CN116226674B

CN116226674B - Layout model training method, layout method and device for frame beams

Info

Publication number: CN116226674B
Application number: CN202310499116.3A
Authority: CN
Inventors: 方长建; 康永君; 刘宜丰; 龙丹冰; 刘冠军; 赖逸峰; 赵一静; 陈其铧; 黄扬; 叶波
Original assignee: China Southwest Architectural Design and Research Institute Co Ltd
Current assignee: China Southwest Architectural Design and Research Institute Co Ltd
Priority date: 2023-05-06
Filing date: 2023-05-06
Publication date: 2023-09-05
Anticipated expiration: 2043-05-06
Also published as: CN116226674A

Abstract

The application relates to a training method, a training method and a training device for a layout model of a frame beam, and aims to improve the construction efficiency of training data of layout of a frame structure, reduce the degree of freedom of model training and training cost and realize application in actual engineering. The method constructs a non-complete sensitivity theory of the arrangement of the frame structure based on the characteristic that the arrangement logic of the frame structure is basically consistent in a certain sensitivity range, a certain sensitivity distance and a certain sensitivity angle. By adopting a step and repeat technology, a data regularization processing technology and a frame element cleaning and rotation augmentation technology, an incomplete graph training sample can be constructed, so that a large amount of high-efficiency training data can be obtained from fewer engineering cases, and a trained frame beam layout model can be obtained. In executing the method for arranging the frame beams, the arrangement of the frame beams can be automatically completed by using the model under the condition that the positions of the frame columns are known. The patent has wide application value.

Description

Layout model training method, layout method and device for frame beams

Technical Field

The application relates to the technical field of data processing, in particular to a training method, a training method and a training device for a layout model of a frame beam.

Background

The frame beam arrangement of the building engineering frame structure is an important link in structural design, has close relation with building plane arrangement, functional arrangement, structural design concept, load size and the like, and is mainly manually completed by relying on engineering experience of a designer in the traditional design method, so that the time consumption is long and the arrangement result is single.

In recent years, a method for introducing deep learning artificial intelligence into structural design is presented, but the training quantity is insufficient in the frame beam arrangement of a specific frame structure, the training quality is general, and the problem of difficulty in meeting the actual design needs is solved. The frame structure is flexible and changeable, besides the conventional orthogonal axis network, most projects are non-orthogonal irregular axis networks, the number of single frame columns is large, besides the conventional 4 beams, a large number of 3, 5, 6 and other conditions exist, the space between the columns is also quite large, the complexity and diversity of the frame structure are caused, the conventional training method is required to depend on a large number of various training data of the layout type, the large number of the training data are difficult to directly acquire in the actual projects, the training result is also poor, and the training result is difficult to directly apply to the actual design.

Disclosure of Invention

The application aims to provide a training method, a training method and a training device for an arrangement model of a frame beam, which are used for constructing an incomplete sensitivity theory of the arrangement of the frame structure based on the characteristic that the arrangement logic of the frame structure is basically consistent in a certain sensitivity range, a certain sensitivity distance and a certain sensitivity angle, adopting a technical response range insensitivity by using step and repeat, adopting a data regularization processing technology to respond to the distance insensitivity and the angle insensitivity, matching with a frame element cleaning and rotation amplification technology, completing the construction of an incomplete graph training sample, acquiring a large amount of high-efficiency training data from fewer engineering cases, greatly improving the construction efficiency of the training data, reducing the training freedom degree and the training cost, effectively improving the generation quality of artificial intelligence, and being capable of obtaining different arrangement results for different project conditions through setting a basic arrangement score threshold value and being applied to the ground in practical engineering.

In a first aspect, a method for training an arrangement model of a frame beam is provided, and the method may include:

obtaining design drawings of different frame structures; the design drawing comprises position information of a frame column, position information of a frame beam and structural information of the frame column and the frame beam;

Traversing each frame column in any design drawing by adopting a preset step-and-repeat sampling technology to obtain initial drawing data corresponding to each frame column; the initial map data comprise initial polar coordinate vectors of the frame columns and polar coordinate positions of corresponding frame beams;

performing data regularization processing on the initial graph data of each frame column to obtain regular graph data, wherein the regular graph data comprises regular polar coordinate vectors of each frame column and polar coordinate positions of corresponding frame beams;

carrying out data cleaning on the regular graph data of each frame column by adopting a preset frame element cleaning technology to obtain cleaning graph data;

performing rotation augmentation on the cleaning map data of each frame column by adopting a preset map data rotation augmentation technology to obtain current map data;

determining an incomplete graph training sample meeting a preset condition based on a current polar coordinate vector of each frame column and a polar coordinate position of a corresponding frame beam in the current graph data aiming at any graph data; the preset condition is that the included angle between any frame column and a frame beam formed by the rest frame columns is larger than a preset angle;

and taking each current diagram data and a corresponding incomplete diagram training sample as training data, and performing iterative training on the layout model of the frame beams to be trained to obtain a trained layout model of the frame beams, wherein the layout model of the frame beams is used for outputting the layout scores of the frame beams between every two frame columns in the incomplete diagram training samples.

In a second aspect, a method for arranging frame beams is provided, which may include:

acquiring position information of each frame column in a current design drawing of a scheme to be built;

after determining any frame column as a central frame column of the current design drawing, calculating the distances between other frame columns in the current design drawing except the central frame column and the central frame column respectively based on the position information of each frame column in the current design drawing;

carrying out data regularization processing on the polar coordinate vector of the frame column with the distance smaller than the range sensitivity constant to obtain a current polar coordinate vector of the corresponding frame column;

acquiring data of a to-be-input graph corresponding to the center frame column and meeting preset conditions based on a current polar coordinate vector of the frame column with a distance smaller than a range sensitivity constant; the preset condition is that the included angle between any frame column and a frame beam formed by the rest frame columns is larger than a preset angle;

inputting the to-be-input graph data into a frame beam layout model trained based on the frame beam layout model training method of the first aspect to obtain the frame beam layout score of the frame beam between every two frame columns in the to-be-input graph data;

If any frame beam in the image data to be input does not have a preset local arrangement score threshold, acquiring a first appointed frame beam with a frame beam arrangement score greater than a preset basic arrangement score threshold, and determining the first appointed frame beam as the frame beam to be drawn;

if the target frame beam in the graph data to be input has a preset local arrangement score threshold value,

acquiring a second designated frame beam with a frame beam arrangement score greater than the local arrangement score threshold, and determining the second designated frame beam as the frame beam to be drawn;

and drawing the frame beams to be drawn of the image data to be input by adopting a CAD technology so as to realize automatic arrangement of the frame beams.

In a third aspect, a layout model training apparatus for a frame beam is provided, the apparatus may include:

the acquisition unit is used for acquiring design drawings of different frame structures; the design drawing comprises position information of a frame column, position information of a frame beam and structural information of the frame column and the frame beam;

the sampling unit is used for traversing each frame column in any design drawing by adopting a preset step-and-repeat sampling technology to obtain initial drawing data corresponding to each frame column; the initial map data comprise initial polar coordinate vectors of the frame columns and polar coordinate positions of corresponding frame beams;

The processing unit is used for carrying out data regularization processing on the initial graph data of each frame column to obtain regular graph data, wherein the regular graph data comprises regular polar coordinate vectors of each frame column and polar coordinate positions of corresponding frame beams;

the cleaning unit is used for carrying out data cleaning on the regular graph data of each frame column by adopting a preset frame element cleaning technology to obtain cleaning graph data;

an amplifying unit, configured to rotationally amplify the cleaning map data of each frame column by using a preset map data rotation amplifying technology, so as to obtain current map data;

the determining unit is used for determining incomplete graph training samples meeting preset conditions according to any graph data based on the current polar coordinate vector of each frame column in the graph data and the polar coordinate position of the corresponding frame beam; the preset condition is that the included angle between any frame column and a frame beam formed by the rest frame columns is larger than a preset angle;

the training unit is used for carrying out iterative training on the layout model of the frame beams to be trained by taking the image data and the corresponding incomplete image training samples as training data to obtain a trained layout model of the frame beams, and the layout model of the frame beams is used for outputting the layout scores of the frame beams between every two frame columns in the incomplete image training samples.

