Indoor environment simulation method, system, electronic device, and storage medium

Abstract

Disclosed in embodiments of the present invention are an indoor environment simulation method, a system, an electronic device, and a storage medium. The method comprises: acquiring first information and second information, wherein the first information is used for describing an indoor environment, and the second information is used for describing a simulation target of the indoor environment; inputting the first information and the second information to a generative artificial intelligence model; receiving a first script and a second script from the generative artificial intelligence model, wherein the first script is adapted to generate a computer-aided design model of the indoor environment, and the second script is adapted to perform computer-aided engineering simulation on the computer-aided design model to generate a simulation result corresponding to the simulation target; on the basis of the first script, generating a computer-aided design model of the indoor environment; and on the basis of the second script, performing computer-aided engineering simulation on the computer-aided design model to generate a simulation result. Indoor environment simulation can be achieved without the need to manually build a computational fluid dynamics model of an indoor environment, improving applicability.

Description

Indoor environment simulation method, system, electronic equipment and storage medium Technical Field

[0001] The present invention relates to the technical field of environmental control, in particular to a method, system, electronic equipment and storage medium for simulating an indoor environment. Background Art

[0002] In modern commercial building operations, interior room layouts often need to be rearranged to meet user needs (e.g., due to tenant requirements). Changing the location of hardware such as air vents is often costly, while simply reconfiguring interior spaces with partition walls can result in uneven indoor temperature distribution, low indoor comfort, and inefficient air conditioning systems. These issues can also arise in other environmentally sensitive scenarios, such as updating the layout of server racks in a data center room.

[0003] To proactively assess the impact of layout updates on the environment (for example, indoor temperature-related KPIs or air conditioning system efficiency), a computational fluid dynamics (CFD) simulation model of the indoor environment can be constructed to estimate the environmental state after the layout change. However, constructing a CFD simulation model of the indoor environment requires considerable time and effort by professionals, hindering its widespread application.

[0004] Summary of the Invention

[0005] The embodiments of the present invention provide a method, system, electronic device and storage medium for simulating an indoor environment.

[0006] In a first aspect, a method for simulating an indoor environment is provided, comprising:

[0007] Acquire first information and second information, wherein the first information is used to describe an indoor environment, and the second information is used to describe a simulation target of the indoor environment;

[0008] Inputting the first information and the second information into a generative artificial intelligence (AI) model;

[0009] receiving a first script and a second script from the generative AI model, wherein the first script is adapted to generate a computer-aided design (CAD) model of the indoor environment, and the second script is adapted to perform a computer-aided engineering (CAE) simulation on the CAD model to generate a simulation result corresponding to the simulation objective;

[0010] generating a CAD model of the indoor environment based on the first script;

[0011] Based on the second script, a CAE simulation is performed on the CAD model to generate the simulation result.

[0012] In a second aspect, a simulation system for an indoor environment is provided, including:

[0013] A human-computer interaction device, configured to obtain first information and second information, wherein the first information is used to describe an indoor environment, and the second information is used to describe a simulation target of the indoor environment;

[0014] A script generation device, configured to input the first information and the second information into a generative AI model; and receive a first script and a second script from the generative AI model, wherein the first script is adapted to generate a CAD model of the indoor environment, and the second script is adapted to perform a CAE simulation on the CAD model to generate a simulation result corresponding to the simulation target.

[0015] A CAD device, configured to generate a CAD model of the indoor environment based on the first script;

[0016] A CAE simulation device is used to perform CAE simulation on the CAD model based on the second script to generate the simulation result.

[0017] According to a third aspect, an electronic device is provided, including:

[0018] processor;

[0019] a memory for storing executable instructions of the processor;

[0020] The processor is configured to read the executable instructions from the memory and execute the executable instructions to implement the above indoor environment simulation method.

[0021] In a fourth aspect, a computer-readable storage medium is provided, on which computer instructions are stored. When the computer instructions are executed by a processor, the above indoor environment simulation method is implemented.

[0022] In a fifth aspect, a computer program product is provided, comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements the above indoor environment simulation method.

[0023] Thus, the embodiments of the present invention provide a simulation solution for indoor environments. By using a generative AI model to generate a first script for creating a CAD model of the indoor environment and a second script for performing CAE simulation on the CAD model, indoor environment simulation can be achieved without manually building a CFD model of the indoor environment, reducing the need for modeling expertise and improving applicability. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, so that those skilled in the art will understand the above and other features and advantages of the present invention more clearly. In the accompanying drawings:

[0025] FIG1 is an exemplary flow chart of a method for simulating an indoor environment according to an embodiment of the present invention.

[0026] FIG2 is a first exemplary schematic diagram of a simulation process of an indoor environment according to an embodiment of the present invention.

[0027] FIG3 is a second exemplary schematic diagram of a simulation process of an indoor environment according to an embodiment of the present invention.

[0028] FIG4 is an exemplary schematic diagram of a fine-tuned generative AI model according to an embodiment of the present invention.

[0029] FIG. 5 is an exemplary diagram of simulation results according to an embodiment of the present invention.

[0030] FIG6 is an exemplary structural diagram of an indoor environment simulation system according to an embodiment of the present invention.

[0031] FIG. 7 is an exemplary structural diagram of an electronic device according to an embodiment of the present invention.

