CN117150622A - Archaized roof cornice characteristic optimization method based on GAN algorithm and BIM technology - Google Patents

Archaized roof cornice characteristic optimization method based on GAN algorithm and BIM technology Download PDF

Info

Publication number
CN117150622A
CN117150622A CN202311155030.5A CN202311155030A CN117150622A CN 117150622 A CN117150622 A CN 117150622A CN 202311155030 A CN202311155030 A CN 202311155030A CN 117150622 A CN117150622 A CN 117150622A
Authority
CN
China
Prior art keywords
angle
data
generator
gan
attribute
Prior art date
Legal status (The legal status 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 status listed.)
Granted
Application number
CN202311155030.5A
Other languages
Chinese (zh)
Other versions
CN117150622B (en
Inventor
李明照
徐建宁
王广浩
刘伟
秦龙
邵华
巩传明
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
China Construction Second Engineering Bureau Co Ltd
Original Assignee
China Construction Second Engineering Bureau Co Ltd
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 China Construction Second Engineering Bureau Co Ltd filed Critical China Construction Second Engineering Bureau Co Ltd
Priority to CN202311155030.5A priority Critical patent/CN117150622B/en
Publication of CN117150622A publication Critical patent/CN117150622A/en
Application granted granted Critical
Publication of CN117150622B publication Critical patent/CN117150622B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

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
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
    • G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/27—Regression, e.g. linear or logistic regression
    • 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
    • G06N3/0475—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
    • G06N3/094—Adversarial learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2111/00—Details relating to CAD techniques
    • G06F2111/04—Constraint-based CAD
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2111/00—Details relating to CAD techniques
    • G06F2111/06—Multi-objective optimisation, e.g. Pareto optimisation using simulated annealing [SA], ant colony algorithms or genetic algorithms [GA]

Landscapes

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

Abstract

The application provides an archaized roof cornice feature optimization method based on a GAN algorithm and a BIM technology, which realizes the optimal design of the feature cornice by converting BIM processing diagram data into an input format suitable for a machine learning model and combining with a GAN network to construct a mathematical model. According to the method, eave size data are firstly obtained from the BIM processing diagram, and high-dimensional size data are mapped into low-dimensional continuous vector representation through feature extraction and encoding. And then, generating the dimension attribute of the pseudo-classic roof cornice by utilizing the trained GAN network, and realizing angle constraint and design optimization by adopting an angle prediction and segmentation template strategy. And mapping the size attribute into an actual warping angle through a linear regression model, and selecting a proper template segmentation strategy according to the prediction angle. The method has remarkable advantages in the aspects of environmental protection, energy conservation, economic benefit and engineering quality, and realizes sustainable green construction concept through the recycling of the waste templates and the reduction of construction cost. Has wide application prospect in the field of archaize building design.

