A method and device for training and generating a small sample of a high-rise residential building plane in a cold region based on a diffusion model

Abstract

The application provides a diffusion model-based cold region high-rise residential building plane small sample training and generation method and device. The method comprises the following steps: collecting and sorting the original data set of standard plane graphs of cold region high-rise residential buildings, analyzing to obtain the function partition, house type contour graph and entrance information data set corresponding to each standard plane graph, and generating a text description of statistical information corresponding to each house type plane to serve as an original training data set; training a generation model from a house type contour to a house type plane graph based on a diffusion model framework; developing a tool platform capable of intelligently designing and automatically generating a house type plane graph, which can generate a house type plane graph meeting a condition from basic house type contours and entrance and exit graphic information in combination with specific requirement text information of each function room. The application provides an efficient and high-accuracy automatic residential building plane scheme generation method and device.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of high-rise residential building plane generation, and in particular to a high-rise residential building plane small sample training and generation method and device based on a diffusion model. BACKGROUND

[0002] Residential buildings are important places for people to live and live. With the improvement of living standards and the improvement of individual aesthetic needs, people pay more and more attention to the level and convenience of their own living and living houses, and the demand for residential building layout design is also more diversified. Different users have different needs for residential building layout design, and it is particularly important to generate healthy, livable and comfortable high-rise residential building layout design scheme to meet people's diversified demand for living space by combining the needs of different families or groups.

[0003] Due to the vast territory and complex geographical environment of China, there are significant differences in the general sense of residential building planes in the north and south, and the harsh climate in the north requires higher requirements for building thermal performance such as heat preservation and cold prevention, so it is necessary to design a plane more suitable for high-rise buildings in the cold region.

[0004] The current most building design thinking is a "top-down" thinking mode based on the designer as the main body and making design decisions based on the subjective judgment of the designer. At present, to generate a residential building plane, mainly through manual use of computer-aided design tools by designers to obtain through human-computer interaction. The main disadvantage of using traditional CAD design is low design deepening efficiency, and the entire process almost involves manual participation, and the quality and rate of generation are greatly dependent on the experience and business level of the participants, and residential building plane design drawings cannot be generated in batches.

[0005] Another disadvantage is a high error rate. For some designers who lack design experience, there may be problems such as not meeting design specifications and standards when creating a residential building plane. Moreover, building creation errors can lead to design rework, extended construction period and other subsequent effects, and high error correction costs.

[0006] In addition, since the design process is based on the designer as the main body, there is often a problem of low design personalization. The design decisions made by the subjective judgment of the designer are not personalized to some extent, and cannot meet the diversified demand for residential building planes.

[0007] Therefore, how to provide a high-rise residential building plane generation design method that fits the regional climate and meets the personalized multi-cloud demand is a key technical problem to be solved in the field. SUMMARY

[0008] The application provides a small sample training and generation method and device for high-rise residential building planes in cold regions based on a diffusion model.

[0009] The application is implemented by the following technical solutions, and provides a small sample training and generation method for high-rise residential building planes in cold regions based on a diffusion model.

[0010] Obtain a high-rise residential building plane in a cold region, analyze the corresponding residential building area, house type area, house type layout, house type contour and residential structure size information of each standard plane to obtain original training data;

[0011] Organize the above plane information into a text description, form a mapping relationship with the corresponding picture, and use a color block labeling method to label the indoor function layout as picture input conditions; the text data and picture data are used as input conditions, and a low-order adaptive modeling LoRA of a large language model is used to train a high-rise residential building plane diffusion model in a cold region;

[0012] After the diffusion model training is completed, the target plane house type text description and contour map are used as input parameters, and the high-rise residential building plane in a cold region generated by the trained diffusion model is used as an output end.

[0013] Further, the method further comprises: evaluating whether the generated high-rise residential building plane in a cold region meets the specification requirements by using an evaluation model, and if not, re-generating until the requirements are met.

