Computationally assisted decision-making method and system for climate-adaptive building cavity design

Abstract

The present invention relates to the technical field of natural ventilation in buildings, and in particular to a computationally assisted decision-making method and system for climate-adaptive building cavity design. The method comprises: using a Delaunay triangulation method to generate an initial mesh, using a Laplace operator-based mesh refinement method to adaptively refine the initial mesh, so as to obtain an adaptive mesh system, wherein the adaptive mesh system is used for dynamically adjusting the mesh density of conditional PINNs; constructing a multi-task learning framework within the conditional PINNs, wherein the multi-task learning framework is used for jointly predicting a plurality of physical field variables within the conditional PINNs; and iteratively training the conditional PINNs, and using a gradient descent algorithm to minimize a loss function until a predetermined number of training iterations or loss convergence is reached, thereby generating a key physical field variable prediction model in building cavity design. The present invention can implement efficient and accurate prediction of key physical quantities in building cavity design, and can be adapted to different building layouts and functional space characteristics.

Description

A computational decision-making aid method and system for designing weather-adaptive building cavities. Technical Field

[0001] This invention relates to the field of building natural ventilation technology, and in particular to a computational auxiliary decision-making method and system for designing weather-appropriate building cavities. Background Technology

[0002] In the vast field of modern architectural design, cavity design, as an innovative passive green technology, is gradually becoming a key strategy for improving the environmental performance and energy efficiency of public buildings. This technology cleverly utilizes the physical principles of nature—especially the thermo-pressure ventilation mechanism—to create natural airflow channels through the rational layout of the building's internal and external spaces. It achieves effective circulation and renewal of indoor air without relying on additional energy consumption, thereby significantly improving indoor air quality, reducing the load on air conditioning systems, and ultimately achieving the goal of energy conservation and emission reduction.

[0003] While cavity design demonstrates significant potential in green building, the complexity and precision required in its design process present numerous challenges. Traditional cavity design methods primarily rely on empirical formulas and computational fluid dynamics (CFD) simulations. While these methods have their value, they have revealed several limitations in practical applications.

[0004] 1. Low computational efficiency: Traditional CFD simulations often use fixed mesh systems, which are difficult to adapt to the complex and ever-changing geometry and flow characteristics inside building cavities. This results in a huge amount of computation and long processing time, especially when dealing with large-scale or high-precision building models.

[0005] 2. Long design cycle: Due to low computational efficiency, the design iteration cycle is lengthened, making it difficult to meet the requirements of quickly responding to market demands or adjusting design schemes. This not only increases design costs but may also delay project progress.

[0006] 3. Sensitive to initial parameters: The accuracy of CFD simulation results is highly dependent on the setting of initial and boundary conditions, and even small changes in these parameters can have a significant impact on the simulation results, increasing the uncertainty and debugging difficulty in the design process.

[0007] 4. Physical quantity separation simulation: Traditional methods often require separate simulation of different physical quantities such as wind speed and temperature, ignoring their interactions and influences, making it difficult to comprehensively and accurately reflect the real fluid dynamics behavior inside the building cavity.

[0008] 5. Insufficient capture of complex flow details: The limitations of fixed grids are also reflected in the difficulty of accurately capturing complex flow phenomena such as eddies and backflows inside building cavities, which are of great significance for optimizing ventilation and reducing energy consumption. Summary of the Invention

[0009] The purpose of this invention is to provide a computational auxiliary decision-making method and system for weather-adaptive building cavity design, which can efficiently and accurately predict key physical quantities in building cavity design, including wind speed, temperature, pressure distribution and PMV value, and can adapt to different building layouts and functional space characteristics, so as to solve at least one of the above-mentioned prior art problems.

[0010] In a first aspect, the present invention provides a computational auxiliary decision-making method for the design of weather-adaptive building cavities, the method specifically comprising:

[0011] Obtain the building cavity design parameters, and construct a conditional PINNs network based on the building cavity design parameters, physical field variables, and physical laws;

[0012] An initial mesh is generated using the Delaunay triangulation method, and then adaptively refined using a Laplacian operator-based mesh refinement method to obtain an adaptive mesh system. This adaptive mesh system is used to dynamically adjust the mesh density of the conditional PINNs network.

[0013] A multi-task learning framework is constructed in the conditional PINNs network, which is used to collaboratively predict multiple physical field variables in the conditional PINNs network.

[0014] Set a loss function, which includes a data error loss function and a physical law residual loss function. The data error loss function is used to measure the difference between the model prediction value and the actual observation value, and the physical law residual loss function is used to measure the difference between the model prediction value and the physical law.

[0015] The conditional PINNs network is iteratively trained, and the loss function is minimized using the gradient descent algorithm until the predetermined number of training iterations or loss convergence is reached, thereby generating a prediction model for key physical field variables in building cavity design.

[0016] Obtain the building structure diagram, and generate a building cavity thermal diagram based on the prediction results of the key physical field variable prediction model in the building cavity design and the building structure diagram.

[0017] Secondly, the present invention provides a computational auxiliary decision-making system for the design of weather-adaptive building cavities, the system specifically comprising:

[0018] The first decision module is used to obtain the building cavity design parameters and construct a conditional PINNs network based on the building cavity design parameters, physical field variables and physical laws.

[0019] The second decision module is used to generate an initial mesh using the Delaunay triangulation method, and to adaptively refine the initial mesh using a mesh refinement method based on the Laplacian operator to obtain an adaptive mesh system. The adaptive mesh system is used to dynamically adjust the mesh density of the conditional PINNs network.

[0020] The third decision module is used to construct a multi-task learning framework in the conditional PINNs network, and the multi-task learning framework is used to collaboratively predict multiple physical field variables in the conditional PINNs network.

[0021] The fourth decision module is used to set the loss function, which includes a data error loss function and a physical law residual loss function. The data error loss function is used to measure the difference between the model prediction value and the actual observation value, and the physical law residual loss function is used to measure the difference between the model prediction value and the physical law.