Description

Archaized roof cornice characteristic optimization method based on GAN algorithm and BIM technology
Technical Field
The application relates to the fields of building engineering, digital design, machine learning and computer vision, in particular to an archaized roof cornice feature optimization method based on a GAN algorithm and a BIM technology.
Background
In the field of construction, pursuing green construction and resource reuse has become an important trend in the industry. However, the conventional construction technology has problems of low resource utilization efficiency and considerable environmental impact. The waste of resources is difficult to avoid in manual drawing and template making, and a large amount of paper, drawing tools and model materials are consumed, thus bringing unnecessary burden to the environment. In addition, the design limitation causes that complicated archaized special eave elements are difficult to accurately express, communication efficiency is low, and design verification is difficult. The application provides an innovative technology under the background, combines a generated countermeasure network (GAN) algorithm and a BIM technology, and performs optimal design on the characteristic eave of the archaized roof. At present, the traditional construction mode often causes resource waste and environmental pressure.
Disclosure of Invention
In order to solve the technical problems, the application provides an archaized roof cornice feature optimization method based on a GAN algorithm and a BIM technology, which realizes the fine design of the archaized roof cornice by converting BIM processing diagram data into an input format suitable for a machine learning model and constructing a network model by means of a GAN network.
In order to achieve the above purpose, the technical scheme adopted by the application is as follows:
the characteristic optimization method for the archaized roof cornice based on the GAN algorithm and the BIM technology comprises the following steps:
(1) Collecting and preprocessing eave data;
obtaining eave size data from a BIM processing diagram, mapping the data into low-dimensional continuous vectors by adopting a feature extraction and coding method, reserving a size trend, reducing the data dimension, converting the size data into a machine learning model by utilizing a principal component analysis dimension reduction technology, inputting the machine learning model, and obtaining input data of a GAN network by using a standardized and archaized roof cornice principal component analysis formula;
(2) Constructing a GAN network;
the GAN model consists of a generator and a discriminator, wherein the generator receives potential space vectors to generate artistic building size data, the discriminator distinguishes between generation and real size attributes, the training process comprises parameter setting, definition of a loss function and optimizer setting, the generator generates attributes through the potential vectors, loss and parameter updating of the generator and the discriminator are performed, the consistency of the generated data and expected angles is ensured by adopting an angle constraint formula, network target giving is performed by utilizing a total loss formula, finally, the generator gradually generates vivid archaized roof cornice attributes through iterative training, the discriminator more accurately distinguishes the data, and the training termination condition is that the preset times or satisfactory generation results are achieved;
(3) Predicting a linear regression model;
further analyzing the angle of the generated archaized roof cornice design scheme by adopting a linear regression model formula through the trained GAN network, converting the generated attribute into a predicted angle through a mapping model, and adopting a segmentation strategy according to the angle, wherein the mapping model converts the dimension attribute into an actual warping angle through training the attribute and actual angle data;
(4) Benefit analysis;
the pre-assembled template is used as a material of the corner and eave formwork support in the construction preparation stage, and after the pre-assembled template is removed, the template is transferred to other projects for use.
As a further improvement of the application, the analysis formula of the principal components of the pseudo-classic roof cornice in the step 1) is expressed as follows:
wherein, the analysis formula of the main component of the archaized roof cornice is expressed as:
in the whole data preprocessing process, the matrix Y represents a dimension-reduced data matrix, each row corresponds to one sample, each column corresponds to one principal component, X is a data matrix, and the average value vector is mu= (mu) 1 ，μ 2 ，μ 3 ，μ 4 ) The method comprises the steps of carrying out a first treatment on the surface of the Normalized vector: sigma= (sigma) 1 ，σ 2 ，σ 3 ，σ 4 )，λ C Is a characteristic value.
As a further improvement of the present application, the angle constraint formula in the step 2) is expressed as:
wherein the angle constraint formula is expressed as follows:
l angle ＝||θ generated -θ actual ||
the difference between the attributes generated by the generator and the actual warp angle is incorporated into a loss function, thereby guiding the generator to more accurately generate data meeting the angle requirement, the angle loss is defined as the angle loss, wherein theta generated Represents the angle, θ, of the generation actual Representing the actual upturned angle, |·| representing the euclidean distance, and this loss term provides additional constraints for the generator by calculating the euclidean distance between the angle at which the data is generated and the true angle.