[0014] Further, the standards in the evaluation process of the evaluation model include:

[0015] a, whether the adjacent function areas are reasonable; b, whether the area of a certain function area is reasonable; c, whether it is generated according to the input contour; d, whether all the required functions are generated;

[0016] If the evaluation standards are not met, re-generate until the requirements are met.

[0017] Further, the plane information is organized into a text description, and a corresponding relationship is formed with the corresponding picture, specifically:

[0018] The plane information is organized into a standard format of “x room x hall x toilet x kitchen x balcony - x square meter” for machine recognition, and a corresponding relationship is formed with the corresponding picture as text training data set;

[0019] The indoor function layout is marked by color block marking method as a picture input condition, specifically:

[0020] The function area, contour and entrance of the collected plan are marked with different colors to form a picture training data set;

[0021] The specific marking method is that the main entrance is red, the balcony is orange, the bedroom is yellow, the kitchen is green, the dining room is blue, and the bathroom is purple.

[0022] Further, the target plan house type text description and contour map are used as input parameters, specifically:

[0023] According to the format of the plan text description, the requirement of "x room x hall x bathroom x kitchen x balcony - x square meter" is input as a text input condition;

[0024] The target plan house contour map is input, and the control network in the diffusion model is used to perform edge detection on the contour map to convert it into an edge line map as a picture input condition.

[0025] Further, the output picture that meets the evaluation requirements is vectorized, specifically:

[0026] The picture that meets the evaluation requirements is imported into the vectorization module, and the irrelevant information in the image is eliminated through image preprocessing to recover the useful true information;

[0027] In the vectorization module, content recognition is performed through edge detection, contour tracking and contour expression, the data of walls, windows and doors are layered, and the vectorization of the sub-layer graph is completed.

[0028] Further, the method further comprises:

[0029] The specific function area of the standard plan of the high-rise residential building in cold regions is obtained, and the generated house layout of the high-rise residential building in cold regions is trained to obtain an AI evaluation model of the specific function area;

[0030] According to the specific function area AI evaluation model, the specific function area in the generated high-rise residential building in cold regions is evaluated within a reasonable range, the generated high-rise residential building in cold regions is consistent with the input target plan house contour map, and the high-rise residential building plan in cold regions is confirmed and output.

[0031] The present application proposes a kind of based on diffusion model's high-rise residential building plan in cold regions small sample training and generation device, the device includes:

[0032] Image data processing module: This module processes the plan data of high-rise residential buildings in cold regions during the training process. The image correction unit corrects and scales the images to a uniform resolution of 512×512 pixels. The data annotation module then annotates the functional areas, outlines, and entrances of the corrected plan with different colors. The specific annotation method is: the main entrance is red, the balcony is orange, the bedroom is yellow, the kitchen is green, the dining room is blue, and the bathroom is purple.

[0033] Text information annotation module: used to annotate text information describing the characteristics of high-rise residential buildings in cold regions. Based on the information extraction unit's acquisition of the standard floor area, unit area, unit layout, and unit outline data corresponding to the high-rise residential building floor plan, the information annotation unit uses the standard format of "x room x hall x bathroom x kitchen x balcony - x square meters" as the image description for machine recognition and serves as the text training dataset;

[0034] Small sample training module: used to train the cold-region high-rise residential building plan feature model based on a small sample size. It uses a small amount of text and image training data sets as input for training and constructs a cold-region high-rise residential building plan diffusion model.

[0035] Building plan generation module: used to generate the plan of a high-rise residential building in cold regions based on text feature description. It takes the outline of the target unit type and the location of the entrance and exit of the target unit type of the high-rise residential building in cold regions as the conditional image input unit, and the text description of the unit type of the high-rise residential building in cold regions as the input, uses the trained diffusion model as the black box processing module, and outputs the plan of the high-rise residential building in cold regions;

[0036] Post-evaluation module: used to evaluate whether the floor plan generated by the floor plan generation module meets the requirements of residential building design. If not, it will be regenerated until it meets the requirements.

