Intelligent management and control method for building energy-saving wall and lattice structure wall applying method

Abstract

The invention relates to an intelligent management and control method for a building energy-saving wall and a lattice structure wall applying the method, and solves the problems that although a traditional heat insulation material can reduce heat conduction, the function is single, and heat cannot be actively adjusted. Setting weights for the input data, and calculating the sum of the input data by adopting a weighted average method; according to a mapping relation between an interval range in which the sum of the input data falls and a grade, analyzing and determining the grade corresponding to the thermal working condition; according to the mapping relation between the thermal working condition grade and the overall regulation and control scheme, the overall regulation and control scheme is analyzed and determined, the determined overall regulation and control scheme is executed, and the overall regulation and control scheme comprises a phase change material regulation and control scheme, a lattice structure regulation and control scheme and a heat exchange fluid regulation and control scheme. The method has the following effects that the heat condition of the building is accurately regulated and controlled, the heat performance of the wall is improved, and efficient energy saving and intelligent management are achieved.

Description

Technical Field

[0001] The present invention relates to the field of building energy conservation, and in particular to an intelligent control method for building energy conservation walls and a lattice structure wall using the method. Background Art

[0002] Against the backdrop of increasingly severe global energy situations and people's increasing requirements for building environmental comfort, the development of building energy-saving technologies is of vital importance. As an important area of ​​energy consumption, buildings account for a high proportion of total social energy consumption. Therefore, improving the level of building energy conservation has become a key link in alleviating the energy crisis and achieving sustainable development. As the core part of the building envelope, the wall plays a vital role in building energy conservation, and its thermal performance is directly related to the quality of the indoor thermal environment and the energy consumption of the building.

[0003] At present, there are mainly two types of technical means in terms of building wall energy-saving technology. One is the traditional thermal insulation technology, which uses thermal insulation materials such as polystyrene boards and rock wool to increase the thermal resistance of the wall to reduce heat conduction. These materials can reduce the cold and heat load of the building to a certain extent and are widely used in the field of building energy conservation. The other is the application technology of phase change materials in walls. Phase change materials (PCMs) are introduced into building walls due to their characteristics of absorbing and releasing a large amount of latent heat during the solid-liquid phase change process. They are used to alleviate indoor temperature fluctuations, reduce air conditioning and heating energy consumption, and thus improve indoor thermal comfort.

[0004] However, these related technologies have many drawbacks in practical applications. Although traditional thermal insulation materials can reduce heat conduction, their functions are relatively simple and they cannot actively regulate heat. At the same time, the current application of phase change materials in building walls lacks intelligent control methods and cannot be dynamically adjusted and replaced according to seasonal changes, indoor temperature requirements, and the attenuation of phase change material performance, which seriously restricts the improvement of wall thermal performance. Summary of the invention

[0005] In order to accurately control the thermal conditions of buildings, improve the thermal performance of walls, and achieve efficient energy saving and intelligent management, the present application provides a method for intelligent control of energy-saving walls in buildings and a lattice structure wall using the method.

[0006] In the first aspect, the present application provides a method for intelligent control of building energy-saving walls, which adopts the following technical solutions: A method for intelligent management and control of building energy-saving walls, comprising: Obtain environmental data and user behavior data, and perform data preprocessing; Input environmental data and user behavior data into the pre-built heat demand prediction model to output building heat demand information; Collect and obtain real-time monitoring data of heat transfer related parameters, and input the real-time monitoring data into the pre-built multi-physics field coupled heat transfer model. The multi-physics field coupled heat transfer model calculates the heat transfer process of the fine units of the wall structure based on the finite element analysis method, and obtains the heat transfer model analysis results by solving the multi-physics field coupling equations. The heat transfer model analysis results include the temperature distribution at different locations of the wall at different times, the heat transfer rate, and the phase change state of the phase change material; Standardize the data included in the building heat demand information and the data included in the heat transfer model analysis results to make them uniform in scope; According to different building types, usage scenarios and seasons, weights are set for each input data, and the total input data is calculated using the weighted average method; According to the mapping relationship between the interval range that the sum of input data falls into and the level, the level corresponding to the thermal condition is analyzed and determined; According to the mapping relationship between the thermal condition level and the overall control plan, the overall control plan is analyzed and determined, and the determined overall control plan is executed. The overall control plan includes a phase change material control plan, a lattice structure control plan, and a heat exchange fluid control plan.

[0007] Optionally, the steps for building a heat demand prediction model are as follows: Collect environmental data and user behavior data, and perform data preprocessing; Build a Transformer model based on spatiotemporal decoupling, randomly initialize the model parameters, and determine the model hyperparameters; Divide the preprocessed data into a training set and a validation set according to a preset ratio; The training set data is input into the Transformer model based on spatiotemporal decoupling, and the prediction results are calculated through forward propagation. The weighted mean square error loss function is used to calculate the difference between the prediction results and the actual heat demand value. In addition, a Dropout operation is performed during each forward propagation to obtain the prediction results under different parameter samples. Based on the results of the loss function, the gradient is calculated through the back-propagation algorithm to update the model parameters; After each training cycle is completed, the validation set loss is calculated. If the change in validation set loss for consecutive preset cycles is less than the set threshold, the training is considered completed.

[0008] Optionally, the steps to build a multiphysics coupled heat transfer model are as follows: Collect and obtain multi-source data and perform data preprocessing; Build a model architecture at the micro, meso and macro levels; According to the correspondence between the building thermal management scenario and the physical field coupling mode and intensity setting scheme, the physical field coupling mode and intensity of the current building thermal management scenario are analyzed and determined as the starting state of the model training; The nonequilibrium heat flow equation Integrate into the model heat transfer calculation system, where is the heat flux density, is the thermal conductivity, is the temperature gradient, is the pressure gradient, is the substance concentration gradient, is the pressure-dependent heat transfer coefficient, is the heat transfer coefficient related to the species concentration; A deep neural network with a multilayer perceptron structure of 3 hidden layers and 100 neurons in each layer was selected and integrated into the model using a residual connection fusion strategy. The input layer of the deep neural network received temperature, humidity, and building structure parameter data, the hidden layer used the ReLU function to extract features, and the output layer predicted heat transfer related parameters. Input the preprocessed data into a model that has a well-built architecture, a determined coupling method, incorporates non-equilibrium equations, and integrates a deep neural network; During training, the mean square error loss function is used to calculate the deviation between the model prediction value and the true value. The stochastic gradient descent method is used to iteratively update the model parameters at a learning rate of 0.01 to reduce the loss function value. After every 10 parameter updates, the weight of the influence of the coupling strength of each physical field on the loss function is calculated based on the model prediction results and real data under the latest parameters. According to the weight, the coupling strength coefficient is adjusted by 10% to reduce the loss function value. According to the dynamic change results of the scene within each interval time range, the physical field coupling strength is changed through a preset coupling strength adjustment function; The mean square error is continuously monitored. When the mean square error change value is less than the preset change value in consecutive preset rounds of training, the training is considered complete.

[0009] In a second aspect, the present application provides a lattice structure wall, which adopts the following technical solution: A lattice structure wall, comprising: Lattice structure framework; Phase change material, which is filled in the internal space of the lattice structure frame and is paraffin or hydrated salt, is used to store and release heat; A macro packaging structure for packaging the phase change material within a lattice structure framework; Concrete is poured into the formwork embedded with the lattice structure frame, so that the lattice and concrete are tightly combined.

[0010] Optionally, the lattice structure frame is a double-layer lattice structure frame, the inner lattice provides a stable filling space for the phase change material, the outer lattice is arranged around the inner layer, and a heat exchange fluid circulation channel is provided, and the channel is filled with heat exchange fluid. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 It is a flow chart of the intelligent control method of energy-saving building walls in the embodiment of the present application.

[0012] Figure 2 It is a schematic diagram of a single-layer lattice structure in an embodiment of the present application.

[0013] Figure 3 yes Figure 2 Enlarged schematic diagram at point a in the middle.

[0014] Figure 4 It is a schematic diagram of a double-layer lattice structure in an embodiment of the present application.

[0015] Figure 5 yes Figure 4 Enlarged schematic diagram at point b in the middle. DETAILED DESCRIPTION

[0016] The present application is further described in detail below in conjunction with the accompanying drawings.

[0017] Reference Figure 1 , is a building energy-saving wall intelligent control method disclosed in this application, comprising: Step S100, obtaining environmental data and user behavior data, and performing data preprocessing.

[0018] Environmental data refers to data related to the external environment of the building, including but not limited to external temperature, humidity, sunlight intensity, etc. These data will affect the building's thermal demand and the heat transfer process of the wall. Environmental data is obtained by reasonably arranging various sensors around the building. For example, temperature sensors are used to measure the external temperature, humidity sensors collect humidity data, and sunlight sensors monitor sunlight intensity. These sensors transmit the real-time monitored data to the data acquisition device.

[0019] User behavior data: data related to users’ activities in the building, such as entry and exit times, usage of indoor equipment (such as air conditioning on / off times, frequency of electrical appliance use), etc. These behaviors will change the indoor thermal environment.

[0020] The general process is as follows: various sensors and monitoring equipment continuously collect environmental data and user behavior data. After the data acquisition device collects these raw data, it first performs preliminary screening to eliminate erroneous or abnormal data, such as the obvious deviation value of the temperature sensor. Then, the data is formatted and standardized, and data of different types and dimensions are converted into a form suitable for subsequent analysis. Finally, pre-processed data is obtained to provide reliable data support for subsequent steps such as predicting building heat demand.

[0021] Step S200: inputting environmental data and user behavior data into a pre-built heat demand prediction model, and outputting building heat demand information.

[0022] Heat demand prediction model: Based on the Transformer model with time-space decoupling, it can learn the complex relationship between environmental data, user behavior data and building heat demand. Through training on a large amount of historical data, the model can predict the heat demand information of the building in the future based on the input current environment and user behavior data.

[0023] Building heat demand information: refers to the amount of heat that a building needs within a certain period of time to maintain a suitable indoor thermal environment, including heating, cooling and other needs. This information is crucial for building thermal management and energy-saving regulation.