[0032] The accompanying drawings are numerals as follows: DETAILED DESCRIPTION

[0033] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention is further described in detail with reference to the following examples.

[0034] For the sake of brevity and intuitiveness in description, the solution of the present invention is explained below by describing several representative implementations. A large number of details in the implementations are only used to help understand the solution of the present invention. However, it is obvious that the technical solution of the present invention may not be limited to these details when implemented. In order to avoid unnecessarily obscuring the solution of the present invention, some implementations are not described in detail, but only a framework is given. Hereinafter, "including" means "including but not limited to", and "according to..." means "at least according to..., but not limited to only according to...". Due to the language habits of Chinese, when the number of a component is not specifically specified below, it means that the component can be one or more, or can be understood as at least one.

[0035] The embodiments of the present invention establish an automatic construction system for simulation models, which enables non-professionals to quickly, accurately and cost-effectively generate simulation models of various layouts in indoor environments, thereby significantly reducing the evaluation cost of indoor environment layout plans, accelerating the realization of efficient and zero-carbon buildings, and supporting the generation of the building indoor environment metaverse.

[0036] The applicant found that in the prior art, in order to use CFD models to analyze the rearrangement of rooms inside a building, geometric modeling software is usually used to construct a detailed three-dimensional model of the building and its interior space. This three-dimensional model is then imported into CAE software, which applies algorithms to solve the governing equations for fluid flow and heat transfer. Engineers can analyze airflow patterns and temperature distribution within the building to optimize the design. However, generating a CFD model of a building's interior space can be a time-consuming and complex process that requires a lot of expertise. In addition to collecting detailed information about the geometry, engineers also need to use specialized software to create a three-dimensional model of the building and its interior space, and then apply CAE software to pre-process the three-dimensional model, including meshing, algorithm selection, boundary condition definition, and building material property definition. In general, the process of generating a CFD model for the rearrangement of rooms inside a building requires a lot of time and expertise.

[0037] Parametric modeling is another approach used to reduce the time required to generate CFD models of building interiors. It involves defining a set of parameters that control the CFD model's geometry and boundary conditions, then using software tools to automatically generate a parametric model based on these parameters. This approach can significantly reduce the time and effort required to create CFD models and enable engineers to quickly evaluate a range of design options. However, parametric modeling also has significant limitations. First, parametric models are generally not universal and must be customized for each building geometry, which can also result in high costs in terms of time and resources. Furthermore, if the input parameters are not clearly defined or are inaccurate, the resulting CFD model may not accurately represent the actual building performance.

[0038] In an embodiment of the present invention, a generative AI model (for example, a large-scale pre-trained model, such as a GPT-4 model, a CodeGeeX model, etc.) is used to simulate the indoor environment. This can achieve indoor environment simulation without the need for professional knowledge of establishing a CFD model or a parametric model of the indoor environment, thereby improving applicability.

[0039] FIG1 is an exemplary flow chart of a method for simulating an indoor environment according to an embodiment of the present invention. As shown in FIG1 , the method includes:

[0040] Step 101: Acquire first information and second information, wherein the first information is used to describe an indoor environment, and the second information is used to describe a simulation target of the indoor environment.

[0041] Here, the indoor environment is preferably implemented as an indoor environment in an environmentally sensitive industry. For example, indoor environments may include: commercial buildings, office buildings, industrial parks, data centers (such as substations in the power industry, computer control centers in the telecommunications industry, and monitoring centers in the broadcasting industry), greenhouses, cold storage, clean rooms, or operating rooms, etc.

[0042] Here, both the first information and the second information can be represented as statements described using natural language. The specific type of the natural language can be any language, such as Chinese, English, or Japanese. The first information uses natural language to describe a specific indoor environment. For example, the first information describes the number of rooms in the indoor environment, the area and orientation of each room, the specific equipment in each room, and the layout location of each equipment, etc. Based on the first information, the generative AI model can generate the basic code or all the code for constructing a CAD model of the indoor environment.

[0043] For example, the first information may be implemented as any one or more of the following sentences in any combination:

[0044] (1): “build a Flovent room CFD model with 2 interior rooms”;

[0045] (2): "Establish a CFD model of a ventilation chamber with two internal rooms";

[0046] (3): "interior room 1 has 6 rows rack inside";

[0047] (4): “There are 6 rows of cabinets inside interior room 1”;

[0048] (5): "the room size of room1 is 14.1m in x direction and 14.4m in z direction";

[0049] (6): “The room dimensions of room 1 are 14.1 meters in the x direction and 14.4 meters in the z direction”;

[0050] (7): "the wall thickness is 0.2m";

[0051] (8): “The thickness of the wall is 0.2 meters”;

[0052] (9): "there are 45 racks in total in room1.the number of racks in each row are 9,6,6,8,8,8.the rows are all in x direction";

[0053] (10): "There are 45 cabinets in room 1. The number of cabinets in each row is 9, 6, 6, 8, 8. The rows are all in the x direction";

[0054] (11): "interior room2 has 5 people inside";

[0055] (12): “There are 5 people in the inner room 2”.

[0056] The second information uses natural language to describe the simulation target of the indoor environment. For example, the simulation target can be implemented as the overall temperature distribution, local temperature distribution, overall temperature maximum, overall temperature minimum, local temperature maximum, local temperature minimum, etc. in the indoor environment.

