、系统及生成模型的构建方法
CN116680778B (zh) * 2023-04-27 2024-03-12 清华大学 建筑结构布置生成方法、系统及生成模型的构建方法

Also Published As

Publication number Publication date
CN111666617A (zh) 2020-09-15
CN111666617B (zh) 2021-09-14
US20220207205A1 (en) 2022-06-30

Similar Documents

Publication Publication Date Title
WO2021238113A1 (zh) 基于对抗生成网络的剪力墙结构布置方法和装置
CN113987637B (zh) 基于生成对抗网络的楼盖结构设计方法和装置
Liao et al. Automated structural design of shear wall residential buildings using generative adversarial networks
WO2023029432A1 (zh) 物理-数据驱动的智能化剪力墙建筑结构设计方法和装置
CN110689011A (zh) 多尺度联合卷积神经网络的太阳能电池板缺陷检测方法
CN107680678A (zh) 基于多尺度卷积神经网络甲状腺超声图像结节自动诊断系统
TW201022708A (en) Method of change detection for building models
CN106127756A (zh) 一种基于多特征信息融合技术的绝缘子识别检测方法
CN108053412B (zh) 融合多定量评价指标的遥感影像最优分割结果确定方法
CN103489193B (zh) 基于融合策略的面向对象的高分辨率遥感影像变化检测方法
CN115409069A (zh) 村镇建筑的识别方法、分类方法、装置、电子设备和介质
WO2021248335A1 (zh) 一种基于街景图片及机器学习的城市内部贫困空间测度方法及系统
Ni et al. An improved Faster R-CNN for defect recognition of key components of transmission line
CN113569788B (zh) 一种建筑物语义分割网络模型训练方法、系统及应用方法
CN111797188A (zh) 一种基于开源地理空间矢量数据的城市功能区定量识别方法
CN107616796A (zh) 基于深度神经网络的肺结节良恶性检测方法及装置
Zhao et al. A 3D modeling method for buildings based on LiDAR point cloud and DLG
CN116186825B (zh) 基于图节点分类图神经网络的剪力墙设计方法和装置
Sun et al. Automatic building age prediction from street view images
CN115661071A (zh) 基于深度学习的复合材料加工表面缺陷检测及评价方法
Song et al. Developing urban building energy models for shanghai city with multi-source open data
Shi et al. An innovative approach to determine building window-to-wall ratios for urban energy simulation
CN116150837B (zh) 基于边特征预测图神经网络的剪力墙设计方法和装置
CN113255530B (zh) 基于注意力的多通道数据融合网络架构及数据处理方法
CN112508958B (zh) 一种轻量多尺度的生物医学图像分割方法

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20937441

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20937441

Country of ref document: EP

Kind code of ref document: A1