In the training process, minimizing the angle loss causes the generator to generate archaized roof cornice size data with more accurate angle attribute, thereby enhancing the reality and quality of the generated data.
As a further improvement of the present application, the total loss formula in the step 2) is expressed as:
L GAN ＝E s，x [logD(s，x)]+E s [log(1-D(s，G(s)))]
L data ＝L GAN (G，d)+λL angle E s，x [logD(s，x)]this part is the training target of the arbiter. Wherein s is a real data sample such as a real archaized roof cornice size attribute, x is an attribute matched with s and comprises a real data attribute, D (s, x) represents the output of the discriminator on the real data sample, namely the probability that the discriminator considers the data to be real, E s [log(1-D(s，G(s)))]This part is the training target of the generator, where s is a potential vector randomly sampled from the potential space, G(s) is the archaized roof cornice size attribute generated by the generator from this potential vector, D (s, G (s)) represents the output of the data generated by the arbiter to the probability that the arbiter considers the data as generated, L GAN Loss value of generator, L angle Then is the angle loss function, L data Then the total data loss value.
As a further improvement of the present application, the linear model regression prediction model in the step 3) is formulated as:
wherein the linear model regression prediction model is formulated as:
wherein, the angle prediction uses a trained generator G to input a random potential space vector Z so as to generate the dimension attribute of the pseudo-classic roof corniceAnd inputting the attribute value into a trained mapping model to obtain a predicted flip angle +.>Judging an angle segmentation strategy according to the predicted upwarp angle +.>Make a judgment if->Then a 3 DEG split template strategy is adopted; />A 4.5 ° split template strategy is employed. Compared with the prior art, the application has the beneficial effects that:
(1) The method combines the generation countermeasure network (GAN) algorithm with the Building Information Model (BIM) technology, and realizes the optimal design of the eave features of the archaized building. Through data conversion and mathematical model construction, the archaized roof cornice size attribute meeting the design requirement can be quickly generated, so that the design period is greatly shortened, and the design efficiency is improved.
(2) The patent adopts angle prediction and segmentation template strategies, and realizes the angle constraint on the design scheme. And mapping the size attribute into a practical warping angle through a linear regression model, and selecting a proper template segmentation strategy according to a prediction angle to ensure the accuracy of the design scheme in angle. The accurate angle control not only enhances the artistic quality of the building, but also improves the engineering quality.
(3) This patent is positive to respond to the environmental protection theory, realizes green construction and resource recycling. Through the reuse of pre-assembled templates, the waste materials are reduced, the construction cost is reduced, and meanwhile, the resource waste is reduced, so that the building construction is more environment-friendly and sustainable.
Drawings
Fig. 1: according to the whole flow chart of the method for optimizing the characteristic eave of the archaized roof based on the GAN algorithm and the BIM technology, which is provided by the embodiment of the application;
FIG. 2 is a detailed flow chart of an archaized roof cornice feature roof optimization method based on a GAN algorithm and a BIM technology, which is provided by the embodiment of the application.
Detailed Description
The application is described in further detail below with reference to the attached drawings and detailed description:
as a specific embodiment of the present application, the present application provides an overall flowchart as shown in FIG. 1, and the flowchart is shown in FIG. 2, and specific steps are as follows:
step S1, eave data collection and pretreatment
The patent provides a novel method for optimizing the characteristic eave of the archaized roof based on a GAN algorithm and a BIM technology, namely, processing diagram data obtained by the BIM technology are converted into a specific input format, and a corresponding mathematical model is constructed by using the GAN algorithm in machine learning, so that the purpose of optimizing the characteristic eave is realized. In this method, the present application first obtains eave dimension data, such as length L, width W, and height H, from the BIM process map, which are typically present in a structured form in the drawing or data source. However, to accommodate the needs of machine learning models, it is necessary to convert these dimensional data into a continuous, numerical vector representation. Thus, the present patent application employs a feature extraction and encoding method, i.e., the original size data is mapped from a high-dimensional space to a continuous vector representation of lower dimensions through both feature extraction and encoding steps during data preprocessing. The processing method effectively reduces the dimension of the data while keeping the variation trend of the dimension data so as to construct a machine learning model more effectively. In the application, the feature in eave data is extracted by adopting Principal Component Analysis (PCA) of a dimension reduction technology, so that the aim of data coding is fulfilled. The PCA is used for carrying out feature extraction and coding processing on the original size data, at the moment, the low-dimensional vector can be used as an input variable of a machine learning model, and a GAN algorithm is used for generating an archaized roof cornice design scheme or processing other archaized roof cornice related tasks.