[0037] The present invention also proposes an electronic device comprising a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, it implements the steps of a method for training and generating a small sample of cold-region high-rise residential building plans based on a diffusion model.

[0038] The present invention also proposes a computer-readable storage medium for storing computer instructions, which, when executed by a processor, implement the steps of a method for training and generating a small sample of cold-region high-rise residential building plans based on a diffusion model.

[0039] Beneficial effects of the present invention:

[0040] Compared with the existing method, the method provided by the application has the following innovative points: 1) breaking through the limitation of the designer's personal experience accumulation in the existing high-rise building layout design, realizing the generation of a high-rise residential building layout plane that meets the regional characteristics of the cold region and diversified needs by fusing deep learning modeling technology; 2) compared with the existing method that fuses deep learning modeling, the method provided by the application realizes small sample training through low-order adaptive modeling of a large language model, solves the problem that the existing training method needs a large-scale training data set and the training data is difficult to obtain, and avoids the bottleneck problem that the existing method has a large workload and low efficiency in the large-scale data set labeling process; 3) the method introduced in the method significantly improves the interaction efficiency of the designer and the model by introducing multi-modal information of “text-image” cooperation, reduces the promotion threshold of the method in the engineering practice design process, and has strong generalizability. BRIEF DESCRIPTION OF DRAWINGS

[0041] In order to more clearly illustrate the technical solutions in the embodiments of the application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description only constitute the embodiments of the application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of the provided drawings.

[0042] Figure 1 is a deep learning model training process schematic diagram for automatically generating a high-rise residential building plane in the cold region in the embodiment of the application;

[0043] Figure 2 is a generation design process schematic diagram for automatically generating a high-rise residential building plane in the cold region in the embodiment;

[0044] Figure 3 is a running principle diagram of a diffusion model for automatically generating a high-rise residential building plane model in the cold region applied in the embodiment;

[0045] Figure 4 is a whole process schematic diagram of training and generating a plane diagram containing a layout description word in the embodiment;

[0046] Figure 5 is an original plane diagram example of a high-rise residential building in the cold region as training data;

[0047] Figure 6 is a labeled high-rise residential building plane diagram in the cold region;

[0048] Figure 7 is a high-rise residential building layout plane diagram in the cold region automatically generated by a diffusion model in the embodiment;

[0049] Figure 8A functional module schematic diagram of a small sample training and generating device for a diffusion model-based high-rise residential building plane in cold regions.

[0050] Figure 9 A module internal function unit schematic diagram of the small sample training and generating device for the diffusion model-based high-rise residential building plane in cold regions. DETAILED DESCRIPTION

[0051] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0052] The small sample training and generating method and device for the diffusion model-based high-rise residential building plane in cold regions automatically generate a high-rise residential building plane by fusing deep learning modeling technology, the input is an apartment layout external contour, an apartment layout main entrance position and an apartment layout text description, and the output is a high-rise residential building plane map meeting user needs. The user can perform vector processing on the generated result to realize deepening design of the high-rise residential building plane. The method for automatically generating a residential building plane in the embodiment is obtained by training a diffusion model through selecting a small number of high-quality standard plane maps as a data set in more than 1,000 high-rise residential building planes in cold regions. The output result of the model meets the standard specification while ensuring diversity. For example, Figures 1 to 6 Figure 1 is a deep learning model training process schematic diagram for automatically generating a high-rise residential building plane in cold regions in the embodiment of the present application; Figure 2 is a generating design process schematic diagram for automatically generating a high-rise residential building plane in cold regions in the embodiment; Figure 3 is a running principle diagram of a diffusion model for automatically generating a high-rise residential building plane in cold regions in the embodiment; Figure 4 is a whole process schematic diagram of training and generating a plane map including apartment layout description words; Figure 5 is an original plane map example of a high-rise residential building in a cold region range as training data; Figure 6 is a labeled high-rise residential building plane map in cold regions.