In a fourth aspect, an arrangement of frame beams is provided, which may comprise:

the acquisition unit is used for acquiring the position information of each frame column in the current design drawing of the scheme to be constructed;

the calculating unit is used for calculating the distances between other frame columns except the central frame column in the current design drawing and the central frame column respectively based on the position information of each frame column in the current design drawing after determining any frame column as the central frame column of the current design drawing;

the processing unit is used for carrying out data regularization processing on the polar coordinate vectors of the frame columns with the distance smaller than the range sensitivity constant to obtain the current polar coordinate vectors of the corresponding frame columns;

the acquisition unit is further used for acquiring the data of the to-be-input graph corresponding to the center frame column and meeting the preset condition based on the current polar coordinate vector of the frame column with the distance smaller than the range sensitivity constant; the preset condition is that the included angle between any frame column and a frame beam formed by the rest frame columns is larger than a preset angle;

inputting the to-be-input graph data into a frame beam layout model trained by a frame beam layout model training method to obtain the frame beam layout score of the frame beam between every two frame columns in the to-be-input graph data;

A determining unit, configured to obtain a first designated frame beam with a frame beam arrangement score greater than a preset basic arrangement score threshold if any frame beam in the map data to be input does not have a preset local arrangement score threshold, and determine the first designated frame beam as a frame beam to be drawn;

if the target frame beams in the image data to be input have a preset local arrangement score threshold, acquiring a second designated frame beam with the frame beam arrangement score larger than the local arrangement score threshold, and determining the second designated frame beam as the frame beam to be drawn;

and the drawing unit is used for drawing the frame beams to be drawn of the image data to be input by adopting a CAD technology so as to realize automatic arrangement of the frame beams.

In a fifth aspect, there is provided a computer-readable storage medium having stored therein a computer program which, when executed by a processor, implements the frame beam arrangement model training method described in the first aspect or the frame beam arrangement method described in any one of the second aspects.

After obtaining design drawings of different frame structures including frame columns and frame beam information, the method for training the layout model of the frame beams adopts a preset step-and-repeat sampling technology to traverse each frame column in any design drawing so as to obtain initial diagram data corresponding to each frame column; the initial map data comprise initial polar coordinate vectors of all frame columns and polar coordinate positions of corresponding frame beams; carrying out data regularization processing on the initial graph data of each frame column to obtain regular graph data, wherein the regular graph data comprises regular polar coordinate vectors of each frame column and polar coordinate positions of corresponding frame beams; carrying out data cleaning on the regular graph data of each frame column by adopting a preset frame element cleaning technology to obtain cleaning graph data; carrying out rotation augmentation on cleaning map data of each frame column by adopting a preset map data rotation augmentation technology to obtain current map data; determining an incomplete graph training sample meeting a preset condition based on the current polar coordinate vector of each frame column and the polar coordinate position of the corresponding frame beam in the current graph data aiming at any graph data; the preset condition is that the included angle between any frame column and a frame beam formed by the rest frame columns is larger than a preset angle; and taking each current diagram data and the corresponding incomplete diagram training sample as training data, and performing iterative training on the layout model of the frame beams to be trained to obtain a trained layout model of the frame beams, wherein the layout model of the frame beams is used for outputting the layout scores of the frame beams between every two frame columns in the incomplete diagram training sample. Therefore, when the method for arranging the frame beams is executed, the arrangement of the frame beams can be automatically completed by utilizing the arrangement model of the frame beams under the condition of the position of the frame column in the known design scheme, so that the design cost is reduced, and the design efficiency is improved.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings that are needed in the embodiments of the present application will be briefly described below, it should be understood that the following drawings only illustrate some embodiments of the present application and should not be considered as limiting the scope, and other related drawings can be obtained according to these drawings without inventive effort for a person skilled in the art.

FIG. 1 is a schematic diagram of a directed graph, an undirected graph, and a full graph according to an embodiment of the present application;

fig. 2 is a schematic structural diagram of a neural network according to an embodiment of the present application;

fig. 3 is a schematic flow chart of a training method for an arrangement model of a frame beam according to an embodiment of the present application;

FIG. 4 is a schematic diagram of an implementation effect of a step-and-repeat sampling technique according to an embodiment of the present application;

FIG. 5 is a schematic diagram illustrating an implementation effect of another step-and-repeat sampling technique according to an embodiment of the present application;

FIG. 6 is a schematic diagram of graph data including a single column according to an embodiment of the present application;

FIG. 7 is a schematic diagram of graph data including anti-seismic slot columns according to an embodiment of the present application;

FIG. 8 is a schematic diagram of the data of the diagram after deleting the anti-seismic slot columns according to an embodiment of the present application;

FIG. 9 is a schematic diagram of horizontal map data and oblique map data according to an embodiment of the present application;

FIG. 10 is a schematic diagram of graph data after changing a rotation angle according to an embodiment of the present application;

fig. 11 is a schematic flow chart of a method for arranging frame beams according to an embodiment of the present application;

fig. 12 is a schematic structural diagram of an arrangement model training device for frame beams according to an embodiment of the present application;

fig. 13 is a schematic structural view of an arrangement device for frame beams according to an embodiment of the present application;

fig. 14 is a schematic structural diagram of an electronic device according to an embodiment of the present application.

Detailed Description

The following description of the embodiments of the present application will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments, but not all embodiments of the present application. All other embodiments, which can be made by those skilled in the art based on the embodiments of the application without making any inventive effort, are intended to be within the scope of the application.

It should be noted that the terms "first," "second," and the like in the description and the claims of the present application and the above figures are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used may be interchanged where appropriate such that the embodiments of the application described herein may be implemented in sequences other than those illustrated or otherwise described herein. The implementations described in the following exemplary examples do not represent all implementations consistent with the application. Rather, they are merely examples of apparatus and methods consistent with aspects of the application as detailed in the accompanying claims.

The frame beam arrangement model training method and the frame beam arrangement method provided by the embodiment of the application can be executed by terminal equipment or a server for providing service for the terminal equipment.

The terminal device may be a mobile terminal, a fixed terminal or a portable terminal, such as a mobile handset, a site, a unit, a device, a multimedia computer, a multimedia tablet, an internet node, a communicator, a desktop computer, a laptop computer, a notebook computer, a netbook computer, a tablet computer, a personal communication system device, a personal navigation device, a personal digital assistants, an audio/video player, a digital camera/camcorder, a positioning device, a television receiver, a radio broadcast receiver, an electronic book device, a game device, or any combination thereof, including the accessories and peripherals of these devices, or any combination thereof. It is also contemplated that the terminal device can support any type of interface (e.g., wearable device) for the user, etc.