The application further describes the preprocessing process of the archaized roof cornice data by combining related formulas. Firstly, size data of the archaized roof cornices are collected from the BIM structure chart, wherein the size data comprise length L, width W, height H and angle A. The data are organized into a matrix, with each row in the matrix representing an eave sample data and each column representing a tree property (length, width, height and angle). The data of the different dimensions are then normalized, i.e. at each dimension attribute (length, width, height and angle) the mean (mean) and standard deviation (std) of the n samples are calculated. It is assumed that n samples and 4 size attributes are observable in this application, and an n×4 original data matrix X is formed, where each row is one sample and each column is one size attribute. And (3) carrying out standardization processing on the original data matrix X to obtain a standardized data matrix Z, and applying a PCA dimension reduction method to the matrix Z. The specific implementation steps are as follows: the mean and standard deviation of each attribute are calculated, and the mean vector μ is calculated with the normalized vector σ. In addition, a covariance matrix is calculated, principal components are selected, and the first k principal components are selected according to the sequence from the large characteristic value to the small characteristic value, wherein k is the dimension to which dimension reduction is expected. Projecting the matrix W to a low-dimensional space, forming a matrix W by the first k eigenvectors, projecting the standardized data matrix Z to the low-dimensional space, and further obtaining a matrix Y after dimension reduction by adopting an archaized roof cornice principal component analysis formula.
Wherein, the analysis formula of the main component of the archaized roof cornice is expressed as:
in the whole data preprocessing process, the matrix Y represents a data matrix after dimension reduction, each row of the data matrix corresponds to one sample, and each column corresponds to one main component. Matrix Y represents the dimensionality reduced data matrix during the whole data preprocessing process, each row corresponds to one sample, each column corresponds to one principal component, X is the data matrix, and the mean vector is μ= (μ) 1 ，μ 2 ，μ 3 ，μ 4 ) The method comprises the steps of carrying out a first treatment on the surface of the Normalized vector: sigma= (sigma) 1 ，σ 2 ，σ 3 ，σ 4 )，λ C Is a characteristic value.
Step S2: GAN network construction
The present application builds a generation of a antagonism network (GAN) model. The network structure consists of a generator and a discriminator, so that the application can generate the characteristic proposal of the archaized roof cornice through the antagonism training. The generator network: the task of the generator is to receive a potential space vector and then to generate the dimensional attributes of the archaized roof cornices. The generator trains and learns random potential space formed by BMI observation data, and extracts building size data with artistic and archaizing characteristics such as length, width, height, angle and the like. The aim of the discriminator in the discriminator network is to distinguish the archaized roof cornice size attribute generated by the generator from the real building size attribute. It receives the generated data from the generator and input from the real building data and attempts to determine which are real and which are generated. The training process is illustrated in fig. 2, where the training parameters are set to define a loss function, where the loss function in the arbiter and generator may be used to guide the training process. The training step of generator G is to generate a random potential spatial vector (z) as input. And (3) passing the potential space vector through a generator G to obtain the generated archaized roof cornice size attribute G (z). The generated attribute is input to the discriminator D, and the output D (G (z)) of the discriminator is obtained. The parameters of the generator are updated and the loss of the generator is minimized by the Adam optimizer. And randomly selecting a real archaized roof cornice size attribute sample (x). Generating a random potential space vector (z), and obtaining the generated archaized roof cornice size attribute G (z) through a generator. The real attribute sample (x) and the generated attribute sample G (z) are input to the discriminator (D), respectively. The loss of the discriminators is calculated and its ability to classify correctly on the true and generated data is measured. The training steps of the generator and the arbiter are repeatedly performed, and a plurality of iterations are performed. As the number of iterations increases, the generator gradually learns to generate more realistic archaized roofing cornice size attributes, and the arbiter gradually becomes more adept at distinguishing the real and generated data. After a predetermined number of exercises is reached or the output of the generator reaches a satisfactory level, the exercises are stopped.
In addition, the patent also innovatively proposes a new constraint formula based on angle conditions, which is used for measuring the difference between the generator attribute and the actual warp angle.
Wherein the angle constraint formula is expressed as follows:
l angle ＝||θ generated -θ actual || (2)