[0024] Step S300, collect and obtain real-time monitoring data of heat transfer related parameters, and input the real-time monitoring data into a pre-built multi-physics field coupled heat transfer model. The multi-physics field coupled heat transfer model calculates the heat transfer process of the fine units of the wall structure based on the finite element analysis method, and obtains the heat transfer model analysis results by solving the multi-physics field coupling equations. The heat transfer model analysis results include the temperature distribution at different locations of the wall at different times, the heat transfer rate, and the phase change state of the phase change material.

[0025] Real-time monitoring data of heat transfer related parameters: refers to various data directly related to heat transfer collected in real time during the heat transfer process of the building wall, such as the temperature at different locations of the wall, heat flux density, phase change state of phase change materials, etc. These data reflect the real-time dynamic situation of wall heat transfer.

[0026] Multi-physics coupled heat transfer model: A heat transfer model that comprehensively considers the interaction of multiple physical fields. In the present invention, the model integrates physical processes such as heat conduction, convection, radiation, and the phase change process of phase change materials to accurately simulate the heat transfer behavior of the wall structure. It is based on the finite element analysis method and divides the wall structure into fine units for calculation.

[0027] Finite element analysis method: Discretize the continuous wall structure into a finite number of units, perform mechanical and thermal analysis on each unit, and then combine these units to solve the numerical calculation method of the mechanical and thermal response of the entire structure. In this step, it is used to calculate the heat transfer process of the fine unit of the wall structure.

[0028] Heat transfer model analysis results: The results calculated by the multi-physics field coupled heat transfer model include temperature distribution at different locations of the wall at different times, heat transfer rate, and phase change state of phase change materials. These results provide key basis for building thermal management.

[0029] The specific steps for obtaining the heat transfer model analysis results can refer to step S310 to step S350, which will not be described in detail here.

[0030] Step S400, standardizing the data included in the building heat demand information and the data included in the heat transfer model analysis result to unify their ranges.

[0031] Standardization: A data processing method that aims to convert data with different characteristics, dimensions and value ranges into a unified scale or range through specific mathematical transformations to facilitate subsequent comparison, analysis and calculation. In this step, the purpose is to eliminate the interference of data differences on subsequent weighted calculations and thermal condition level analysis.

[0032] Step S500, according to different building types, usage scenarios and seasons, weights are set for various input data, and the sum of the input data is calculated using a weighted average method. Various input data refer to various types of data related to building thermal management in the intelligent management and control method for building energy-saving walls, including environmental data, user behavior data, real-time monitoring data of heat transfer related parameters, data included in building thermal demand information, and data included in the analysis results of the heat transfer model.

[0033] Building type: Classification based on the building's use function, structural characteristics, etc., such as residential, office building, industrial plant, etc. Different types of buildings have different thermal demands and heat transfer characteristics. For example, residential buildings mainly meet the comfort needs of residents, while industrial plants may pay more attention to the appropriate environment for equipment operation, which makes the degree of attention to various data in thermal management different.

[0034] Usage scenario: refers to the specific situation of the building during actual use, including population density, equipment usage frequency, indoor activity types, etc. For example, during meetings, the conference room is crowded with people and equipment is frequently used, so its heat demand will be higher than during normal office hours; the heat demand of shopping malls during business hours and non-business hours is also very different. These scenario differences will affect the thermal management strategy, and then affect the weight setting of input data.

[0035] Season: Different time periods of the year, such as spring, summer, autumn and winter. The ambient temperature, sunshine duration and other factors vary significantly in different seasons, which have a great impact on the thermal demand of the building and the heat transfer process of the wall. For example, in the hot summer, the building has a large demand for cooling; in the cold winter, the heating demand is high, which makes the importance of various input data to thermal management decisions in different seasons different.

[0036] Weight: A value used to measure the relative importance of each input data when calculating the sum of the input data. Depending on the building type, usage scenario and season, the impact of each data on thermal management decisions will vary, and the weight will be adjusted accordingly. For example, in the summer, the ambient temperature has a greater impact on cooling demand, and its weight may be relatively high.

[0037] Step S600, analyzing and determining the level corresponding to the thermal condition according to the mapping relationship between the interval range into which the sum of the input data falls and the level.

[0038] Range: To facilitate the classification of thermal conditions, a preset numerical range of the sum of input data is set. Different ranges represent different thermal management conditions, and each range corresponds to a thermal condition level.

[0039] Grade: A classification mark for thermal conditions, used to distinguish different states of building thermal management. There is a corresponding relationship between the thermal condition grade and the range of the sum of input data. The thermal condition grade is determined by judging the range to which the sum of input data belongs, so that corresponding control measures can be taken later.

[0040] Thermal conditions: describes the thermal status of a building at a certain moment, including indoor temperature, wall heat transfer, etc. The classification of thermal conditions helps to quickly understand the thermal management needs of the building, so as to formulate targeted control strategies.

[0041] The general process is as follows: obtain the sum of the input data obtained in step S500, compare it with the preset interval range, determine which interval the sum of the input data falls into, and then determine the corresponding thermal condition level.

[0042] Step S700: Analyze and determine the overall control scheme according to the mapping relationship between the thermal condition level and the overall control scheme, and execute the determined overall control scheme.

[0043] Thermal condition level: A classification identifier determined based on the range in which the sum of the input data falls. It is used to characterize the current thermal state of the building and reflects the degree of demand for thermal management of the building. Different levels correspond to different thermal management strategies.

[0044] Overall control plan: A series of control measures formulated for different thermal conditions, including phase change material control plan, lattice structure control plan, and heat exchange fluid control plan. It aims to achieve efficient thermal management and meet indoor comfort and energy-saving requirements by adjusting relevant parameters of building walls.

[0045] Phase change material regulation scheme: mainly focuses on the phase change materials filled in the wall lattice structure. Phase change materials can absorb or release a large amount of latent heat during the solid-liquid phase change process, which has a significant impact on the thermal performance of the wall. This scheme chooses to replace phase change materials with different melting points according to seasonal changes, changes in external temperature, and the thermal needs of the building. For example, in the high temperature in summer, a low melting point phase change material is selected to quickly absorb heat when the temperature rises and lower the indoor temperature; in winter, it is replaced with a high melting point phase change material, which is conducive to storing heat and slowly releasing it to maintain indoor warmth.

[0046] Lattice structure control scheme: Focus on optimizing and adjusting the lattice structure of 3D printing. The lattice structure provides a stable filling space for the phase change material, and its geometric shape and structural parameters directly affect the thermal contact area between the phase change material and the outside world and the heat conduction efficiency. The control scheme may include changing the pore size of the lattice, the arrangement of the lattice units, etc., so as to enhance the heat conduction effect and improve the heat storage and heat dissipation capacity of the wall.

[0047] Heat exchange fluid control scheme: Precisely control the fluid in the outer lattice heat exchange fluid circulation channel. The fluid circulates in the channel and plays an important role in transferring heat. This scheme achieves efficient transfer and dissipation of wall heat by adjusting parameters such as the fluid circulation rate and temperature. In summer, increasing the fluid circulation rate and lowering the fluid temperature can quickly take away the excess heat absorbed by the wall; in winter, appropriately lowering the circulation rate and raising the fluid temperature can assist the phase change material in releasing heat and keep the indoor temperature stable.

[0048] The general process is as follows: When the system obtains the thermal condition level determined by step S600, it will immediately search the database for the corresponding overall control plan. For example, if the thermal condition level shows that the current building is in a high temperature and high load state, the overall control plan found by the system may be: in terms of phase change material control, select low melting point hydrated salt phase change materials for replacement; in terms of lattice structure control, by adjusting the parameters of the 3D printing equipment, increase the pores of the lattice to increase the thermal contact area; in terms of heat exchange fluid control, start the circulation pump, increase the fluid circulation rate to the maximum value, and reduce the fluid temperature. After determining the control plan, the system will automatically send control instructions to the relevant execution equipment. For phase change material replacement, control the special filling and discharge device to discharge the original phase change material and inject the new phase change material that meets the requirements; for lattice structure adjustment, control the 3D printing device (if it has real-time adjustment function) or manufacture the lattice structure according to the new parameters in the subsequent production process; in terms of heat exchange fluid control, adjust the fluid circulation rate by controlling the motor speed of the circulation pump, and adjust the fluid temperature by using the heating or cooling device. By implementing these control measures, the thermal performance of building walls can be optimized, so that the building can achieve good thermal management effects under different thermal conditions and meet the dual needs of indoor comfort and energy saving.

[0049] The steps to build the heat demand forecasting model are as follows: Step S201, collect environmental data and user behavior data, and perform data preprocessing.

[0050] Step S202, building a Transformer model based on spatiotemporal decoupling, randomly initializing the parameters of the model, and determining the hyperparameters of the model.

[0051] Transformer model based on spatiotemporal decoupling: The Transformer model is a deep learning model based on the attention mechanism, which is widely used in fields such as natural language processing. In the present invention, the Transformer model based on spatiotemporal decoupling is improved for the building heat demand prediction scenario. Spatiotemporal decoupling refers to the separate processing of time and space features, which can more effectively capture the changing patterns of environmental data and user behavior data in time series, as well as the correlation of data in different spatial locations (such as different areas of a building), thereby improving the accuracy of building heat demand prediction.

[0052] Model parameters: These are the values ​​that the model needs to learn and adjust during the training process, such as weights and biases in a neural network. These parameters determine the function and performance of the model. By continuously optimizing the parameters, the model can better fit the training data and accurately predict the building heat demand.

[0053] Hyperparameters: are parameters that are set before model training and cannot be learned directly through training. Hyperparameters affect the structure, training process, and performance of the model. For example, the number of layers, number of neurons in the hidden layer, and learning rate of the model are all hyperparameters. Reasonable setting of hyperparameters is crucial to the training effect of the model.

[0054] The general process is as follows: First, select a suitable deep learning framework (such as TensorFlow or PyTorch) as the basis for model building. Then, according to the design principle of the Transformer model based on spatiotemporal decoupling, write code in the framework to build the model structure. During the construction process, define the input layer, hidden layer, and output layer of the model, where the hidden layer adopts the structural unit of the Transformer and incorporates the processing logic of spatiotemporal decoupling. After the model is built, use the functions provided by the framework to randomly initialize the model parameters so that the model has an initial state at the beginning of training. Finally, based on the understanding and experience of the building heat demand prediction task, set the model's hyperparameters, such as determining the number of layers of the model, the number of neurons in the hidden layer, the learning rate, etc., to prepare for subsequent model training.