[0022] The fifth decision module is used to iteratively train the conditional PINNs network and minimize the loss function using the gradient descent algorithm until a predetermined number of training iterations or loss convergence is reached, thereby generating a prediction model for key physical field variables in building cavity design.

[0023] The sixth decision module is used to obtain the building structure diagram and generate a building cavity thermal diagram based on the prediction results of the key physical field variable prediction model in the building cavity design and the building structure diagram.

[0024] Thirdly, the present invention provides a computer device comprising: a memory and a processor, and a computer program stored in the memory, wherein when the computer program is executed on the processor, it implements a computational aid decision-making method for weather-appropriate building cavity design as described in any of the above methods.

[0025] Fourthly, the present invention provides a computer-readable storage medium, characterized in that it stores a computer program thereon, which, when executed by a processor, implements a computational auxiliary decision-making method for the design of weather-appropriate building cavities as described in any of the above methods.

[0026] Compared with the prior art, the present invention has at least one of the following technical effects:

[0027] 1. By combining adaptive mesh refinement, multi-task learning framework and conditional PINNs, this invention aims to improve the prediction accuracy and generalization ability of the model, thereby achieving performance optimization design of building cavities adapted to climate characteristics.

[0028] 2. This invention uses adaptive mesh refinement technology to refine the mesh only in key areas, reducing unnecessary computation and improving simulation accuracy.

[0029] 3. The multi-task learning framework of this invention can predict all key physical quantities simultaneously, providing more comprehensive analysis results and reducing the number of design iterations.

[0030] 4. The real-time feedback and multi-task prediction capabilities of this invention enable designers to quickly evaluate different design schemes, thus accelerating the design decision-making process.

[0031] 5. This invention reduces reliance on traditional CFD simulation and shortens the design cycle through an automated PINN model.

[0032] 6. This invention utilizes the PINN model to predict the natural ventilation effect of building cavities, including key parameters such as wind speed and temperature distribution. Combined with the predicted ventilation effect parameters, it evaluates and optimizes indoor thermal comfort (PMV) to achieve a suitable living and working environment.

[0033] 7. This invention improves natural ventilation efficiency and reduces reliance on air conditioning and mechanical ventilation systems by optimizing building cavity design, thereby reducing building energy consumption.

[0034] 8. The PINN model trained by this invention has good generalization ability and can adapt to different architectural design scenarios, providing architects with a fast and accurate design evaluation tool. Attached Figure Description

[0035] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0036] Figure 1 is a flowchart illustrating a computational auxiliary decision-making method for weather-adaptive building cavity design according to an embodiment of the present invention;

[0037] Figure 2 is a schematic diagram of a computational auxiliary decision-making system for weather-adaptive building cavity design according to an embodiment of the present invention;

[0038] Figure 3 is a schematic diagram of the structure of a computer device provided in an embodiment of the present invention. Detailed Implementation

[0039] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0040] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0041] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0042] As used in this application specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if detected [the described condition or event]" may be interpreted, depending on the context, as meaning "once determined," "in response to determination," "once detected [the described condition or event]," or "in response to detection [the described condition or event]."

[0043] Furthermore, in the description of this application and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0044] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0045] In this embodiment of the application, the executing entity of the process includes a terminal device. This terminal device includes, but is not limited to, devices capable of executing the methods disclosed in this application, such as servers, computers, smartphones, and tablets. Figure 1 shows a schematic flowchart of a computational auxiliary decision-making method for weather-appropriate building cavity design disclosed in the first embodiment of the present invention, detailed below:

[0046] S101, Obtain the building cavity design parameters, and construct a conditional PINNs network based on the building cavity design parameters, physical field variables, and physical laws.

[0047] In this embodiment, to more accurately simulate and optimize ventilation effects in building cavity design, it is necessary to comprehensively consider the geometric parameters, physical field variables, and physical laws of the building cavity. Traditional methods often suffer from problems such as large computational load and insufficient accuracy when dealing with these complex factors. Therefore, this embodiment proposes to utilize conditional PINNs networks to optimize the design parameters of the building cavity.

[0048] The design parameters for building cavities include building geometry, environmental information, other information, and predicted values. Building geometry includes building plan dimensions, number of floors, floor height, cavity cross-sectional dimensions, ventilation opening dimensions on each floor, opening dimensions at the top of the cavity, and number of opening layers. Environmental information includes regional latitude and longitude, regional average temperature during the transitional season, and regional average humidity during the transitional season. Other information includes air density, aerodynamic viscosity, gravity, specific heat capacity of air, thermal conductivity of air, source term, metabolic rate, external power, mean radiant temperature, thermal conductivity of clothing, surface temperature of clothing, and convective heat transfer coefficient. Predicted values ​​include specific values, actual measured values, and traditional CFD simulation values.

[0049] In this embodiment, compared with traditional methods, conditional PINNs networks can complete the optimization process of design parameters more quickly, reduce the number of design iterations and computation time, and shorten the overall design cycle. By introducing physical laws as constraints, conditional PINNs networks can more accurately simulate and predict the fluid flow behavior within building cavities, thereby improving the accuracy and reliability of design parameters.

[0050] In some embodiments, in step S101 above, the building cavity design parameters include building layout information, material property information, and environmental condition information; the construction of a conditional PINNs network based on the building cavity design parameters, physical field variables, and physical laws specifically includes:

[0051] The building layout information, the material property information, and the environmental condition information are transformed into a first feature vector, a second feature vector, and a third feature vector, respectively. The first feature vector, the second feature vector, and the third feature vector are then concatenated to form a comprehensive feature vector.

[0052] Design a conditional PINNs network, which includes an input layer, hidden layers, and an output layer. The input layer receives the synthesized feature vector and the physical field variables, and the output of the hidden layer is h. (l) =σ(W (l) h (l-1) +b (l) ), where σ represents the ReLU activation function, W (l) and b (l) These are the weight matrix and bias vector of the l-th hidden layer, respectively, h (l) and h (l-1) These represent the activation values ​​of the l-th and (l-1)-th layers in the network, respectively. The output layer is used to output the predicted values ​​of the physical field variables.