the difference between the attribute generated by the generator and the actual warp angle can be included in the loss function, so that the generator is guided to generate data meeting the angle requirement more accurately. The angular loss can be defined as where θ generated Represents the angle, θ, of the generation actual The actual upturned angle is shown, and I.I. is shown as the Euclidean distance. This loss term provides additional constraints to the generator by calculating the euclidean distance between the angle at which the data is generated and the true angle. In the training process, minimizing the angle loss causes the generator to generate archaized roof cornice size data with more accurate angle attribute, thereby enhancing the reality and quality of the generated data.
At this time, the total loss formula is:
L GAN ＝E s，x [logD(s，x)]+E s [log(1-D(s，G(s)))] (3)
L data ＝L GAN (G，D)+λL angle (4)
E s，x [logD(s，x)]this part is the training target of the arbiter. Where s is a real data sample (e.g., a real attribute of the size of the cornice of the archaized roof), x is an attribute (e.g., an attribute of real data) matched with s, and D (s, x) represents the output of the arbiter to the real data sample (the probability that the arbiter considers the data to be real). E (E) s [log(1-D(s，G(s)))]This part is the training target of the generator. Where s is a potential vector randomly sampled from the potential space, G(s) is an archaized roof cornice size attribute generated by the generator according to the potential vector, and D (s, G (s)) represents the output of the arbiter to the generated data (the probability that the arbiter considers the data to be generated). L (L) GAN Loss value of generator, L angle Then is the angle loss function, L data Then the total data loss value.
Step S3: detection of GAN network model generation architectural design
In the step, a linear regression model is adopted to further optimize the angle of generating eave design scheme by the GAN network model. Firstly, a trained GAN network is used, random potential space vectors are input, archaized roof cornice size attributes are generated, and the generated attributes are input into a mapping model. For simplicity of calculation, the mapping model employed here is a simple linear regression model. Secondly, training a mapping model, and considering attribute characteristic data and angle data in training set data in order to learn the relation between the size attribute and the actual warping angle. And inputting the generated dimension attribute of each archaized roof cornice into a trained mapping model. Here, the mapping model will output a predicted tilt angle, representing a real value.
And applying a segmentation template strategy according to the predicted real value of the upward warp angle. If the predicted upturned angle is within 30 degrees, a template is segmented by 3 degrees; if the predicted flip angle is greater than 30, a piece of template is segmented by 4.5. The linear regression model is as follows: data preparation and preprocessing the attribute and the actual warp angle are here represented as vectors respectively: attribute vector S and angle vector a. A mapping model is established, and the linear regression model for the patent application is A=θ map S defines a mapping model, and uses the attribute vector S as an input value to initially input and output a predicted upward warp angleTraining a mapping model, training by using an attribute vector S in a training data set and a corresponding upward warping angle A, and performing angle evaluation by using a linear model regression prediction model formula.
Wherein the linear model regression prediction model is formulated as:
wherein, the angle prediction uses a trained generator G to input a random potential space vector Z so as to generate the dimension attribute of the pseudo-classic roof corniceAnd inputting the attribute value into a trained mapping model to obtain a predicted flip angle +.>Judging an angle segmentation strategy according to the predicted upwarp angle +.>Make a judgment if->Then a 3 DEG split template strategy is adopted; />A 4.5 ° split template strategy is employed.
Notably, the linear regression model serves as a conversion of dimensional properties to actual rocker angles. By training the parameter theta in a linear regression model map The relationship between the generated dimensional attribute and the actual angle is found. And then, inputting the newly generated size attribute into the linear regression model, further predicting the corresponding upturned angle, and determining the adopted template segmentation strategy according to the angle.
Step S4 benefit analysis
The template is used as a raw material in the construction preparation stage of the patent application, so that the concept of green construction and recycling of waste resources is satisfied. After the corner roof eave formwork support meets the dismantling requirement, the pre-assembled templates can be transferred to other corner slope eave formwork supports for use, so that the recycling of waste template materials is realized, and the resource waste is further reduced. By the method provided by the application, the installation precision and the molding effect of the pre-assembled templates meet the design requirements, and the construction cost is reduced by recycling the waste templates and recycling the pre-assembled templates. In addition, due to shortening of the construction period, improvement of the construction efficiency is directly brought, and therefore the construction cost is further reduced. Taking project of raising eave as an example, compared with the common construction process, the combined tool type template construction technology brings remarkable economic benefit.
The above description is only of the preferred embodiment of the present application, and is not intended to limit the present application in any other way, but is intended to cover any modifications or equivalent variations according to the technical spirit of the present application, which fall within the scope of the present application as defined by the appended claims.