[0053] In combination with Figures 1-9 , the present application proposes a small sample training and generating method for a diffusion model-based high-rise residential building plane in cold regions, which comprises the following steps:

[0054] ​Obtain floor plans of high-rise residential buildings in cold regions, analyze and obtain the residential building area, unit area, unit layout, unit outline, and residential structural dimensions corresponding to each standard floor plan as the original training data;

[0055] The above floor plan information is organized into text descriptions and mapped to the corresponding images as text input conditions. The indoor functional layout is annotated using the color block annotation method as image input conditions. Using the text data and image data as input conditions, a low-rank adaptation of large language models (LoRA) is used to train a planar diffusion model for high-rise residential buildings in cold regions.

[0056] After the diffusion model training is completed, the target floor plan text description and outline map are used as input parameters, and the cold-region high-rise residential building plan generated by the trained diffusion model is used as the output end; specifically: the cold-region high-rise residential building plan diffusion model constructed above is applied, with the cold-region high-rise residential building target floor plan, target floor plan entrance and exit locations, and high-rise residential building floor plan text description as input, and the cold-region high-rise residential building plan generated by the trained diffusion model is used as the output.

[0057] Furthermore, the method also includes: evaluating whether the generated cold-region high-rise residential building plan meets the specification requirements through an evaluation model, and if not, regenerating it until it meets the requirements.

[0058] Furthermore, the evaluation criteria of the evaluation model during the evaluation process include:

[0059] a. Whether the adjacent functional areas are reasonable; b. Whether the area of ​​a functional area is reasonable; c. Whether it is generated completely according to the input outline; d. Whether all required functions are generated;

[0060] If it does not meet the evaluation criteria, it will be regenerated until it meets the requirements.

[0061] Furthermore, the floor plan information is organized into text descriptions and corresponding to the corresponding pictures, specifically:

[0062] The floor plan information is organized into a standard format for machine recognition, such as "x room x hall x bathroom x kitchen x balcony - x square meters", and a corresponding relationship is formed with the corresponding image as a text training dataset;

[0063] Use the color block annotation method to annotate the indoor functional layout as the image input condition, specifically:

[0064] The functional areas, outlines, and entrances of the collected floor plans are labeled with different colors to form an image training dataset;

[0065] The specific marking method is: main entrance is red, balcony is orange, bedroom is yellow, kitchen is green, dining room is blue, and bathroom is purple.

[0066] Furthermore, the target floor plan text description and outline are used as input parameters, specifically:

[0067] Enter the requirement "x rooms x halls x bathrooms x kitchens x balconies - x square meters" in the format of the floor plan text description as the text input condition;

[0068] Input the target floor plan outline map, use the control network in the diffusion model to perform edge detection on the outline map and convert it into an edge line map as the image input condition.

[0069] Furthermore, the output images that meet the evaluation requirements are vectorized. Specifically:

[0070] Import the images that meet the evaluation requirements into the vectorization module, and then perform image preprocessing to eliminate irrelevant information in the image and restore useful real information;

[0071] In the vectorization module, content recognition is performed through edge detection, contour tracking, and contour expression, and the data of walls, windows, and doors are layered to complete the vectorization of layered graphics.

[0072] Furthermore, the method further comprises:

[0073] Obtain the area of ​​specific functional areas in the standard floor plan of high-rise residential buildings in cold regions, and train it with the corresponding generated apartment layout of high-rise residential buildings in cold regions to obtain an AI evaluation model for the area of ​​specific functional areas;

[0074] According to the AI ​​evaluation model of the specific functional area area, the specific functional area of ​​the generated cold-region high-rise residential building is evaluated to be within a reasonable range, the outer contour of the generated cold-region high-rise residential building is consistent with the input target plane apartment outline map, and the cold-region high-rise residential building plane is confirmed and output.