[0053] Each physical law is embedded into the conditional PINNs network, and the residuals of the physical laws are calculated through an automatic differentiation mechanism.

[0054] In this embodiment, a conditional PINNs network is constructed, and the feature vector c output by the conditional encoder and the physical field variables (wind pressure p, temperature T, wind speed) are input. (and PMV). The network structure can be divided into the following modules:

[0055] Input layer: Accepts physical field variables and conditional eigenvectors c. Assume the physical field variable is x, and the input layer input is [x, c].

[0056] Hidden layers: Several hidden layers, using a fully connected network (FCN), each layer using the ReLU activation function; assuming a total of L hidden layers, the output of each layer is: h (l) =σ(W (l) h (l-1) +b (l) )

[0057] Output layer: Outputs predicted values ​​of physical field variables, such as wind pressure p, temperature T, and wind speed. And PMV.

[0058] Conditional PINNs models can automatically adjust the network structure and parameters according to different building layouts and functional space characteristics.

[0059] Furthermore, the comprehensive feature vector is c = [b, m, e] ∈ R. n+m+k+p Where b represents the first eigenvector and b∈R n+m m represents the second eigenvector and m∈R k Let e ​​represent the third eigenvector and e∈R pR represents the weight matrix in the neural network, n represents the number of rooms in the building layout information, m represents the number of ventilation openings in the building layout information, k represents the number of material properties, and p represents the number of environmental conditions.

[0060] In this embodiment, when designing a conditional PINNs network, the input parameter information first needs to be encoded into a feature vector using a conditional encoder, employing the following method:

[0061] Building layout coding: Encoding building layout information (such as room location, size, and vent location) into a multi-dimensional vector. Assuming a building has n rooms and m vents, the location and size of each room, as well as the location of each vent, can be used as features to form a feature vector b∈R. n+m .

[0062] Material property encoding: Encode material properties (such as the thermal conductivity of wall materials and insulation materials) into a feature vector m∈R. k , where k is the number of material properties.

[0063] Environmental condition encoding: Encoding environmental conditions (such as external temperature) into a feature vector e∈R p , where p is the number of environmental conditions.

[0064] Finally, the above feature vectors are concatenated to form a comprehensive feature vector c = [b, m, e] ∈ R. n+m+k+p .

[0065] By introducing a conditional encoder, conditional information such as building layout and functional space characteristics is encoded into feature vectors and fused with the output of a shared encoder to adapt to different design conditions.

[0066] Furthermore, the physical laws include the wind pressure formula, temperature formula, wind speed formula, and thermal comfort formula;

[0067] Wherein, the wind pressure formula is ρ represents air density (kg / m3), and v represents the velocity field (m / s);

[0068] The temperature formula is T represents temperature (K), ρ represents air density (kg / m³). 3 ), c p This represents the specific heat capacity of air (J / (kg·K)). Let S represent the velocity field (m / s), k represent the thermal conductivity of air (W / (m·K)), and S represent the velocity field. T The term represents the source term, including internal heat sources, radiation, etc. (W / m²) 3 ), This represents the Laplace operator (the Laplace operator is a second-order differential operator used to calculate the divergence of the gradient of a function in space, i.e., the local rate of change of the function). It represents the rate of change of the temperature field T in space, i.e., the divergence of the temperature gradient;

[0069] The formula for wind speed is Let ρ represent the velocity field (m / s), and ρ represent the air density (kg / m³). 3 ), p represents pressure (Pa), μ represents the dynamic viscosity of air (Pa·s), This represents external physical force (such as gravity) (N / m) 3 );

[0070] The formula for thermal comfort is: M represents metabolic rate (W / m²) 2 W represents external power (W / m) 2 ), p a T represents the partial pressure of water vapor (Pa). a T represents air temperature (°C). r f represents the average radiant temperature (°C). cl T represents the thermal conductivity coefficient of clothing. cl This indicates the surface temperature of the clothing (°C).

[0071] In this embodiment, physical law formulas for wind pressure, temperature, wind speed, and PMV are embedded in the network, and the residuals of the physical laws are calculated through an automatic differentiation mechanism. Specifically, the derivative of the velocity field is calculated using the automatic differentiation mechanism, and wind pressure and wind speed formulas are embedded in the velocity field predicted by the network to calculate wind pressure and wind speed respectively; the time and spatial derivatives of the temperature field are calculated using automatic differentiation, and temperature formulas are embedded in the temperature field predicted by the network to calculate temperature; and PMV values ​​are calculated based on the temperature predicted by the network, radiation temperature, water vapor partial pressure, etc.

[0072] Through the above design and implementation, it can be ensured that the conditional PINNs network can effectively combine building condition information and physical laws, thereby improving the prediction accuracy and adaptability of the model.

[0073] S102, an initial mesh is generated using the Delaunay triangulation method, and the initial mesh is adaptively refined using a mesh refinement method based on the Laplacian operator to obtain an adaptive mesh system. The adaptive mesh system is used to dynamically adjust the mesh density of the conditional PINNs network.

[0074] In this embodiment, the generation of the initial mesh and its subsequent adaptive refinement are crucial for improving the accuracy and efficiency of network prediction when constructing a Conditional PINNs (Physical Information Neural Network). The Delaunay triangulation method is widely used in mesh generation because it can generate triangular meshes that conform to certain geometric criteria. Meanwhile, the Laplacian operator-based mesh refinement method can effectively adjust the mesh density according to local features, achieving adaptive mesh refinement. This embodiment combines these two methods to construct an adaptive mesh system for dynamically adjusting the mesh density of a Conditional PINNs network.

[0075] The Delaunay triangulation method is used to generate the initial mesh. First, based on the geometry and boundary conditions of the region to be analyzed, a series of discrete points (which can be uniformly distributed or distributed according to actual needs) are obtained. Then, by calling existing numerical computation libraries (such as the Delaunay functions in SciPy), the discrete points are triangulated to generate the initial triangular mesh. Delaunay triangulation ensures that the generated triangular mesh satisfies the empty circle property, meaning that the circumcircle of any triangle does not contain any other vertices.