The server can be an independent physical server, a server cluster or a distributed system formed by a plurality of physical servers, and can also be a cloud server for providing cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, basic cloud computing services such as big data and artificial intelligent platforms and the like.

For convenience of understanding, the terms involved in the embodiments of the present application are explained below:

(1) The design drawing of the frame structure stores correct arrangement information of frame columns and frame beams, and concentrates important information such as design specifications, enterprise design experience and the like.

(2) Graph data (Graph) is made up of two sets, one set of vertices V that is non-empty, but finite, and the other set of vertices E that describes the relationship between vertices, i.e., edges. The graph data may be represented as g= (V, E). Each vertex may have its own eigenvector, each edge being a vertex pair (V, w) and V, w e V. The feature vector is used to describe the properties of the vertex itself. Wherein the graph data can be divided into directed graphs and undirected graphs. In the directed graph, vertex pairs (v, w) are ordered, the vertex pairs (v, w) are referred to as one edge of vertex v to vertex w, and (v, w) and (w, v) are two different edges. As shown on the left side of fig. 1 as a directed graph. In the undirected graph, the vertex pair (v, w) is unordered, and the vertex pair (v, w) is referred to as an edge where the vertex v is associated with the vertex w, and the edge has no specific direction, (v, w) and (w, v) are equivalent edges. As shown in fig. 1, the middle is an undirected graph. The following examples of the present application relate to the drawing data as undirected drawings.

(3) The complete graph data is a simple undirected graph in which each pair of different vertices are connected by exactly one edge, as shown on the right side of fig. 1.

(4) The graph SAGE is a generalized framework and has the characteristic of generalized learning. The advantage of inductive learning is that the information of known nodes can be used to generate embeddings for unknown nodes. The running flow of graphSAGE can be divided into three steps:

the first step: sampling the neighbor nodes of each node in the graph;

a certain number of neighbor nodes are sampled for each node as nodes of information to be aggregated (of course, information aggregation can be performed on all neighbor nodes of each node, and thus the obtained information is complete).

And a second step of: aggregating information contained by neighbor nodes;

the graphSAGE can have multiple layers, in each layer, each node aggregates the sampling result of the previous layer and its own characteristics to obtain the characteristic information of the layer, as shown in the left diagram in FIG. 2, wherein the characteristic of the bottom layer is the original input characteristic. By stacking K convolutional layers, each node can aggregate information into its K-th order domain. As shown in the right-hand diagram of fig. 2.

And a third step of: obtaining vector representations of nodes in the graph for downstream tasks;

After the graph SAGE performs K-layer aggregation, each node feature vector in the graph data is updated. At this time, the new feature vector of each node contains the feature information of the node itself and the feature information of the surrounding nodes in the K-order field. At this time, the graphSAGE completes feature extraction, and the extracted result can be used by downstream tasks.

The training method for the layout model of the frame beam, provided by the embodiment of the application, is applied to the stage before the building scheme is created, the training model of the layout model of the frame beam is used for extracting the characteristic information in the data and carrying out induction summary, and new data can be predicted according to the induction result model, such as: when the model learns a plurality of drawings, the arrangement information of the frame columns and the frame beams in the drawings is extracted, the arrangement rules are summarized, and then the frame beams can be correctly arranged when the model faces to new frame columns according to the summarized arrangement rules.

The frame beam arranging method provided by the embodiment of the application is applied to the stage of creating a building scheme, and when only the position information of the frame column exists in the current design drawing of the scheme to be built, the frame beam arranging method provided by the embodiment of the application can automatically arrange the frame beams according to the position information of the frame column and a pre-trained deep learning model, namely a frame beam arranging model training model, so that the problem of repeated inefficiency in the scheme creating stage is solved.

The preferred embodiments of the present application will be described below with reference to the accompanying drawings of the specification, it being understood that the preferred embodiments described herein are for illustration and explanation only, and not for limitation of the present application, and embodiments of the present application and features of the embodiments may be combined with each other without conflict.

Fig. 3 is a schematic flow chart of a training method for an arrangement model of a frame beam according to an embodiment of the present application. As shown in fig. 3, the method may include:

step S310, obtaining design drawings of different frame structures.

The design drawing may include position information of the frame columns, position information of the frame beams, and structural information of the frame columns and the frame beams, such as which frame beams exist between the frame columns. The position information of the frame beam may include a start point coordinate and an end point coordinate of the frame beam.

The start point coordinates of the frame beams can be obtained by using a polyline. Startpoint attribute in the CAD API, the end point coordinates of the frame beams can be obtained by using a polyline. Endpoint attribute, and the position coordinates of the frame columns can be obtained by using an entity. Position attribute.

Step 320, traversing each frame column in any design drawing by adopting a preset step-and-repeat sampling technology, and obtaining initial drawing data corresponding to different design drawings.

Because a large number of training samples are needed for training a deep learning model with higher precision, but because of the industry characteristics, the number of design drawings of a frame structure owned by a reconnaissance design enterprise is not large, so that the application needs to perform step-and-repeat sampling technical processing on each collected design drawing to extract more data sets, namely a plurality of initial drawing data corresponding to each design drawing.

In specific implementation, in the execution of the preset step-and-repeat sampling technology, each frame column in any design drawing is traversed, and for the traversed current frame column, the following steps are executed:

and step 1, determining the current frame column as a central frame column of the design drawing.

And 2, calculating first distances between other frame columns in the design drawing except the center frame column and the center frame column respectively based on the position information of each frame column and the position information of the frame beams in the design drawing, and calculating second distances between each frame beam in the design drawing and the center frame column respectively.

Specifically, based on the position information of each frame column in the design drawing, acquiring first distances between other frame columns and the central frame column respectively;

The second distance calculation method may include the following two methods:

mode 1, firstly, determining the center position of a corresponding frame beam based on the starting point coordinate and the end point coordinate of each frame beam, and then calculating a second distance between the center position and a center frame column;

mode 2, calculating the starting point distance between the starting point coordinates of the frame beams and the central frame column and the ending point distance between the ending point coordinates of the frame beams and the central frame column aiming at any frame beam; and determining the maximum distance between the starting point distance and the end point distance as the second distance between the frame beam and the central frame column.

In combination with mode 2, in a specific implementation example, for any frame beam, calculating a starting distance between a starting point coordinate and a central frame column position coordinate, and calculating an ending distance between an ending point coordinate and the central frame column position coordinate by using a math object; and then taking the larger of the two distances as the second distance of the frame beam from the center post.

It is to be understood that the calculation manner of the second distance may be set according to actual requirements, which is not limited herein.

And step 3, acquiring a first frame column with a first distance smaller than the range sensitivity constant and a first frame beam with a second distance smaller than the range sensitivity constant.

The embodiment of the sampling technology can obtain the sampling result (corresponding to a small area diagram of the design drawing, namely the diagram data of the local drawing) with the central frame column as the polar coordinate center and the range sensitivity constant as the radius in the design drawing, so that for a design drawing containing a plurality of frame columns, the diagram data corresponding to a plurality of central frame columns can be provided.

For a single frame structure design drawing, a plurality of sampling results, namely a plurality of regional plots, can be obtained after the sampling technology is executed, and a design drawing P shown in fig. 4 can be obtained after the sampling technology is executed.

And 4, acquiring initial graph data taking the central frame column as a polar coordinate center based on the structural information of the first frame column and the first frame beam. The initial map data may include a polar coordinate vector of the first frame column and a polar coordinate vector of the center frame column, and a polar coordinate position of the first frame beam. The initial map data is stored in csv format.

