A building passive evaporative cooling system based on capillary water transport and interface evaporation

Abstract

The present application relates to the technical field of building cooling and energy saving, in particular to a building passive evaporation cooling system based on capillary water transport and interface evaporation, comprising a data acquisition module, a physical state sensing engine, a neural network prediction module, a water supply scheduling module and an execution control module. The physical state sensing engine is used to predict the interface dry time, and the neural network prediction module is used to generate a water content distribution heat map with physical consistency of Richards equation and Fourier heat conduction law. The water supply scheduling module dynamically corrects the water supply time according to the heat map and the dry time, and drives the intermittent water supply of the micro water pump. The present application realizes the deep coupling of physical mechanism and deep learning, solves the problem of imbalance between water supply and heat load matching under complex microclimate, and improves the evaporation cooling efficiency and energy utilization accuracy of the building facade.

Description

Technical Field

[0001] This invention relates to the field of building cooling and energy-saving technology, specifically to a passive evaporative cooling system for buildings based on capillary water transport and interfacial evaporation. Background Technology

[0002] Against the backdrop of global warming and the intensification of the urban heat island effect, building cooling energy consumption has increased significantly in summer. Utilizing the latent heat of vaporization of water for passive cooling has become an important means to reduce air conditioning load and improve building energy efficiency.

[0003] Traditional evaporative cooling technologies for building facades primarily rely on timed spraying or wet curtain systems. Existing control methods are mostly based on single hard threshold temperature and humidity triggers, lacking a deep understanding of the heat and moisture transport mechanisms within the porous media of building exterior walls. In actual operation, moisture migration within porous materials exhibits a highly nonlinear nature and is severely affected by microclimate factors such as solar radiation and wind speed, making it difficult for the system to accurately determine the physical critical point of moisture evaporation transitions. Due to the lack of effective physical mechanism constraints and spatiotemporal prediction capabilities, existing technologies are prone to problems such as insufficient localized cooling, excessive moisture leading to structural leakage, or water waste in complex environments. Furthermore, they struggle to coordinate the dynamic balance between building thermal inertia and energy supply status.

[0004] How to achieve accurate perception and spatiotemporal prediction of the evaporation process at the interface of unsaturated porous media on building facades, in order to solve the dynamic imbalance between moisture supply and transient heat load under complex microclimate disturbances, is an urgent technical problem to be solved.

[0005] To address this, a passive evaporative cooling system for buildings based on capillary water transport and interfacial evaporation is proposed. Summary of the Invention

[0006] The purpose of this invention is to provide a passive evaporative cooling system for buildings based on capillary water transport and interfacial evaporation, which achieves accurate perception and on-demand scheduling of hydrothermal status through the coupling of physical mechanisms and deep learning.

[0007] To achieve the above objectives, the present invention provides the following technical solution: A passive evaporative cooling system for buildings based on capillary water transport and interfacial evaporation includes: Data acquisition module: Real-time acquisition of building exterior wall surface temperature, humidity of porous absorbent material layer, and building microclimate environmental parameters composed of solar radiation intensity and wind speed; Physical state perception engine: Based on the Richards equation, a moisture migration model for unsaturated porous media is constructed. The humidity is used to calculate the capillary flux of liquid water permeating from the vertical capillary network to the surface of the porous absorbent material layer. By identifying the critical physical point of transition from the constant evaporation rate period to the decreasing evaporation rate period, the interface drying time is predicted. Neural network prediction module: The Richards equation, Fourier's law of heat conduction, and the convective heat transfer boundary conditions determined by the building microclimate environmental parameters are used as regularization constraints to embed the loss function. Combined with the temperature, the measurement point data obtained by the data acquisition module are nonlinearly mapped to generate a heat map of moisture content distribution covering the surface of the building exterior wall. Water supply scheduling module: connects the physical state perception engine and the neural network prediction module, and dynamically adjusts the preset water supply duration based on the water content distribution heat map and the interface drying time. Execution control module: Based on the water supply scheduling module, it drives the micro water pumps in the water supply unit to intermittently supply water to the capillary network on the facade.

[0008] Preferably, the data acquisition module includes a preprocessing unit, which uses an extended Kalman filter algorithm to smooth the acquired measurement point data. The preprocessing unit improves the confidence weight of the measurement point data in the current sampling period by introducing multiple decay factors with a weighted decay memory index, and suppresses nonlinear measurement noise caused by fluctuations in building microclimate environmental parameters.

[0009] Preferably, the unsaturated porous medium moisture migration model includes a capillary suction calculation unit, an evaporation phase change coupling unit, and a hydrothermal coupling evolution unit: the capillary suction calculation unit calculates the matrix potential gradient inside the porous absorbent material layer according to the Richards equation, determines the spontaneous migration flux of liquid water driven by capillary force, and obtains the capillary flux; the evaporation phase change coupling unit simulates the mass exchange at the gas-liquid interface, takes the surface moisture evaporation rate determined by the building microclimate environmental parameters as a negative mass source term, and performs mass conservation superposition with the capillary flux to predict the water content evolution trend of the gas-liquid interface; it determines the moment when the water content evolution trend decreases to a preset drying threshold and predicts the drying time of the interface; the hydrothermal coupling evolution unit simultaneously calculates the equivalent thermal conductivity of the porous absorbent material layer according to the water content evolution trend, and feeds back the equivalent thermal conductivity and the gas-liquid interface state determined by the evaporation phase change coupling unit to the neural network prediction module.

[0010] Preferably, the step of embedding the Richards equation, Fourier's law of heat conduction, and the convective heat transfer boundary conditions determined by the building microclimate environmental parameters as regularization constraints into the loss function includes: using an automatic differentiation method to calculate the derivatives of the water content output by the neural network prediction module with respect to time and spatial coordinates, constructing a water migration residual term that conforms to the Richards equation, and constructing a heat conduction residual term according to the Fourier law of heat conduction; transforming the convective heat transfer boundary conditions into an interface energy balance equation, calculating the deviation between the predicted temperature gradient and the environmental heat transfer determined by wind speed and solar radiation intensity, forming a boundary residual term; and using the water migration residual term, the heat conduction residual term, and the boundary residual term as regularization terms, weighted and summed with the mean square error term of the measurement point data to obtain a loss function used to update the neural network weights.

[0011] Preferably, the process of generating a moisture content distribution heatmap is as follows: a three-dimensional coordinate grid model of the building's exterior wall surface is established as the input coordinate set of the neural network prediction module; the preprocessed measurement point data is used as the true value label, and the neural network prediction module is used to perform spatiotemporal interpolation and nonlinear feature extraction to initially map and obtain the initial moisture content prediction values ​​of all grid nodes; the initial moisture content prediction values ​​are iteratively optimized using the loss function to ensure that the predicted values ​​of all grid nodes satisfy the conservation constraint of the moisture migration residual term in spatial distribution and the balance constraint of the heat conduction residual term and the boundary residual term in temporal evolution, thereby generating a continuous moisture content distribution heatmap with physical consistency.

[0012] Preferably, the specific steps for dynamically correcting the preset water supply duration include: extracting the predicted values ​​of each grid node in the moisture content distribution heatmap, calculating the spatial integral of the deviation between the predicted values ​​and the preset saturated moisture content threshold to obtain the global moisture deficit of the porous absorbent material layer at the current moment; calculating the water consumption rate under the current evaporation condition based on the interface drying time output by the physical state perception engine, and determining the theoretical water replenishment required to maintain a constant evaporation rate in combination with the global moisture deficit; mapping the theoretical water replenishment to the single pulse water supply time of the micro water pump, and adjusting the water supply frequency of the micro water pump according to the spatial non-uniformity of the moisture content distribution heatmap to update the preset water supply duration.

[0013] Preferably, the water supply scheduling module further includes an energy balance correction unit: The energy balance correction unit acquires the real-time remaining power of solar power and the predicted value of future light intensity, and outputs the predicted building heat load distribution for the next 24 hours based on historical meteorological sequences and wall thermal response characteristics. If a high-intensity heat load is predicted in the future and the current remaining power is sufficient, the energy balance correction unit instructs the water supply scheduling module to pre-wet the facade capillary network and use the thermal and humid inertia of the porous water-absorbing material layer for cold storage; if the current remaining power is lower than a preset threshold, the energy balance correction unit reduces the value of the deviation penalty weight of the water supply scheduling to reduce the single pulse water supply time to a minimum value that maintains basic evaporation requirements.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention introduces the Richards equation through a physical state perception engine, establishing a precise mathematical description of moisture migration within unsaturated porous media. Compared to traditional systems that rely solely on temperature and humidity threshold control, this invention can accurately identify the critical physical point where the evaporation rate transitions from a constant evaporation rate period to a decreasing evaporation rate period. This mechanism-based predictive capability allows the system to intervene with water replenishment before the evaporation interface completely dries out, ensuring that the porous absorbent material is always under optimal evaporative cooling conditions. This solves the water resource waste problems commonly found in traditional passive cooling systems caused by delayed water supply response or indiscriminate water supply.

[0015] 2. This invention embeds the Richards equation, Fourier's law of heat conduction, and convective heat transfer boundary conditions as regularization constraints into the neural network loss function to construct a physical information neural network prediction module. This architecture utilizes automatic differentiation technology to force the network prediction results to conform to the fundamental laws of thermodynamics and fluid mechanics, effectively overcoming the "physical illusion" caused by traditional data-driven models when sensor measurement points are sparse or affected by meteorological noise. By generating a physically consistent continuous water content distribution heat map, the system achieves a high-fidelity mapping from local measurement points to the global facade state, improving the robustness and sensing accuracy of the system operation in complex microclimate environments.

[0016] 3. This invention incorporates solar power supply status, light forecast, and the building's heat load distribution over the next 24 hours into the scheduling logic through an energy balance correction unit. The system utilizes the unique physical thermal and moisture inertia of the porous absorbent material layer to execute a "cold storage" pre-wetting strategy before power is sufficient and high loads occur, achieving spatiotemporal shift of cooling operations. Under energy-scarce conditions, it achieves simplified water supply by dynamically adjusting deviation penalty weights, minimizing active energy consumption while ensuring building thermal comfort. This proactive, forward-looking optimization strategy, compared to traditional reactive control, not only reduces pump operating power consumption but also achieves a multi-objective Pareto optimal balance between building energy efficiency, water conservation goals, and energy sustainability. Attached Figure Description

[0017] Figure 1 This is a system structure diagram of a passive evaporative cooling system for buildings based on capillary water transport and interfacial evaporation proposed in this invention; Figure 2This is a system flow diagram of a passive evaporative cooling system for buildings based on capillary water transport and interfacial evaporation proposed in this invention. Figure 3 This is a structural diagram of the unsaturated porous media moisture migration model proposed in this invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Example 1 Please see Figures 1 to 3 This invention provides a passive evaporative cooling system for buildings based on capillary water transport and interfacial evaporation, the technical solution of which is as follows: A passive evaporative cooling system for buildings based on capillary water transport and interfacial evaporation, such as Figures 1-2 As shown, it includes: Data acquisition module: Real-time acquisition of building exterior wall surface temperature, humidity of porous absorbent material layer, and building microclimate environmental parameters composed of solar radiation intensity and wind speed; Physical state perception engine: Based on the Richards equation, a moisture migration model for unsaturated porous media is constructed. The humidity is used to calculate the capillary flux of liquid water permeating from the vertical capillary network to the surface of the porous absorbent material layer. By identifying the critical physical point of transition from the constant evaporation rate period to the decreasing evaporation rate period, the interface drying time is predicted. Neural network prediction module: The Richards equation, Fourier's law of heat conduction, and the convective heat transfer boundary conditions determined by the building microclimate environmental parameters are used as regularization constraints to embed the loss function. Combined with the temperature, the measurement point data obtained by the data acquisition module are nonlinearly mapped to generate a heat map of moisture content distribution covering the surface of the building exterior wall. Water supply scheduling module: connects the physical state perception engine and the neural network prediction module, and dynamically adjusts the preset water supply duration based on the water content distribution heat map and the interface drying time. Execution control module: Based on the water supply scheduling module, it drives the micro water pumps in the water supply unit to intermittently supply water to the capillary network on the facade.

[0020] Furthermore, the data acquisition module includes a preprocessing unit, which uses an extended Kalman filter algorithm to smooth the acquired measurement point data. The preprocessing unit introduces a multiple decay factor with a weighted decay memory exponent to increase the confidence weight of the measurement point data in the current sampling period and suppress nonlinear measurement noise caused by fluctuations in building microclimate environmental parameters.

[0021] The extended Kalman filter algorithm introduces multiple decay factors during state updates. For the prediction error covariance matrix The correction is made, and the calculation formula is as follows: ; in, This represents the prediction error covariance matrix. Here is the state transition matrix. The process noise covariance matrix; This is the sampling sequence number at the current moment.

[0022] The multiple attenuation factors The construction method is as follows: ; Among them, the attenuation component of each dimension Obtained through the following formula: ; This is the residual vector for the current sampling period (the difference between the predicted value and the measured data at the actual measurement point). To measure the noise covariance matrix; The observation matrix; This refers to the weighted decay memory index, whose preset value range is: This index is obtained by fitting historical experimental data and is used to control the weight of the influence of historical data on the correction of the current state; The preset weight adjustment coefficients for each channel are obtained by sensitivity analysis of the response frequency of the building facade under different wind speed intensities.

[0023] The weight adjustment coefficients of each channel Satisfy normalization constraints .

[0024] In a set of typical embodiments of the present invention, for a standard terracotta panel facade, the weighting coefficient of each channel is... Typical value ranges are as follows: Temperature Channel The value is usually between 0.35 and 0.45. As the most direct indicator of energy state, its weight remains the highest under all operating conditions. Humidity Channel The value is usually taken as 0.2 to 0.3. When predicting the drying time of the interface, its weight increases dynamically as the moisture content of the material decreases. Solar radiation channel The value is usually between 0.15 and 0.25, and it mainly affects the heat flow balance prediction in the boundary residual term. Wind speed channel Its weights exhibit high sensitivity to wind speed fluctuations; when wind speed... hour, It is recommended to raise it to above 0.15 to correct the calculation deviation of the convective heat transfer coefficient.

[0025] Based on the above sensitivity analysis results, technicians can fine-tune the parameters for specific building geometries.

[0026] This invention improves the system's response accuracy to transient changes in building facades by introducing multiple attenuation factors with weighted attenuation memory exponents into the extended Kalman filter. The algorithm dynamically adjusts the prediction error covariance matrix based on the residual vector of the current sampling period, automatically increasing the confidence weight of real-time measurement data and suppressing nonlinear measurement noise when the building's microclimate environment fluctuates drastically. Compared to traditional filtering algorithms, this scheme effectively solves the problem of accumulated errors caused by model linearization, ensuring the smoothness and robustness of the input source to the neural network prediction module, and laying a data foundation for achieving high-fidelity moisture content heatmap mapping.