Claims (5)

1. The characteristic optimization method for the archaized roof cornice based on the GAN algorithm and the BIM technology is characterized by comprising the following steps of: the method comprises the following steps:
(1) Collecting and preprocessing eave data;
obtaining eave size data from a BIM processing diagram, mapping the data into low-dimensional continuous vectors by adopting a feature extraction and coding method, reserving a size trend, reducing the data dimension, converting the size data into a machine learning model by utilizing a principal component analysis dimension reduction technology, inputting the machine learning model, and obtaining input data of a GAN network by using a standardized and archaized roof cornice principal component analysis formula;
(2) Constructing a GAN network;
the GAN model consists of a generator and a discriminator, wherein the generator receives potential space vectors to generate artistic building size data, the discriminator distinguishes between generation and real size attributes, the training process comprises parameter setting, definition of a loss function and optimizer setting, the generator generates attributes through the potential vectors, loss and parameter updating of the generator and the discriminator are performed, the consistency of the generated data and expected angles is ensured by adopting an angle constraint formula, network target giving is performed by utilizing a total loss formula, finally, the generator gradually generates vivid archaized roof cornice attributes through iterative training, the discriminator more accurately distinguishes the data, and the training termination condition is that the preset times or satisfactory generation results are achieved;
(3) Predicting a linear regression model;
further analyzing the angle of the generated archaized roof cornice design scheme by adopting a linear regression model formula through the trained GAN network, converting the generated attribute into a predicted angle through a mapping model, and adopting a segmentation strategy according to the angle, wherein the mapping model converts the dimension attribute into an actual warping angle through training the attribute and actual angle data;
(4) Benefit analysis;
the pre-assembled template is used as a material of the corner and eave formwork support in the construction preparation stage, and after the pre-assembled template is removed, the template is transferred to other projects for use.
2. The method for optimizing the characteristics of the archaized roof cornice based on the GAN algorithm and the BIM technology according to claim 1, which is characterized in that:
the analysis formula of the main component of the pseudo-classic roof cornice in the step 1) is expressed as follows:
wherein, the analysis formula of the main component of the archaized roof cornice is expressed as:
in the whole data preprocessing process, the matrix Y represents a dimension-reduced data matrix, each row corresponds to one sample, each column corresponds to one principal component, X is a data matrix, and the average value vector is mu= (mu) 1 ，μ 2 ，μ 3 ，μ 4 ) The method comprises the steps of carrying out a first treatment on the surface of the Normalized vector: sigma= (sigma) 1 ，σ 2 ，σ 3 ，σ 4 )，λ C Is a characteristic value.
3. The method for optimizing the characteristics of the archaized roof cornice based on the GAN algorithm and the BIM technology according to claim 1, which is characterized in that: the angle constraint formula in the step 2) is expressed as follows:
wherein the angle constraint formula is expressed as follows:
l angle ＝||θ generated -θ actual ||
incorporating differences between the attributes generated by the generator and the actual warp angle into the loss function, thereby guiding the generator to more accurately generate complianceData of angle requirement, angle loss is defined as, where θ generated Represents the angle, θ, of the generation actual Representing the actual upturned angle, |·| representing the euclidean distance, and this loss term provides additional constraints for the generator by calculating the euclidean distance between the angle at which the data is generated and the true angle.
4. The method for optimizing the characteristics of the archaized roof cornice based on the GAN algorithm and the BIM technology according to claim 1, which is characterized in that: the total loss formula in the step 2) is expressed as:
L GAN ＝E s，x [logD(s，x)]+E s [log(1-D(s，G(s)))]
L data ＝L GAN (G，D)+λL angle
E sx [logD(s，x)]this part is the training target of the arbiter. Wherein s is a real data sample such as a real archaized roof cornice size attribute, x is an attribute matched with s and comprises a real data attribute, D (s, x) represents the output of the discriminator on the real data sample, namely the probability that the discriminator considers the data to be real, E s [log(1-D(s，G(s)))]This is partly the training target of the generator, where s is a potential vector randomly sampled from the potential space, G(s) is the archaized roof cornice size attribute generated by the generator from this potential vector, D (s, G (s)) represents the output of the data generated by the arbiter to the probability that the arbiter considers the data as generated, L GAN Loss value of generator, L angle Then it is the angle loss function that,
L data then the total data loss value.
5. The method for optimizing the characteristics of the archaized roof cornice based on the GAN algorithm and the BIM technology according to claim 1, which is characterized in that: the linear model regression prediction model formula in the step 3) is expressed as:
wherein the linear model regression prediction model is formulated as:
wherein, the angle prediction uses a trained generator G to input a random potential space vector Z so as to generate the dimension attribute of the pseudo-classic roof corniceAnd inputting the attribute value into a trained mapping model to obtain a predicted flip angle +.>Judging an angle segmentation strategy according to the predicted upwarp angle +.>Make a judgment if->Then a 3 DEG split template strategy is adopted;a 4.5 ° split template strategy is employed.
CN202311155030.5A 2023-09-08 2023-09-08 Archaized roof cornice characteristic optimization method based on GAN algorithm and BIM technology Active CN117150622B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202311155030.5A CN117150622B (en) 2023-09-08 2023-09-08 Archaized roof cornice characteristic optimization method based on GAN algorithm and BIM technology