The arrangement of the first frame beam, the first frame column and the central frame column is represented as one piece of initial graph data, and the first frame beam can be used as a side of the initial graph data with the first frame column and the central frame column as nodes of the initial graph data. The specific initial graph data (hereinafter referred to as "graph data") is prepared as follows:

The first step: and obtaining a Draw list, and taking out one Draw object instance, wherein the Draw object instance is not processed in the following second step to the sixth step.

And a second step of: and obtaining a 'center frame column' graphic element of the Draw object instance, and establishing a polar coordinate system by taking the position of the 'center frame column' graphic element as a pole.

And a third step of: a list of Nodes in the Draw object instance is obtained, the list storing a plurality of frame column primitives. Calculating the coordinates of all frame column primitives in a polar coordinate systemWherein->Represents the pole diameter of the frame pole, < >>Representing the polar angle of the frame posts.

Fourth step: the number of the primitive of the central frame column is 0, and the numbers of other frame columns are sequentially increased by 1. The feature vector of the "center frame column" primitive is (0, 0), and the feature vectors of the other frame columns are the coordinates of each in polar coordinates. Creating initial graph data: vertices of the initial graph data are generated by numbers of frame columns or "center frame column" primitives, and feature vectors of the vertices of the initial graph data are represented by feature vectors of the frame columns or "center frame column" primitives.

Fifth step: a list of Beams in the Draw object instance is obtained, where a plurality of frame beam primitives are stored. And obtaining the starting point coordinates and the end point coordinates of each frame beam.

Sixth step: and inquiring the frame columns corresponding to the starting point coordinates and the end point coordinates, acquiring the number S of the frame column corresponding to the starting point coordinates and the number E of the frame column corresponding to the end point coordinates according to the number result in the fourth step, and combining the number S of the frame column corresponding to the starting point coordinates and the number E of the frame column corresponding to the end point coordinates into undirected edges (S, E) of the initial diagram data. Thereby obtaining a complete initial map data.

Seventh step: and (3) ending the current Draw processing, judging whether the Draw is not processed in the second step to the sixth step, if yes, returning to the first step, and if not, ending the flow.

In one example, the preset step-and-repeat sampling technique may be implemented by the following procedure:

the first step: after all the framework structure design drawings in dwg file format are obtained, as shown in (1) in fig. 5, a range sensitivity constant and a list draw are set, and the list draw is used for storing local sampling junctions.

And a second step of: for a single design drawing that is not labeled "handled", as in (2) of fig. 5, the positional information of all frame columns and the positional information of all frame beams of the design drawing are acquired.

And a third step of: from all frame columns in the second step, a single frame column not labeled "column handled" is removed and labeled "center frame column". And sequentially calculating the distances from the central frame column to other frame Columns except the central frame column in all frame Columns in the design drawing, if the distances are smaller than the range sensitivity constant, performing data regularization on the polar coordinate vectors of the frame Columns with the distances smaller than the range sensitivity constant to obtain the current polar coordinate vector of the corresponding frame column, and storing the frame Columns through a list Columns.

Fourth step: and (3) inheriting the central frame column in the third step, calculating the distance between all frame Beams and the central frame column, and storing the polar coordinate vector of the frame Beams through list Beams if the distance is smaller than the range sensitivity constant.

The third step and the fourth step can obtain the position information and the structure information of the frame column and the frame beam around one frame column (namely the central frame column) in the design drawing, namely the sampling result with the central frame column as the polar coordinate center and the range sensitivity constant as the radius.

Fifth step: a new list object Draw is declared, representing a region small graph (local) in a drawing, the "center frame column", the frame Columns stored in list Columns and the frame Beams stored in list Beams are stored to the list object Draw, and finally, the list object Draw is added to the list Draw for storage. And fifthly, storing the obtained sampling result.

Sixth step: the "center frame column" labeled frame column is labeled "column handled" and the "center frame column" label added to the frame column by the third step is cleared.

Seventh step: judging whether all frame columns of the single design drawing obtained in the second step have marks of frame columns which are not 'column handled', if so, returning to the third step, re-determining a 'center frame column' and obtaining a sampling result corresponding to the 'center frame column', and if not, marking the current drawing as 'handled'. The stored sampling results may be as shown in (3), (4) to (n) of fig. 5.

Eighth step: judging whether the mark of the design drawing is not 'handle', if so, returning to the second step, and if not, ending the flow.

The above flow inputs the local disk storage path set of a plurality of frame structure design drawings, and outputs a list drags. Multiple Draw objects, i.e., multiple region minidrawings, are stored in the Draw. The materialized representation of multiple Draw objects as shown in the storage row of fig. 5, a single Draw object stores one frame column and all frame columns and frame beams within a range of 15000 (i.e., a range sensitivity constant of 15000) around that frame column. Note that the frame columns and frame beams stored in the Draw object are both independent dwg formatted units.

The preset step-and-repeat sampling technique has 4 advantages as follows:

i. the number of frame columns and frame beams in a single regional plot manufactured by the sampling technology is greatly reduced, and the difficulty of a deep learning model in generalizing the regional plot can be greatly reduced;

through the sampling technology, about 50 area small drawings can be manufactured on average by a single frame structure design drawing, and the quantity of corresponding training data can be greatly improved;

and thirdly, the regional small map manufactured by the sampling technology only keeps the arrangement information of the frame columns and the frame beams, and eliminates irrelevant information.

The data set composed of the graph data obtained in step S320 is a tag data set of the training model, and is represented in the format of the graph data and stored in the csv format. These map data can express the ideas of building designers and structural engineers in the layout of frame columns during the design phase of the architectural plan.

And step S330, performing data regularization processing on the polar coordinate vector of the first frame column and the polar coordinate vector of the central frame column in the initial graph data to obtain regular graph data.

The regular graph data may include regular polar coordinate vectors of each frame column and polar coordinate positions of corresponding frame beams.

Specifically, acquiring polar coordinate vectors of all frame columns (including a first frame column and a central frame column in each regional small diagram (namely initial diagram data)) in each regional small diagram;

dividing the polar diameter of the polar coordinate vector of any frame column in the first frame column and the central frame column by a pre-configured distance sensitive constant to obtain a first quotient; determining a first quotient as a polar diameter regular result of the corresponding frame column; the distance sensitive constant is a distance constant customized according to actual conditions;

dividing the polar angle of the polar coordinate vector of any frame column in the first frame column and the central frame column by a pre-configured angle sensitivity constant to obtain a second quotient; determining a second quotient as a polar angle regular result of the corresponding frame column; the angle sensitivity constant is a self-defined angle constant according to actual conditions;

And determining the polar diameter regular result and the polar angle regular result as the current polar coordinate vector of the corresponding frame column.

The regular graph data is determined based on the regular polar coordinate vectors of the corresponding frame columns and the polar coordinate positions of the corresponding frame beams.

It should be noted that, the same current polar coordinate vector (i.e., the polar coordinate vector after the regularization process) corresponds to one frame column, and the polar coordinate vector before the regularization process corresponding to the same current polar coordinate vector is the range in which the frame column can exist.

And step 340, adopting a preset frame element cleaning technology to perform data cleaning on the regular graph data of each frame column to obtain cleaning graph data.