[0055] Step S203, dividing the preprocessed data into a training set and a validation set according to a preset ratio.

[0056] Preset ratio: It is the ratio of the training set and the validation set that is preset before model training. Common preset ratios include 70:30, 80:20, etc. It determines the distribution relationship between the amount of data used to train the model and the amount of data used to evaluate the model performance.

[0057] Training set: a data subset separated from the preprocessed data for model training. During the training process, the model continuously adjusts its own parameters by learning from the training set data to improve the accuracy of building heat demand prediction. Validation set: a data subset used to evaluate the model training effect. During the model training process, the validation set is used to calculate the model loss value at regular training cycles. By observing the changes in the validation set loss, it is determined whether the model is overfitting or whether a good training effect has been achieved.

[0058] In step S204, the training set data is input into the Transformer model based on spatiotemporal decoupling, the prediction results are calculated through forward propagation, the weighted mean square error loss function is used to calculate the difference between the prediction results and the actual heat demand value, and a Dropout operation is performed during each forward propagation to obtain the prediction results under different parameter samples.

[0059] Among them, forward propagation: In a deep learning model, data enters from the input layer, passes through the calculation and transformation of each hidden layer in turn, and finally reaches the output layer to generate a prediction result. In this process, the model performs a series of mathematical operations on the input data according to the current parameters, such as matrix multiplication, activation function operation, etc.

[0060] Prediction result: The result output by the model after forward propagation. In this scenario, it refers to the predicted value of the building heat demand based on the Transformer model based on spatiotemporal decoupling. This value will be compared with the actual heat demand value.

[0061] Weighted mean square error loss function: A function used to measure the difference between the model prediction result and the true value. The mean square error is the average of the squares of the difference between the predicted value and the true value, while the weighted mean square error is to give different weights to different data points when calculating the mean square error to highlight the importance of certain data. In the building heat demand forecasting, the parameters of the model are optimized by minimizing this loss function.

[0062] Dropout operation: A technique to prevent model overfitting. During model training, Dropout randomly "discards" (i.e. sets the neuron output to 0) some neurons in the hidden layer, so that the model cannot rely too much on certain specific neuron connections, thereby enhancing the generalization ability of the model and reducing overfitting.

[0063] The general process is as follows: Assume that the Transformer model based on spatiotemporal decoupling has been built and there is a training set containing 100 data. First, take a batch of data from the training set (such as 10 data as a batch) and input it into the model. The data enters the model from the input layer and passes through the hidden layer calculation in turn according to the forward propagation calculation rules. For example, the neurons in the hidden layer will perform weighted summation on the input data and transform it through the activation function, and finally get the prediction result in the output layer. This prediction result is an estimate of the building's heat demand.

[0064] Step S205, based on the result of the loss function, the gradient is calculated by the back propagation algorithm to update the model parameters.

[0065] Loss function: In step S204, a weighted mean square error loss function is used to measure the difference between the model prediction result and the actual heat demand value. The value of the loss function reflects the accuracy of the current model prediction. The smaller the value, the closer the model prediction result is to the actual value.

[0066] Back propagation algorithm: It is the core algorithm used to optimize model parameters in deep learning model training. It is based on the chain derivation rule. Starting from the output layer, it backpropagates the gradient of the loss function to the parameters of each layer and calculates the gradient value of the loss function to each parameter. These gradient values ​​indicate how much impact a small change in the parameter will have on the loss function.

[0067] Gradient: Mathematically, a gradient is a vector that points in the direction in which the function changes most rapidly. In model training, the gradient of the loss function with respect to the model parameters indicates in which direction the parameter adjustment can minimize the value of the loss function.

[0068] The general process is as follows: 1. Receive the loss value: Obtain the loss value calculated by the weighted mean square error loss function from step S204. This value reflects the deviation between the model prediction and the actual heat demand. 2. Start back propagation: Start back propagation with the model output layer as the starting point. Because the loss function is calculated based on the output layer prediction result, the impact of each parameter on the loss can be directly calculated from here. 3. Calculate the output layer gradient: Use the chain derivative rule to calculate the gradient of the loss function to the output layer neuron input, that is, the gradient of the output layer. 4. Propagate the gradient to the hidden layer: Back propagate the output layer gradient to the hidden layer, and calculate the gradient of the layer in combination with the hidden layer activation function. 5. Multi-layer hidden layer processing: If there are multiple hidden layers, back propagate the gradient to each layer in turn until the input layer. 6. Calculate the parameter gradient: According to the gradient of each layer of neurons, calculate the gradient of the loss function to all parameters (weights and biases) of the model. 7. Set the learning rate: The learning rate controls the parameter update step size, and the appropriate value needs to be determined in advance based on experience or experiments. If the learning rate is too large, the optimal solution may be missed, and if it is too small, the training will be slow. 8. Update model parameters: Update model parameters using the gradient descent algorithm based on parameter gradients and learning rates. Repeat steps S204 and S205 to continuously optimize model parameters and improve the accuracy of heat demand prediction.

[0069] Step S206, after each training cycle, the validation set loss is calculated. If the change in the validation set loss for consecutive preset cycles is less than a set threshold, the training is considered completed.

[0070] Among them, training cycle (Epoch): During the model training process, the entire training set data is fully input into the model and the process of completing a forward propagation and back propagation to update the parameters is called a training cycle. After each training cycle, the model has completed a round of learning of the training set data.

[0071] Validation set loss: Use validation set data to input the trained model, and calculate the loss value by calculating the difference between the predicted result and the true value of the validation set. It is usually calculated using the weighted mean square error loss function. Validation set loss is used to evaluate the performance of the model on unseen data and reflects the generalization ability of the model.

[0072] Preset number of cycles: A pre-set integer used to measure the change in validation set loss over a number of consecutive training cycles to determine whether the model has converged. For example, a preset value of 5 means observing the change in validation set loss over 5 consecutive training cycles.

[0073] Set threshold: A predetermined small value is used to determine whether the change in validation set loss is small enough. If the change in validation set loss is less than this threshold for consecutive preset cycles, it is considered that the model training has achieved a good effect and training can be stopped. The setting of this threshold needs to be determined based on specific problems and experience, and is generally a small value, such as 0.001.

[0074] The steps to build a multiphysics coupled heat transfer model are as follows: Step S301, collect and obtain multi-source data and perform data preprocessing.

[0075] Among them, multi-source data refers to various types of data related to the heat transfer process of building walls. These data come from a wide range of sources, including environmental data (such as external temperature, humidity, sunlight intensity, etc.), building structure data (such as wall materials, thickness, lattice structure parameters, etc.), phase change material property data (such as phase change latent heat, melting point, thermal conductivity, etc.) and heat transfer related monitoring data (such as temperature and heat flux density at different locations of the wall). These data affect the heat transfer process of the wall from different aspects and are the basis for building a multi-physics field coupled heat transfer model.

[0076] Data preprocessing: A series of processing operations are performed on the collected raw multi-source data to improve the data quality and make it more suitable for model training and analysis. It mainly includes data cleaning (removing outliers and duplicate values), data completion (filling missing data), data standardization (unifying data of different dimensions and different value ranges into a specific interval, such as [0,1]), etc., to ensure the accuracy, completeness and consistency of the data.

[0077] Step S302, build a model architecture at the micro, meso and macro levels.

[0078] Micro-level model architecture: focuses on the microstructure and physical phenomena of materials, and is used to describe the heat transfer mechanism at the atomic and molecular scales, such as the influence of thermal motion of molecules inside phase change materials and lattice vibration on heat transfer, explaining the microscopic process of heat transfer from a fundamental level.

[0079] Mesoscopic model architecture: It lies between the microscopic and macroscopic levels. It mainly studies the relationship between the microscopic structural characteristics of materials and heat transfer, such as the distribution of phase change materials in the lattice structure, the influence of tiny pores or defects on heat conduction, etc. It is a bridge connecting microscopic and macroscopic heat transfer phenomena.

[0080] Macro-level model architecture: describes the heat transfer behavior of the building wall as a whole, taking into account macro factors such as the overall geometry of the wall, boundary conditions, and heat exchange with the external environment, such as the overall temperature distribution of the wall, the flow of heat in and out, etc., reflecting the performance of heat transfer at the macro scale.

[0081] Take the construction of a heat transfer model for a lattice structure wall containing phase change materials as an example.

[0082] At the microscopic level, based on the principles of materials science, the thermal motion model of phase change material molecules is determined. For example, it is assumed that the intermolecular forces and motions follow a certain classical mechanics model to describe how heat is transferred between phase change material molecules at the microscopic scale, as well as the impact of molecular state changes on heat transfer during the phase change process.

[0083] At the mesoscopic level, the distribution of the phase change material in the lattice is analyzed according to the characteristics of the 3D printed lattice structure. For example, the size and shape of the lattice pores and the contact between the phase change material and the lattice wall are considered to construct a corresponding heat conduction model to simulate the heat transfer path and efficiency through the lattice structure and phase change material at the mesoscopic scale.

[0084] At the macro level, the entire wall is considered as a whole, and the size, position, orientation, and heat exchange boundary conditions between the wall and the indoor and outdoor environments are considered. For example, the convection heat transfer coefficient between the outside of the wall and the external environment, the heat exchange method between the inside and the indoor air, etc. are set to establish a macro heat transfer equation to describe the heat balance and temperature change trend of the entire wall. By integrating the model architecture of these three levels, a complete model framework is formed that can describe the wall heat transfer process from different scales, laying the foundation for subsequent model calculation and analysis.

[0085] Step S303, according to the correspondence between the building thermal management scenario and the physical field coupling mode and strength setting scheme, analyze and determine the physical field coupling mode and strength of the current building thermal management scenario as the starting state of model training.

[0086] Building thermal management scenarios: refers to the thermal conditions of buildings under different usage and environmental conditions, covering a variety of factors such as building type (residential, office building, shopping mall, etc.), usage time (day, night, weekdays, holidays), season (spring, summer, autumn, winter), and indoor and outdoor environmental parameters (temperature, humidity, light, etc.). The thermal demand and heat transfer process of buildings in different scenarios vary significantly.