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202311155030.5A CN117150622B (en) 2023-09-08 2023-09-08 Archaized roof cornice characteristic optimization method based on GAN algorithm and BIM technology

Publications (2)

Publication Number Publication Date
CN117150622A true CN117150622A (en) 2023-12-01
CN117150622B CN117150622B (en) 2024-05-07

Family

ID=88883975

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202311155030.5A Active CN117150622B (en) 2023-09-08 2023-09-08 Archaized roof cornice characteristic optimization method based on GAN algorithm and BIM technology

Country Status (1)

Country Link
CN (1) CN117150622B (en)

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108197376A (en) * 2017-12-28 2018-06-22 中铁二局集团有限公司 A kind of temporary building design and modeling method and system based on BIM
CN111608332A (en) * 2020-04-10 2020-09-01 河南七建工程集团有限公司 Construction method of cast-in-place concrete roof wing-shaped angle cornice
CN114239116A (en) * 2021-12-21 2022-03-25 盈嘉互联（北京）科技有限公司 BIM design recommendation method based on style migration
CN114693064A (en) * 2022-02-25 2022-07-01 山东建筑大学 Building group scheme generation performance evaluation method
CN115099521A (en) * 2022-07-15 2022-09-23 中用科技有限公司 Building energy consumption optimization method fusing BIM and deep neural network