[0057] Based on the second information, the generative AI model can generate basic code or all code for performing CAE simulation on the CAD model of the indoor environment to obtain simulation results corresponding to the simulation target. For example, assume that the simulation target described in the second information is the maximum temperature of a room in the indoor environment. The specific process of performing CAE simulation on the CAD model of the indoor environment includes: (1): simulating the temperature field of the room using the CAD model of the indoor environment; (2): traversing the temperature field of the room; (3): determining the maximum temperature of the room based on the traversal results. The generative AI model can generate code to execute the simulation process.

[0058] For example, the second information can be implemented as any one or more of the following sentences in any combination:

[0059] 1: "Maximum temperature value of room 1";

[0060] 2: "Maximum temperature value for room 1";

[0061] 3. “Temperature distribution map of the entire floor”;

[0062] 4: "Temperature distribution diagram of the whole floor".

[0063] The above exemplary descriptions are typical examples of the first information and the second information. Those skilled in the art will appreciate that such descriptions are merely exemplary and are not intended to define the scope of protection of the embodiments of the present invention.

[0064] In one embodiment, obtaining the first information and the second information in step 101 includes obtaining input data including the first information and the second information via a human-computer interaction interface. In this embodiment, the first information and the second information can be provided at once using the input data provided by the user.

[0065] In one embodiment, obtaining the first information and the second information in step 101 includes obtaining a predefined template, wherein the predefined template includes the first information and the second information. Here, the predefined template is used to quickly obtain the first information and the second information without user input.

[0066] Predefined templates are predefined prompts used to guide large models in performing tasks and provide necessary reference information during task execution. The predefined data in the predefined templates can include default parameter values, reasonable value ranges, or correspondences between parameters. Predefined templates support high tolerance for requirement descriptions, that is, the first script and the second script can be generated when the input is insufficient. The default / reference values ​​stored in the predefined templates can detect incorrect input parameters, thereby increasing the flexibility of the system. In order to improve the performance of the model when generating different tasks, the system can integrate multiple predefined templates. This enables the automatic selection of the appropriate predefined template based on user input. In addition, new predefined templates can be generated based on user input and added to the library.

[0067] For example, an example of a predefined template could be: "Generate a data center room model in SimCenter NX and perform thermal simulation in SimCenter StarCCM+; the default dimensions of the rack are 0.6 x 1.2 x 2 meters; the default power of the rack is 2000 watts; the maximum and minimum temperatures of the rack are derived from the simulation results;"

[0068] In one embodiment, obtaining the first information and the second information in step 101 includes: obtaining a predefined template, the predefined template including predefined data for the first information and predefined data for the second information; obtaining edit data for the first information and the edit data for the second information via a human-computer interaction interface; determining the first information based on the predefined data for the first information and the edit data for the first information; and determining the second information based on the predefined data for the second information and the edit data for the second information. In this embodiment, the first information and the second information can be jointly determined using the edit data provided by the user in the predefined template.

[0069] In one embodiment, based on the predefined data of the first information and the edited data of the first information, it is determined that the first information includes at least one of the following:

[0070] (1): Based on the addition processing of the indoor environment description represented by the edited data of the first information, the description of the indoor environment is added based on the predefined data of the first information to determine the first information.

[0071] For example, the predefined data of the first information indicates that air conditioner a is arranged in room abc, and the edited data of the first information indicates that air conditioner b is added to room abc. Then the description of room abc is changed to: air conditioners a and b are arranged in room abc.

[0072] (2): Based on the reduction processing of the indoor environment description represented by the edited data of the first information, the description of the indoor environment is reduced based on the predefined data of the first information to determine the first information.

[0073] For example, the predefined data of the first information indicates that air conditioners a and b are arranged in room abc, and the edited data of the first information indicates that air conditioner b is deleted in room abc, so the description of room abc is changed to: air conditioner a is arranged in room abc.

[0074] (3): Based on the change processing of the indoor environment description represented by the edited data of the first information, the description of the indoor environment is changed based on the predefined data of the first information to determine the first information.

[0075] For example, the predefined data of the first information indicates that air conditioners a and b are arranged in room abc. The edited data of the first information indicates that air conditioner c is arranged in room abc. Then the description of room abc is changed to: air conditioner c is arranged in room abc.

[0076] In one embodiment, based on the predefined data of the second information and the edited data of the second information, it is determined that the second information includes at least one of the following:

[0077] (1): Based on the addition processing of the simulation target description represented by the editing data of the second information, the description of the simulation target is added based on the predefined data of the second information to determine the second information.

[0078] For example, the predefined data of the second information represents the simulation target as: the maximum temperature in room a; the edited data of the second information represents the simulation target added with: the maximum temperature in room b; then the simulation target changes to: the maximum temperature in room a and the maximum temperature in room b.

[0079] (2): Based on the reduction processing of the simulation target description represented by the editing data of the second information, the description of the simulation target is reduced based on the predefined data of the second information to determine the second information.

[0080] For example, the predefined data of the second information represents the simulation target as: the maximum temperature in room a and the maximum temperature in room b; the edited data of the second information represents the simulation target reduced as: the maximum temperature in room b; then the simulation target is changed to: the maximum temperature in room a.

[0081] (3): Based on the change processing of the simulation target description represented by the editing data of the second information, the description of the simulation target is changed based on the predefined data of the second information to determine the second information.

[0082] For example, the predefined data of the second information represents the simulation target as: the maximum temperature in room a and the maximum temperature in room b; the edited data of the second information represents the simulation target changed to: the temperature distribution in room d; then the simulation target is changed to: the temperature distribution in room d.

[0083] The above examples describe typical examples of the first information and the second information. Those skilled in the art will appreciate that such descriptions are merely exemplary and are not intended to limit the scope of protection of the embodiments of the present invention.

[0084] Step 102: Input the first information and the second information into a generative AI model.

[0085] Generative AI is an AI technology that leverages machine learning models and deep learning techniques to generate new content, whether text, images, audio, or video, by studying patterns in historical data. Rather than generating output based on given rules or data, generative AI autonomously generates entirely new content. Generative AI models have the ability to generate code for input tasks. Generative AI models can be implemented as large-scale pre-trained models, such as the GPT-4 model and the CodeGeeX model.

[0086] Step 103: Receive a first script and a second script from the generative AI model, wherein the first script is adapted to generate a CAD model of the indoor environment, and the second script is adapted to perform CAE simulation on the CAD model to generate simulation results corresponding to the simulation target.

[0087] In one embodiment, a first script can be directly generated based on the first information, and a second script can be directly generated based on the second information. The first script can be specifically implemented as a code in the output language format of a generative AI model, such as a code in Python or Java format. When the first script is executed by a CAD tool, a CAD model of the indoor environment described by the first information can be generated. The second script is used to perform CAE simulation on the CAD model generated by the first script to determine the specific numerical value (i.e., simulation result) of the simulation target described by the second information. The second script can be specifically implemented as a code in the output language format of a generative AI model, such as a code in Python or Java format. When the second script is executed by a CAE tool, a simulation can be performed on the CAD model of the indoor environment to determine the simulation result.

[0088] In one embodiment, the first script is generated based on the first sample information and the second sample information, and the second script is generated based on the first sample information and the second sample information. In this embodiment: the simulation target described in the second information not only affects the second script, but also affects the CAD model of the indoor environment (for example, different simulation targets can affect the type, attribute, or quantity of the corresponding CAD model), thereby affecting the first script. Moreover, the overall environment described in the first information not only affects the first script, but also affects the simulation target (for example, different types, attributes, or quantities of CAD models may affect different simulation targets), thereby affecting the second script.

[0089] Step 104: Generate a CAD model of the indoor environment based on the first script.

[0090] For example, the first script is input into CAD software to run the first script in the CAD software, thereby obtaining a CAD model of the indoor environment.

[0091] According to the top-down hierarchical structure, the CAD model required in the simulation process can include: (1) CAD model files; (2) standardized functional modules corresponding to the physical components in the CAD model files; (3) the underlying elements / codes that constitute the corresponding standardized functional modules. Each component should have corresponding annotations to indicate the simulation type supported by the corresponding component, the required inputs, and the necessary constraints, where:

[0092] CAD model file annotations include: (1) simulation type / function, which is used to describe the supported CAE simulation types; (2) the input required to describe the physical components of the CAD model, the physical parameters of the physical components (including dimensions, material properties), the operating parameters of the physical components, and the spatial physical relationships between the physical components; and (3) necessary constraints to specify the units and ranges of the parameters input in subsequent CAE simulations.

[0093] The physical components are generated by standardized functional modules, so the standardized functional module annotations include: (1) simulation type / function, which is used to describe the supported CAE simulation types and CAD simulation models; (2) the inputs required to describe the physical parameters (including dimensions, material properties) of the physical components and their operating parameters, which need to be generated by the standardized functional modules; (3) the necessary constraints that specify the parameter units and input ranges to achieve the physical component generation.

[0094] The standardized function module consists of the underlying element code, and the underlying element / code annotations include: (1) simulation type usage / function, describing the supported CAE simulation types and standardized function modules; (2) describing the input required to drive the underlying elements / code to form the parameters of the standardized function module; (3) describing the unit constraints and scope constraints of the supported standardized function modules.

[0095] Therefore, the first script corresponds to the following key descriptions: supported simulation types; the physical components of the CAD model and their physical parameters (including physical dimensions, material properties) and operating parameters; the spatial physical relationships between the physical components of the CAD model; and the units and ranges of the parameters input in the CAE simulation. Unlike traditional text generation, the generation of the first script requires a generative AI model based on key descriptions, combining the hierarchical structure of the simulation model, top-down CAD component parsing, and bottom-up annotation hierarchy calls.

[0096] Step 105: Based on the second script, perform CAE simulation on the CAD model to generate simulation results corresponding to the simulation target.

[0097] The second script controls the CAE software to simulate the CAD model obtained in step 104, thereby obtaining simulation results corresponding to the simulation objectives. The second script is used to open the CAE simulation software, load the CAD simulation model to generate the CAE simulation model, write the CAE simulation input into the CAE simulation model, initialize the CAE simulation model, run the CAE simulation model, and output the specific results of the defined simulation objectives upon completion of the simulation. According to the execution logic sequence, the second script contains the following structure from top to bottom:

[0098] (a) CAE software calling module: includes the code for loading the software name, version, path, and related functions to realize the opening of CAE software;

[0099] (b) CAD model loading module: including the code for writing the CAD model name, type and path and loading the model;

[0100] (c) CAE simulation input writing module: including the relevant codes for writing and loading CAE simulation input file name, type, and path;

[0101] (d) CAE simulation initialization module: including relevant codes for CAE simulation initialization based on CAD models and CAE simulation data, CAE model generation, and preprocessing;

[0102] (e) CAE simulation operation module: including the relevant codes for starting CAE simulation and monitoring CAE simulation;

[0103] (f) Simulation output module: includes the relevant codes for implementing output data and format definition, and output window or file creation.

[0104] Therefore, the second script contains the following key descriptions: CAE software, CAD model, input parameter file and output parameter information.

[0105] In one embodiment, the first script and the second script have a universal language format, wherein the universal language format is the output language format of the generative AI model (e.g., Python or Java format). The method shown in Figure 1 includes: before step 104, converting the universal language format of the first script into a CAD format, for example, converting the first script into a .cpp file format, a .xml file format, a .CATscript file format, or a .inp file format, etc. Before step 105, converting the universal language format of the second script into a CAE format. For example, converting the second script into a .xml file format or a .bat file format. After the format conversion, the first script and the second script can both be conveniently executed.

[0106] The above exemplary descriptions are of typical formats after the first script and the second script are converted. Those skilled in the art will appreciate that such descriptions are merely exemplary and are not intended to limit the scope of protection of the embodiments of the present invention.

[0107] Considering that generative AI models (such as large-scale pre-trained models such as GPT-4 and CodeGeeX) have been developed and widely used, they have demonstrated the ability to generate code for general tasks. However, the ability to use them to directly generate scripts that conform to CAD and / or CAE formats is still not good enough. This is because those scripts used in CAD / CAE software are developed for specific needs, have special syntax or API connections, and are generally not included in open training datasets. Therefore, it is preferable to fine-tune generative AI models to adapt to the needs of generating scripts in CAD and / or CAE formats.

[0108] In one embodiment, prior to step 102, the process includes fine-tuning the generative AI model based on a sample, where the sample includes first sample information, second sample information, a first sample script, and a second sample script. The first sample information is used to describe a sample indoor environment, the second sample information is used to describe a sample simulation target of the sample indoor environment, the first sample script is adapted to generate a CAD model of the sample indoor environment, and the second sample script is adapted to perform CAE simulation on the CAD model of the sample indoor environment to determine a sample simulation result corresponding to the sample simulation target. For example, the first sample script and the second sample script contain user-manufactured code based on the first sample information and the second sample information.

[0109] In one embodiment, fine-tuning includes: inputting first sample information and second sample information into a generative AI model; obtaining a first output script and a second output script from the generative AI model, wherein: the first output script is adapted to generate a CAD model of a sample indoor environment, and the second output script is adapted to perform CAE simulation on the CAD model of the sample indoor environment to determine a sample simulation result corresponding to a sample simulation target; determining the loss function value of the generative AI model based on the difference between the first output script and the first sample script and the difference between the second output script and the second sample script; configuring the model parameters of the generative AI model so that the loss function value is lower than a preset threshold.

[0110] In one embodiment, the method includes: displaying the simulation results via a human-computer interaction interface. Therefore, by displaying the simulation results on the human-computer interaction interface, it is convenient for users to know the specific results of the simulation target in a timely manner.

[0111] FIG2 is a first exemplary schematic diagram of a simulation process of an indoor environment according to an embodiment of the present invention.

[0112] In Figure 2, input data including first and second information is received from a user via a human-computer interaction interface 10. Here, the first and second information can be provided simultaneously using the user-provided input data. The first information is used to describe the indoor environment, while the second information is used to describe the simulation target of the indoor environment.

[0113] The simulation model automatic generation system 11 includes a fine-tuned generative AI model 13 and a format conversion module 14. Based on the first information and the second information, the fine-tuned generative AI model 13 generates a first script and a second script in a universal language format, wherein the first script includes code (in a universal language format) adapted to generate a CAD model of an indoor environment after execution, and the second script includes code (in a universal language format) adapted to perform a CAE simulation on the CAD model to generate simulation results corresponding to the simulation target after execution. The format conversion module 14 converts the first script into a format supported by the CAD tool. The format conversion module 14 converts the second script into a format supported by the CAE tool. The converted first script 15 is provided to the CAD tool 17. Based on the converted first script 15, the CAD tool 17 generates a CAD model of the indoor environment. The converted second script 16 is provided to the CAE tool 18. The CAD model of the indoor environment is provided to the CAE tool 18. Based on the converted second script, the CAE tool 18 simulates the CAD model of the indoor environment and determines the simulation results corresponding to the simulation target.

[0114] FIG3 is a second exemplary schematic diagram of a simulation process of an indoor environment according to an embodiment of the present invention.

[0115] In Figure 3, a predefined template containing predefined data for first information and predefined data for second information is retrieved from template library 12. Furthermore, edited data containing the first information and the second information is received from a user via human-computer interaction interface 10. The first information is determined based on the predefined data for the first information and the edited data for the first information. The second information is determined based on the predefined data for the second information and the edited data for the second information. The first information is used to describe the indoor environment, and the second information is used to describe the simulation target of the indoor environment.

[0116] The simulation model automatic generation system 11 includes a fine-tuned generative AI model 13 and a format conversion module 14. The fine-tuned generative AI model 13 generates a first script and a second script in a universal language format based on the first information and the second information. The first script includes code (in a universal language format) that, when executed, is adapted to generate a CAD model of an indoor environment, and the second script includes code (in a universal language format) that, when executed, is adapted to perform a CAE simulation on the CAD model to generate simulation results corresponding to the simulation target. The format conversion module 14 converts the first script into a format supported by the CAD tool. The format conversion module 14 converts the second script into a format supported by the CAE tool. The converted first script 15 is provided to the CAD tool 17. The CAD tool 17 generates a CAD model of the indoor environment based on the converted first script 15. The converted second script 16 is provided to the CAE tool 18. The CAD model of the indoor environment is provided to the CAE tool 18. The CAE tool 18 simulates the CAD model of the indoor environment based on the converted second script and determines the simulation results corresponding to the simulation target.

[0117] Large-scale pre-trained models can generate high-quality code for general purposes, but they may not have sufficient training data to generate specialized scripts for CAD or CAE software. However, by using the fast-tuning algorithms of embodiments of the present invention, significant improvements can be achieved in the ability to generate specialized CAD / CAE scripts with relatively little training effort. Each fine-tuned large model can be designed to solve one or more relevant CAD / CAE problems. One or more fine-tuned large models can be integrated into the system, and the most appropriate and relevant large model can be selected for code generation based on the target task or application software.

[0118] FIG4 is an exemplary schematic diagram of a fine-tuned generative AI model according to an embodiment of the present invention. FIG4 shows the process involved in generating a fine-tuned large-scale generative AI model. A fine-tuning process 21 is performed on a generative AI model 22 using a sample 20 to obtain a fine-tuned generative AI model 23. The sample 20 can be implemented in the form of a prompt template. Specifically, the fine-tuning process includes: first, generating a prompt template, and then applying the prompt template to fine-tune the AI ​​model 22 generated by large-scale pre-training, which can enhance the ability of large-scale pre-trained models to generate specific CAD and CAE scripts. The detailed process includes:

[0119] (1) Construct a prompt template. The function of the prompt template is to reconstruct the input simulation parameters and simulation requirement description and the output script (including the CAD generation script (i.e., the first script) and the CAE control operation script (i.e., the second script)) into a new text with a mask slot. The implementation methods of constructing the prompt template include: (1): Manual template: different templates and label words are constructed based on human experience knowledge; (2): Heuristic-based template: appropriate templates are constructed through rules, heuristic search, etc.; (3): Generative template: appropriate templates are generated based on the training data of a given task; (4): Word vector fine-tuning (word embedding): templates of discrete characters are explicitly defined. The word vectors of these template characters involve gradient descent during the training process. The initially defined discrete characters are used as vector initialization; (5): Pseudo-token: discrete templates are not explicitly defined, and templates are used as trainable parameters.

[0120] The example of template generation is given. A paragraph (including (a): a description of the key parameters required to generate a CAD file, such as the dimensions and device layout in the CAD file; (b): the content of the CAE simulation to be performed and the simulation output); a [mask] token and a placeholder [Prompt] for the template to be generated. The predicted probability of the generated token is fed into the generative model and used as feedback. Based on the current feedback, the decision function is trained using the policy gradient method of reinforcement learning. Based on this decision function, discrete template words can be generated for [prompt]. The decision function is essentially a pre-trained language model that generates tokens through the LM head. After generating the template word, the above steps are repeated until a certain number of template words are generated.

[0121] (2) Constructing a prompt-answer space mapping (Verbalizer), that is, constructing a mapping function to map the output script label words with the description text label words. Implementation methods may include: (1) Manual design: label words are manually specified based on experience; (2) Domain knowledge-guided search for discrete label words, where words related to a given label word are obtained from a word list using existing dictionary tools; then, the relevance of each word to the label word is measured through a corresponding heuristic method, ultimately obtaining a specified number of suitable label words; (3) Dynamic generation of prototype networks: The expected representation of all samples of each class is used as the prototype vector of the class, and the prototype vector is used to replace the continuous label words.

[0122] Take the example of domain knowledge-guided search for discrete label words. Each text has a corresponding label, and related words are selected (including word embedding, concept network and WordNet). Nodes represent single words, and edges represent the correlation scores between words. Select the node corresponding to each label, and obtain the adjacent nodes corresponding to all edges above the threshold as label word candidates. Since there may be some noise, refinement is performed to obtain the prior probability of each label word, and label words below a certain threshold are eliminated. For partially labeled data, the importance of each label word is obtained through training, and weights are assigned to each label word by learning a weight vector to avoid the influence of noise. Finally, each label will correspond to multiple label words. Assuming that each label word has the same importance, the probability of the label can be considered to be the average of the probabilities corresponding to all label words being predicted, and finally the probability is selected from all labels with the highest probability.

[0123] (3) Replace the text into the generated template to form a prompt template consisting of the prompt template template, the pre-trained model (plm) and the output mapping (promptVeralizer), and use the pre-trained language model for prediction. The following pre-trained language models can be selected: (1) autoregressive model; (2) autoencoding model; (3) multimodal model; (4) retrieval-based model; etc.

[0124] (4) Map the prediction results back to the labels and export the fine-tuned generative AI model. The fine-tuning process is then completed and the fine-tuned model is applied to the automatic generation process of the simulation model.

[0125] In summary, the advantages of the embodiments of the present invention include at least the following: (1) Time and cost savings: The automatic generation and modification of CAD / CAE models can significantly reduce the time and effort required to build simulation models. This can improve efficiency and reduce the costs associated with evaluating and optimizing building designs and improvements (such as interior layout adjustments); (2) Low threshold: Various types of inputs can be used in the CAD / CAE model generation and modification process, such as natural language or general descriptions from other software systems. This eliminates the obstacles for non-CAE experts to obtain guidance from simulation results and support building owners' decision-making; (3) Higher tolerance for requirement descriptions: CAD / CAE scripts can be generated in the absence of sufficient input, and default / reference values ​​are stored in the task templates. Incorrect input parameters can be detected during the process, thereby improving the flexibility of the system.

[0126] Fig. 5 is an exemplary schematic diagram of simulation results according to an embodiment of the present invention. In Fig. 5, taking the temperature distribution diagram in a room as an example, the simulation results of the simulation target based on the natural language description input are exemplarily described.

[0127] FIG6 is an exemplary structural diagram of an indoor environment simulation system according to an embodiment of the present invention. The indoor environment simulation system includes:

[0128] The human-computer interaction device 31 is configured to obtain first information and second information, wherein the first information is used to describe the indoor environment, and the second information is used to describe a simulation target of the indoor environment;

[0129] A script generation device 32 is used to input the first information and the second information into a generative AI model; a first script and a second script are received from the generative AI model, wherein the first script is adapted to generate a CAD model of an indoor environment, and the second script is adapted to perform CAE simulation on the CAD model to generate a simulation result corresponding to a simulation target; a CAD device 33 is used to generate a CAD model of an indoor environment based on the first script; a CAE simulation device 34 is used to perform CAE simulation on the CAD model based on the second script to generate a simulation result corresponding to the simulation target.

[0130] The human-computer interaction interface (HCI) provided by the HCI device 31 is the interface through which users interact with the simulation system. It allows users to enter requirements and descriptions, select predefined templates, and view simulation results. The HCI can be a standalone application, a website service, or integrated into other software systems.

[0131] In one embodiment, it also includes: a template library 35 for storing predefined templates, the predefined templates including predefined data of the first information and predefined data of the second information; a human-computer interaction device 31 for obtaining the editing data of the first information and the editing data of the second information via a human-computer interaction interface; determining the first information based on the predefined data of the first information and the editing data of the first information; and determining the second information based on the predefined data of the second information and the editing data of the second information.

[0132] The embodiment of the present invention also proposes an electronic device with a processor-memory architecture. Figure 7 is an exemplary structural diagram of an electronic device with a processor-memory architecture according to an embodiment of the present invention. As shown in Figure 7, the electronic device 700 includes a processor 701, a memory 702, and a computer program stored on the memory 702 and runnable on the processor 701. When the computer program is executed by the processor 701, it implements any of the above methods for simulating an indoor environment. Among them, the memory 702 can be specifically implemented as a variety of storage media such as an electrically erasable programmable read-only memory (EEPROM), a flash memory (Flash memory), and a programmable read-only memory (PROM). The processor 901 can be implemented to include one or more central processing units or one or more field programmable gate arrays, wherein the field programmable gate array integrates one or more central processing unit cores. Specifically, the central processing unit or the central processing unit core can be implemented as a CPU or an MCU or a DSP, and so on.

[0133] It should be noted that not all steps and modules in the above processes and structure diagrams are required, and certain steps or modules can be omitted based on actual needs. The execution order of the steps is not fixed and can be adjusted as needed. The division of the modules is merely for the convenience of describing the functional division adopted. In actual implementation, a module can be implemented by multiple modules, and the functions of multiple modules can be implemented by the same module. These modules can be located in the same device or in different devices.

[0134] The hardware modules in each embodiment can be implemented mechanically or electronically. For example, a hardware module may include a specially designed permanent circuit or logic device (such as a dedicated processor, such as an FPGA or ASIC) for performing a specific operation. The hardware module may also include a programmable logic device or circuit (such as a general-purpose processor or other programmable processor) temporarily configured by software to perform a specific operation. As for whether to implement the hardware module mechanically, or using a dedicated permanent circuit, or using a temporarily configured circuit (such as configured by software), it can be decided based on cost and time considerations.

[0135] The above description is only a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included in the scope of protection of the present invention.

Claims

1. A method for simulating an indoor environment, characterized in that: include: Acquiring (101) first information and second information, wherein the first information is used to describe an indoor environment, and the second information is used to describe a simulation target of the indoor environment; Inputting (102) the first information and the second information into a generative artificial intelligence model; receiving (103) a first script and a second script from the generative artificial intelligence model, wherein the first script is adapted to generate a computer-aided design model of the indoor environment, and the second script is adapted to perform a computer-aided engineering simulation on the computer-aided design model to generate a simulation result corresponding to the simulation target; Based on the first script, generating (104) a computer-aided design model of the indoor environment; Based on the second script, a computer-aided engineering simulation is performed (105) on the computer-aided design model to generate the simulation result.