[0087] Physical field coupling mode: In the process of building heat transfer, there are multiple physical field interactions, such as heat conduction, convection, radiation, and phase change process of phase change materials. The physical field coupling mode describes the specific forms of the mutual relationship and influence between these physical fields. For example, the coupling of heat conduction and convection is manifested as heat conduction in solid materials while heat exchange is carried out through convection with the surrounding fluid; the coupling of heat conduction and radiation involves heat conduction inside the material while heat is transferred through radiation on the surface of the object.

[0088] Physical field coupling strength: an indicator that measures the degree of interaction between different physical fields. For example, in the coupling of heat conduction and convection, if the convection effect is strong and the heat transfer is mainly dominated by convection, the coupling strength is high; conversely, if heat conduction plays a major role, the coupling strength is relatively low. The coupling strength determines the relative importance of each physical field in the heat transfer process.

[0089] Correspondence: The association rules between building thermal management scenarios and physical field coupling methods and strength setting schemes are obtained through theoretical research, experimental testing, and actual project experience. It provides a basis for determining the appropriate physical field coupling method and strength according to specific building thermal management scenarios.

[0090] Step S304, integrating the non-equilibrium heat flow equation into the model heat transfer calculation system.

[0091] The non-equilibrium heat flow equation is: ,in, is the heat flux density, is the thermal conductivity, is the temperature gradient, is the pressure gradient, is the substance concentration gradient, is the pressure-dependent heat transfer coefficient, is the heat transfer coefficient that is dependent on the species concentration.

[0092] Among them, the non-equilibrium heat flow equation is a mathematical expression used to describe the law of heat transfer under non-equilibrium conditions. In the building heat transfer scenario, the actual situation is often not in an ideal equilibrium state. This equation can more accurately reflect the heat transfer process. For example, in the given equation, the heat flux density is affected by many factors such as material properties, temperature change rate, pressure gradient, material concentration gradient, etc. Compared with the traditional equilibrium heat transfer equation, it considers the effects of more dynamic factors on heat flow.

[0093] Model heat transfer calculation system: It is a collection of rules, formulas and algorithms used to calculate the heat transfer process in the multi-physics field coupled heat transfer model. It integrates the basic equations of different physical fields (heat conduction, convection, radiation, etc.), and performs comprehensive calculations based on the physical field coupling mode and strength to simulate the heat transfer behavior of structures such as building walls. The incorporation of non-equilibrium heat flow equations is intended to improve the calculation system and enhance the model's simulation accuracy for complex heat transfer processes.

[0094] The general process is as follows: Suppose a multi-physics coupling model is constructed to simulate the heat transfer of a new phase change material wall. First, the researchers obtained a suitable non-equilibrium heat flow equation from professional heat transfer literature, which fully considers the influence of temperature, pressure and material concentration changes on heat flow during the phase change process of the phase change material. Then, the established model heat transfer calculation system is analyzed to determine the location and method of integrating the equation. For example, in the macro-level heat transfer calculation, the traditional equilibrium heat conduction equation was originally used to calculate the heat conduction of the entire wall. Now, the non-equilibrium heat flow equation is integrated with the original heat conduction, convection, radiation and other equations. For a certain tiny unit in the wall, when calculating its heat flux density, the various parameters in the non-equilibrium heat flow equation are introduced, such as determining the correlation coefficient based on the material properties, and monitoring the temperature change rate, pressure gradient and material concentration gradient in real time through sensors, and substituting them into the equation for calculation. In this way, the calculation system that originally only considered steady-state heat transfer can more accurately simulate the non-equilibrium heat transfer process caused by changes in the external environment and the dynamic characteristics of phase change materials in actual use of the wall due to the integration of non-equilibrium heat flow equations, thereby improving the accuracy and reliability of the model's simulation of wall heat transfer.

[0095] Step S305, a deep neural network with a multilayer perceptron structure having 3 hidden layers and 100 neurons in each layer is selected and integrated into the model using a residual connection fusion strategy. The input layer of the deep neural network receives temperature, humidity, and building structure parameter data, the hidden layer uses the ReLU function to extract features, and the output layer predicts heat transfer related parameters.

[0096] Specifically, a deep neural network with a multi-layer perceptron (MLP) structure: a feed-forward neural network consisting of an input layer, multiple hidden layers, and an output layer. In this step, there are 3 hidden layers with 100 neurons in each layer, and the neurons are connected by weights. It can perform nonlinear transformations on the input data and predict heat transfer related parameters by learning patterns in a large amount of data. Neurons in different layers can learn features at different levels of the data, gradually abstracting the original input features into more advanced and representative features for the final prediction task.

[0097] Residual connection fusion strategy: A connection method used to solve the gradient vanishing or gradient exploding problems that may occur during the training of deep neural networks, and to improve the learning efficiency of the network. Residual connection allows the network to directly learn the residual between input and output, that is, the goal of network learning becomes the difference between "output-input" rather than directly learning the output itself. In this step, the different layers of the multi-layer perceptron are fused through residual connections, so that information can flow more efficiently in the network, which helps the model converge faster and improves the model's ability to model complex heat transfer processes.

[0098] ReLU function: Rectified Linear Unit, is a commonly used activation function. Using the ReLU function in the hidden layer can introduce nonlinear factors into the neural network. Because the linear model has limited expressive power, through the ReLU function, when the input is greater than 0, the neuron outputs the input value normally; when the input is less than 0, the output is 0, so that the neural network can learn more complex functional relationships and better extract data features.

[0099] Heat transfer related parameters: various physical quantities closely related to the heat transfer process of building walls, such as heat flux density, temperature distribution, thermal conductivity changes of materials, etc. These parameters are the targets that need to be accurately predicted by the multi-physics field coupled heat transfer model. Through the analysis and processing of input data by deep neural networks, the prediction results of these parameters are output, providing key information for the model to fully simulate the heat transfer process.

[0100] Step S306, input the preprocessed data into a model that has a well-built architecture, determined coupling mode, integrated into non-equilibrium equations, and integrated with a deep neural network.

[0101] Model with a well-constructed architecture: The model architecture of the microscopic, mesoscopic and macroscopic levels built in step S302 comprehensively describes the heat transfer mechanism from the atomic and molecular scale, the material microstructure to the entire building wall, and provides a basic framework for the model.

[0102] Determine the coupling model: According to step S303, based on the correspondence between the building thermal management scenario and the physical field coupling mode and intensity setting scheme, determine the coupling mode between physical fields such as heat conduction, convection, and radiation in the current scenario, such as how heat conduction and convection interact with each other, the correlation between heat conduction and radiation, etc., so that the model can reflect the physical field relationship in the actual heat transfer process.

[0103] Model incorporating non-equilibrium equations: In step S304, the non-equilibrium heat flow equation is incorporated into the model heat transfer calculation system. The equation considers the effects of factors such as material properties, temperature change rate, pressure gradient, and material concentration gradient on heat flux density, thereby improving the model's simulation accuracy for complex non-equilibrium heat transfer processes.

[0104] Model integrating deep neural network: In step S305, a deep neural network with a multi-layer perceptron structure of 3 hidden layers and 100 neurons in each layer is integrated into the model using a residual connection fusion strategy. This deep neural network inputs data such as temperature, humidity, and building structure parameters, extracts features through the hidden layer ReLU function, and predicts heat transfer related parameters in the output layer to enhance the prediction and analysis capabilities of the model.

[0105] Step S307: During training, the mean square error loss function is used to calculate the deviation between the model prediction value and the true value, and the stochastic gradient descent method is used to iteratively update the model parameters at a learning rate of 0.01 to reduce the loss function value.

[0106] Mean square error loss function: a common function used to measure the difference between the model's predicted value and the true value. It reflects the degree of deviation by calculating the average of the square of the difference between the predicted value and the true value. In the training of multi-physics field coupled heat transfer models, the accuracy of the model's prediction of heat transfer related parameters (such as heat flux density and temperature distribution) can be quantified. The smaller the loss value, the closer the model prediction is to the true value.

[0107] Model prediction value: After step S306, the pre-processed data is input into the constructed multi-physics field coupled heat transfer model, and the model outputs the results of the heat transfer related parameters, such as the predicted value of the heat flux density of the wall at a certain moment.

[0108] True value: The accurate value of the parameters related to heat transfer in the building wall obtained through actual measurement, experiment or authoritative data sources, such as the temperature and heat flux density data obtained by real-time monitoring by installing high-precision sensors on the wall, which serves as a reference standard for model training.

[0109] Stochastic gradient descent method: An optimization algorithm used to update parameters to minimize the value of the loss function during model training. Each time, it randomly selects a small batch of data samples from the training data set to calculate the gradient, and then updates the model parameters based on the gradient and the set learning rate. It has high computational efficiency and can quickly converge to a better parameter value on a large-scale data set.

[0110] Learning rate: A hyperparameter in the stochastic gradient descent method that controls the step size of each parameter update. This step is set to 0.01, which means that each time the parameters are updated, the model parameters move 0.01 times the gradient value in the opposite direction of the gradient. If the learning rate is too large, the model training may skip the optimal solution and fail to converge; if the learning rate is too small, the training will be slow, consuming a lot of time and computing resources.

[0111] The general process and examples are as follows: Take the training of a heat transfer simulation model for the wall of a commercial building as an example. First, obtain a batch of real data on the heat transfer of the building wall from the data acquisition system, covering parameter values ​​such as temperature and heat flux density at different locations. At the same time, run the constructed model, input the corresponding preprocessed data, and obtain the model's predicted values ​​for these heat transfer parameters. Then use the mean square error loss function to calculate the deviation between the predicted value and the true value. Then use the stochastic gradient descent method to update the model parameters at a learning rate of 0.01. By repeating this process continuously, a small batch of data is randomly selected from the training data set each time to calculate the loss and update the parameters, gradually reducing the loss function value, making the model prediction value closer to the true value, and improving the accuracy of the model's simulation of the heat transfer process of the commercial building wall.

[0112] Step S308: After every 10 parameter updates, the weight of the coupling strength of each physical field on the loss function is calculated based on the model prediction results and real data under the latest parameters. According to the weight, the coupling strength coefficient is adjusted by 10% to reduce the loss function value.

[0113] Parameter update: During the model training process, the model parameters (such as weights, biases, etc.) are adjusted according to the gradient calculated by the loss function based on the optimization algorithm (such as the stochastic gradient descent method in step S307). Each parameter update is intended to make the model prediction result closer to the true value and reduce the loss function value.