[0076] Define the criteria for mesh refinement. This is typically based on local error estimates or gradient changes in physical field variables. For example, it can be set that mesh refinement is required for a region when the gradient of physical field variables exceeds a certain threshold.

[0077] The initial mesh is iteratively refined. In each refinement round, the Laplacian operator for each vertex (i.e., the weighted average of the position differences between that vertex and its adjacent vertices) is calculated, and the vertex positions are adjusted based on the value of the Laplacian operator, automatically refining the mesh in the areas requiring refinement. During the refinement process, the degree and effect of refinement can be controlled by setting different refinement parameters (such as the number of iterations, the weight of the Laplacian operator, etc.).

[0078] An adaptive mesh system is constructed by combining the initial mesh generated by Delaunay triangulation with a Laplacian operator-based mesh refinement method. This system can automatically adjust the mesh density according to changes in physical field variables, improving the mesh resolution in critical regions.

[0079] When constructing conditional PINNs networks, this adaptive grid system is used to dynamically adjust the network's grid density. This allows the network to perform more detailed simulations and predictions of physical processes in key regions while maintaining overall computational efficiency.

[0080] In this embodiment, an adaptive mesh generation and refinement module is introduced into the model's input layer. Based on the gradient or error distribution of the physical field, local mesh refinement is implemented in key areas (such as shaft entrances, corners, and top openings) to improve simulation accuracy in these areas while reducing computational load. The adaptive mesh system can automatically adjust the mesh density according to changes in physical field variables, resulting in higher mesh resolution in areas where physical processes change drastically, thereby improving simulation accuracy and precision. Compared to a globally uniform mesh, the adaptive mesh system reduces unnecessary computation and improves overall computational efficiency by refining the mesh only in areas requiring refinement. This method offers strong flexibility, allowing for customization of mesh refinement standards and parameters based on different physical field characteristics and simulation requirements, making it suitable for mesh generation and optimization in various complex scenarios.

[0081] In some embodiments, step S102 above, which involves generating an initial mesh using the Delaunay triangulation method and adaptively refining the initial mesh using a Laplacian operator-based mesh refinement method to obtain an adaptive mesh system, specifically includes:

[0082] An initial mesh was generated using the Delaunay triangulation method, and a preliminary simulation was performed on the initial mesh using the finite element method to calculate the gradients of the velocity and temperature fields.

[0083] A mesh refinement standard is set, and a refinement algorithm based on the Laplacian operator is used to determine the region to be refined on the initial mesh according to the mesh refinement standard.

[0084] The region to be refined is meshed using the red-green refinement method to obtain an adaptive mesh system.

[0085] The angle of the grid cells in the adaptive grid system is checked, and the position of the grid nodes in the adaptive grid system is adjusted using the Laplacian smoothing algorithm.

[0086] In this embodiment, an adaptive mesh generation algorithm based on error estimation is employed. An initial mesh is generated using the Delaunay triangulation method, and adaptive refinement is performed using a Laplacian operator-based mesh refinement method. The specific implementation steps are as follows:

[0087] Initial mesh generation: The initial mesh is generated using the Delaunay triangulation method, ensuring mesh uniformity and good geometry. This is achieved by calling existing numerical computation libraries (Delaunay functions in SciPy).

[0088] Error estimation: A preliminary simulation is performed on the initial mesh to calculate the gradients of the velocity and temperature fields. The preliminary simulation is conducted using the finite element method, employing existing finite element software (COMSOL). The calculation formulas are as follows:

[0089] Where φ is a physical field (such as velocity or temperature), It is a unit vector.

[0090] Mesh refinement criteria: Set refinement criteria, such as refining the local mesh when the velocity field gradient exceeds 0.1 m / s² or the temperature change rate exceeds 1 K / m.

[0091] Refinement Algorithm: A Laplacian operator-based refinement algorithm is employed. The region requiring refinement is determined by calculating the Laplacian operator values ​​at the grid nodes. The Laplacian operator calculation formula is as follows:

[0092] Where φ is the physical field and N is the number of neighboring nodes of node 0.

[0093] Mesh refinement: In the defined areas requiring refinement, mesh nodes are added to generate finer mesh cells. The red-green refinement method is used for mesh refinement to ensure that the refined mesh retains good geometry.

[0094] Mesh quality control: Ensure the refined mesh system has good quality, such as orthogonality, uniformity, and minimum angle limits. Specific measures include:

[0095] Check the angles of the grid cells to ensure that all angles are within a reasonable range (30° to 150°).

[0096] Adjust the grid node positions to improve grid orthogonality and uniformity. Use the Laplacian smoothing algorithm:

[0097] Where, x i It is the new position of node i, x j is the position of the neighboring nodes of node i, and N is the number of neighboring nodes.

[0098] S103, a multi-task learning framework is constructed in the conditional PINNs network, the multi-task learning framework being used to collaboratively predict multiple physical field variables in the conditional PINNs network.

[0099] In this embodiment, a multi-task learning framework is constructed in a conditional PINNs (Physical Information Neural Network) network, aiming to achieve synergistic optimization and mutual enhancement among multiple physical field variables by simultaneously learning and predicting them.

[0100] Specifically, consider a building cavity design problem that requires the simultaneous prediction of multiple physical field variables, such as wind speed, temperature, and pressure. These physical field variables have certain interdependencies; for example, changes in wind speed affect temperature distribution, and changes in temperature affect pressure.

[0101] In conditional PINNs networks, a shared feature extractor is first designed. This extractor is responsible for extracting general features from the input data that are helpful for predicting multiple physical field variables. These features may include the geometry of the building cavity, boundary conditions, initial conditions, etc.

[0102] Following the shared feature extractor, multiple parallel task-specific layers (also known as "towers" or "heads") are designed, each responsible for predicting a specific physical field variable. These task-specific layers may share some parameters or have independent parameters, depending on the correlation between tasks and the complexity of the network design.

[0103] To optimize the prediction performance of multiple tasks simultaneously, a multi-task loss function is designed. This function is typically a weighted sum of the loss functions for each task, and the weights can be adjusted based on the importance of the tasks or the characteristics of the dataset.

[0104] The conditional PINNs network is trained using preprocessed data. During training, the network optimizes its parameters by minimizing a multi-task loss function. Because multiple tasks share a partial feature extractor, the network learns more generalized feature representations, thereby improving overall prediction performance.

[0105] In this embodiment, through a multi-task learning framework, the conditional PINNs network can simultaneously learn and predict multiple physical field variables, leveraging their interdependencies to improve prediction accuracy. Compared to single-task learning models, multi-task learning models can learn more generalized and robust feature representations. The multi-task learning framework reduces computation by sharing feature extractors, avoiding the overhead of training a separate model for each task. Furthermore, because multiple tasks share some parameters, the network can converge to the optimal solution faster during training. By simultaneously learning multiple related tasks, the conditional PINNs network can learn more general feature representations, thereby improving the model's generalization ability under different working conditions and circumstances. This is of great significance for handling complex and variable architectural cavity design problems.

[0106] Specifically, the multi-task learning framework includes a shared encoder structure and a task-specific decoder structure;

[0107] The shared encoder is used to encode the input physical field variables and the integrated feature vector into a shared latent representation;

[0108] The task-specific decoder structure is used to design a separate decoder for each physical field variable. The decoder is used to take the output of the shared encoder as the output and output the predicted value of the corresponding physical field variable.

[0109] In this embodiment, a deep neural network (DNN) is used as a shared encoder to encode the input physical field variables and conditional feature vectors into a shared latent representation. Assuming the input is , the output of the shared encoder is .

[0110] The shared encoder consists of several hidden layers, with the hidden layer structure described in step three, employing an FCN structure. Each layer uses the ReLU activation function. Assume the shared encoder has L hidden layers, and the output of each layer is: h (l) =σ(W (l) h (l-1) +b (l) )

[0111] Where σ is the ReLU activation function, W (l) and b (l) These are the weight matrix and bias vector of the l-th layer, respectively. The output h of the last layer... (L) That is, the output h of the shared encoder shared .

[0112] For each physical quantity (wind pressure p, temperature T, wind speed) Design an independent decoder for each physical quantity (PMV). Each decoder takes the output of the shared encoder as input and outputs the predicted value of the corresponding physical quantity.

[0113] The task-specific decoder also consists of several hidden layers, with the same structure as described above. Assume the task-specific decoder has M hidden layers, and the output h of the last layer... (M) This is the predicted value of the corresponding physical quantity.

[0114] S104, Set the loss function, which includes a data error loss function and a physical law residual loss function. The data error loss function is used to measure the difference between the model prediction value and the actual observation value, and the physical law residual loss function is used to measure the difference between the model prediction value and the physical law.

[0115] In this embodiment, when dealing with the multi-physics variable prediction problem in building cavity design, setting a comprehensive loss function that includes a data error loss function and a physical law residual loss function is an effective method, which can improve the prediction accuracy and physical consistency of the model. Specifically, the data error loss function typically uses mean squared error (MSE) or mean absolute error (MAE) as a metric to calculate the difference between the model-predicted physical variable values ​​(such as wind speed, temperature, pressure, etc.) and the actual observed values. The physical law residual loss function is used to evaluate whether the model-predicted physical variables satisfy known physical laws, and it is usually implemented by constructing residual terms based on physical laws.

[0116] Furthermore, the data error loss function is: Where K represents the number of physical field variables, λ k These are the weighting coefficients for each physical quantity. This represents the data error loss for the k-th physical quantity, where N is the number of samples. and These are the actual observed value and the predicted value of the i-th sample, respectively;

[0117] The physical law residual loss function is: Where, μ m These are the weighting coefficients for each physical law. This represents the residual loss for the m-th physical law, where N is the sample size. It is the physical law residual of the i-th sample;

[0118] The loss function is: Here, α and β are weighting coefficients for data error loss and physical law residual loss, used to balance the importance of the two.

[0119] In this embodiment, data error loss is used to measure the difference between model predictions and actual observations. In this embodiment, multiple physical quantities (wind pressure p, temperature T, wind speed) are considered. Therefore, the data error loss can be defined as the weighted sum of the errors of each physical quantity (including PMV).

[0120] For K physical quantities, the predicted value for each physical quantity is... The actual observed value is y k Then data error loss It can be defined as:

[0121] Where, λ k These are the weighting coefficients for each physical quantity. This is the data error loss for the k-th physical quantity. In this embodiment, K = 4. Specifically, the mean square error (MSE) is used as the data error loss for each physical quantity.

[0122] Where N is the number of samples, and These are the actual observed value and the predicted value of the i-th sample, respectively.

[0123] The physical law residual loss measures the difference between the model's predicted values ​​and the physical laws. In this scenario, considering the physical law formulas for wind pressure, temperature, wind speed, and PMV, the physical law residual loss can be defined as the weighted sum of the residuals of each physical law. For M physical laws (M=4), the residual for each physical law is r. m Then the residual loss of physical laws It can be defined as:

[0124] Where, μ m These are the weighting coefficients for each physical law. This is the residual loss for the m-th physical law. Specifically, the mean squared error (MSE) is used as the residual loss for each physical law:

[0125] Where N is the number of samples, It is the physical law residual of the i-th sample.

[0126] Total loss function It is a weighted sum of data error loss and physical law residual loss:

[0127] Here, α and β are weighting coefficients for data error loss and physical law residual loss, used to balance the importance of the two.

[0128] S105, the conditional PINNs network is iteratively trained, and the loss function is minimized by using the gradient descent algorithm until the predetermined number of training iterations is reached or the loss converges, thereby generating a prediction model for key physical field variables in building cavity design.

[0129] In this embodiment, in each iteration, the gradient of the comprehensive loss function with respect to the network parameters is calculated using automatic differentiation techniques, including the partial derivatives of the data error loss function and the physical law residual loss function with respect to the parameters.

[0130] The weights and biases of the network are updated using the calculated gradients obtained from gradient descent algorithms (such as batch gradient descent, stochastic gradient descent, or mini-batch gradient descent). The update rule typically follows a certain step size in the opposite direction of the gradient.

[0131] Set a predetermined number of training iterations or a loss convergence condition as the criterion for stopping training. After each iteration, check whether the stopping condition is met; if it is, stop training; otherwise, continue to the next iteration.

[0132] Through iterative training, conditional PINNs networks can gradually learn the complex mapping relationships of key physical field variables in building cavity design, thereby improving the accuracy of predictions.

[0133] In some embodiments, step S105 above, which involves iteratively training the conditional PINNs network while minimizing the loss function using a gradient descent algorithm, specifically includes:

[0134] Initialize the network parameters of the conditional PINNs network according to the Xavier initialization method;

[0135] The conditional PINNs network is trained iteratively, and in each iteration, forward propagation is used to calculate the predicted value of each physical field variable.

[0136] The loss of the network parameters of the conditional PINNs network is calculated using the loss function.

[0137] Set up an Adam optimizer to minimize the loss function. The Adam optimizer is Adam(η,β1,β2), where η represents the learning rate, β1 represents the exponential decay rate of the first moment estimate, and β2 represents the exponential decay rate of the second moment estimate.

[0138] In this embodiment, the Xavier initialization method is used to initialize the network parameters to ensure that the network parameters maintain an appropriate scale in the early stages of training, avoiding gradient vanishing or exploding. For the weights W...

[0139] Where u represents a uniform distribution, n in and n out These are the number of input and output units for the current layer, respectively. The bias vector b is initialized to zero (b = 0).

[0140] The Adam optimizer is chosen as the gradient descent algorithm because it combines the advantages of momentum and adaptive learning rate, resulting in more stable convergence. The parameters of the Adam optimizer include the learning rate η, the exponential decay rate β1 for the first moment estimate, and the exponential decay rate β2 for the second moment estimate. Adam(η,β1,β2)

[0141] The default parameters are set to η = 0.001, β1 = 0.9, and β2 = 0.999.

[0142] In each training iteration, forward propagation is performed first to calculate the predicted value of each physical quantity.

[0143] Shared encoder: The input [x,c] is passed through a shared encoder to obtain h. shared .

[0144] Task-specific decoder: h shared The values ​​are input into each task-specific decoder to obtain the predicted value of each physical quantity.

[0145] Calculate the data error loss based on the predicted and actual observed values. And calculate the physical law residual loss based on the predicted values ​​and physical law formulas. Finally, the total loss was calculated using the formula.

[0146] Calculate the gradient of the total loss with respect to all parameters, and update the parameters using the Adam optimizer.

[0147] Calculate the gradient:

[0148] Update parameters:

[0149] Repeat the above process of forward propagation, loss calculation, and backpropagation until the predetermined number of training iterations is reached or the loss converges.

[0150] S106, Obtain the building structure diagram, and generate a building cavity thermal diagram based on the prediction results of the key physical field variable prediction model in the building cavity design and the building structure diagram.

[0151] In this embodiment, the building cavity heat map can intuitively display the distribution of various physical field variables inside the building, providing architects, engineers, and others with an intuitive reference. By combining the predictive model and the heat map, the distribution of various physical field variables inside the building cavity can be predicted during the design phase, thereby identifying and resolving potential problems in advance and improving design efficiency and accuracy.

[0152] In some embodiments, step S106 above, generating a thermal map of the building cavity based on the prediction results of the prediction model for key physical field variables in the building cavity design and the building structure diagram, specifically includes:

[0153] Physical quantity data are obtained from the prediction results of the key physical field variable prediction model in the building cavity design. The physical quantity data includes Z... p Z T , and Z PMV Where each Z represents a two-dimensional array, Zp Z T , and Z PMV These represent the distribution of wind pressure, temperature, wind speed, and thermal comfort on the building plan, respectively.

[0154] Z p Z T , and Z PMV Each array C is mapped to a color space, and an initial heatmap corresponding to the physical field variables is generated based on each array C.

[0155] Each initial heatmap is overlaid with the building structure diagram to form the target heatmap.

[0156] In this embodiment, the physical quantity data to be displayed is extracted from the model prediction results, assuming they are Z... p Z T , and Z PMV Each Z is a two-dimensional array representing the distribution of the corresponding physical quantity on the building plane.

[0157] Choose a suitable color mapping scheme (Jet) to map the value of each physical quantity to a color space. The color mapping function C can be defined as: C: Z→C

[0158] Here, C is a two-dimensional array of the same size as Z, and each element corresponds to a color value.

[0159] Heatmap generation: Heatmaps are generated using a color-mapped array C. Each physical quantity corresponds to one heatmap, for a total of four. Each heatmap is overlaid with the building plan and cavity location annotations to form the heatmap for the corresponding physical quantity. The four heatmaps are displayed on the same page using a sub-map layout (2x2 grid) for easy comparison and analysis.

[0160] To facilitate design interaction and optimization, the visualization module should support user input and real-time updates of heatmaps. The design interface should allow users to input different design parameters (building layout, cavity locations, etc.) and rerun the model prediction based on these parameters to obtain new physical quantity data Z. p Z T , and Z PMV The system generates and updates heatmaps using new physical quantity data, while also updating the display of building plan structure and cavity locations. By observing changes in the heatmaps, users can intuitively assess the impact of different design parameters on the distribution of physical quantities, thereby optimizing the design. Based on the results of the observation and analysis, users can adjust the design parameters and rerun model predictions and heatmap updates.