In the specific implementation, in the graph data production stage, a plurality of regular graph data are automatically produced from a framework structure design drawing of a survey design enterprise by using a CAD API through a C# code. The application can preprocess the obtained regular graph data by the following data:

frame columns in the frame structure design drawing are divided into 3 types: common column, single column and anti-seismic slit column. The individual columns are single frame columns without any beam attached, such as the frame columns in the dashed boxes in fig. 6. The anti-seismic slot columns are frame columns which are arranged on two sides of the anti-seismic slot due to the requirement of the anti-seismic slot, such as 3 pairs of frame columns which correspond up and down in a dotted line frame in fig. 7. The common column is a frame column except for a single column and an anti-seismic seam column. Because the single column, the anti-seismic slit column and the corresponding edge of the anti-seismic slit column in the regular graph data can seriously influence the effect of the deep learning model, the effect needs to be removed in the obtained regular graph data.

In the specific implementation, after regular graph data taking a central frame column as a polar coordinate center is obtained, detecting whether a single column exists in structural information of a first frame column and a first frame beam in the regular graph data, if so, removing the single column to obtain removed first graph data, wherein the single column is a frame column which does not exist a frame beam with the rest frame columns in the structural information;

detecting whether an anti-seismic seam side column exists in the structural information; if the regular graph data exists, removing the anti-seismic slit side columns, obtaining second graph data after removal, and combining with fig. 7, obtaining the regular graph data shown in fig. 8 after removing the anti-seismic slit columns. The anti-seismic seam side columns are frame columns, and the distance between two frame columns in the structural information is smaller than the minimum distance of the configured frame columns;

cleaning map data is determined based on the removed map data, which may include the removed first map data and/or the removed second map data.

And step 350, adopting a preset graph data rotation augmentation technology to carry out rotation augmentation on the cleaning graph data of each frame column so as to obtain current graph data.

The application can divide the regular graph data into two categories: the left side of the horizontal drawing data and the oblique drawing data as shown in fig. 9 is the oblique drawing data, and the right side is the horizontal drawing data. The arrangement of the frame columns is based on the axle net, the arrangement of the frame columns under the horizontal axle net and the arrangement of the frame columns under the inclined axle net being greatly different. Because the distribution of the image data is uneven, the number of the cleaning image data of the frame columns which are obliquely arranged may be far smaller than that of the cleaning image data of the frame columns which are horizontally arranged, so that the whole arrangement of the obtained cleaning image data can be inclined by a certain angle in order to further improve the accuracy of the model, and the inclination image data with different inclination degrees can be obtained.

In the specific implementation, graph data rotation augmentation processing is carried out on the cleaning graph data to obtain graph data with updated polar angles; the graph data rotation augmentation processing refers to processing of updating polar angles in polar coordinate vectors of all first frame columns in graph data according to target rotation angles randomly selected in an angle range of (0 DEG, 360 DEG), so as to obtain graph data with updated polar angles; the number of target rotation angles is at least one;

the current map data is determined based on the map data after the polar angle update.

In one embodiment, for a regular graph data, the specific implementation method of the rotation angle is as follows:

the first step: acquiring characteristic vector of each frame column in the graph data，/>，/>……；

And a second step of: the initialization returns to execution number k=0.

And a third step of: a rotation angle a is randomly selected from (0 °,360 °).

Fourth step: the polar angle coordinates of the feature vector of each frame column are added with the selected rotation angle, and the corresponding feature vector is obtained，/>，/>… …, and the value of K is +1.

Fifth step: judging whether K is smaller than 40 (preset return execution times), if so, returning to the third step, and if so, returning to the first step, wherein the rotation angle selected in each return execution process can be different. With the above embodiment, when k=40, the current map data output after the rotation angle of 40 times in fig. 10 can be obtained in combination with the regular map data shown in fig. 9.

Step S360, aiming at any graph data, determining an incomplete graph training sample meeting preset conditions based on the current polar coordinate vector of each frame column and the polar coordinate position of the corresponding frame beam in the current graph data.

The preset condition is that the included angle between any frame column and a frame beam formed by the rest frame columns is larger than a preset angle.

In specific implementation, numbering each frame column based on the current polar coordinate vector of each frame column to obtain the frame column number of each frame column;

taking edges between any two frame column numbers based on the frame column numbers of all frame columns to serve as candidate edges;

for any frame column number, detecting whether the included angle of any two candidate edges corresponding to the frame column number is larger than a preset angle, such as 5 degrees;

if not, deleting the candidate edge with longer edge length from the two candidate edges; and determining the processed candidate edges as frame beams corresponding to the frame column numbers;

if yes, determining the two candidate edges as frame beams corresponding to the frame column numbers;

and then, based on the current polar coordinate vector of each frame column and the polar coordinate position of the corresponding frame beam, acquiring an incomplete graph training sample X which corresponds to the center frame column and meets the preset condition.

Specifically, the incomplete graph training sample X satisfying the preset condition may be produced by the following procedure:

the first step: from all the obtained current map data, single current map data is obtained. The map data is not processed in the following second to seventh steps.

And a second step of: all frame column numbers in the single current diagram data are acquired, and the frame column numbers are arranged in the sequence from small to large: 1,2, … …, n.

And a third step of: a document F is created for storing the information of the frame beams.

Fourth step: obtaining the frame column number b with the smallest penultimate number, inquiring the frame column number a with the smaller number, and then carrying out frame beam, namely representing candidate edges: (a, b).

Fifth step: obtaining the frame column number c with the third smallest reciprocal, inquiring the frame column numbers a and b smaller than the frame column number c, and then carrying out frame beam, namely representing candidate edges: (a, c), (b, c). Calculating the included angle between (a, c) and (b, c), and if the included angle between the two candidate edges is smaller than 5 degrees, removing the longer candidate edge with the included angle smaller than 5 degrees. That is, for the undirected graph, for the frame column number c, if the included angle between (a, c) and (b, c) is smaller than 5 °, deleting the candidate edge with the longer length in (a, c) and (b, c), and determining the processed candidate edge as the frame beam corresponding to the corresponding frame column number; and if the included angle of any two candidate edges is not smaller than 5 degrees, determining the candidate edges as frame beams corresponding to the corresponding frame column numbers.

Sixth step: obtaining the frame column number d with the fourth last small, inquiring the frame column numbers a, b and c smaller than the frame column number d, and then carrying out frame beam, namely representing candidate edges: (a, d), (b, d), (c, d). Calculating the included angles between every two of the three candidate edges, and if the included angle between any two candidate edges is smaller than 5 degrees, removing the longer candidate edge with the included angle smaller than 5 degrees. That is, for the undirected graph, for the frame column number d, if the included angle between any two candidate edges of (a, d), (b, d) and (c, d) is smaller than 5 °, deleting the candidate edge with the longer length of the two candidate edges, and determining the processed candidate edge as the frame beam corresponding to the corresponding frame column number; and if the included angle of any two candidate edges is not smaller than 5 degrees, determining the candidate edges as frame beams corresponding to the corresponding frame column numbers.

Seventh step: according to the method, frame column numbers are obtained from small to large in sequence until n is over.

Eighth step: all candidate edges are saved into document F as reserved frame beams.

Ninth step: and judging that all the current diagram data are processed, if not, returning to the first step, and if so, ending the flow, thereby obtaining a diagram data training sample of single current diagram data, namely, a diagram data training sample which corresponds to a central frame column in the single current diagram data and meets the preset condition.

And step S370, performing iterative training on the layout model of the frame beam to be trained by taking each current diagram data and the corresponding incomplete diagram training sample as training data to obtain a trained layout model of the frame beam.