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108197376A (en) * 2017-12-28 2018-06-22 中铁二局集团有限公司 A kind of temporary building design and modeling method and system based on BIM
CN111608332A (en) * 2020-04-10 2020-09-01 河南七建工程集团有限公司 Construction method of cast-in-place concrete roof wing-shaped angle cornice
CN114239116A (en) * 2021-12-21 2022-03-25 盈嘉互联（北京）科技有限公司 BIM design recommendation method based on style migration
CN114693064A (en) * 2022-02-25 2022-07-01 山东建筑大学 Building group scheme generation performance evaluation method
CN115099521A (en) * 2022-07-15 2022-09-23 中用科技有限公司 Building energy consumption optimization method fusing BIM and deep neural network

Non-Patent Citations (5)

* Cited by examiner, † Cited by third party
Title
伍小粦 等: "一种大角度仿古建筑坡屋面瓦片干挂施工技术", 建筑科学与工程, 17 March 2021 (2021-03-17), pages 1 - 5 *
庄兴 等: "大型重檐攒尖顶式仿古建筑施工技术", 广东土木与建筑, no. 06, 15 December 2017 (2017-12-15), pages 1 - 5 *
石若明 等: "府学胡同36号院古建筑信息模型建立研究", 北京建筑大学学报, no. 04, 31 December 2014 (2014-12-31), pages 1 - 7 *
肖进胜 等: "基于生成对抗网络的雾霾场景图像转换算法", 计算机学报, no. 01, 11 September 2019 (2019-09-11), pages 1 - 7 *
闻江 等: "匠心唐韵冠长安 穿越千年筑鲁班——中建二局三公司陕西大剧院项目创鲁班奖纪实", 建筑, no. 6, 20 March 2019 (2019-03-20), pages 1 - 3 *

Also Published As

Publication number Publication date
CN117150622B (en) 2024-05-07

Similar Documents

Publication Publication Date Title
Liao et al. Automated structural design of shear wall residential buildings using generative adversarial networks
CN113139453B (en) Orthoimage high-rise building base vector extraction method based on deep learning
CN104361611B (en) Group sparsity robust PCA-based moving object detecting method
CN110223324A (en) A kind of method for tracking target of the twin matching network indicated based on robust features
CN110580501A (en) Zero sample image classification method based on variational self-coding countermeasure network
CN111046462A (en) Drawing display system and method for outdoor building design
CN103345744B (en) A kind of human body target part automatic analytic method based on many images
CN112613097A (en) BIM rapid modeling method based on computer vision
CN106650820B (en) Matching and recognizing method for handwritten electric component symbol and standard electric component symbol
CN101863088A (en) Method for forecasting Mooney viscosity in rubber mixing process
CN106709509B (en) Satellite telemetry data clustering method based on time series special points
WO2022256460A1 (en) Systems for rapid accurate complete detailing and cost estimation for building construction from 2d plans
CN117150622B (en) Archaized roof cornice characteristic optimization method based on GAN algorithm and BIM technology
CN111079826A (en) SLAM and image processing fused construction progress real-time identification method
CN116029037A (en) IFC-based assembly type construction engineering amount calculating method
CN116541912A (en) Method and device for generating wiring diagram based on CAD image recognition
CN110705355A (en) Face pose estimation method based on key point constraint
CN113835964B (en) Cloud data center server energy consumption prediction method based on small sample learning
CN106875101B (en) Energy management system control method and control device
CN111460974B (en) Scattered point cloud data global feature extraction method based on optimization
CN111259981B (en) Automatic classification system after remote sensing image processing
CN114372181A (en) Intelligent planning method for equipment production based on multi-mode data
CN112906244A (en) Multipoint geostatistical modeling parameter optimization method based on connectivity function
CN111626376A (en) Domain adaptation method and system based on discrimination joint probability
Lu et al. Automated Structural Design of Shear Wall Residential Buildings Using GAN-Based Machine Learning Algorithm

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant