Moisture content driven heat and humidity distribution simulation and artificial intelligence prediction method for building envelope

Abstract

The invention relates to a moisture content-driven building envelope heat and humidity distribution simulation and artificial intelligence prediction method, which comprises the following steps of: 1) obtaining basic data of a building envelope and collecting early-stage information, including parameters such as a construction drawing, building material heat and humidity physical properties and indoor and outdoor environments; 2) constructing a heat and moisture coupling transfer theoretical model based on moisture content driving potential, including establishment of a control equation, processing of a material interface and setting of boundary conditions; 3) performing numerical solution of the theoretical model based on a finite element method, and verifying precision and efficiency; 4) changing or combining the basic information of the building, carrying out multiple rounds of numerical solution, generating a large number of heat and humidity distribution nephograms, and taking the heat and humidity distribution nephograms as a training set; and 5) evaluating and applying the heat and humidity distribution of the building envelope based on the artificial intelligence model. According to the method, real-time prediction of heat and humidity distribution in the enclosure structure in the building operation stage is achieved, and technical support is provided for energy consumption optimization, humidity risk early warning and intelligent regulation and control.

Description

Technical Field

[0001] The present invention relates to a simulation method for the thermal and moisture distribution of building envelopes driven by moisture content and an artificial intelligence prediction method, belonging to the field of building performance simulation and prediction.

Background Art

[0002] According to statistics, the energy consumption and carbon emissions during the building operation stage have reached as high as 70% and 59% respectively, and will continue to increase as China gradually enters a new stage of urbanization. In order to reduce the cooling and heating loads during this stage and further control the energy consumption of heating, ventilation, and air conditioning (HVAC) systems, passive design methods represented by improving the thermal and moisture performance of building envelopes are widely adopted, and the basis for their implementation lies in accurately analyzing the heat and moisture transfer process inside the building envelope during the design stage.

[0003] Solving the heat and moisture coupled transfer process by means of finite element analysis software is the main way for industry insiders to master the temperature and humidity field distribution of building envelopes, and the basis for its implementation lies in accurately establishing the theoretical model. After nearly decades of research, the theoretical model of heat and moisture coupled transfer has been developed by leaps and bounds, and there is a broad consensus on "using temperature as the heat driving potential". However, since moisture transfer includes two parts: gaseous water diffusion and liquid water penetration, involving multiple transport mechanisms, the industry has different opinions on the choice of moisture driving potential.

[0004] Currently, moisture content, relative humidity, and capillary pressure are regarded as three typical and main choices of moisture driving potential, each having specific theoretical bases and application scenarios: 1) Moisture content directly represents the moisture content inside the material, with clear physical meaning and intuitive transfer process; however, its discontinuity at the material interface limits the applicability to multi-layer building envelopes and may cause numerical calculation instability; 2) Relative humidity depends on the equilibrium relationship with moisture content, can intuitively describe the humidity level, and has good interpretability in low-humidity regions; while its gradient change is weak in high-humidity regions, reducing the calculation accuracy, and its non-linear characteristics exacerbate the solving complexity and affect the model stability; 3) Capillary pressure has a large value in low-humidity regions and can accurately depict the liquid water migration process, but it significantly decreases in high-humidity regions, affecting the solving accuracy; while its high calculation cost will reduce the efficiency of large-scale simulations, and its weak ability to intuitively express the moisture level limits its application in condensation risk assessment and mold growth prediction.

[0005] The above characteristics emphasize that choosing moisture content as the moisture driving potential has certain advantages in theory, that is, it can effectively describe the moisture level in porous media, and its linearity is significantly better than other parameters; however, the discontinuity problem at its interface limits the applicability and stability of the model, and innovative solutions are urgently needed to optimize the theoretical model.

[0006] In addition, although existing finite element numerical solution methods can provide high-precision contour maps of the thermal and moisture distribution of building envelopes, there are still problems such as high computing power requirements and long solution times in practical applications. It is difficult to meet the need for rapid prediction during the building operation stage, and it is even more difficult to support the real-time monitoring and active control of intelligent building systems. Although significant progress has been made in building energy consumption prediction and intelligent control of HVAC systems by artificial intelligence in recent years, its research on the prediction of the temperature and humidity field distribution of building envelopes is still relatively limited. The technical system is still in the initial exploration stage, and it is urgent to further expand its modeling and reasoning capabilities in complex thermal and moisture environments.

[0007] Different from building energy consumption, due to the spatial concealment and distribution complexity of the temperature and humidity fields inside building envelopes, it is difficult to achieve comprehensive coverage with traditional measurement methods - existing contact and destructive technologies such as embedded sensor measurement or material sampling analysis can only obtain data at local or discrete points; this limitation not only affects the accurate assessment of the overall thermal and moisture characteristics of building envelopes, but also restricts the training accuracy of data-driven models. In addition, the traditional training mode relying on historical data lacks a deep understanding of the thermal and moisture transfer mechanism. When facing extreme meteorological conditions, material property changes or complex boundary conditions, its generalization ability is insufficient, and the prediction accuracy is easily affected, making it difficult to effectively support the needs of intelligent building operation management.

[0008] Therefore, to solve the above problems, it is indeed necessary to provide an innovative moisture content-driven simulation of the thermal and moisture distribution of building envelopes and an artificial intelligence prediction method, which can break through the deficiency that the thermal and moisture coupling transfer model with moisture content as the driving potential is difficult to be used for multi-layer building envelopes, and realize the real-time prediction of the temperature and humidity distribution inside the building envelope with the help of artificial intelligence technology to overcome the defects in the existing technology.

Summary of the Invention

[0009] The purpose of the present invention is to provide a moisture content-driven simulation of the thermal and moisture distribution of building envelopes and an artificial intelligence prediction method. This method establishes a theoretical model of thermal and moisture coupling transfer with moisture content as the moisture driving potential, and then learns the contour map of the thermal and moisture distribution of the building envelope obtained by numerical solution through an artificial intelligence model with a convolutional long short-term memory network (ConvLSTM) architecture, and finally realizes the real-time prediction of the temperature and humidity distribution inside the building envelope during the operation stage.

[0010] To achieve the above purpose, the technical solution adopted by the present invention is: a moisture content-driven simulation of the thermal and moisture distribution of building envelopes and an artificial intelligence prediction method, which includes the following process steps:

[0011] 1), Obtaining basic data of the building envelope and collecting preliminary information, specifically:

[0012] 1-1) Collect the construction drawings of the target building envelope and clarify the structural dimensions and construction information;

[0013] 1 - 2), Test the thermal and moisture physical properties of building materials to obtain key physical indicators;

[0014] 1 - 3), Collect relevant indoor and outdoor environmental parameters of the building to clarify the environmental characteristics of the building;

[0015] 2), Construct a theoretical model of heat and moisture coupled transfer based on the driving potential of moisture content, specifically:

[0016] 2 - 1), Establish a control equation according to the transfer mechanism of heat and moisture in porous media;

[0017] 2 - 2), Solve the discontinuity problem of moisture content at the interface according to the heat and moisture transfer characteristics at the interface;

[0018] 2 - 3), Analyze the heat and moisture transfer characteristics between the target building and the actual environment and set boundary conditions;

[0019] 3), Numerical solution of the theoretical model based on the finite element method and verification of accuracy and efficiency, specifically:

[0020] 3 - 1), Input the building information in step 1) into the finite element simulation platform to complete the preliminary settings;

[0021] 3 - 2), Mesh the building geometric model in the finite element simulation platform to ensure the calculation accuracy;

[0022] 3 - 3), Apply the theoretical model in step 2), configure the solver, and complete the preliminary preparation for numerical solution;

[0023] 3 - 4), Verify the applicability, accuracy, and efficiency of the theoretical model with the help of benchmark cases;

[0024] 3 - 5), Obtain the numerical simulation results of the heat and moisture distribution of the building envelope under specific working conditions;

[0025] 4), Multiple rounds of numerical solution of the heat and moisture distribution of the building envelope and construction of the training set, specifically:

[0026] 4 - 1), Change or combine the building information in step 1) to create multiple types of cases;

[0027] 4 - 2), Conduct multiple rounds of numerical solution to generate cloud maps of the heat and moisture distribution of the building envelope under multiple working conditions;

[0028] 4 - 3), Perform data pre - processing to construct a training data set for the artificial intelligence model;

[0029] 5), Evaluation and application of the heat and moisture distribution of the building envelope based on the artificial intelligence model, specifically:

[0030] 5-1), Select the convolutional long short-term memory network structure to carry out artificial intelligence model training;

[0031] 5-2), Verify the model prediction performance, optimize the calculation efficiency, and implement model deployment;

[0032] 5-3), Apply the artificial intelligence model to predict the thermal and moisture distribution.

[0033] The method for simulating the thermal and moisture distribution of the building envelope driven by moisture content and artificial intelligence prediction of the present invention is further as follows: In the step 1-1), the obtained construction drawings of the target building envelope include at least: the geometric dimensions, structural layers, and building materials selected for the building envelope;

[0034] In the step 1-2), the obtained key physical indicators include at least: the apparent density ρ under the dry condition, the specific heat capacity at constant pressure c p , the thermal conductivity λ at different moisture contents, the equilibrium relative humidity at different moisture contents the gaseous water permeability coefficient δ at different moisture contents v , the liquid water diffusion coefficient D driven by moisture content ω or the liquid water permeability coefficient K l ;

[0035] In the step 1-3), the collection duration of the relevant indoor and outdoor environmental parameters of the building is at least one year, the collection interval is no longer than 1 hour, and at least includes: the indoor temperature T of the building in , the indoor relative humidity of the building the outdoor environmental temperature T out , the outdoor environmental relative humidity the outdoor environmental wind speed v, the attack angle θ between the outdoor environmental wind and the building envelope, the total radiation I received by the building envelope including the long-wave part and the short-wave part.

[0036] The method for simulating the thermal and moisture distribution of the building envelope driven by moisture content and artificial intelligence prediction of the present invention is further as follows: The control equation in the step 2-1) is specifically:

[0037]

[0038] Among them, ρ is the apparent density under the dry condition, c p is the specific heat capacity at constant pressure, ω is the moisture content in the building envelope, c p,l is the specific heat capacity at constant pressure of liquid water, T is the temperature of the building envelope, λ is the effective thermal conductivity at different moisture contents, h lat is the latent heat of vaporization of water, δ v is the gaseous water permeability coefficient, ξ is the slope of the water vapor saturation pressure-temperature curve, is the relative humidity of the building envelope, psat is the saturated vapor pressure, and is the slope of the moisture content - relative humidity curve.

[0039] The moisture content - driven simulation method for heat and moisture distribution in building envelopes and the artificial intelligence prediction method of the present invention further are: The solution to the problem of discontinuous moisture content at the interface in step 2 - 2) is specifically: Disconnect building materials A and B at their interface to remove the transfer process driven by moisture content at the boundary; Add additional heat flux and moisture flux to the original interfaces of building materials A and B respectively so that the two interfaces have exactly the same temperature and humidity change processes;

[0040] Among them, the additional heat flux q toA for building material A and the additional heat flux q toB for building material B are specifically calculated according to the following formulas respectively: q toA = h sr (T sr,B - T sr,A ), q toB = h sr (T sr,A - T sr,B ), where hsr is the imaginary convective heat transfer coefficient at the interface of building materials, and T sr,A and T sr,B are the temperatures on the interfaces of building material A and building material B respectively;

[0041] The additional moisture flux g toA for building material A and the additional moisture flux g [[ID=�9]] toB for building material B are specifically calculated according to the following formulas respectively: g toA = h m,sr (p v,sr,B - p v,sr,A ), g toB = h m,sr (p v,sr,A - p v,sr,B ), where hm,sr is the imaginary convective mass transfer coefficient at the interface of building materials, and p v,sr,A and p v,sr,B are the partial vapor pressures on the interfaces of building material A and building material B respectively;

[0042] The imaginary convective heat transfer coefficient and convective mass transfer coefficient at the interface of the building materials satisfy: The convective heat transfer and mass transfer resistance between the building materials are less than the heat conduction and moisture transfer resistance inside them, and the values of h sr and h m,sr are respectively greater than the order of magnitude of 10 5 and 10 2 2 2 order of magnitude.

[0043] The simulation method for the thermal and moisture distribution of the building envelope driven by moisture content and the artificial intelligence prediction method further includes: in the step 2-3), the thermal and moisture transfer characteristics between the target building and the actual environment are specifically the moisture transfer and heat transfer processes actually occurring on the surface of the building envelope;

[0044] Among them, the moisture transfer processes on the inner and outer surfaces of the building envelope are respectively: Where g in And g out Are the moisture transfer fluxes on the inner and outer surfaces respectively, h m,in And h m,out Are the convective mass transfer coefficients on the inner and outer sides respectively, p sat,in And p sat,out Are the saturated water vapor pressures of the air on the inner and outer sides respectively, And Are the relative humidities of the air on the inner and outer sides respectively, R WDR Is the amount of rainwater falling on the surface of the building envelope caused by the wind-driven rain phenomenon; p sat Is the water vapor saturation pressure, Is the relative humidity of the building envelope;

[0045] The heat transfer processes on the inner and outer surfaces of the building envelope are respectively: Among them, q in And q out Are the heat transfer fluxes occurring on the inner and outer surfaces respectively, h in And h out Are the convective heat transfer coefficients on the inner and outer sides respectively, T in And T out Are the temperatures of the air on the inner and outer sides respectively, T is the temperature of the building envelope, h lat Is the latent heat of vaporization of water, h m,in And h m,out Are the convective mass transfer coefficients on the inner and outer sides respectively, p sat,in And p sat,out Are the saturated water vapor pressures of the air on the inner and outer sides respectively, And Are the relative humidities of the air on the inner and outer sides respectively, R WDR Is the amount of rainwater falling on the surface of the building envelope caused by the wind-driven rain phenomenon, p sat Is the water vapor saturation pressure, Is the relative humidity of the building envelope, c p,l Is the specific heat capacity at constant pressure of liquid water, T dp Is the dew point temperature of the ambient air, α is the absorptivity of the outdoor surface, and I is the total radiation received by the building envelope.

[0046] The method for simulating the thermal and moisture distribution of the building envelope driven by moisture content and artificial intelligence prediction of the present invention further includes: the finite element simulation platform in step 3-1) at least includes the following functions: supporting the coupled solution of multiple physical fields, having a flexible material property definition function, supporting complex geometric modeling and mesh generation, allowing the setting of different environmental boundary conditions, having an efficient solution algorithm, and supporting data export and post-processing;

[0047] In step 3-2), the mesh divided for the building geometric model at least meets the following conditions: the mesh type adapts to the calculation requirements, the mesh is encrypted at key parts, the mesh quality meets the requirements of numerical stability, the mesh characteristics take into account both calculation accuracy and efficiency, and it is adapted to subsequent finite element analysis;

[0048] The building geometric model is drawn according to the construction drawings of the building envelope in step 1-1);

[0049] In step 3-3), the application of the theoretical model in step 2) includes the following process: input the physical properties parameters of building materials in step 1-2), input the indoor and outdoor environmental parameters in step 1-3), input the control equations and boundary conditions in step 2), and set the initial conditions;

[0050] Configuring the solver at least includes the following process: selecting a suitable numerical method, optimizing the time step, adjusting the mesh adaptability, setting the convergence criterion, and enabling transient solution;

[0051] In step 3-4), the benchmark case is a numerical solution case, an analytical solution case provided by widely recognized standards or reports at home and abroad, or the thermal and moisture transfer process on the surface and inside of the building envelope obtained through experimental monitoring.

[0052] The method for simulating the thermal and moisture distribution of the building envelope driven by moisture content and artificial intelligence prediction of the present invention further includes: in step 4-1), the changes or combinations made to the building information obtained in step 1) are specifically: adjusting the size and structure of the building envelope to achieve changes in the geometric characteristics of the building envelope and increase the applicable range of data; modifying the thermal and moisture physical properties parameters of building materials to achieve changes in the building envelope building material parameters and simulate the influence of different materials on the thermal and moisture distribution; setting different indoor and outdoor environmental parameters to achieve a combination of multiple boundary conditions to cover different climate zones and usage scenarios; adding small-range random perturbations to the building material parameters and boundary conditions to improve the generalization ability of the artificial intelligence model;

[0053] In step 4-2), the heat and moisture distribution cloud map is made to have a linear color gradient change through a standardized color mapping method. The color mapping scheme can be selected from Jet, Viridis, or Turbo according to different visual perception requirements, and the color resolution is 32-bit true color. The upper and lower limit ranges of the temperature distribution cloud map and the relative humidity distribution cloud map can be set as -20 to 60 °C and 20% to 100% respectively, and it is ensured that the color scale mapping is consistent with the numerical ratio. The cloud map resolution matches the physical size of the building envelope to ensure that the resolution per unit length is not less than 1000 px·m -1 , and it is output in a vector format or a high-bit-depth raster format. The cloud map export time interval is no longer than 1 hour and matches the simulation time step;

[0054] In step 4-3), the data preprocessing includes the following steps: data cleaning, data normalization, and data augmentation. The training set construction is based on the division of the training set and the test set, and the division ratio is 80:20.

[0055] The convolutional long short-term memory network structure in step 5-1) of the moisture-driven heat and moisture distribution simulation and artificial intelligence prediction method of the present invention further includes a convolutional layer, a long short-term memory layer, a pooling layer, and a fully connected layer. The number of convolutional layers is not less than 3 layers. After each convolutional layer, batch normalization and a rectified linear unit activation function are added. The convolutional kernel size of the convolutional layer is between 3×3 and 5×5. The pooling layer uses max pooling, the window size is not greater than 2×2, and the stride is not greater than 2. The number of long short-term memory layers is 2 layers, and the number of hidden units is between 64 and 256. The fully connected layer has no less than 2 layers, the number of neurons in each layer is between 128 and 512, and a dropout method is used to reduce the risk of overfitting. The dropout ratio of the dropout method is not less than 0.2 and not higher than 0.5;

[0056] The artificial intelligence model combines an adaptive moment estimation optimizer and a learning rate decay strategy. The learning rate of the adaptive moment estimation optimizer is not greater than 0.0005, the decay rate of the first moment estimation is not less than 0.9, and the decay rate of the second moment estimation is not less than 0.999. The decay rate of the learning rate decay strategy is not greater than 0.1, and the learning rate is dynamically adjusted during training;

[0057] The model training uses a batch size between 16 and 64, the maximum number of training epochs is not less than 50 epochs, and an early stopping strategy with a tolerance of no more than 10 epochs is used to optimize the use of computing resources.

[0058] The method for simulating the thermal and moisture distribution of the building envelope driven by moisture content and the artificial intelligence prediction further includes: the verification of the model prediction ability in step 5-2) is based on the test set data. The thermal and moisture distribution cloud map generated by the artificial intelligence model is compared with the high-precision numerical simulation results pixel by pixel, and the image similarity index is calculated. The image similarity index specifically includes the structural similarity index, the peak signal-to-noise ratio, and the mean square gradient error. When they are not less than 0.85, not less than 30 dB, and not more than 0.02 respectively, it is considered that the difference between the model prediction result and the standard numerical solution is within an acceptable range.

[0059] The calculation efficiency of the optimized model is achieved through model pruning and model quantization. The model pruning adopts weight pruning or structural pruning to balance storage optimization and calculation acceleration, and ensures that the accuracy loss of the pruned model does not exceed 2%. The model quantization adopts static quantization and dynamic quantization, and selects the optimal strategy according to the hardware characteristics to quantize the model parameters from FP32 to INT8, ensuring that the accuracy loss after quantization does not exceed 1%.

[0060] The method for simulating the thermal and moisture distribution of the building envelope driven by moisture content and the artificial intelligence prediction of the present invention further includes: the thermal and moisture distribution prediction in step 5-3) is based on the actual measurement data, dynamically calculates the temperature and humidity fields inside the building envelope, and predicts the distribution conditions at future times. The actual measurement data includes indoor and outdoor environmental parameters. The dynamic calculation is based on the historical and real-time acquired data, and combines the time series analysis method to calculate the thermal and moisture distribution cloud map at the next moment or multiple future time steps.

[0061] Compared with the prior art, the present invention has the following beneficial effects:

[0062] 1. The method for simulating the thermal and moisture distribution of the building envelope driven by moisture content and the artificial intelligence prediction of the present invention combines the physical model and data driving, balances interpretability and calculation efficiency, effectively improves the real-time performance of prediction, and can be applied to multiple fields such as building performance analysis, moisture risk warning, and intelligent control.

[0063] 2. The method for simulating the thermal and moisture distribution of the building envelope driven by moisture content and the artificial intelligence prediction of the present invention uses moisture content as the moisture driving potential. While improving the numerical solution efficiency and accuracy, it combines the interface treatment method to overcome the discontinuity problem of moisture content and enhances the calculation applicability of multi-layer building envelopes.

[0064] 3. The method for simulating the thermal and moisture distribution of the building envelope driven by moisture content and the artificial intelligence prediction of the present invention uses the finite element simulation to generate cloud maps to train the artificial intelligence model, without relying on measured data, avoiding the insufficient generalization ability caused by limited measurement data, and improving the adaptability of the model to different building envelopes and meteorological conditions.

[0065] 4. The moisture content-driven simulation of the thermal and moisture distribution of building envelopes and the artificial intelligence prediction method of the present invention use the structural similarity index, peak signal-to-noise ratio, and mean square gradient error to evaluate the prediction cloud map and the reference cloud map, ensuring numerical accuracy and spatial distribution consistency, and avoiding the problem that the traditional point-to-point error evaluation ignores local deviations.

Description of the Drawings

[0066] Figure 1 is a flowchart of the moisture content-driven simulation of the thermal and moisture distribution of building envelopes and the artificial intelligence prediction method of the present invention.

[0067] Figure 2 is a detailed drawing of a typical node of the building envelope of a nearly zero-energy building in the hot summer and cold winter regions in an embodiment of the present invention.

[0068] Figure 3 is a schematic diagram of various environmental parameters outside the building in an embodiment of the present invention.

[0069] Figure 4 is a schematic diagram of the moisture transfer process at the interface of building materials when the driving potential is the moisture content in an embodiment of the present invention.

[0070] Figure 5 is a schematic diagram of the solution to the problem of discontinuous moisture content at the interface in an embodiment of the present invention.

[0071] Figure 6 is the error comparison between the numerical simulation results and the reference case in an embodiment of the present invention.

[0072] Figure 7 is the numerical simulation result of the thermal and moisture distribution of the building envelope under specific working conditions in an embodiment of the present invention.

Detailed Embodiments

[0073] Please refer to the attached drawings of the specification Figure 1 As shown, the present invention is a moisture content-driven simulation of the thermal and moisture distribution of building envelopes and an artificial intelligence prediction method, which includes the following process steps:

[0074] 1), Obtaining the basic data of the building envelope and collecting the preliminary information, specifically:

[0075] 1-1), Collecting the construction drawings of the target building envelope and clarifying the structural dimensions and construction information. Among them, the construction drawings of the target building envelope should at least include the geometric dimensions, construction layers, and building materials selected.

[0076] In the embodiment of the present invention, the detailed drawing of a typical node of the building envelope of a nearly zero-energy building in the hot summer and cold winter regions that meets the "Technical Standard for Nearly Zero-Energy Buildings" is as Figure 2 shown, which includes information such as geometric dimensions, construction layers, and materials used.

[0077] 1-2), Test the thermo-hygroscopic physical properties of building materials to obtain key physical indicators. Among them, the key physical indicators should at least include: the apparent density (ρ, kg·m -3 ) under the dry condition, the specific heat capacity at constant pressure (c p , J·kg -1 ·K -1 ) under the dry condition, the thermal conductivity (λ, W·m -1 ·K -1 ) at different moisture contents, the equilibrium relative humidity at different moisture contents The gaseous water permeability coefficient (δ v , kg·m -1 ·s -1 ·Pa -1 ) at different moisture contents, the liquid water diffusion coefficient (D ω , m 2 ·s -1 ) or the liquid water permeability coefficient (K l , kg·m -1 ·s -1 ·Pa -1 ) driven by moisture content.

[0078] In the embodiments of the present invention, the key physical indicators of each building material are shown in Table 1 below.

[0079]

[0080] 1-3), Collect relevant indoor and outdoor environmental parameters of the building to clarify the environmental characteristics of the building. Specifically, the collection duration of the relevant indoor and outdoor environmental parameters of the building is at least one year, the collection interval is no longer than 1 hour, and it at least includes: the indoor temperature of the building (T in , K), the indoor relative humidity of the building The outdoor environmental temperature (T out , K), the outdoor environmental relative humidity The outdoor environmental wind speed (v, m·s -1 ), the attack angle (θ, °) between the outdoor environmental wind and the enclosure structure, the total radiation (I, W·m -2 ) received by the enclosure structure including the long-wave part and the short-wave part.

[0081] In the embodiments of the present invention, the environmental parameters of each building outdoors are as Figure 3As shown, the indoor temperature and humidity are generated according to the following method: the attenuation coefficient of the indoor and outdoor air temperature and water vapor partial pressure is 1.8, and the delay time is 5 hours; when the 24-hour average temperature of the outdoor air is higher than 26°C or lower than 8°C, the indoor cooling or heating system is operated, and the operation time of the above system is limited to 20:00-8:00; when the cooling or heating system is started, the set relative humidity in the room is 70%, and the set temperature is 26°C and 18°C, respectively. Therefore, the indoor water vapor partial pressure is 2351Pa and 1445Pa, respectively.

[0082] 2) Construct a theoretical model of heat and moisture coupling transfer based on the moisture content driving potential, which is as follows:

[0083] 2-1), establish the control equation based on the heat and moisture transfer mechanism in porous media.

[0084] Among them, in the process of constructing the theoretical model of heat and moisture coupling transfer of envelope structures, the following assumptions are made: liquid water and gaseous water in building materials are continuous media, and volume changes caused by pressure changes are ignored; gaseous water is an ideal gas, and intermolecular forces and molecular volume are ignored; the temperature of the envelope structure is usually higher than 0°C, and the phase change between liquid or gaseous water and solid water is not considered; the moisture content characteristics of porous materials do not change with temperature; the hysteresis effect of moisture absorption or desorption is ignored, and the average value of the equilibrium moisture absorption and desorption curve is used in the calculation process; there is no thermal resistance and moisture resistance at the interface between different building materials; chemical reactions in building materials, such as concrete hydration or mold corrosion, are ignored. The transfer processes of heat, gaseous water, and liquid water in the porous medium of the envelope structure follow Fourier's law, Darcy's law, and Fick's law, respectively, that is, they satisfy the following expressions: Among them, q and g v 、g l are energy, gaseous water mass, and liquid water mass diffusion fluxes (W·m -2 , kg·m -2 ·s -1 , kg·m -2 ·s -1 ), T, p v 、p c are the temperature (K), water vapor partial pressure (Pa), and capillary pressure (Pa) of the enclosure structure. The moisture transfer governing equation with moisture content as the driving force and the heat transfer governing equation with temperature as the driving force can be derived from the above equations, as shown below: Where ω is the moisture content in the enclosure (kg·m -3 ), ξ is the slope of the water vapor saturation pressure-temperature curve (Pa·K -1 ), p sat is the water vapor saturation pressure (Pa), ζ is the slope of the moisture content-relative humidity curve (kg·m-3 ), c p,l is the specific heat capacity of liquid water at constant pressure (J·kg -1 ·K -1 );h lat is the latent heat of evaporation of water (J·kg -3 ).

[0085] The control equation of the heat and moisture coupled transfer theoretical model based on the moisture content driving potential can be obtained by combining the moisture transfer and heat transfer control equations, as shown in the following equation:

[0086] Where ρ is the apparent density under absolute dry conditions, c p is the specific heat capacity at constant pressure, ω is the moisture content in the enclosure structure, c p,l is the constant pressure specific heat capacity of liquid water, T is the temperature of the enclosure structure, λ is the effective thermal conductivity under different moisture contents, h lat is the latent heat of evaporation of water, δ v is the gaseous water permeability coefficient, ξ is the slope of the water vapor saturation pressure-temperature curve, is the relative humidity of the enclosure structure, p sat is the water vapor saturation pressure, is the slope of the moisture content-relative humidity curve.

[0087] 2-2): Aiming at the heat and moisture transfer characteristics of the interface, solve the problem of discontinuity of moisture content at the interface. Without considering the thermal resistance and moisture resistance, the two porous media surfaces in contact have completely consistent thermal and moisture change processes. Specifically, the changes in parameters such as the two surface temperatures, relative humidity, water vapor partial pressure, and capillary pressure are completely consistent. The problem of discontinuity of moisture content at the interface can be illustrated by the following example: In an adiabatic and insulated system, the air and the building materials A and B in contact with each other have the same and uniform temperature and relative humidity, so the entire system (including gaseous water diffusion and liquid water penetration) is in a dynamic equilibrium state, and the thermal and moisture parameters remain unchanged. However, if Figure 4 As shown in the figure, if the moisture content of building material A is greater than that of building material B at the same relative humidity, the control equation with moisture content as the driving force will cause moisture to be transferred from building material A to building material B, causing the water vapor partial pressure inside the two to decrease and increase respectively (process I); then, the gaseous water diffusion balance between the building materials and the air is broken, resulting in a water cycle transfer (process II), which violates the second law of thermodynamics.

[0088] Specifically, the solution to the moisture discontinuity problem at the interface is to disconnect building materials A and B at the interface to remove the transfer process driven by moisture content on the boundary; add additional heat flux and moisture flux to the original interface between building materials A and B, so that the two interfaces have completely consistent temperature and humidity change processes (see Figure 5)。The additional heat fluxes (q toA and q toB , W·m -2 ) for building materials A and B can be calculated respectively by the following formulas: q toA = h sr (T sr,B - T sr,A ), q toB = h sr (T sr,A - T sr,B ), where hsr is the hypothetical convective heat transfer coefficient at the interface of building materials (W·m -2 ·K -1 ), T sr,A and T sr,B are the temperatures (K) at the interfaces of building materials A and B respectively; the additional moisture fluxes (g oA and g toB , kg·m -2 ·s -1 ) for building materials A and B can be calculated respectively by the following formulas: g toA = h m,sr (p v,sr,B - p v,sr,A ), g toB = h m,sr (p v,sr,A - p v,sr,B ), where hm,sr is the hypothetical convective mass transfer coefficient at the interface of building materials (kg·m -2 ·s -1 ·Pa -1 ), p v,sr,A and p v,sr,B are the partial pressures of water vapor (Pa) at the interfaces of building materials A and B respectively. The hypothetical convective heat transfer coefficient and convective mass transfer coefficient at the interface of building materials should satisfy the following characteristics: the convective heat transfer and mass transfer resistances between building materials are much smaller than the heat conduction and moisture transfer resistances inside them. Generally speaking, setting the values of h sr and h m,sr to be greater than the order of magnitude of 10 5 and 10 2 respectively can meet this requirement.

[0089] 2 - 3): Analyze the heat and moisture transfer characteristics between the target building and the actual environment, and set the boundary conditions. The heat and moisture transfer characteristics between the target building and the actual environment are the heat and moisture transfer processes actually occurring on the surface of the enclosure structure. The moisture transfer process includes convective mass transfer driven by the difference in partial pressure of water vapor and liquid water transfer caused by wind-driven rain. The moisture transfer processes on the inner and outer surfaces of the enclosure structure are: where g in and g out are the moisture transfer fluxes (kg·m-2 ·s -1 ),h m,in and h m,out are the convective mass transfer coefficients on the inner and outer sides respectively (kg·m -2 ·s -1 ·Pa -1 ), p sat,in and p sat,out are the saturated water vapor pressures of the air on the inner and outer sides respectively (Pa), and are the relative humidities of the air on the inner and outer sides respectively, R WDR is the amount of rainwater falling on the surface of the building envelope caused by the wind-driven rain phenomenon, kg·m -2 ·s -1 . The heat transfer process includes convective heat transfer driven by temperature difference, latent heat released or absorbed during the adsorption or desorption process of gaseous water, sensible heat of the rainwater absorbed by the surface of the building envelope along with the wind-driven rain, and heat exchange caused by radiation. The heat transfer process on the outer and inner surfaces of the building envelope is as follows:

[0090] where q in and q out are the heat transfer fluxes occurring on the inner and outer surfaces respectively, h in and h out are the convective heat transfer coefficients on the inner and outer sides respectively, T in and T out are the temperatures of the air on the inner and outer sides respectively, T is the temperature of the building envelope, h lat is the latent heat of vaporization of water, h m,in and h m,out are the convective mass transfer coefficients on the inner and outer sides respectively, p sat,in and p sat,out are the saturated water vapor pressures of the air on the inner and outer sides respectively, and are the relative humidities of the air on the inner and outer sides respectively, R WDR is the amount of rainwater falling on the surface of the building envelope caused by the wind-driven rain phenomenon, p sat is the saturated water vapor pressure, is the relative humidity of the building envelope, c p,l is the specific heat capacity at constant pressure of liquid water, T dp is the dew point temperature of the ambient air, α is the absorptivity of the outdoor surface, and I is the total radiation received by the building envelope.

[0091] 3), numerical solution of the theoretical model based on the finite element method and verification of its accuracy and efficiency, specifically:

[0092] 3-1): Input the building information in step 1) into the finite element simulation platform and complete the preliminary settings. Among them, the finite element simulation platform should at least support multi-physics field coupled solution, have a flexible material property definition function, support complex geometric modeling and mesh generation, allow setting different environmental boundary conditions, have an efficient solution algorithm, and support data export and post-processing.

[0093] In the embodiment of the present invention, the finite element simulation platform used is the commercial software COMSOL Multiphysics.

[0094] 3-2), Generate a mesh for the building geometric model in the finite element simulation platform to ensure the calculation accuracy. Among them, the mesh should at least meet the following conditions: the mesh type adapts to the calculation requirements, the mesh is refined at key parts, the mesh quality meets the requirements of numerical stability, the mesh characteristics balance calculation accuracy and efficiency, and it is adapted to subsequent finite element analysis.

[0095] 3-3): Apply the theoretical model in step 2), configure the solver, and complete the preliminary preparation for numerical solution. That is, input the physical properties of building materials in step 1-2), input the indoor and outdoor environmental parameters in step 1-3), input the control equations and boundary conditions in step 2), set the initial conditions, select a suitable numerical method, optimize the time step, adjust the mesh adaptability, set the convergence criterion, and enable transient solution.

[0096] 3-4), Verify the applicability, accuracy, and efficiency of the theoretical model with the help of a benchmark case. Among them, the benchmark case should be a numerical solution case, an analytical solution case provided by widely recognized standards or reports at home and abroad, or a heat and moisture transfer process on the surface and inside of the enclosure structure obtained through experimental monitoring (experimental case). Based on the same model settings, when the theoretical model disclosed in the present invention is used for numerical simulation, if the solution results show that the theoretical model is applicable under various meteorological conditions throughout the year, its applicability can be proved; if the solution results are consistent with the benchmark case within the error range provided by the benchmark case, its accuracy can be proved; if the solution time is significantly shorter than that of other similar models, its efficiency can be proved.

[0097] In the embodiment of the present invention, an analytical solution case from the British standard "Thermal and Moisture Performance of Building Components - Numerical Simulation for the Assessment of Moisture Transfer" BS EN 15026-2007 is used for verification and compared with a theoretical model with relative humidity and capillary pressure as the moisture driving potential. The comparison between the error ranges provided by the benchmark case and the numerical simulation results is shown in Table 2 below. As can be seen from the table, at different sampling time nodes, all numerical simulation results fall within the error range provided by the benchmark case, indicating that the choice of moisture driving potential has little impact on the calculation results of this verification case and all have relatively good accuracy.

[0098]

[0099] In an embodiment of the present invention, the 1st case (numerical solution case) from the report of the HAMSTAD project (Determination of moisture transfer characteristics in porous building materials and establishment of numerical evaluation methods) is also used for verification. The comparison between the numerical simulation results and the error range provided by this benchmark case is as shown in Figure 6 (a) to (d) below. As can be seen from the figure, under the same mesh division and solver configuration, the numerical simulation can consistently fall within the error range provided by the benchmark case at different sampling time periods and positions only when the moisture content is used as the moisture driving potential, indicating its better accuracy. In addition, considering that the simulation period of this case covers various meteorological conditions throughout the year, the applicability of the model disclosed in the present invention is also verified.

[0100] In an embodiment of the present invention, the mesh division in step 3-2) and the solver configuration in step 3-3) are further adjusted, and the solution efficiency is verified by comparing the time used for each numerical simulation under different settings. The specific situation is shown in Table 3 below. As can be seen from the table, when the number of meshes is small, the highest solution efficiency can be obtained by using relative humidity as the moisture driving potential in the numerical solution process; while when the number of meshes is large, the highest solution efficiency can be obtained by using moisture content as the moisture driving potential. Thus, the better efficiency of the model disclosed in the present invention when the number of meshes is large is also verified.

[0101]

[0102] In an embodiment of the present invention, the numerical simulation results of the thermal and moisture distribution of the building envelope under specific working conditions obtained through step 3-5) are as shown in Figure 7 below, which respectively show the temperature and humidity distribution at 0:00 on January 1st, April 1st, July 1st, and October 1st.

[0103] 4), multiple rounds of numerical solution of the thermal and moisture distribution of the building envelope and construction of the training set, specifically:

[0104] 4-1): Change or combine the building information in step 1) to create multiple types of calculation examples. That is, the changes or combinations made to the building information obtained in step 1) should include the following aspects: Adjust the size, structure, etc. of the building envelope to achieve changes in the geometric characteristics of the building envelope and increase the applicable range of the data; Modify the thermal and moisture physical properties parameters of the building materials to achieve changes in the building envelope material parameters and simulate the influence of different materials on the thermal and moisture distribution; Set different indoor and outdoor environmental parameters to achieve combinations of multiple boundary conditions to cover different climate zones and usage scenarios; Add small-range random perturbations to the building material parameters and boundary conditions to improve the generalization ability of the artificial intelligence model.

[0105] 4-2): Conduct multiple rounds of numerical solutions to generate heat and moisture distribution contour maps of the enclosure structure under multiple working conditions. The heat and moisture distribution contour maps should show a linear color gradient through a standardized color mapping method. Color mapping schemes such as Viridis and Turbo can be selected according to different visual perception requirements, and the color resolution is 32-bit true color; the upper and lower limit ranges of the temperature distribution contour map and the relative humidity distribution contour map can be set as -20 to 60 °C and 20% to 100% respectively, and ensure that the color scale mapping is consistent with the numerical ratio; the contour map resolution should match the physical size of the enclosure structure and be output in vector format or high-depth raster format; the time interval for exporting the contour map is no longer than 1 hour and matches the simulation time step.

[0106] 4-3): Perform data preprocessing to construct a training dataset for the artificial intelligence model. Among them, data preprocessing includes data cleaning, data normalization, and data augmentation. Data cleaning is for the contour map data, using outlier removal and image filtering methods based on physical constraints to remove abnormal pixel points and data that do not conform to the characteristics of building heat and moisture transfer, ensuring the integrity and physical consistency of the input data; data normalization uses the Min-Max normalization method to normalize the temperature and humidity channels in the contour map respectively, so that the data maintains a consistent numerical range, improving the stability of model training and the efficiency of gradient optimization; data augmentation is based on image processing methods, including random perturbation, rotation, scaling, etc., to improve the generalization ability of the model to diverse working conditions. The training set construction is based on the division of the training set and the test set, and the division ratio is 80:20, ensuring that the coverage of the training set and the test set in terms of time, space, and environmental conditions is balanced, avoiding data leakage, and effectively verifying the prediction accuracy and robustness of the model.

[0107] 5), Evaluation and application of the heat and moisture distribution of the building enclosure structure based on the artificial intelligence model, specifically:

[0108] 5-1), Select a convolutional long short-term memory network structure to conduct artificial intelligence model training. Among them, the convolutional long short-term memory network should include a convolutional layer, a long short-term memory layer, a pooling layer, and a fully connected layer; the number of convolutional layers should be no less than 5 layers; after each convolutional layer, batch normalization and a rectified linear unit activation function should be attached; the convolutional kernel size of the convolutional layer should be between 3×3 and 5×5; the pooling layer should use max pooling, with a window size not greater than 2×2 and a stride not greater than 2; the number of long short-term memory layers is 2 layers, and the number of hidden units is between 64 and 256; the fully connected layer should have no less than 2 layers, and the number of neurons in each layer should be between 128 and 512, and the dropout method should be used to reduce the risk of overfitting; the dropout ratio of the dropout method is not less than 0.2 and not higher than 0.5. The artificial intelligence model should combine an adaptive moment estimation optimizer and a learning rate decay strategy to improve training stability; the learning rate of the adaptive moment estimation optimizer is not greater than 0.0005, the decay rate of the first moment estimation is not less than 0.9, and the decay rate of the second moment estimation is not less than 0.999; the decay rate of the learning rate decay strategy is not greater than 0.1, and the learning rate can be dynamically adjusted during training to prevent unstable convergence or gradient vanishing. The artificial intelligence model should use a loss function to ensure that the prediction result is numerically accurate while maintaining the spatial consistency of the heat and moisture distribution cloud map. The model training should use a batch size between 16 and 64, the maximum number of training epochs should be no less than 50 epochs, and an early stopping strategy with a tolerance of no more than 10 epochs should be used to optimize the use of computing resources.

[0109] 5-2), Verify the model prediction performance, optimize the computing efficiency, and implement model deployment. Among them, the verification of the model prediction ability should be based on the test set data. Compare the heat and moisture distribution cloud map generated by the artificial intelligence model with the high-precision numerical simulation results pixel by pixel, and calculate the image similarity index; when the structural similarity index is not less than 0.85, the peak signal-to-noise ratio is not less than 30 dB, and the mean square gradient error does not exceed 0.02, it is considered that the difference between the model prediction result and the standard numerical solution is within an acceptable range. Optimize the model computing efficiency through model pruning and model quantization; model pruning can use weight pruning or structural pruning to balance storage optimization and computing acceleration, and ensure that the accuracy loss of the pruned model does not exceed 2%; model quantization can use static quantization and dynamic quantization, and select the optimal strategy according to the hardware characteristics to quantize the model parameters from FP32 to INT8 or lower bits, reduce the computing complexity, and ensure that the accuracy loss after quantization does not exceed 1%. The model deployment should be compatible with TensorFlow Lite, ONNX, or other lightweight deep learning frameworks, support efficient operation on edge devices, cloud computing platforms, or embedded systems, and adapt to hardware acceleration of tensor processing units or neural network processing units; the deployment plan should support batch inference and streaming inference to meet the computing requirements of different application scenarios and improve the real-time performance and computing efficiency of building intelligent management.

[0110] In an embodiment of the present invention, a total of 17,520 cloud maps of the thermal and moisture distribution of the envelope structure are generated using the deployed artificial intelligence model, and are compared one by one with the high-precision numerical simulation results in the test set. The comparison results show that the annual average values of the adopted evaluation indicators, namely, the structural similarity index, the peak signal-to-noise ratio, and the mean square gradient error, are 0.97, 34.4 dB, and 0.01 respectively, and the passing rates reach 97.6%, 96.2%, and 95.8% respectively, indicating that the model proposed by the present invention has good prediction accuracy and stability, and can meet the efficient evaluation requirements for the thermal and moisture distribution of the envelope structure during the building operation stage.

[0111] 5-3): Apply the artificial intelligence model to perform thermal and moisture distribution prediction. Among them, the thermal and moisture distribution prediction refers to dynamically calculating the temperature and humidity fields inside the envelope structure based on actual measurement data, and predicting the distribution conditions at future moments; the actual measurement data includes indoor and outdoor environmental parameters; the dynamic calculation should be based on historical and real-time acquired data, and combined with time series analysis methods, to calculate the cloud maps of the thermal and moisture distribution at the next moment or multiple future time steps.

[0112] In summary, the moisture content-driven thermal and moisture distribution simulation and artificial intelligence prediction method of the present invention improves the accuracy and efficiency of numerical solution by optimizing the theoretical model, and introduces a data-driven method. On the basis of maintaining the clarity of physical constraints and the heat and moisture transfer mechanism, it breaks through the limitations of large computing power requirements and low operation efficiency of conventional methods, and finally realizes the real-time prediction of the thermal and moisture distribution inside the envelope structure during the building operation stage, providing technical support for energy consumption optimization, moisture risk warning, and intelligent control.

[0113] The above specific implementation manners are only the preferred embodiments of this creation, and are not used to limit this creation. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of this creation shall be included within the protection scope of this creation.

Claims

1. Moisture content-driven simulation of heat and moisture distribution in building envelopes and artificial intelligence prediction method, characterized in that: It includes the following technological steps: 1), Obtaining basic data of the building envelope structure and collecting preliminary information, specifically: 1-1) Collecting construction drawings of the target building envelope structure to clarify the structural dimensions and construction information; 1-2), Testing the thermo-hygroscopic physical property parameters of building materials to obtain key physical indicators; 1-3), Collecting relevant indoor and outdoor environmental parameters of the building to clarify the environmental characteristics of the building; 2), Constructing a theoretical model of thermo-hygroscopic coupled transfer based on the driving potential of moisture content, specifically: 2-1), Establishing a control equation according to the transfer mechanism of heat and moisture in porous media; 2-2), Solving the discontinuous problem of moisture content at the interface according to the thermo-hygroscopic transfer characteristics at the interface; 2-3), Analyzing the thermo-hygroscopic transfer characteristics between the target building and the actual environment and setting boundary conditions; 3), Numerical solution of the theoretical model based on the finite element method and verification of accuracy and efficiency, specifically: 3-1), Inputting the building information in step 1) into the finite element simulation platform to complete the preliminary settings; 3-2), Conducting mesh division on the building geometric model in the finite element simulation platform to ensure calculation accuracy; 3-3), Applying the theoretical model in step 2) and configuring the solver to complete the preliminary preparation for numerical solution; 3-4), Verifying the applicability, accuracy, and efficiency of the theoretical model with the help of benchmark cases; 3-5), Obtaining the numerical simulation results of the thermo-hygroscopic distribution of the envelope structure under specific working conditions; 4), Multiple rounds of numerical solution of the thermo-hygroscopic distribution of the envelope structure and construction of the training set, specifically: 4-1), Changing or combining the building information in step 1) to create multiple types of calculation examples; 4-2), Conducting multiple rounds of numerical solution to generate cloud maps of the thermo-hygroscopic distribution of the envelope structure under multiple working conditions; 4-3), Conducting data preprocessing to construct a training data set for the artificial intelligence model; 5), Evaluation and application of the thermo-hygroscopic distribution of the building envelope structure based on the artificial intelligence model, specifically: 5-1), Selecting the convolutional long short-term memory network structure to conduct training of the artificial intelligence model; 5-2), Verifying the prediction performance of the model, optimizing the calculation efficiency, and realizing model deployment; 5-3), Applying the artificial intelligence model to conduct thermo-hygroscopic distribution prediction.

2. The moisture content-driven simulation method for heat and moisture distribution in building envelopes and the artificial intelligence prediction method according to claim 1, characterized in that: In the above step 1-1), the obtained construction drawings of the target building envelope structure at least include: the geometric dimensions, construction layers, and selected building materials of the envelope structure; In the step 1-2), the obtained key physical indexes at least include: the apparent density ρ under the dry condition and the specific heat capacity at constant pressure c under the dry condition p , the thermal conductivity λ at different moisture contents and the equilibrium relative humidity at different moisture contents , the gaseous water permeability coefficient δ at different moisture contents v , the liquid water diffusion coefficient D with the moisture content as the driving potential ω or the liquid water permeability coefficient K l ; In the said step 1 - 3), the acquisition duration of relevant indoor and outdoor environmental parameters of the building is at least one year, the acquisition interval is no longer than 1 hour, and it includes at least: indoor temperature T of the building in , indoor relative humidity of the building outdoor environmental temperature T out , outdoor environmental relative humidity outdoor environmental wind speed v, the angle of attack θ between the outdoor wind and the building envelope, and the total radiation I received by the building envelope including the long - wave part and the short - wave part.

3. The moisture content-driven simulation method for thermal and moisture distribution of building envelopes and the artificial intelligence prediction method according to claim 1, characterized in that: The control equation in the above step 2-1) is specifically: where ρ is the apparent density under the dry condition, c p is the specific heat capacity at constant pressure, ω is the moisture content in the building envelope, c p,l is the specific heat capacity at constant pressure of liquid water, T is the temperature of the building envelope, λ is the effective thermal conductivity at different moisture contents, h lat is the latent heat of vaporization of water, δ v is the permeability coefficient of gaseous water, ξ is the slope of the water vapor saturation pressure-temperature curve, is the relative humidity of the building envelope, p sat is the water vapor saturation pressure, is the slope of the moisture content-relative humidity curve.

4. The moisture content-driven simulation method for heat and moisture distribution in building envelopes and the artificial intelligence prediction method according to claim 1, characterized in that: The solution to the discontinuous problem of moisture content at the interface in the above step 2-2) is specifically: Disconnecting building materials A and B at their interface to remove the transfer process driven by moisture content at the boundary; Adding additional heat flux and moisture flux to the original interfaces of building materials A and B respectively to make the temperature and humidity change processes of the two interfaces exactly the same; Among them, the additional heat flux q of building material A toA and the additional heat flux q of building material B toB are specifically calculated according to the following formulas respectively: q toA = h sr (T sr,B - T sr,A ), q toB = h sr (T sr,A - T sr,B ), where h sr is the imaginary convective heat transfer coefficient at the interface of building materials, and T sr,A and T sr,B are the temperatures on the interfaces of building materials A and B respectively; The additional moisture flux g for building material A toA and the additional moisture flux g for building material B toB are specifically calculated according to the following formulas: g toA = h m,sr (p v,sr,B - p v,sr,A ), g toB = h m,sr (p v,sr,A - p v,sr,B ), where h m,sr is the imaginary convective mass transfer coefficient at the building material interface, and p v,sr,A and p v,sr,B are the partial pressures of water vapor on the interfaces of building materials A and B, respectively; The hypothetical convective heat transfer coefficient and convective mass transfer coefficient at the building material interface satisfy that the convective heat transfer and mass transfer resistance between building materials are less than the heat conduction and moisture transfer resistance inside them, h sr and h m,sr are respectively greater than 10 5 and 10 2 orders of magnitude.

5. The moisture content-driven simulation method for thermal and moisture distribution in building envelopes and the artificial intelligence prediction method according to claim 1, characterized in that: In the above step 2-3), the thermo-hygroscopic transfer characteristics between the target building and the actual environment are specifically the actual wet transfer and heat transfer processes occurring on the surface of the envelope structure; Among them, the wet transfer processes on the inner and outer surfaces of the envelope structure are respectively: where g in and g out are the wet transfer fluxes on the inner and outer surfaces respectively, h m,in and h m,out are the convective mass transfer coefficients on the inner and outer sides respectively, p sat,in and p sat,out are the saturated water vapor pressures of the air on the inner and outer sides respectively, and are the relative humidities of the air on the inner and outer sides respectively, R WDR is the amount of rain falling on the surface of the building envelope caused by the wind-driven rain phenomenon; p sat is the water vapor saturation pressure, is the relative humidity of the building envelope; The heat transfer processes on the inner and outer surfaces of the enclosure structure are as follows: Among them, q in and q out are the heat transfer fluxes occurring on the inner and outer surfaces respectively, h in and h out are the convective heat transfer coefficients on the inner and outer sides respectively, T in and T out are the temperatures of the air on the inner and outer sides respectively, T is the temperature of the enclosure structure, h lat is the latent heat of vaporization of water, h m,in and h m,out are the convective mass transfer coefficients on the inner and outer sides respectively, p sat,in and p sat,out are the saturated water vapor pressures of the air on the inner and outer sides respectively, and are the relative humidities of the air on the inner and outer sides respectively, R WDR is the amount of rainwater falling on the surface of the enclosure structure caused by the wind-driven rain phenomenon, p sat is the saturated water vapor pressure, is the relative humidity of the enclosure structure, c p,l is the specific heat capacity at constant pressure of liquid water, T dp is the dew point temperature of the ambient air, α is the absorptivity of the outdoor surface, and I is the total radiation received by the enclosure structure.

6. The moisture content-driven simulation method for heat and moisture distribution in building envelopes and the artificial intelligence prediction method according to claim 1, characterized in that: The finite element simulation platform in step 3-1) at least includes the following functions: supporting multi-physics field coupled solution, having a flexible material property definition function, supporting complex geometric modeling and mesh generation, allowing setting of different environmental boundary conditions, having an efficient solution algorithm, and supporting data export and post-processing; In step 3-2), the mesh divided for the building geometric model at least meets the following conditions: the mesh type adapts to the calculation requirements, the mesh of key parts is encrypted, the mesh quality meets the requirements of numerical stability, the mesh characteristics take into account both calculation accuracy and efficiency, and it is adapted to subsequent finite element analysis; The building geometric model is drawn according to the construction drawings of the building envelope structure in step 1-1); In step 3-3), the application of the theoretical model in step 2) includes the following processes: inputting the physical property parameters of building materials in step 1-2), inputting the indoor and outdoor environmental parameters in step 1-3), inputting the control equations and boundary conditions in step 2), and setting the initial conditions; Configuring the solver at least includes the following processes: selecting a suitable numerical method, optimizing the time step, adjusting the mesh adaptability, setting the convergence criterion, and enabling transient solution; In step 3-4), the benchmark case is a numerical solution case, an analytical solution case provided by widely recognized standards or reports at home and abroad, or the heat and moisture transfer process on the surface and inside of the envelope structure obtained through experimental monitoring.

7. The moisture content-driven simulation method for heat and moisture distribution in building envelopes and the artificial intelligence prediction method according to claim 1, characterized in that: In step 4-1), the changes or combinations made to the building information obtained in step 1) are specifically as follows: adjusting the size and structure of the envelope structure to achieve changes in the geometric characteristics of the envelope structure and increase the applicable range of data; modifying the thermal and moisture physical property parameters of building materials to achieve changes in the building material parameters of the envelope structure and simulate the influence of different materials on the heat and moisture distribution; setting different indoor and outdoor environmental parameters to achieve a combination of various boundary conditions to cover different climate zones and usage scenarios; adding small-range random perturbations to the building material parameters and boundary conditions to improve the generalization ability of the artificial intelligence model; In step 4-2), the heat and moisture distribution cloud map is made to have a linear color gradient change through a standardized color mapping method. The color mapping scheme can be selected as Jet, Viridis, or Turbo according to different visual perception requirements, and the color resolution is 32-bit true color; the upper and lower limit ranges of the temperature distribution cloud map and the relative humidity distribution cloud map can be set as -20 to 60 °C and 20% to 100% respectively, and it is ensured that the color scale mapping is consistent with the numerical ratio; the cloud map resolution matches the physical size of the building envelope to ensure that the resolution per unit length is not less than 1000 px·m –1 , and it is output in a vector format or a high-bit-depth raster format; the cloud map export time interval is no longer than 1 hour and matches the simulation time step; In step 4-3), data preprocessing includes the following steps: data cleaning, data normalization, and data augmentation; the training set construction is based on the division of the training set and the test set, and the division ratio is 80:

20.

8. The moisture content-driven simulation method for thermal and moisture distribution in building envelopes and the artificial intelligence prediction method according to claim 1, characterized in that: The convolutional long short-term memory network structure in step 5-1) includes a convolutional layer, a long short-term memory layer, a pooling layer, and a fully connected layer; the number of convolutional layers is not less than 5 layers; after each convolutional layer, batch normalization and a rectified linear unit activation function are attached; the convolutional kernel size of the convolutional layer ranges between 3×3 and 5×5; the number of long short-term memory layers is 2 layers, and the number of hidden units ranges between 64 and 256; the pooling layer uses max pooling, the window size is not greater than 2×2, and the stride is not greater than 2; the fully connected layer has no less than 2 layers, the number of neurons in each layer ranges between 128 and 512, and the dropout method is used to reduce the risk of overfitting; the dropout ratio of the dropout method is not less than 0.2 and not higher than 0.5; The artificial intelligence model combines the adaptive moment estimation optimizer and the learning rate decay strategy; the learning rate of the adaptive moment estimation optimizer is not greater than 0.0005, the decay rate of the first moment estimation is not less than 0.9, and the decay rate of the second moment estimation is not less than 0.999; the decay rate of the learning rate decay strategy is not greater than 0.1, and the learning rate is dynamically adjusted during training; The model training uses a batch size between 16 and 64, the maximum number of training epochs is not less than 50 epochs, and an early stopping strategy with a tolerance of no more than 10 epochs is used to optimize the utilization of computing resources.

9. The moisture content-driven simulation method for heat and moisture distribution in building envelopes and the artificial intelligence prediction method according to claim 1, characterized in that: The verification of the model prediction ability in step 5-2) is based on the test set data. The heat and moisture distribution cloud map generated by the artificial intelligence model is compared pixel by pixel with the high-precision numerical simulation results, and the image similarity index is calculated; the image similarity index is specifically the structural similarity index, the peak signal-to-noise ratio, and the mean square gradient error. When they are not less than 0.85, not less than 30 dB, and not more than 0.02 respectively, it is considered that the difference between the model prediction result and the standard numerical solution is within an acceptable range; The optimization of the model calculation efficiency is achieved through model pruning and model quantization; the model pruning uses weight pruning or structural pruning to balance storage optimization and calculation acceleration, and ensures that the accuracy loss of the pruned model does not exceed 2%; the model quantization uses static quantization and dynamic quantization, and selects the optimal strategy according to the hardware characteristics to quantize the model parameters from FP32 to INT8, ensuring that the accuracy loss after quantization does not exceed 1%.

10. The moisture content-driven simulation method for heat and moisture distribution in building envelopes and the artificial intelligence prediction method according to claim 1, characterized in that: The heat and moisture distribution prediction in step 5-3) is based on the actual measurement data, dynamically calculates the temperature and humidity fields inside the enclosure structure, and predicts the distribution situation at future times; the actual measurement data includes indoor and outdoor environmental parameters; the dynamic calculation is based on historical and real-time acquired data, and combines time series analysis methods to calculate the heat and moisture distribution cloud map at the next time step or multiple future time steps.

Images