The frame beam arrangement model is used for outputting frame beam arrangement scores of frame beams between every two frame columns in the incomplete graph training sample X.

The layout model of the frame beams can comprise a graph neural network and a prediction network; the graph neural network can be three layers of graph SAGE, and is used for extracting features of an incomplete graph training sample X and outputting extracted graph data features; after feature extraction, the new feature vector of each node contains the feature information of the new feature vector and the feature information of surrounding nodes. I.e. the new vector of the frame column represented by each node has both the position information of the own frame column and the position information of other frame columns around the own frame column. The prediction network may be configured by a plurality of ResNET residual blocks, such as 4 ResNET residual blocks, and is configured to predict the arrangement of the frame beams based on each current graph data and the extracted graph data feature, and output a frame beam arrangement score of the frame beams between every two frame columns in the incomplete graph training sample X.

The training process of the layout model of the frame beam is as follows:

inputting an incomplete graph training sample X into a graph neural network, and outputting new current graph data through sampling and aggregation processing of the graph neural network, wherein the characteristic vector of each node in the new current graph data not only comprises the position information of a self frame column, but also comprises the position information of other frame columns around the self frame column.

Inputting new current diagram data into a prediction network, and predicting the score of the edge between any two frame columns, namely the frame beam arrangement score of the frame beam;

if the frame beam arrangement score is larger than the configured score threshold value, the frame beam arrangement score indicates that the frame beam is required to be arranged between the frame columns corresponding to the two nodes, and if the frame beam arrangement score is smaller than the configured score threshold value, the frame beam arrangement score indicates that the frame beam is not required to be arranged between the frame columns corresponding to the two nodes.

And then, updating model parameters of the layout model of the frame beam by adopting the current graph data and the configured loss function which correspond to the incomplete graph training sample X and are used as the label data, so as to obtain the trained layout model of the frame beam.

Wherein, the loss function is:

wherein,,representing two nodes in the incomplete graph training sample X; / >A frame beam arrangement score representing an edge between two nodes in the current graph data as label data, respectively: if there is an edge between two nodes in the current graph data, +.>=1, if there is no edge between two nodes in the current graph data, +.>=0；/>Representing a predictive score of an arrangement model of a frame beam for an edge between two nodes, if the arrangement model of the frame beam predicts an edge between two nodes, then +.>>0, if frame beamIf the placement model of (1) predicts that no edge exists between two nodes, then-><0，/>The larger the model of the layout of the frame beams, the higher the likelihood that an edge exists between the two nodes.

The accuracy of the layout model of the frame beams can be expressed as:

fig. 11 is a flow chart of a method for arranging frame beams according to an embodiment of the present application. As shown in fig. 11, the method may include:

step 1110, obtaining position information of each frame column in the current design drawing of the scheme to be built.

Step S1120, after determining any frame column as the central frame column of the current design drawing, calculating the distances between other frame columns in the current design drawing except the central frame column and the central frame column based on the position information of each frame column in the current design drawing.

And step S1130, performing data regularization processing on the polar coordinate vector of the frame column with the distance smaller than the range sensitivity constant to obtain the current polar coordinate vector of the corresponding frame column.

Step S1140, acquiring to-be-input graph data corresponding to the center frame column and meeting a preset condition based on the current polar coordinate vector of the frame column with a distance smaller than the range sensitivity constant.

The preset condition is that the included angle between any frame column and the frame beam formed by the rest frame columns is larger than the preset angle.

Step S1150, inputting the data of the to-be-input diagram into a frame beam arrangement model trained by a frame beam arrangement model training method, and obtaining the frame beam arrangement score of the frame beam between every two frame columns in the data of the to-be-input diagram.

Step S1160, determining the frame beams to be drawn in the data of the graph to be input based on the preset local arrangement score threshold and the preset local arrangement score threshold.

Before the step is executed, a basic layout score threshold value is required to be preset for the frame beams, at least one important frame column in the image data to be input can be determined according to actual layout requirements, and a local layout score threshold value is preset for the frame beams corresponding to each important frame column, wherein each local layout score threshold value can be different or the same and is required to be set according to the actual layout requirements, and the application is not limited herein.

In the implementation, if any frame beam in the image data to be input does not have a preset local arrangement score threshold, acquiring a first appointed frame beam with a frame beam arrangement score greater than a preset basic arrangement score threshold, and determining the first appointed frame beam as the frame beam to be drawn;

if a preset local arrangement score threshold value exists in the target frame beams to be input into the graph data, a second designated frame beam with the frame beam arrangement score larger than the local arrangement score threshold value is obtained, and the second designated frame beam is determined to be the frame beam to be drawn.

And step S1170, drawing the frame beams to be drawn of the image data to be input by adopting a CAD technology so as to realize automatic arrangement of the frame beams.

In some embodiments, after the drawing result is obtained in step S1170, the preconfigured local arrangement score threshold may be modified according to a plurality of deviation values in the preset service requirement range to obtain a plurality of new local arrangement score thresholds, and for each new local arrangement score threshold, step S1160 is performed back to obtain a frame beam to be drawn corresponding to each new local arrangement score threshold, so as to obtain a drawing result corresponding to the corresponding frame beam to be drawn. The target painting result, i.e. the automatic arrangement of the target frame beams, may then be determined from the plurality of painting results. The preset service demand range is a range of a plurality of deviation values determined according to demand deviation among local arrangement score thresholds corresponding to service demands of a plurality of adjacent historical time periods.

In some embodiments, configuring the base placement score threshold specifically may include: determining the current building level corresponding to the scheme to be built according to the design level of the current design drawing; and determining a basic arrangement score threshold corresponding to the current building level based on the corresponding relation between the preset building level and the basic arrangement score threshold.

The current building level is determined based on preset earthquake-proof fortification intensity and frame beam live load, and comprises a first-level building, a second-level building, a third-level building, a fourth-level building, a fifth-level building, a sixth-level building and a seventh-level building;

the corresponding relation between the preset earthquake-proof fortification intensity and the movable load of the frame beam and the preset building grade is shown in table 1:

the correspondence between the preset building level and the basic layout score threshold may include that the basic layout score threshold corresponding to the first-level building is 0, the basic layout score threshold corresponding to the second-level building is 0.1, the basic layout score threshold corresponding to the third-level building is 0.2, the basic layout score threshold corresponding to the fourth-level building is 0.3, the basic layout score threshold corresponding to the fifth-level building is 0.4, the basic layout score threshold corresponding to the sixth-level building is 0.5, and the basic layout score threshold corresponding to the seventh-level building is 0.6. The corresponding relation between the preset building level and the basic arrangement score threshold is shown in table 2:

In some embodiments, before the frame beam to be drawn of the image data to be input is drawn, numbering the corresponding frame columns based on the current polar coordinate vector of each frame column in the image data to be input to obtain frame column numbers of each frame column;

drawing the frame beam to be drawn of the image data to be input by adopting a CAD technology, comprising the following steps: acquiring frame column numbers of two frame columns corresponding to any frame beam to be drawn; adopting a CAD technology, aiming at frame column numbers of two frame columns, positioning the corresponding frame columns; and drawing frame beams for the two positioned frame columns.