[0075] The present invention proposes a device for training and generating small sample plans of high-rise residential buildings in cold regions based on a diffusion model, the device comprising:

[0076] Image data processing module: This module processes the plan data of high-rise residential buildings in cold regions during the training process. The image correction unit corrects and scales the images to a uniform resolution of 512×512 pixels. The data annotation module then annotates the functional areas, outlines, and entrances of the corrected plan with different colors. The specific annotation method is: the main entrance is red, the balcony is orange, the bedroom is yellow, the kitchen is green, the dining room is blue, and the bathroom is purple.

[0077] The text information labeling module is used for labeling the characteristic description text information of the high-rise residential building in cold regions, and based on the residential standard layer area, the house type area, the house type layout and the house type contour data corresponding to the high-rise residential building plan obtained by the information extraction unit, the information labeling unit takes the standard format of "x room x hall x toilet x kitchen x balcony - x square meter" as the picture description for machine recognition, as the text training data set;

[0078] The small sample training module is used for training the high-rise residential building plan feature model based on a small sample amount, and the small data amount of text and image training data set is used as input for training to construct the high-rise residential building plan diffusion model in cold regions.

[0079] The building plan generation module is used for generating the high-rise residential building plan in cold regions based on the text feature description, and the target house type contour and the target house type entrance position of the conditional image input unit and the high-rise residential building house type text description of the demand text input unit are used as input, the diffusion model obtained by training is used as the black box processing module, and the high-rise residential building plan in cold regions is used as output.

[0080] The post-evaluation module is used for evaluating whether the plan generated by the plan generation module meets the design requirements of the residential building, and if not, the plan is regenerated until the requirements are met.

[0081] The present application provides a small sample training and generation method and device for high-rise residential building plan in cold regions based on diffusion model, which comprises the following steps: collecting and sorting the standard plan original data set of high-rise residential building in cold regions, analyzing to obtain the function partition, the house type contour graph and the entrance information data set corresponding to each standard plan, and the text description generated by the statistical information of each house type plan as the original training data set; based on the diffusion model framework, the generation model from the house type contour to the house type plan is trained; a tool platform for realizing intelligent design and automatic generation of house type plan is developed, which can generate the house type plan meeting the conditions by combining the specific requirement text information of each function room with the basic house type contour and entrance information. The device comprises an image data processing module, a text information labeling module, a small sample training module, a building plan generation module and a post-evaluation module. The present application provides an efficient and accurate automatic residential building plan generation method and device.

[0082] The present application also provides an electronic device comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the small sample training and generation method for high-rise residential building plan in cold regions based on diffusion model when executing the computer program.

[0083] The application further provides a computer readable storage medium for storing computer instructions, which, when executed by a processor, implement steps of the method for training and generating a small sample of a floor plan of a high-rise residential building in a cold region based on a diffusion model.

[0084] The memory in the embodiments of the present application can be a volatile memory or a non-volatile memory, or can include both volatile and non-volatile memories. The non-volatile memory can be a read only memory (ROM), a programmable ROM (PROM), an erasable PROM (EPROM), an electrically EPROM (EEPROM), or a flash memory. The volatile memory can be a random access memory (RAM) used as an external cache. By way of example, and not limitation, many forms of RAM are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous dynamic RAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchlink DRAM (SLDRAM), and direct rambus RAM (DR RAM). It is to be noted that the memory described herein is intended to include, without being limited to, these and any other suitable types of memory.

[0085] In the above embodiments, all or part of the methods can be implemented by software, hardware, firmware, or any combination thereof. When implemented by software, all or part of the methods can be implemented in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of the present application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions can be transferred from one website, computer, server, or data center to another website, computer, server, or data center through a wired (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (such as infrared, wireless, microwave, etc.) manner. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server, data center, etc. that includes one or more available media sets. The available media can be a magnetic medium (such as a floppy disk, a hard disk, a magnetic tape), an optical medium (such as a high-density digital video disc (DVD)), or a semiconductor medium (such as a solid state disc (SSD)), etc.