[0114] Model prediction results: Based on the current parameters, the model calculates the input data (such as environmental data related to building thermal management, building structure data, etc.) and outputs the predicted values ​​of heat transfer related parameters (such as heat flux density and temperature distribution).

[0115] Real data: Data obtained through actual measurements, experiments or reliable historical data records that accurately reflect the actual situation of heat transfer in building walls and are used to evaluate the accuracy of model predictions.

[0116] Physical field coupling strength: In a multi-physics field coupled heat transfer model, it is an indicator that describes the degree of interaction between different physical fields (such as heat conduction, convection, radiation, and phase change processes). In different building thermal management scenarios, the coupling strength of each physical field is different, which has an important impact on the accuracy of model prediction.

[0117] Impact weight: a value indicating the degree of influence of the coupling strength of each physical field on the loss function. The larger the weight, the more significant the change in the coupling strength of the physical field will be on the change in the loss function value.

[0118] Coupling strength coefficient: A parameter used to quantify the coupling strength of physical fields. By adjusting this coefficient, the coupling strength between physical fields can be changed. In the model, the coupling strength coefficient is associated with the relevant equations or calculation processes of each physical field. Changing the coefficient value can adjust the degree of interaction between physical fields.

[0119] Step S309: according to the dynamic change result of the scene within each interval time range, the physical field coupling strength is changed by a preset coupling strength adjustment function.

[0120] Interval time range: During model training and application, a time interval is set to capture the changes in building thermal management scenarios over time. For example, every 1 hour, 3 hours, etc. is used as an interval time range, and scenario data is collected and analyzed within this range.

[0121] Dynamic scene change results: refers to the changes in various factors in the building thermal management scene within each interval time range, including environmental factors (such as changes in temperature, humidity, and sunlight intensity), user behavior factors (such as equipment usage, changes in personnel activities), and building status factors (such as the increase or decrease of wall temperature over time), etc. These changes will affect the coupling of the physical field.

[0122] Preset coupling strength adjustment function: A predefined mathematical function when building a multi-physics field coupled heat transfer model. This function calculates the corresponding physical field coupling strength adjustment value based on various parameters in the scene dynamic change results (such as temperature change, device power change, etc.) to meet the heat transfer simulation requirements in different scenarios.

[0123] Physical field coupling strength: A quantitative indicator that describes the degree of interaction between multiple physical fields (such as heat conduction, convection, radiation, and phase change processes). In different scenarios, the contribution of each physical field to heat transfer is different, and the coupling strength will change accordingly, which in turn affects the accuracy of the model's simulation of the heat transfer process.

[0124] Step S30A, continuously monitor the mean square error, and when the mean square error change value is less than the preset change value in consecutive preset rounds of training, it is determined that the training is completed.

[0125] Continuous preset rounds of training: a pre-set number of consecutive trainings. Each round of training includes a series of operations such as inputting a batch of data into the model, performing forward propagation to calculate the prediction results, calculating the loss function (here is the mean square error), and updating the model parameters through back propagation. In this step, several consecutive rounds of training are used as the observation cycle to determine whether the model training has reached a stable state.

[0126] Mean square error change: In two consecutive rounds of training, the difference between the mean square error obtained in the latter round of training and the mean square error in the previous round. This value reflects the change in the prediction accuracy of the model after a round of training. If the change value is small, it means that the improvement of the model in this round of training is limited.

[0127] Preset change value: A threshold value set before model training begins based on experience and expectations of model performance. When the mean square error change values ​​in consecutive preset rounds of training are all less than this preset change value, the model is considered to have converged and achieved a good training effect, and training can be stopped.

[0128] The preset coupling strength adjustment function is as follows: ; For Moment, physical field With physical field Adjusted coupling strength coefficient; is the number of scenario factors that affect the coupling strength; is the weight of the kth scenario factor, reflecting the relative importance of this factor to the adjustment of coupling strength; is the kth scene factor At the moment The influence function is used to quantify the influence of the change of this factor on the coupling strength; is the kth scene factor at time The specific state or value of For physical field With physical field The basic coupling strength coefficient.

[0129] The multi-physics coupled heat transfer model calculates the heat transfer process of the fine units of the wall structure based on the finite element analysis method. By solving the multi-physics coupled equations, the heat transfer model analysis results include: Step S310: Fine unit division of the wall structure is performed based on a finite element analysis method according to a preset fine unit division basis.

[0130] Preset basis for fine unit division: determined by the model developer based on the design documents of the building wall, the principles of heat transfer, and previous experience in building similar models. The building design documents contain the material information, geometric dimensions, and other information of the wall, which are an important basis for formulating the basis for division. At the same time, the developer refers to the research results on the heat transfer characteristics of different materials and boundary conditions in heat transfer, and adjusts and improves the basis for division in combination with actual engineering needs. Finally, the determined basis for fine unit division is stored in the form of a text file or configuration file and read during the model building process.

[0131] Finite element analysis method: This can be achieved by using professional finite element analysis software or calling finite element analysis related libraries in programming languages. For example, commercial finite element analysis software such as ANSYS and ABAQUS provide a wealth of functions and tools to facilitate users to perform operations such as model building, unit division, and solution settings. In development based on programming languages ​​(such as Python), finite element analysis libraries such as FEniCS can be used to implement the application of finite element analysis methods by writing code. These software and libraries can be downloaded from the official website, and some commercial software requires the purchase of authorization.

[0132] Step S320: For each fine unit, a heat transfer equation is constructed according to the physical laws of heat conduction, convection, and radiation.

[0133] Heat conduction: The way heat is transferred from the hotter part of an object to the cooler part along the object. It is the main way of heat transfer in solids. It follows Fourier's law, that is, the amount of heat passing through a certain cross section per unit time is proportional to the temperature gradient at the cross section, and is related to the thermal conductivity of the material.

[0134] Convection: The process of heat transfer caused by relative displacement between parts of a fluid (gas or liquid) with different temperatures. In building wall heat transfer, it mainly involves convection heat transfer between the wall surface and the surrounding air. Convection heat transfer follows Newton's law of cooling, that is, the amount of convection heat transfer is related to the temperature difference between the fluid and the solid surface, the convection heat transfer coefficient, and the heat transfer area.

[0135] Radiation: The way an object transfers energy through electromagnetic waves. Thermal radiation does not require any medium and can also occur in a vacuum. The thermal radiation capacity of an object is related to factors such as the object's temperature and surface emissivity, and follows the Stefan-Boltzmann law, that is, the radiation heat flux density per unit surface area of ​​the object is proportional to the fourth power of the object's absolute temperature.

[0136] Heat transfer equation: A mathematical equation that comprehensively considers the three heat transfer modes of heat conduction, convection and radiation to describe the heat transfer law within a fine unit. By constructing the heat transfer equation, the heat transfer process of each fine unit under different boundary conditions and material properties can be quantitatively analyzed, providing a basis for subsequent multi-physics field coupling analysis.

[0137] Step S330, after completing the construction of the heat transfer equation of each tiny unit, construct the field coupling term based on the mutual influence between the physical fields, and add the coupling term to the unit equation.

[0138] Physical field: Here it mainly refers to the three physical fields of heat conduction, convection and radiation, which represent different methods of heat transfer.

[0139] Inter-field coupling term: a mathematical expression that reflects the mutual influence between various physical fields. In the actual heat transfer process, heat conduction, convection and radiation do not exist in isolation, but interact and influence each other. Inter-field coupling term is used to quantify this mutual influence.

[0140] Unit equation: that is, the heat transfer equation constructed in step S320. After adding the field coupling term, it can more accurately describe the heat transfer process of the micro unit under the interaction of multiple physical fields.

[0141] Step S340, assembling the unit equations after adding the coupling terms to form a multi-physics field coupling equation group of the entire wall structure.

[0142] Unit equation after adding coupling terms: refers to the equation obtained by adding field coupling terms to the heat transfer equation constructed by each micro-unit based on heat conduction, convection and radiation in step S330. These equations describe the heat transfer of each micro-unit under the interaction of multiple physical fields.

[0143] Assembly: The process of combining the equations of each tiny unit according to certain rules and order. In finite element analysis, this involves considering the connection relationship between units, boundary conditions and other factors so that the assembled equations can reflect the heat transfer characteristics of the entire wall structure.

[0144] Multi-physics coupling equations: It is composed of multiple small unit equations, which comprehensively consider heat conduction, convection, radiation and the coupling between them, and is used to comprehensively describe the heat transfer process of the entire wall structure under the influence of multiple physical fields. By solving this equation, the temperature distribution of each part of the wall, heat flux density and other heat transfer related parameters can be obtained.

[0145] The general process is as follows: 1. Read equations: Get the heat transfer equations of tiny units after adding coupling terms from the storage location. These equations describe the heat transfer under the interaction of multiple physical fields within the unit. 2. Read structural information: Read the wall structure grid file to clarify the connection relationship and boundary information between units, such as which units are connected through nodes. 3. Equation assembly: Integrate the unit equations according to the finite element assembly rules. For units with shared nodes, merge the related terms involving the same variables (such as temperature variables) at the node. 4. Form a group of equations: Complete the assembly of all unit equations and organize them into a mathematical form suitable for solving, such as the KX=F matrix form commonly used in finite element analysis.

[0146] Step S350, using a finite element solution method based on a preconditioned conjugate gradient method to iteratively solve the equation group to obtain a heat transfer model analysis result.

[0147] Preconditioned conjugate gradient method: an iterative algorithm for solving linear equations. In the multi-physics coupled heat transfer model, the assembled multi-physics coupled equations are usually large-scale linear equations, so direct solution is inefficient. The preconditioned conjugate gradient method introduces a preconditioning matrix to pre-process the original equations and improve the condition number of the equations, thereby accelerating the iterative convergence speed and improving the solution efficiency. This method gradually approaches the exact solution of the equations by calculating the conjugate direction and search step size in each iteration.

[0148] Finite element solution method: Based on the finite element analysis theory, a method that discretizes continuous physical problems into a finite number of units for solution. In the heat transfer model, by dividing the wall structure into small units, constructing unit equations and assembling them into a system of equations, the finite element solution method is used to obtain the numerical solution of the equation, and then the heat transfer related parameters of each part of the wall, such as temperature distribution, heat flux density, etc., are obtained.

[0149] Iterative solution: For complex equations, it is impossible to obtain the exact solution through direct calculation, but the exact solution is gradually approached by iteration. Each iteration updates the solution vector through a specific algorithm based on the result of the previous iteration until a certain convergence condition is met, and it is considered that a sufficiently accurate solution is obtained.

[0150] Heat transfer model analysis results: By solving the multi-physics field coupling equations, various parameters and information about the wall heat transfer process are obtained, such as the temperature values ​​at different locations of the wall, the heat flow transfer path and density distribution in the wall, etc. These results can help analyze the thermal performance of the wall and provide a basis for building thermal management.

[0151] According to different building types, usage scenarios and seasons, weights are set for various input data, including: Step S510, summarize the preprocessed multivariate data, and extract basic features directly related to building type, usage scenario, and season from the multivariate data.

[0152] Feature extraction: Carefully analyze multivariate data to identify data parts that are directly related to building type, usage scenario, and season. For example, building type may be related to information such as the building's structural form and wall materials; usage scenarios may involve data such as equipment operating status and personnel activity frequency; and seasons are related to data such as outdoor temperature and sunshine duration. Accurately extract these directly related data from the aggregated data as basic features.

[0153] Step S520, use a convolutional neural network to process the season-related data to obtain a seasonal feature vector, and use a long short-term memory network to analyze the usage scenario data, capture the characteristics of the usage scenario, and obtain a scenario feature vector.

[0154] Among them, Convolutional Neural Network (CNN) is a deep learning neural network designed specifically for processing grid-structured data (such as images and time series data). It automatically extracts features from input data through components such as convolutional layers, pooling layers, and fully connected layers. When processing seasonal data, convolutional neural networks can capture local patterns and features in the data, such as the periodic changes in temperature, sunshine duration, and other data caused by seasonal changes.

[0155] Season-related data: various data closely related to the seasons, such as environmental data such as average temperature, humidity, sunshine duration, precipitation in different seasons, and seasonal energy consumption pattern data, etc. These data reflect the impact of seasonal changes on the building environment and usage.

[0156] Seasonal feature vector: After being processed by the convolutional neural network, the seasonal data is converted into a numerical vector representation. This vector condenses the key feature information in the seasonal data, making it easier to merge and analyze with other features. Each element in the vector represents a different aspect of seasonal characteristics, and its numerical value reflects the strength of the feature.

[0157] Long Short-Term Memory (LSTM): A special type of recurrent neural network (RNN) that can effectively handle long-term dependencies in time series data. When analyzing usage scenario data, LSTM can remember past information and capture long-term features and trends in usage scenarios based on current input and past memory, such as changes in usage patterns of building equipment and personnel activity patterns over time in different time periods.

[0158] Usage scenario data: data that describes the building under different usage scenarios, including the operating status of equipment in the building (such as the start-up time and power consumption of air conditioners, lighting, elevators and other equipment), personnel activities (such as the number of people, activity areas, and activity times), and data related to the purpose of building use (such as customer flow in shopping malls and workload in offices).

[0159] Scenario feature vector: A numerical vector obtained by analyzing the usage scenario data through a long short-term memory network. It summarizes the key features in the usage scenario data, reflects the characteristics and laws of the usage scenario, and can be used for subsequent comprehensive feature analysis and weight generation.

[0160] Step S530: fuse the season feature vector and the scene feature vector to form a comprehensive feature.

[0161] Comprehensive features: New features formed by combining the seasonal feature vector and the scene feature vector in a certain way. The comprehensive features integrate the key information of both season and usage scenario, and can more comprehensively describe the complex environment in which the building is located, providing richer and more representative information for subsequent input into the dynamic weight generation model to determine the weights of each input data.

[0162] Step S540, input the comprehensive features into a dynamic weight generation model based on an attention mechanism neural network, and output the weights set for each input data.

[0163] Attention Mechanism Neural Network: A special neural network architecture that allows the model to automatically focus on the importance of different parts of the input data when processing input information. In this step, it will analyze the importance of each input data (such as building structure parameters, environmental parameters, usage scenario parameters, etc.) to the final result (such as the accuracy of the heat transfer model, the accuracy of energy consumption prediction, etc.) based on the comprehensive characteristics of the input, and assign different attention weights to different input data.

[0164] Dynamic weight generation model: A model built on an attention mechanism neural network, whose function is to generate dynamic weights for each input data based on the comprehensive characteristics of the input. These weights are not fixed, but will be dynamically adjusted with changes in building type, usage scenario and season to adapt to the differences in the degree of impact of each input data on the results in different situations.

[0165] Weighting of each input data: The model output is the weight value assigned to each input data (such as wall material parameters, indoor and outdoor temperature data, equipment operating power data, etc.) in the multi-physics field coupled heat transfer model or energy management model. The larger the weight, the more significant the impact of the input data on the model results under the current building type, usage scenario and season.

[0166] According to the mapping relationship between the range in which the sum of the input data falls and the level, the levels corresponding to the thermal conditions are analyzed and determined to include: Step S610, the acquired input data sum, building type, usage scenario, season and comprehensive features are input into the trained reinforcement learning model, and the reinforcement learning model selects an action according to the learned strategy to adjust the current input data sum interval range.

[0167] Reinforcement learning model: A machine learning model that continuously learns the optimal strategy based on reward signals fed back by the environment through interaction with the environment. In this step, the reinforcement learning model determines how to adjust the range of the input data sum to more accurately judge the thermal condition level by learning the relationship between the input data sum and thermal conditions under different building types, usage scenarios, seasons, and comprehensive characteristics.

[0168] Action: The decision taken by the reinforcement learning model in the current state (i.e., the state composed of the sum of input data, building type, usage scenario, season, and comprehensive characteristics), such as expanding or reducing the range of the sum of input data, or shifting the range. The model selects an action that it believes is most conducive to accurately judging the thermal condition level based on the learned strategy.

[0169] Suppose a thermal analysis is performed on an office building.

[0170] First, the sum of the current input data is queried from the database after data preprocessing. The sum is obtained by summing up various heat-related data such as indoor and outdoor temperature, office equipment operating power, and lighting equipment power. At the same time, it is read from the architectural design document that the office building belongs to the commercial office building type, and the property management system is learned that the current use scene is during the daytime on weekdays. The current season is determined to be summer through the meteorological data interface, and the comprehensive feature data generated and saved in step S530 is read from the local file system. The data integrates the summer season characteristics and the characteristics of the office building use scene during the daytime on weekdays.

[0171] Then, load the reinforcement learning model that was pre-trained and stored in the file (assuming PyTorch is used). Provide the acquired input data sum, building type, usage scenario, season, and comprehensive features as input information to the reinforcement learning model. The reinforcement learning model analyzes the state represented by the current input information based on the strategy learned from previous training. For example, based on past experience, the model knows that during the daytime on summer working days, office buildings are frequently active, many devices are running, the heat load is large, and the sum of input data is often high. Based on this, the model selects an action, such as expanding the range of the current sum of input data upward by a certain proportion as a whole, to better adapt to the characteristics of this thermal condition, and to prepare for the subsequent accurate judgment of the thermal condition level.

[0172] Step S620: Determine the corresponding thermal condition level as a preliminary judgment result according to the adjusted interval range.

[0173] Adjusted interval range: In step S610, the reinforcement learning model expands, shrinks or translates the original input data sum interval range according to the input information and the learned strategy to obtain a new interval range. This interval range is more suitable for the actual thermal conditions of the current building type, usage scenario and season.

[0174] Thermal condition level: A series of thermal condition classification levels are pre-set to facilitate the evaluation and management of building thermal conditions. For example, thermal condition levels can be divided into comfortable, relatively comfortable, uncomfortable, etc. Each level corresponds to a certain range of indoor and outdoor environmental parameters, energy consumption level, and human thermal perception. Different thermal condition levels reflect the quality of the building's thermal environment and its impact on energy efficiency.

[0175] Preliminary judgment result: According to the adjusted interval range, the corresponding thermal condition level is found in the pre-established interval range and thermal condition level mapping relationship. This result is based on the preliminary analysis of the model, and will be further optimized and confirmed through fuzzy processing and fuzzy reasoning.

[0176] Step S630, fuzzy processing is performed on the preliminary judgment result together with the sum of input data, environmental data, and user behavior data.

[0177] Fuzzification: The process of converting precise numerical values ​​or clear categories (such as preliminary judgment results, specific environmental data values) into fuzzy sets. In fuzzy sets, the degree of membership of an element to a set is not simply "belongs" or "does not belong", but is represented by a value between 0 and 1. Through fuzzification, uncertainty and ambiguity in data can be better handled, providing a suitable data form for subsequent reasoning based on fuzzy rules.

[0178] Step S640, performing reasoning based on a pre-built fuzzy rule base.

[0179] Pre-built fuzzy rule base: A set of rules established based on professional knowledge, practical experience and a large amount of experimental data in the field of building thermal engineering before thermal condition analysis. These rules are presented in the form of "if...then...", for example, "if the outdoor temperature is high and people are active, then the thermal condition may be hot". The fuzzy rule base associates the fuzzified input data (preliminary judgment results, the sum of input data, environmental data, and fuzzy membership information of user behavior data) with possible thermal condition levels to provide a basis for reasoning.

[0180] Reasoning: The process of analyzing and processing the fuzzified input data using the rules in the fuzzy rule base to obtain a more accurate judgment on the thermal condition level. The reasoning process is not a simple logical judgment, but a comprehensive consideration of multiple fuzzy input data and their membership, and the membership distribution of the final thermal condition level is determined based on the matching and calculation of the rules in the rule base.

[0181] Step S650, using a defuzzification method to convert the fuzzy judgment result into a clear thermal condition level.

[0182] Defuzzification method: a method to convert the fuzzy membership distribution of the thermal condition level obtained through fuzzy reasoning into a clear thermal condition level that can be used for practical decision-making. Since the result of fuzzy reasoning is the distribution of multiple thermal condition levels with different memberships, defuzzification is to extract a single and definite thermal condition level from these fuzzy information in order to clearly evaluate the thermal condition of the building.

[0183] Fuzzy judgment result: In step S640, the result is obtained by reasoning the fuzzified input data (preliminary judgment result, input data sum, environmental data, user behavior data) based on the fuzzy rule base. It is expressed as the membership value corresponding to each thermal condition level, reflecting the possibility of each thermal condition level under the current input data conditions.

[0184] Clear thermal condition level: A single thermal condition level determined after defuzzification, such as "comfortable", "uncomfortable", "overheated", etc. This level is a clear and definite description of the current thermal condition of the building, which can be directly used to guide subsequent building thermal management decisions, such as whether the air conditioning system operating parameters need to be adjusted, whether ventilation needs to be strengthened, etc.

[0185] Assuming that the thermal condition analysis of an office building is continued as an example, the fuzzy judgment result obtained after step S640 is {'comfortable': 0.48, 'hot': 0.35, 'uncomfortable': 0.17}.

[0186] If the maximum membership method is used for defuzzification, the maximum membership method selects the thermal condition level with the largest membership as the final clear thermal condition level. In the above fuzzy judgment results, the membership of "comfortable" is 0.48, which is the largest. Therefore, after defuzzification, it is determined that the current clear thermal condition level of the office building is "comfortable".

[0187] If the centroid method is used, a quantitative value must first be set for each thermal condition level. Assume that the quantitative value of "comfortable" is 1, the quantitative value of "hotter" is 2, and the quantitative value of "uncomfortable" is 3. According to the centroid method formula, the numerator is calculated as: 0.48×1+0.35×2+0.17×3=0.48+0.7+0.51=1.69, and the denominator is: 0.48+0.35+0.17=1, then the centroid value is 1.69÷1=1.69. According to the correspondence between the pre-set quantitative value and the thermal condition level, it is judged that 1.69 is closer to the quantitative value 1 corresponding to "comfortable", so the current clear thermal condition level of the office building is determined to be "comfortable". Through such a defuzzification process, the membership distribution of the thermal condition level obtained by fuzzy reasoning is converted into a clear thermal condition level, providing a clear decision-making basis for building thermal management.

[0188] The reinforcement learning model selects an action based on the learned strategy and adjusts the current input data sum range including: Step S611, the reinforcement learning model adopts a multi-layer perceptron structure. The input layer receives a state vector integrated by the sum of input data, building type, usage scenario, season, and comprehensive features extracted in the early stage. After nonlinear transformation of multiple hidden layers, the output layer calculates the corresponding Q value for various predefined actions, where various predefined actions include translation, scaling, splitting and merging.

[0189] Hidden layer: A neural network layer located between the input layer and the output layer. It transforms the input information through a nonlinear activation function, extracts more advanced features, and enhances the expressiveness of the model.

[0190] Nonlinear transformation: The operation of transforming neuron input using nonlinear activation functions (such as ReLU, Sigmoid, etc.) enables the neural network to learn complex nonlinear relationships and improve the model's ability to process data.

[0191] Output layer: The last layer of the multilayer perceptron calculates the Q value corresponding to various predefined actions based on the information output by the hidden layer.

[0192] Predefined actions: Pre-set operations on the range of the total input data interval, including translation (moving the interval as a whole by a certain value), scaling (proportionally expanding or reducing the interval range), splitting and merging (splitting the interval into multiple small intervals or merging multiple intervals). The model evaluates the pros and cons of these actions by calculating Q values.

[0193] Q value: It is used to measure the estimated value of the long-term cumulative reward of taking a predefined action in the current state. The higher the Q value, the more likely it is to get a better long-term result by taking the corresponding action in this state. The model selects the optimal action based on the Q value.

[0194] Step S612, use The policy determines the actions taken.

[0195] in, For the ε-greedy strategy, in order to balance exploration and utilization, the ε-greedy strategy randomly selects an action with a probability of ε, and selects the action with the highest Q value with a probability of 1-ε. Among them, ε is a parameter between 0 and 1, called the exploration rate. For example, when ε=0.1, the model has a 10% probability of randomly selecting an action and a 90% probability of selecting the action with the highest Q value. In this way, the model can not only use the learned knowledge to select the current optimal action, but also explore new actions with a certain probability to avoid falling into the local optimum.

[0196] Assume that a thermal condition analysis is performed on a hotel, step S611 has been completed, and the Q values ​​for predefined actions such as translation, scaling, splitting and merging are obtained, which are 0.7 (translation), 0.5 (scaling), and 0.3 (splitting and merging), respectively.

[0197] Load the reinforcement learning model stored in the file (assuming PyTorch is used). The policy adopted by the model is the ε-greedy policy with ε=0.2.

[0198] The model first generates a random number between 0 and 1, assuming that the generated random number is 0.15. Since this random number is less than the exploration rate ε (0.2), the model will select an action randomly. Randomly select one from the predefined actions (translation, scaling, split and merge), assuming that the scaling action is randomly selected.

[0199] If the generated random number is greater than 0.2, for example, 0.8, the model will follow the greedy strategy and select the action with the highest Q value, that is, the translation action. By using this strategy to determine the action to be taken, we can prepare for the subsequent adjustment of the input data sum interval range according to the action, so as to more accurately analyze the thermal conditions of the hotel.

[0200] Step S613, determining the interval adaptation strategy according to the determined mapping relationship between the action and the interval adaptation strategy, and applying it to the original interval to complete the adjustment of the interval range of the sum of the input data.

[0201] Interval adaptation strategy: For each predefined action, the specific adjustment method and rules are pre-set. For example, for a translation action, the interval adaptation strategy may stipulate that the interval as a whole should be moved to the left or right by a certain value; for a scaling action, it may stipulate that the interval range should be enlarged or reduced by a certain proportional factor; for a split and merge action, it will specify how to split the original interval into multiple small intervals, or how to merge multiple small intervals into a new interval.

[0202] Mapping relationship: The corresponding relationship between the action and the interval adaptation strategy. It associates the action determined in step S612 with the corresponding specific adjustment method, so that the model can find the corresponding interval adaptation strategy according to the selected action, thereby adjusting the input data sum interval range. This mapping relationship has been determined and stored in the model construction and training stage.

[0203] Original interval: The interval range of the sum of input data determined before step S611, which is the basis for adjustment. This interval range is set based on the initial input data and certain rules. As the thermal condition analysis process progresses, it needs to be adjusted based on the learning and judgment of the model.

[0204] Assume that a hotel is subjected to thermal condition analysis, and the action determined in step S612 is "zoom". First, read the mapping relationship between the action and the interval adaptation strategy from the file. Search the interval adaptation strategy corresponding to the "zoom" action in the file, and assume that the description found is "reduce the original interval range by a proportional factor of 0.8". Then, read the original interval range from the file, and assume that the original interval is [100,200]. According to the determined interval adaptation strategy, the original interval is scaled. The lower limit value is adjusted to 100×0.8=80, and the upper limit value is adjusted to 200×0.8=160, and the adjusted interval is [80,160]. Through such a process, according to the action determined in step S612, the corresponding interval adaptation strategy is found and applied to the original interval, completing the adjustment of the interval range of the sum of the input data, providing a data basis for a more accurate analysis of the hotel's thermal conditions in the future. For example, when judging the thermal condition level, the adjusted interval range can better reflect the actual thermal conditions of the current hotel.

[0205] Furthermore, the intelligent control method for building energy-saving walls also includes a distributed control method for building energy-saving walls based on intelligent edge computing, which is as follows: Distributed nodes are deployed on each floor or functional area of ​​the building. Each node collects real-time environmental (temperature, humidity, light, etc.), user behavior (personnel activities, equipment status, etc.) and heat transfer related (wall temperature, heat flux, etc.) parameters through the local sensor network. Each edge computing node independently conducts thermal demand analysis and thermal condition level judgment based on local training models and real-time data, and determines the preliminary control plan from the local strategy library, such as phase change material control on this floor and local lattice structure fine-tuning.

[0206] Every 15 minutes (preset interval), each distributed node sends local environmental data, thermal condition analysis results, and preliminary control decisions to adjacent nodes and the central control platform.

[0207] The central control platform and nodes complete the control decision conflict detection within 1 hour after receiving data based on the pre-built rule base and optimization algorithm. If a conflict is detected, the central control platform will use a multi-objective optimization algorithm (such as the number of genetic algorithm iterations set to 50 times) within 15 minutes to generate a coordinated control plan based on energy saving, comfort and other goals. After the central control platform generates the plan, it will be sent to each distributed node within 3 minutes. The node will complete the local decision adjustment within 2 minutes and feedback the execution readiness status.

[0208] After receiving the coordinated control plan, each distributed node sends a control instruction to the actuator within 5 minutes, and the actuator completes the adjustment of parameters such as phase change materials, lattice structure, and heat exchange fluid within 10 minutes.

[0209] After the adjustment is completed, the sensor collects the adjusted data in real time and feeds back the environment and heat transfer related data to the central control platform and nodes every 30 minutes. After receiving the feedback data, the node and platform complete the adjustment effect evaluation within 15 minutes based on the preset evaluation indicators (energy consumption reduction rate accurate to two decimal places, indoor temperature fluctuation range accurate to 0.1℃, etc.). If it does not meet the standards, repeat the conflict detection and subsequent processes.

[0210] The present application also provides a lattice structure wall, which includes a lattice structure frame and a phase change material filled in the internal space of the lattice structure frame. The lattice structure wall also includes a macro packaging structure that encapsulates the phase change material inside the lattice structure frame and concrete that is cast in a template embedded with the lattice structure frame to tightly combine the lattice with the concrete. The phase change material is paraffin or hydrated salt. The hydrated salt phase change material has a higher phase change latent heat, a greater heat storage density and better thermal conductivity, thereby further improving the heat storage and heat regulation capabilities of the wall. At the same time, compared to paraffin, hydrated salt phase change materials are more environmentally friendly and have lower costs, and are more suitable for wide application in the field of building energy conservation. The macro packaging process simplifies the filling and packaging steps, avoids the complex manufacturing process in microcapsule technology, and reduces production costs.