[0027] Furthermore, such as Figure 3 As shown, the unsaturated porous medium moisture migration model includes a capillary suction calculation unit, an evaporation phase change coupling unit, and a hydrothermal coupling evolution unit: the capillary suction calculation unit calculates the matrix potential gradient inside the porous water-absorbing material layer according to the Richards equation, determines the spontaneous migration flux of liquid water driven by capillary force, and obtains the capillary flux; the evaporation phase change coupling unit simulates the mass exchange at the gas-liquid interface, takes the surface moisture evaporation rate determined by the building microclimate environment parameters as a negative mass source term, and performs mass conservation superposition with the capillary flux to predict the water content evolution trend of the gas-liquid interface; it determines the moment when the water content evolution trend decreases to a preset drying threshold and predicts the drying time of the interface; the hydrothermal coupling evolution unit simultaneously calculates the equivalent thermal conductivity of the porous water-absorbing material layer according to the water content evolution trend, and feeds back the equivalent thermal conductivity and the gas-liquid interface state determined by the evaporation phase change coupling unit to the neural network prediction module.

[0028] In the capillary suction calculation unit, the capillary flux The calculation follows the evolution of Darcy's law in unsaturated media, and the formula is: ;in, This refers to the volumetric water content. The unsaturated permeability coefficient; The matrix potential, determined by the Richards equation, is used to characterize the ability of water to be transported from the internal grid to the evaporation interface.

[0029] In the evaporation phase change coupling unit, the discretized expression of the mass conservation superposition is as follows: ; in, The surface moisture evaporation rate is calculated from the building microclimate environmental parameters using the Bowen ratio energy balance method. To calculate the step size; The thickness of the interface layer feature.

[0030] The preset drying threshold The range of values ​​is to ,in The saturated water content of the material is denoted as α; this threshold is obtained by measuring the critical water content at which the evaporation rate of the porous material deviates significantly from a constant rate under controlled laboratory conditions.

[0031] The hydrothermal coupling evolution unit synchronously calculates the equivalent thermal conductivity using a dynamic thermal bridge correction formula. The formula is as follows: ; in, and The thermal conductivity of the material was experimentally determined in both completely dry and saturated states. The material porosity distribution characteristic index (range of values) ).

[0032] The material pore distribution characteristic index It reflects the connectivity and uniformity of the distribution of micropores within the porous medium.

[0033] For typical building facade materials, parameters The selection criteria are as follows: Porous clay slabs / bricks: Pore distribution is relatively uniform and capillary action is significant; parameters... The recommended value is 1.5 to 1.8; Cement fiberboard / aerated concrete: Due to the presence of numerous isolated micropores or irregular large pores, the rate of thermal bridge formation is relatively slow, and the parameters... The recommended value is 2.0 to 2.5.

[0034] When the moisture content is lower than the preset drying threshold, the equivalent thermal conductivity is directly taken as the experimentally measured value under the dry state; when the moisture content is higher than the threshold, the power function interpolation calculation is performed using the ratio of the difference between the moisture content and the threshold to the effective moisture content range.

[0035] This invention achieves in-depth quantification of the moisture transport mechanism on building facades by constructing a precise physical model incorporating capillary suction, evaporative phase change, and hydrothermal coupling units. Utilizing the Richards equation and dynamic thermal bridge correction formula, the system can accurately capture the nonlinear relationship between capillary flux and equivalent thermal conductivity with moisture content, thereby accurately predicting interface drying time. This mechanism-driven computational approach not only provides physical prior constraints for neural networks but also ensures the scientific rigor and predictability of water supply scheduling in complex microclimate environments, improving the system's evaporative cooling efficiency and operational stability.

[0036] Further, the step of embedding the Richards equation, Fourier's law of heat conduction, and the convective heat transfer boundary conditions determined by the building microclimate environmental parameters as regularization constraints into the loss function includes: using an automatic differentiation method to calculate the derivatives of the water content output by the neural network prediction module with respect to time and spatial coordinates, constructing a water migration residual term that conforms to the Richards equation, and constructing a heat conduction residual term according to the Fourier law of heat conduction; transforming the convective heat transfer boundary conditions into an interface energy balance equation, calculating the deviation between the predicted temperature gradient and the environmental heat transfer determined by wind speed and solar radiation intensity, forming a boundary residual term; and using the water migration residual term, the heat conduction residual term, and the boundary residual term as regularization terms, weighted and summed with the mean square error term of the measurement point data to obtain the loss function used to update the neural network weights.