[0086] In the implementation process, each step of the above method can be completed by integrated logic circuit of hardware in the processor or instruction in the form of software. The steps of the method disclosed in the embodiments of the present application can be directly embodied as hardware processor execution completion, or executed by hardware and software modules in the processor. The software module can be located in a random access memory, a flash memory, a read-only memory, a programmable read-only memory, an electrically erasable programmable memory, a register, or other mature storage media in the art. The storage medium is located in the memory, and the processor reads the information in the memory and combines the hardware to complete the steps of the above method. To avoid repetition, it will not be described in detail here.

[0087] It should be noted that the processor in the embodiments of the present application can be an integrated circuit chip with a signal processing capability. In the implementation process, each step of the method embodiments can be completed by the integrated logic circuit of hardware or the instruction in the form of software in the processor. The processor mentioned above can be a general processor, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components. Each method, step and logic block diagram disclosed in the embodiments of the present application can be implemented or executed. The general processor can be a microprocessor or the processor can also be any conventional processor. The steps of the method disclosed in combination with the embodiments of the present application can be directly embodied as a hardware code processor for execution, or a combination of hardware and software modules in the code processor for execution. The software module can be located in a random access memory, a flash memory, a read only memory, a programmable read only memory or an electrically erasable programmable memory, a register or other mature storage medium in the art. The storage medium is located in the memory, and the processor reads the information in the memory, and combines the hardware to complete the steps of the above method.

[0088] The above describes in detail the method and device for training and generating a small sample of a high-rise residential building plane in cold regions based on a diffusion model. The principles and implementation modes of the present application are described by using specific examples. The above description of the embodiments is only used to help understand the method of the present application and its core idea. Meanwhile, for those skilled in the art, according to the idea of the present application, the specific implementation mode and application range will be changed. In summary, the content of the specification should not be understood as a limitation of the present application.

Claims

1. A diffusion model-based method for training and generating small sample planes of high-rise residential buildings in cold regions, characterized in that, The method comprises: Obtaining a high-rise residential building plan in cold regions, and analyzing to obtain residential building area, house type area, house type layout, house type contour and residential structure size information corresponding to each standard plan as original training data; Organize the above plan information into a text description, form a mapping relationship with the corresponding picture, and use the color block labeling method to label the indoor function layout as the picture input condition; use the color block labeling method to label the indoor function layout as the picture input condition; use the color block labeling method to label the indoor function layout as the picture input condition; the text data and picture data are used as input conditions, and the low-order adaptive modeling LoRA of the large language model is used to train the high-rise residential building plan diffusion model in cold regions; After the diffusion model is trained, the target plan house type text description and contour map are used as input parameters, and the high-rise residential building plan in cold regions generated by the trained diffusion model is used as the output end; Organize the plan information into a text description, and form a corresponding relationship with the corresponding picture, specifically: Organize the plan information into a text description, and form a corresponding relationship with the corresponding picture, specifically: Organize the plan information into a text description, and form a corresponding relationship with the corresponding picture, specifically: The specific marking method is: the main entrance is red, the balcony is orange, the bedroom is yellow, the kitchen is green, the dining room is blue, and the bathroom is purple; The target plan house type text description and contour map are used as input parameters, specifically: According to the format of the plan text description, input the requirement of "x room x hall x bathroom x kitchen x balcony-x square meter" as the text input condition; Input the target plan house type contour map, and use the control network in the diffusion model to perform edge detection on the contour map to convert it into an edge line map as the picture input condition. The method further comprises: evaluating whether the generated high-rise residential building plan in cold regions meets the specification requirements by using an evaluation model, and if not, regenerating until the requirements are met.