[0161] Referring to Figure 2, an embodiment of the present invention provides a computational auxiliary decision-making system 2 for the design of weather-appropriate building cavities, the system 2 specifically comprising:

[0162] The first decision module 201 is used to obtain the building cavity design parameters and construct a conditional PINNs network based on the building cavity design parameters, physical field variables and physical laws.

[0163] The second decision module 202 is used to generate an initial mesh using the Delaunay triangulation method, and to adaptively refine the initial mesh using a mesh refinement method based on the Laplacian operator to obtain an adaptive mesh system. The adaptive mesh system is used to dynamically adjust the mesh density of the conditional PINNs network.

[0164] The third decision module 203 is used to construct a multi-task learning framework in the conditional PINNs network, and the multi-task learning framework is used to collaboratively predict multiple physical field variables in the conditional PINNs network.

[0165] The fourth decision module 204 is used to set the loss function, which includes a data error loss function and a physical law residual loss function. The data error loss function is used to measure the difference between the model prediction value and the actual observation value, and the physical law residual loss function is used to measure the difference between the model prediction value and the physical law.

[0166] The fifth decision module 205 is used to iteratively train the conditional PINNs network and simultaneously use the gradient descent algorithm to minimize the loss function until a predetermined number of training iterations or loss convergence is reached, thereby generating a prediction model for key physical field variables in building cavity design.

[0167] The sixth decision module 206 is used to obtain the building structure diagram and generate a building cavity thermal diagram based on the prediction results of the key physical field variable prediction model in the building cavity design and the building structure diagram.

[0168] It is understood that the content of the computational auxiliary decision-making method embodiment for weather-friendly building cavity design shown in Figure 1 is applicable to the computational auxiliary decision-making system embodiment for weather-friendly building cavity design. The specific functions implemented by the computational auxiliary decision-making system embodiment for weather-friendly building cavity design are the same as those of the computational auxiliary decision-making method embodiment for weather-friendly building cavity design shown in Figure 1, and the beneficial effects achieved are also the same as those achieved by the computational auxiliary decision-making method embodiment for weather-friendly building cavity design shown in Figure 1.

[0169] It should be noted that the information interaction and execution process between the above systems are based on the same concept as the method embodiments of the present invention. For details on their specific functions and technical effects, please refer to the method embodiments section, which will not be repeated here.

[0170] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the system can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0171] Referring to FIG3, an embodiment of the present invention also provides a computer device 3, including: a memory 302 and a processor 301, and a computer program 303 stored in the memory 302. When the computer program 303 is executed on the processor 301, it implements a computational auxiliary decision-making method for weather-appropriate building cavity design as described in any of the above methods.

[0172] The computer device 3 can be a desktop computer, laptop, handheld computer, or cloud server, etc. The computer device 3 may include, but is not limited to, a processor 301 and a memory 302. Those skilled in the art will understand that Figure 3 is merely an example of the computer device 3 and does not constitute a limitation on the computer device 3. It may include more or fewer components than shown, or combine certain components, or different components, such as input / output devices, network access devices, etc.

[0173] The processor 301 may be a Central Processing Unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.

[0174] In some embodiments, the memory 302 may be an internal storage unit of the computer device 3, such as a hard disk or memory of the computer device 3. In other embodiments, the memory 302 may be an external storage device of the computer device 3, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the computer device 3. Further, the memory 302 may include both internal and external storage units of the computer device 3. The memory 302 is used to store the operating system, applications, bootloader, data, and other programs, such as the program code of the computer program. The memory 302 can also be used to temporarily store data that has been output or will be output.

[0175] This invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements a computational auxiliary decision-making method for the design of weather-appropriate building cavities as described in any of the above methods.

[0176] In this embodiment, if the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include at least: any entity or device capable of carrying computer program code to a photographing device / terminal device, a recording medium, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium. Examples include USB flash drives, portable hard drives, magnetic disks, or optical disks. In some jurisdictions, according to legislation and patent practice, computer-readable media cannot be electrical carrier signals or telecommunication signals.

[0177] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0178] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0179] In the embodiments disclosed in this application, it should be understood that the disclosed devices / terminal equipment and methods can be implemented in other ways. For example, the device / terminal equipment embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling or direct coupling or communication connection may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0180] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

Claims

1. A computational decision-aiding method for designing weather-adaptive building cavities, characterized in that, The method specifically includes: Obtain the building cavity design parameters, and construct a conditional PINNs network based on the building cavity design parameters, physical field variables, and physical laws; An initial mesh is generated using the Delaunay triangulation method, and then adaptively refined using a Laplacian operator-based mesh refinement method to obtain an adaptive mesh system. This adaptive mesh system is used to dynamically adjust the mesh density of the conditional PINNs network. A multi-task learning framework is constructed in the conditional PINNs network, which is used to collaboratively predict multiple physical field variables in the conditional PINNs network. Set a loss function, which includes a data error loss function and a physical law residual loss function. The data error loss function is used to measure the difference between the model prediction value and the actual observation value, and the physical law residual loss function is used to measure the difference between the model prediction value and the physical law. The conditional PINNs network is iteratively trained, and the loss function is minimized using the gradient descent algorithm until the predetermined number of training iterations or loss convergence is reached, thereby generating a prediction model for key physical field variables in building cavity design. Obtain the building structure diagram, and generate a building cavity thermal diagram based on the prediction results of the key physical field variable prediction model in the building cavity design and the building structure diagram.

2. The method according to claim 1, characterized in that, The building cavity design parameters include building layout information, material property information, and environmental condition information; the construction of a conditional PINNs network based on the building cavity design parameters, physical field variables, and physical laws specifically includes: The building layout information, the material property information, and the environmental condition information are transformed into a first feature vector, a second feature vector, and a third feature vector, respectively. The first feature vector, the second feature vector, and the third feature vector are then concatenated to form a comprehensive feature vector. Design a conditional PINNs network, which includes an input layer, hidden layers, and an output layer. The input layer receives the synthesized feature vector and the physical field variables, and the output of the hidden layer is h. (l) =σ(W (l) h (l-1) +b (l) ), where σ represents the ReLU activation function, W (l) and b (l) These are the weight matrix and bias vector of the l-th hidden layer, respectively, h (l) and h (l-1) These represent the activation values ​​of the l-th and (l-1)-th layers in the network, respectively. The output layer is used to output the predicted values ​​of the physical field variables. Each physical law is embedded into the conditional PINNs network, and the residuals of the physical laws are calculated through an automatic differentiation mechanism.