Corresponding to the method, the embodiment of the application also provides a training device for the layout model of the frame beam, as shown in fig. 12, which comprises:

an obtaining unit 1201, configured to obtain design drawings of different frame structures; the design drawing comprises position information of a frame column, position information of a frame beam and structural information of the frame column and the frame beam;

the sampling unit 1202 is configured to traverse each frame column in any design drawing by using a preset step-and-repeat sampling technique, so as to obtain initial drawing data corresponding to each frame column; the initial map data comprise initial polar coordinate vectors of the frame columns and polar coordinate positions of corresponding frame beams;

The processing unit 1203 is configured to perform data regularization processing on the initial graph data of each frame column to obtain regular graph data, where the regular graph data includes a regular polar coordinate vector of each frame column and a polar coordinate position of a corresponding frame beam;

the cleaning unit 1204 is configured to perform data cleaning on the regular graph data of each frame column by using a preset frame element cleaning technology, so as to obtain cleaning graph data;

an augmentation unit 1205, configured to rotationally augment the cleaning map data of each frame column by using a preset map data rotation augmentation technique, so as to obtain current map data;

a determining unit 1206, configured to determine, for any graph data, an incomplete graph training sample corresponding to the center frame column and meeting a preset condition, based on a current polar coordinate vector of each frame column in the graph data and a polar coordinate position of a corresponding frame beam; the preset condition is that the included angle between any frame column and a frame beam formed by the rest frame columns is larger than a preset angle;

the training unit 1207 is configured to iteratively train the layout model of the frame beam to be trained by using each graph data and the corresponding incomplete graph training sample as training data, so as to obtain a trained layout model of the frame beam, where the layout model of the frame beam is used to output the layout score of the frame beam between every two frame columns in the incomplete graph training sample.

The method comprises the steps of traversing each frame column in any design drawing by adopting a preset step-and-repeat sampling technology to obtain initial drawing data corresponding to each design drawing, wherein the steps comprise:

traversing each frame column in any design drawing by adopting a preset step-and-repeat sampling technology, and aiming at the traversed current frame column, executing the following steps:

determining the current frame column as a central frame column of the design drawing;

calculating first distances between other frame columns in the design drawing except for the center frame column and the center frame column respectively based on the position information of each frame column and the position information of the frame beam in the design drawing, and calculating second distances between each frame beam in the design drawing and the center frame column respectively;

acquiring a first frame column with the first distance smaller than a range sensitivity constant and a first frame beam with the second distance smaller than the range sensitivity constant;

acquiring initial graph data taking the central frame column as a polar coordinate center based on the structural information of the first frame column and the first frame beam; the initial map data comprises polar coordinate vectors of the first frame column and the central frame column, and polar coordinate positions of the first frame beam;

The method for cleaning the regular graph data of each frame column by adopting a preset frame element cleaning technology comprises the following steps of:

detecting whether a single column exists in the structural information; if so, removing the single column, and obtaining first graph data after removal, wherein the single column is a frame column which does not have frame beams with the rest frame columns in the structural information;

detecting whether an anti-seismic seam side column exists in the structural information; if the structural information exists, removing the anti-seismic seam side columns, and obtaining second graph data after removal, wherein the anti-seismic seam side columns are frame columns, and the distance between two frame columns in the structural information is smaller than the minimum distance of the configured frame columns;

determining cleaning map data based on the removed map data, wherein the removed map data comprises removed first map data and/or removed second map data;

the method for obtaining the cleaning image data of each frame column comprises the following steps of:

performing graph data rotation augmentation treatment on the cleaning graph data to obtain graph data with updated polar angles; the map data rotation augmentation processing refers to processing of updating polar angles in polar coordinate vectors of all first frame columns in the map data according to target rotation angles randomly selected in an angle range of (0 degrees and 360 degrees) to obtain map data with updated polar angles; the number of the target rotation angles is at least one;

The functions of each functional unit of the automatic frame beam arranging device based on the artificial intelligence provided by the embodiment of the application can be realized through the steps of the method, so that the specific working process and the beneficial effects of each unit in the automatic frame beam arranging device based on the artificial intelligence provided by the embodiment of the application are not repeated here.

Corresponding to the method, the embodiment of the application also provides a device for arranging frame beams, as shown in fig. 13, which comprises:

an obtaining unit 1301, configured to obtain position information of each frame column in a current design drawing of a scheme to be built;

a calculating unit 1302, configured to calculate, after determining any frame column as a center frame column of the current design drawing, distances between other frame columns in the current design drawing, except for the center frame column, and the center frame column, respectively, based on position information of each frame column in the current design drawing;

the processing unit 1303 is used for performing data regularization processing on the polar coordinate vectors of the frame columns with the distance smaller than the range sensitivity constant to obtain the current polar coordinate vectors of the corresponding frame columns;

The acquiring unit 1301 is further configured to acquire to-be-input graph data corresponding to the center frame column and meeting a preset condition, based on a current polar coordinate vector of the frame column with a distance less than a range sensitivity constant; the preset condition is that the included angle between any frame column and a frame beam formed by the rest frame columns is larger than a preset angle;

a determining unit 1304, configured to obtain a first designated frame beam with a frame beam arrangement score greater than a preset basic arrangement score threshold, and determine the first designated frame beam as a frame beam to be drawn, if any frame beam in the map data to be input does not have the preset local arrangement score threshold;

And the drawing unit 1305 is used for drawing the frame beams to be drawn of the image data to be input by adopting a CAD technology so as to realize automatic arrangement of the frame beams.

The embodiment of the present application further provides an electronic device, as shown in fig. 14, including a processor 1410, a communication interface 1420, a memory 1430, and a communication bus 1440, where the processor 1410, the communication interface 1420, and the memory 1430 complete communication with each other through the communication bus 1440.

A memory 1430 for storing a computer program;

the processor 1410, when executing the program stored in the memory 1430, performs the following steps:

determining an incomplete graph training sample which corresponds to the center frame column and meets a preset condition based on the current polar coordinate vector of each frame column and the polar coordinate position of the corresponding frame beam in the current graph data aiming at any graph data; the preset condition is that the included angle between any frame column and a frame beam formed by the rest frame columns is larger than a preset angle;

Alternatively, the following steps are implemented:

inputting the to-be-input graph data into a frame beam arrangement model trained based on a frame beam arrangement model training method to obtain frame beam arrangement scores of frame beams between every two frame columns in the to-be-input graph data;

If a preset local arrangement score threshold exists in the target frame beams in the image data to be input, acquiring a second designated frame beam with the frame beam arrangement score larger than the local arrangement score threshold, and determining the second designated frame beam as the frame beam to be drawn;

The communication bus mentioned above may be a peripheral component interconnect standard (Peripheral Component Interconnect, PCI) bus or an extended industry standard architecture (Extended Industry Standard Architecture, EISA) bus, or the like. The communication bus may be classified as an address bus, a data bus, a control bus, or the like. For ease of illustration, the figures are shown with only one bold line, but not with only one bus or one type of bus.

The communication interface is used for communication between the electronic device and other devices.

The Memory may include random access Memory (Random Access Memory, RAM) or may include Non-Volatile Memory (NVM), such as at least one disk Memory. Optionally, the memory may also be at least one memory device located remotely from the aforementioned processor.

The processor may be a general-purpose processor, including a central processing unit (Central Processing Unit, CPU), a network processor (Network Processor, NP), etc.; but also digital signal processors (Digital Signal Processing, DSP), application specific integrated circuits (Application Specific Integrated Circuit, ASIC), field programmable gate arrays (Field-Programmable Gate Array, FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

Since the implementation manner and the beneficial effects of the solution to the problem of each device of the electronic apparatus in the foregoing embodiment may be implemented by referring to each step in the embodiment shown in fig. 3 or fig. 11, the specific working process and the beneficial effects of the electronic apparatus provided by the embodiment of the present application are not repeated herein.