2. The method according to claim 1, characterized in that Before inputting (102) the first information and the second information into a generative artificial intelligence model, the method includes: The generative artificial intelligence model is fine-tuned based on a sample, the sample comprising first sample information, second sample information, a first sample script and a second sample script, the first sample information being used to describe a sample indoor environment, the second sample information being used to describe a sample simulation target of the sample indoor environment, the first sample script being adapted to generate a computer-aided design model of the sample indoor environment, and the second sample script being adapted to perform computer-aided engineering simulation on the computer-aided design model of the sample indoor environment to generate a sample simulation result corresponding to the sample simulation target.

3. The method according to claim 2, characterized in that The fine-tuning includes: Inputting the first sample information and the second sample information into the generative artificial intelligence model; Obtaining a first output script and a second output script from the generative artificial intelligence model, wherein: the first output script is adapted to generate a computer-aided design model of the sample indoor environment, and the second output script is adapted to perform a computer-aided engineering simulation on the computer-aided design model of the sample indoor environment to generate a sample simulation result corresponding to a sample simulation target; Determining a loss function value of the generative artificial intelligence model based on a difference between the first output script and the first sample script and a difference between the second output script and the second sample script; The model parameters of the generative artificial intelligence model are configured so that the loss function value is lower than a preset threshold.

4. The method according to claim 3, characterized in that The first output script is generated based on the first sample information, and the second output script is generated based on the second sample information; or The first output script is generated based on the first sample information and the second sample information, and the second output script is generated based on the first sample information and the second sample information.

5. The method according to claim 1, characterized in that The obtaining (101) of the first information and the second information comprises: Input data including the first information and the second information is obtained through a human-computer interaction interface.

6. The method according to claim 1, characterized in that The obtaining (101) of the first information and the second information comprises: Acquire a predefined template, where the predefined template includes predefined data of the first information and predefined data of the second information; Obtaining, via a human-computer interaction interface, edit data of the first information and edit data of the second information; determining the first information based on predefined data of the first information and edited data of the first information; The second information is determined based on predefined data of the second information and edited data of the second information.

7. The method according to claim 6, characterized in that The determining, based on the predefined data of the first information and the edited data of the first information, that the first information comprises at least one of the following: Based on the addition processing of the indoor environment description represented by the edited data of the first information, the description of the indoor environment is added based on the predefined data of the first information to determine the first information; Based on the reduction processing of the indoor environment description represented by the edited data of the first information, the description of the indoor environment is reduced based on the predefined data of the first information to determine the first information; Based on the change processing of the indoor environment description represented by the edited data of the first information, the description of the indoor environment is changed based on the predefined data of the first information to determine the first information.

8. The method according to claim 6, characterized in that The determining, based on the predefined data of the second information and the edited data of the second information, that the second information comprises at least one of the following: Based on the addition processing of the simulation target description represented by the editing data of the second information, the description of the simulation target is added based on the predefined data of the second information to determine the second information; Based on the reduction processing of the simulation target description represented by the edited data of the second information, the description of the simulation target is reduced based on the predefined data of the second information to determine the second information; Based on the change processing of the simulation target description represented by the editing data of the second information, the description of the simulation target is changed based on the predefined data of the second information to determine the second information.

9. The method according to any one of claims 1 to 8, characterized in that The first script and the first second script have a common language format, wherein the common language format is an output language format of the generative artificial intelligence model; the method comprises: Before generating (104) a computer-aided design model of the indoor environment based on the first script, converting a universal language format of the first script into a computer-aided design language format; Before executing (105) a computer-aided engineering simulation on the computer-aided design model based on the second script to generate the simulation result, the universal language format of the second script is converted into a computer-aided engineering language format.

10. The method according to any one of claims 1 to 8, characterized in that include: The simulation results are displayed via a human-computer interaction interface.

11. A simulation system for indoor environment, characterized in that: include: A human-computer interaction device (31) is used to obtain first information and second information, wherein the first information is used to describe an indoor environment, and the second information is used to describe a simulation target of the indoor environment; A script generating device (32) is used to input the first information and the second information into a generative artificial intelligence model; receive a first script and a second script from the generative artificial intelligence model, wherein the first script is adapted to generate a computer-aided design model of the indoor environment, and the second script is adapted to perform a computer-aided engineering simulation on the computer-aided design model to generate a simulation result corresponding to the simulation target; A computer-aided design device (33) for generating a computer-aided design model of the indoor environment based on the first script; A computer-aided engineering simulation device (34) is used to perform a computer-aided engineering simulation on the computer-aided design model based on the second script to generate the simulation result.

12. The system according to claim 11, characterized in that Also includes: A template library (35), used for storing predefined templates, wherein the predefined templates contain predefined data of the first information and predefined data of the second information; The human-computer interaction device (31) is used to obtain the editing data of the first information and the editing data of the second information via the human-computer interaction interface; determining the first information based on predefined data of the first information and edited data of the first information; The second information is determined based on predefined data of the second information and edited data of the second information.

13. An electronic device, characterized in that: include: Processor (701); A memory (702), configured to store executable instructions of the processor (701); The processor (701) is used to read the executable instructions from the memory (702), and execute the executable instructions to implement the indoor environment simulation method according to any one of claims 1 to 10.

14. A computer-readable storage medium having computer instructions stored thereon, characterized in that: When the computer instructions are executed by a processor, the indoor environment simulation method according to any one of claims 1 to 10 is implemented.

15. A computer program product, characterized in that The invention comprises a computer program, which, when executed by a processor, implements the indoor environment simulation method according to any one of claims 1 to 10.

Images