3. The method according to claim 2, characterized in that, The comprehensive feature vector is c = [b, m, e] ∈ R. n+m+k+p Where b represents the first eigenvector and b∈R n+m m represents the second eigenvector and m∈R k Let e ​​represent the third eigenvector and e∈R p R represents the weight matrix in the neural network, n represents the number of rooms in the building layout information, m represents the number of ventilation openings in the building layout information, k represents the number of material properties, and p represents the number of environmental conditions. n+m+k+p R represents a weight matrix that includes the number of rooms, the number of vents, the number of material properties, and the number of environmental conditions. n+m R represents a weight matrix containing feature information about the number of rooms and the number of air vents. k R represents the weight matrix containing the number of material properties. p This represents a weight matrix that includes the number of environmental conditions.

4. The method according to claim 2, characterized in that, The physical laws mentioned include the wind pressure formula; Wherein, the wind pressure formula is ρ represents air density, and v represents the velocity field.

5. The method according to claim 1, characterized in that, The process of generating an initial mesh using the Delaunay triangulation method and then adaptively refining the initial mesh using a Laplacian operator-based mesh refinement method to obtain an adaptive mesh system specifically includes: An initial mesh was generated using the Delaunay triangulation method, and a preliminary simulation was performed on the initial mesh using the finite element method to calculate the gradients of the velocity and temperature fields. A mesh refinement standard is set, and a refinement algorithm based on the Laplacian operator is used to determine the region to be refined on the initial mesh according to the mesh refinement standard. The region to be refined is meshed using the red-green refinement method to obtain an adaptive mesh system. The angle of the grid cells in the adaptive grid system is checked, and the position of the grid nodes in the adaptive grid system is adjusted using the Laplacian smoothing algorithm.

6. The method according to claim 2, characterized in that, The multi-task learning framework includes a shared encoder structure and a task-specific decoder structure; The shared encoder is used to encode the input physical field variables and the integrated feature vector into a shared latent representation; The task-specific decoder structure is used to design a separate decoder for each physical field variable. The decoder is used to take the output of the shared encoder as the output and output the predicted value of the corresponding physical field variable.

7. The method according to claim 1, characterized in that, The data error loss function is: Where K represents the number of physical field variables, λ k These are the weighting coefficients for each physical quantity. This represents the data error loss for the k-th physical quantity, where N is the number of samples. and These are the actual observed value and the predicted value of the i-th sample, respectively; The physical law residual loss function is: Where, μ m These are the weighting coefficients for each physical law. This represents the residual loss for the m-th physical law, where N is the sample size. It is the physical law residual of the i-th sample, and M represents the number of physical laws; The loss function is: Here, α and β are weighting coefficients for data error loss and physical law residual loss, used to balance the importance of the two.

8. The method according to claim 1, characterized in that, The iterative training of the conditional PINNs network, while minimizing the loss function using a gradient descent algorithm, specifically includes: Initialize the network parameters of the conditional PINNs network according to the Xavier initialization method; The conditional PINNs network is trained iteratively, and in each iteration, forward propagation is used to calculate the predicted value of each physical field variable. The loss of the network parameters of the conditional PINNs network is calculated using the loss function. Set up an Adam optimizer to minimize the loss function. The Adam optimizer is Adam(η,β1,β2), where η represents the learning rate, β1 represents the exponential decay rate of the first moment estimate, and β2 represents the exponential decay rate of the second moment estimate.

9. The method according to any one of claims 1 to 8, characterized in that, The step of generating a thermal map of the building cavity based on the prediction results of the key physical field variable prediction model in the building cavity design and the building structure diagram specifically includes: Physical quantity data are obtained from the prediction results of the key physical field variable prediction model in the building cavity design. The physical quantity data includes Z... p Z T , and Z PMV Where each Z represents a two-dimensional array, Z p Z T , and Z PMV These represent the distribution of wind pressure, temperature, wind speed, and thermal comfort on the building plan, respectively. Take Z p 、Z T 、 and Z PMV Each array C is mapped to a color space, and an initial heatmap corresponding to the physical field variables is generated based on each array C. Each initial heatmap is overlaid with the building structure diagram to form the target heatmap.

10. A computational auxiliary decision-making system for the design of weather-adaptive building cavities, characterized in that, The system specifically includes: The first decision module is used to obtain the building cavity design parameters and construct a conditional PINNs network based on the building cavity design parameters, physical field variables and physical laws. The second decision module is used to generate an initial mesh using the Delaunay triangulation method, and to adaptively refine the initial mesh using a mesh refinement method based on the Laplacian operator to obtain an adaptive mesh system. The adaptive mesh system is used to dynamically adjust the mesh density of the conditional PINNs network. The third decision module is used to construct a multi-task learning framework in the conditional PINNs network, and the multi-task learning framework is used to collaboratively predict multiple physical field variables in the conditional PINNs network. The fourth decision module is used to set the loss function, which includes a data error loss function and a physical law residual loss function. The data error loss function is used to measure the difference between the model prediction value and the actual observation value, and the physical law residual loss function is used to measure the difference between the model prediction value and the physical law. The fifth decision module is used to iteratively train the conditional PINNs network and simultaneously use the gradient descent algorithm to minimize the loss function until a predetermined number of training iterations or loss convergence is reached, thereby generating a prediction model for key physical field variables in building cavity design. The sixth decision module is used to obtain the building structure diagram and generate a building cavity thermal diagram based on the prediction results of the key physical field variable prediction model in the building cavity design and the building structure diagram.

Images