2. The method of claim 1, wherein, The standards of the evaluation model in the evaluation process include:

3. The method of claim 2, wherein, a. Whether the adjacent functional areas are reasonable; b. Whether the area of a certain functional area is reasonable; c. Whether it is generated completely according to the input contour; d. Whether all the required functions are generated; If the evaluation standards are not met, regenerate until the requirements are met. The output picture that meets the evaluation requirements is vectorized, specifically:

4. The method of claim 3, wherein, Import the picture that meets the evaluation requirements into the vectorization module, eliminate irrelevant information in the image through image preprocessing, and restore useful real information; In the vectorization module, the content is recognized through edge detection, contour tracking and contour expression, the walls, windows and doors are layered, and the graph vectorization of the sub-layers is completed. The method further comprises:

5. The method of claim 1, wherein, Obtaining the area of a specific functional area in the standard plan of a high-rise residential building in cold regions, and training the house type layout of the generated high-rise residential building in cold regions to obtain a specific functional area AI evaluation model; ​ According to the specific functional area area AI evaluation model, the generated high-rise residential building in cold regions is evaluated whether the specific functional area area is within a reasonable range, and the generated high-rise residential building in cold regions is consistent with the input target plan layout contour map, and the high-rise residential building plan in cold regions is confirmed and output.

6. A device for training and generating a small sample of a plan of a high-rise residential building in a cold region based on a diffusion model, characterized in that The device comprises: An image data processing module: for processing the high-rise residential building plan data in the training process, the image correction unit corrects and scales the image to a uniform resolution of 512x512 pixels, and then the data labeling module labels the corrected plan function area, contour and entrance with different colors; The specific labeling method is: the main entrance is red, the balcony is orange, the bedroom is yellow, the kitchen is green, the dining room is blue, and the bathroom is purple; A text information labeling module: for labeling the text information of the characteristics of the high-rise residential building in cold regions, based on the residential standard floor area, house type area, house type layout and house type contour data corresponding to the high-rise residential building plan obtained by the information extraction unit, the information labeling unit takes "x room x hall x bathroom x kitchen x balcony-x square meter" as the standard format of the picture description for machine recognition, as the text training data set; A small sample training module: for training a high-rise residential building plan feature model based on a small sample size, using a small amount of text and image training data set as input for training, to build a high-rise residential building plan diffusion model in cold regions; A building plan generation module: for generating a high-rise residential building plan in cold regions based on text feature description, taking the target house type contour of the high-rise residential building in cold regions and the target house type entrance position as input by the conditional image input unit, and taking the high-rise residential building house type text description as input by the demand text input unit, using the trained diffusion model as a black box processing module, and outputting the high-rise residential building plan in cold regions; A post-evaluation module: for evaluating whether the plan generated by the plan generation module meets the design requirements of the residential building, if not, it is regenerated until the requirements are met; The plan information is arranged as a text description, and the corresponding picture is formed as a corresponding relationship, specifically: The plan information is arranged as a standard format for machine recognition of "x room x hall x bathroom x kitchen x balcony-x square meter", and the corresponding picture is formed as a corresponding relationship, as a text training data set; Use color block labeling method to label indoor function layout as picture input condition, specifically: The function area, contour and entrance of the collected plan are labeled with different colors to form a picture training data set; The specific labeling method is: the main entrance is red, the balcony is orange, the bedroom is yellow, the kitchen is green, the dining room is blue, and the bathroom is purple; The target plan house type text description and contour map are taken as input parameters, specifically: According to the format of the plan text description, input "x room x hall x bathroom x kitchen x balcony-x square meter" as the text input condition; Input the target plan house type contour map, use the control network in the diffusion model to perform edge detection on the contour map to convert it into an edge line map as the picture input condition. 7.An electronic device comprising a memory and a processor, the memory storing a computer program, wherein, The processor executes the computer program to realize the steps of the method of any one of claims 1-5.

8. A computer readable storage medium for storing computer instructions, characterized in that, The computer instructions, when executed by a processor, implement the steps of the method of any of claims 1-5.

Images