[0211] When the lattice structure wall is a single-layer lattice structure, its composition can refer to Figure 2 As shown, the lattice frame is manufactured using 3D printing technology, and a structure with high space utilization is formed through precise geometric design. The interior of the lattice structure is filled with phase change materials (such as paraffin or hydrated salts), which are injected through a specially designed inlet (upper side) and discharged or replaced through an independent outlet (lower side), forming an efficient dynamic regulation loop. The arrangement of the inlet and outlet ensures that the phase change material can be evenly distributed in the lattice structure, providing stable heat storage and release capabilities.

[0212] Figure 3 A partial enlarged image of the entrance of a single-layer lattice structure is shown. The lattice is filled with phase change materials (such as paraffin or hydrated salts) for efficient heat storage and release. The lattice provides a stable filling space for the phase change material and increases the thermal contact area with the outside world through optimized geometry, thereby improving the efficiency of heat storage and heat release. The lattice design simplifies the packaging process of the phase change material while ensuring its stability and efficiency in thermal management, which is suitable for various building energy-saving scenarios.

[0213] Furthermore, the lattice structure wall can also be a double-layer lattice structure. In this case, the lattice structure frame is a double-layer lattice structure frame. The inner lattice provides a stable filling space for the phase change material. The outer lattice is arranged around the inner layer and is provided with a heat exchange fluid circulation channel. The channel is filled with a heat exchange fluid. The heat exchange fluid is water or ethylene glycol solution, etc. For details, please refer to Figure 4 , Figure 4 The composition and functional design of the double-layer lattice structure are demonstrated.

[0214] Figure 5 The enlarged view of the entrance of the double-layer lattice structure is shown. The lattice is divided into an inner layer and an outer layer. The inner layer is filled with phase change materials (such as paraffin or hydrated salts); the outer layer is filled with heat exchange medium (such as water), and the circulation of the fluid is used to achieve rapid heat conduction and dissipation. The inner lattice has the same function as the single-layer lattice structure, and the outer lattice is designed as a fluid channel to accommodate and guide the flow of the heat exchange fluid to ensure that the heat can be quickly transferred to the external environment of the wall or the internal space.

[0215] The embodiments of this specific implementation method are all preferred embodiments of the present application, and are not intended to limit the protection scope of the present application. Therefore, all equivalent changes made based on the structure, shape, and principle of the present application should be included in the protection scope of the present application.

Claims

1. A method for intelligent management and control of building energy-saving walls, characterized in that: include: Obtain environmental data and user behavior data, and perform data preprocessing; Input environmental data and user behavior data into the pre-built heat demand prediction model to output building heat demand information; Collect and obtain real-time monitoring data of heat transfer related parameters, and input the real-time monitoring data into the pre-built multi-physics field coupled heat transfer model. The multi-physics field coupled heat transfer model calculates the heat transfer process of the fine units of the wall structure based on the finite element analysis method, and obtains the heat transfer model analysis results by solving the multi-physics field coupling equations. The heat transfer model analysis results include the temperature distribution at different locations of the wall at different times, the heat transfer rate, and the phase change state of the phase change material; Standardize the data included in the building heat demand information and the data included in the heat transfer model analysis results to make them uniform in scope; According to different building types, usage scenarios and seasons, weights are set for each input data, and the total input data is calculated using the weighted average method; According to the mapping relationship between the interval range that the sum of input data falls into and the level, the level corresponding to the thermal condition is analyzed and determined; According to the mapping relationship between the thermal condition level and the overall control plan, the overall control plan is analyzed and determined, and the determined overall control plan is executed. The overall control plan includes a phase change material control plan, a lattice structure control plan, and a heat exchange fluid control plan.

2. The intelligent management and control method for energy-saving building walls according to claim 1, characterized in that: The steps to build the heat demand forecasting model are as follows: Collect environmental data and user behavior data, and perform data preprocessing; Build a Transformer model based on spatiotemporal decoupling, randomly initialize the model parameters, and determine the model hyperparameters; Divide the preprocessed data into a training set and a validation set according to a preset ratio; The training set data is input into the Transformer model based on spatiotemporal decoupling, and the prediction results are calculated through forward propagation. The weighted mean square error loss function is used to calculate the difference between the prediction results and the actual heat demand value. In addition, a Dropout operation is performed during each forward propagation to obtain the prediction results under different parameter samples. Based on the results of the loss function, the gradient is calculated through the back-propagation algorithm to update the model parameters; After each training cycle is completed, the validation set loss is calculated. If the change in validation set loss for consecutive preset cycles is less than the set threshold, the training is considered completed.

3. The intelligent management and control method for energy-saving building walls according to claim 1, characterized in that: The steps to build a multiphysics coupled heat transfer model are as follows: Collect and obtain multi-source data and perform data preprocessing; Build a model architecture at the micro, meso and macro levels; According to the correspondence between the building thermal management scenario and the physical field coupling mode and intensity setting scheme, the physical field coupling mode and intensity of the current building thermal management scenario are analyzed and determined as the starting state of the model training; The nonequilibrium heat flow equation Integrate into the model heat transfer calculation system, where is the heat flux density, is the thermal conductivity, is the temperature gradient, is the pressure gradient, is the substance concentration gradient, is the pressure-dependent heat transfer coefficient, is the heat transfer coefficient related to the species concentration; A deep neural network with a multilayer perceptron structure of 3 hidden layers and 100 neurons in each layer was selected and integrated into the model using a residual connection fusion strategy. The input layer of the deep neural network received temperature, humidity, and building structure parameter data, the hidden layer used the ReLU function to extract features, and the output layer predicted heat transfer related parameters. Input the preprocessed data into a model that has a well-built architecture, a determined coupling method, incorporates non-equilibrium equations, and integrates a deep neural network; During training, the mean square error loss function is used to calculate the deviation between the model prediction value and the true value. The stochastic gradient descent method is used to iteratively update the model parameters at a learning rate of 0.01 to reduce the loss function value. After every 10 parameter updates, the weight of the influence of the coupling strength of each physical field on the loss function is calculated based on the model prediction results and real data under the latest parameters. According to the weight, the coupling strength coefficient is adjusted by 10% to reduce the loss function value. According to the dynamic change results of the scene within each interval time range, the physical field coupling strength is changed through a preset coupling strength adjustment function; The mean square error is continuously monitored. When the mean square error change value is less than the preset change value in consecutive preset rounds of training, the training is considered complete.

4. A building energy-saving wall intelligent management and control method according to claim 3, characterized in that: The preset coupling strength adjustment function is as follows: ； For Moment, physical field With physical field Adjusted coupling strength coefficient; is the number of scenario factors that affect the coupling strength; is the weight of the kth scenario factor, reflecting the relative importance of this factor in adjusting the coupling strength; is the kth scene factor At the moment The influence function is used to quantify the influence of the change of this factor on the coupling strength; is the kth scene factor at time The specific state or value of For physical field With physical field The basic coupling strength coefficient.

5. A building energy-saving wall intelligent management and control method according to claim 4, characterized in that: The multi-physics coupled heat transfer model calculates the heat transfer process of the fine units of the wall structure based on the finite element analysis method. By solving the multi-physics coupled equations, the heat transfer model analysis results include: According to the preset fine unit division basis, the wall structure is divided into fine units based on the finite element analysis method; For each fine unit, the heat transfer equation is constructed based on the physical laws of heat conduction, convection, and radiation; After completing the construction of the heat transfer equation for each tiny unit, the field coupling terms are constructed based on the mutual influence between the physical fields, and the coupling terms are added to the unit equation; Assemble the unit equations after adding coupling terms to form a multi-physics field coupling equation group of the entire wall structure; The finite element solution method based on the preconditioned conjugate gradient method is used to iteratively solve the equations and obtain the analysis results of the heat transfer model.

6. A building energy-saving wall intelligent management and control method according to claim 5, characterized in that: According to different building types, usage scenarios and seasons, weights are set for various input data, including: Aggregate the preprocessed multivariate data and extract basic features directly related to building type, usage scenario, and season from the multivariate data; The convolutional neural network is used to process the season-related data to obtain the season feature vector. At the same time, the long short-term memory network is used to analyze the usage scenario data, capture the characteristics of the usage scenario, and obtain the scenario feature vector. The seasonal feature vector and the scene feature vector are fused to form a comprehensive feature; The comprehensive features are input into the dynamic weight generation model based on the attention mechanism neural network, and the weights set for each input data are output.

7. A method for intelligent management and control of building energy-saving walls according to claim 6, characterized in that: According to the mapping relationship between the range in which the sum of the input data falls and the level, the levels corresponding to the thermal conditions are analyzed and determined to include: The acquired input data sum, building type, usage scenario, season and comprehensive features are input into the trained reinforcement learning model. The reinforcement learning model selects an action based on the learned strategy and adjusts the current input data sum range. According to the adjusted interval range, determine the corresponding thermal condition level as the preliminary judgment result; The preliminary judgment result is fuzzified together with the sum of input data, environmental data, and user behavior data; Reasoning based on a pre-built fuzzy rule base; The defuzzification method is used to transform the fuzzy judgment results into clear thermal condition levels.

8. The intelligent management and control method for energy-saving building walls according to claim 7, characterized in that: The reinforcement learning model selects an action based on the learned strategy and adjusts the current input data sum range including: The reinforcement learning model adopts a multi-layer perceptron structure. The input layer receives the state vector integrated by the sum of input data, building type, usage scenario, season, and comprehensive features extracted in the early stage. After nonlinear transformation of multiple hidden layers, the output layer calculates the corresponding Q value for various predefined actions, including translation, scaling, splitting and merging. use The strategy determines the actions to be taken; According to the mapping relationship between the determined action and the interval adaptation strategy, the interval adaptation strategy is determined and applied to the original interval to complete the adjustment of the interval range of the sum of the input data.

9. A lattice structure wall, characterized in that: According to the method described in any one of claims 1 to 8, the lattice structure wall comprises: Lattice structure framework; Phase change material, which is filled in the internal space of the lattice structure frame and is paraffin or hydrated salt, is used to store and release heat; A macro packaging structure for packaging the phase change material within a lattice structure framework; Concrete is poured into the formwork embedded with the lattice structure frame, so that the lattice and concrete are tightly combined.

10. The lattice structure wall according to claim 9, characterized in that: The lattice structure frame is a double-layer lattice structure frame, the inner lattice provides a stable filling space for the phase change material, the outer lattice is arranged around the inner layer, and a heat exchange fluid circulation channel is provided, and the channel is filled with heat exchange fluid.

Images