In yet another embodiment of the present application, there is also provided a computer-readable storage medium having instructions stored therein, which when run on a computer, cause the computer to perform the method for training a layout model of a frame beam or the method for layout of a frame beam according to any of the above embodiments.

In a further embodiment of the present application, a computer program product comprising instructions, which when run on a computer, causes the computer to perform the method of training a layout model of a frame beam or the method of layout of a frame beam according to any of the embodiments described above is also provided.

It will be appreciated by those skilled in the art that embodiments of the present application may be provided as methods, systems, or computer program products. Accordingly, embodiments of the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, embodiments of the present application may take the form of a computer program product on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.

Embodiments of the present application are described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

While preferred embodiments of the present application have been described, additional variations and modifications in those embodiments may occur to those skilled in the art once they learn of the basic inventive concepts. It is therefore intended that the following claims be interpreted as including the preferred embodiment and all such alterations and modifications as fall within the scope of the embodiments of the application.

It will be apparent to those skilled in the art that various modifications and variations can be made in the embodiments of the present application without departing from the spirit or scope of the embodiments of the application. Thus, if such modifications and variations of the embodiments of the present application fall within the scope of the claims and the equivalents thereof, it is intended that such modifications and variations be included in the embodiments of the present application.

Claims

1. A method of training an arrangement model of a frame beam, the method comprising:

performing iterative training on the layout model of the frame beams to be trained by taking the current drawing data and the corresponding incomplete drawing training samples as training data to obtain a trained layout model of the frame beams, wherein the layout model of the frame beams is used for outputting the layout scores of the frame beams between every two frame columns in the incomplete drawing training samples;

2. The method for training an arrangement model of a frame beam according to claim 1, wherein the position information of the frame beam includes a start point coordinate and an end point coordinate of the frame beam;

Calculating a second distance between each frame beam in the design drawing and the central frame column respectively, wherein the second distance comprises the following steps:

calculating a starting point distance between a starting point coordinate of the frame beam and the central frame column and a finishing point distance between an ending point coordinate of the frame beam and the central frame column aiming at any frame beam;

and determining the maximum distance between the starting point distance and the ending point distance as a second distance between the frame beam and the central frame column.

3. The method for training the layout model of the frame beam according to claim 1, wherein performing data regularization on the initial graph data of each frame column to obtain regular graph data comprises:

dividing the polar diameter of the polar coordinate vector of any frame column in the first frame column and the central frame column by a pre-configured distance sensitive constant to obtain a first quotient; determining a first quotient as a polar diameter regular result of the corresponding frame column;

dividing the polar angle of the polar coordinate vector of any frame column in the first frame column and the central frame column by a pre-configured angle sensitivity constant to obtain a second quotient; determining a second quotient as a polar angle regular result of the corresponding frame column;

Determining a polar radius regular result and a polar angle regular result as regular polar coordinate vectors of the corresponding frame columns; the regular graph data is determined based on the regular polar coordinate vectors of the corresponding frame columns and the polar coordinate positions of the corresponding frame beams.

4. The method for training an arrangement model of frame beams according to claim 1, wherein determining an incomplete map training sample satisfying a preset condition based on a current polar coordinate vector of each frame column in the current map data and a polar coordinate position of a corresponding frame beam, comprises:

numbering each frame column based on the current polar coordinate vector of each frame column to obtain the frame column number of each frame column;

taking edges between any two frame column numbers based on the frame column numbers of the frame columns to serve as candidate edges;

for any frame column number, detecting whether the included angle of any two candidate sides corresponding to the frame column number is larger than a preset angle;

if not, deleting the candidate edge with longer edge length from the two candidate edges, and determining the processed candidate edge as the frame beam corresponding to the frame column number;

And acquiring an incomplete graph training sample which corresponds to the center frame column and meets a preset condition based on the polar coordinate vector of each frame column and the polar coordinate position of the corresponding frame beam.

5. The method of training an arrangement model of frame beams according to claim 1, wherein the arrangement model of frame beams comprises a graph neural network and a prediction network;

the graph neural network performs feature extraction on the incomplete graph training sample and outputs extracted graph data features;

and the prediction network predicts the arrangement of the frame beams based on the current image data and the extracted image data characteristics, and outputs the frame beam arrangement scores of the frame beams between every two frame columns in the incomplete image training sample.

6. A method of arranging frame beams, the method comprising:

inputting the to-be-input graph data into a frame beam layout model trained based on the frame beam layout model training method according to any one of claims 1-5 to obtain the frame beam layout score of each two frame beams between the frame columns in the to-be-input graph data;

7. The method for arranging frame beams according to claim 6, wherein before the frame beams to be drawn of the map data to be input are drawn by using CAD technology, the method further comprises:

the step of configuring the basic placement score threshold includes:

determining the current building level corresponding to the scheme to be built according to the design level of the current design drawing;

and determining a basic arrangement score threshold corresponding to the current building level based on the corresponding relation between the preset building level and the basic arrangement score threshold.

8. The method of arranging frame beams according to claim 7, wherein the current building level is determined based on a preset earthquake-proof intensity and a frame beam live load, the current building level including a primary building, a secondary building, a tertiary building, a quaternary building, a fifth building, a sixth building, and a seventh building;

the corresponding relation between the preset building level and the basic layout score threshold value comprises that the basic layout score threshold value corresponding to the first-level building is 0, the basic layout score threshold value corresponding to the second-level building is 0.1, the basic layout score threshold value corresponding to the third-level building is 0.2, the basic layout score threshold value corresponding to the fourth-level building is 0.3, the basic layout score threshold value corresponding to the fifth-level building is 0.4, the basic layout score threshold value corresponding to the sixth-level building is 0.5, and the basic layout score threshold value corresponding to the seventh-level building is 0.6.

9. The method for arranging frame beams according to claim 6, wherein before the frame beams to be drawn of the map data to be input are drawn by using CAD technology, the method further comprises:

numbering corresponding frame columns based on the current polar coordinate vector of each frame column in the to-be-input graph data to obtain frame column numbers of each frame column;

drawing the frame beam to be drawn of the image data to be input by adopting a CAD technology, wherein the drawing comprises the following steps:

acquiring frame column numbers of two frame columns corresponding to any frame beam to be drawn;

adopting a CAD technology, aiming at frame column numbers of the two frame columns, positioning the corresponding frame columns;

and drawing frame beams for the two positioned frame columns.

10. An arrangement model training device for frame beams, the device comprising:

the training unit is used for iteratively training the layout model of the frame beams to be trained by taking the image data and the corresponding incomplete image training samples as training data to obtain a trained layout model of the frame beams, wherein the layout model of the frame beams is used for outputting the layout scores of the frame beams between every two frame columns in the incomplete image training samples;

11. An arrangement of frame beams, characterized in that the arrangement comprises:

inputting the to-be-input graph data into a frame beam layout model trained based on the frame beam layout model training method according to any one of claims 1-5, so as to obtain the frame beam layout score of each two frame beams between the frame columns in the to-be-input graph data;

12. A computer readable storage medium, characterized in that a computer program is stored in the computer readable storage medium, which computer program, when being executed by a processor, implements the method for training the layout model of frame beams according to any one of claims 1-5 or the method for arranging frame beams according to any one of claims 6-9.