[0037] The moisture migration residual term Based on the Richards equation, grid points are calculated using automatic differentiation. The partial derivative at point is given by the formula: ; in, The water content is the output of the neural network; The unsaturated permeability coefficient is fed back by the physical state perception engine; This represents the matrix potential.

[0038] In a single training iteration (r-th round) of a neural network, the unsaturated permeation coefficient required by the loss function and matrix potential A delayed update strategy is adopted. Specifically, the physical state awareness engine performs mechanism calculations based on the full-grid water content prediction values ​​output by the neural network in the previous round (round r-1), and feeds back the calculated spatial distribution parameters to the residual term construction of the current round. In the initial training phase (round 0), the system performs physical state initialization: The value is uniformly taken as the permeability coefficient constant under the material's saturated state. The value is the matrix potential mapping value corresponding to the initial water content. This asynchronous alternating update strategy transforms the solution of complex nonlinear coupled partial differential equations into an asymptotic optimization process of "prediction-verification-re-prediction", effectively avoiding circular dependency deadlock in forward propagation. With adaptive learning rate adjustment, it ensures smooth convergence of the water migration residual term under nonlinear physical property constraints.

[0039] Thermal conduction residual term Based on Fourier's law, the formula is: ; in, Density of porous materials; Specific heat capacity; The equivalent thermal conductivity is fed back in real time by the hydrothermal coupling evolution unit; The density of liquid water. Internal evaporation rate. It is characterized by the rate of change of moisture content over time.

[0040] Boundary residuals Constructed using the following formula: ; in, The convective heat transfer coefficient is determined by a preset formula. get, For measured wind speed, constant , , Determined by fitting data from building wind tunnel experiments; Solar radiation intensity; The latent heat of vaporization of water; This represents the real-time evaporation rate. Among them, The predicted surface temperature of the building facade by a neural network; The equivalent blackbody radiation temperature of the sky is determined by environmental parameters.

[0041] In a set of typical embodiments of the present invention, for a flat building absorbent material layer facade: constant The value of is typically between 5.0 and 6.0, representing the baseline for natural convection under windless conditions; constant The value of is typically between 3.0 and 4.5, representing the gain coefficient of forced convection; constant. The wind speed power exponent is typically taken as 0.8 to 1.0 based on the flow field characteristics.

[0042] The loss function The overall expression is: ; in, For the weight of the data item, The weights are for the physical constraint terms. This is a data-driven loss term used to measure the degree of dispersion between the predicted output of the neural network and the measured values ​​of the sensor (i.e., the preprocessed measurement data).

[0043] Constructing the loss function At that time, the data-driven loss term Quantization is performed using a mean squared error algorithm. In practice, this measure simultaneously covers the deviation between the moisture content predicted by the neural network and the datum temperature at the measurement point. Before calculating the degree of dispersion, the moisture content deviation and temperature deviation are weighted and combined using a preset normalization coefficient. The principle of this weighted combination is to use the reciprocal of the sensor's range as the initial weight, ensuring that the two types of errors are of equal magnitude, thereby guiding the neural network to consider the fitting accuracy of both the water and thermal physical fields during training.

[0044] The weight of the data item Weights of physical constraint terms A dynamic balancing strategy is adopted. In the early stage of training, the ratio of the gradients of each residual term is dynamically adjusted by calculating the norm of the gradient. This ensures that the physical constraint term will not cause the model to ignore the measured data due to its initial magnitude being too large, nor will it degenerate into a purely data-driven model due to its magnitude being too small. This ensures that the generated distribution heatmap has physical consistency.

[0045] This invention constructs a loss function with physical mechanism constraints by transforming the Richards equation, Fourier's law of heat conduction, and interface energy balance equation into physical residual terms. The design utilizes automatic differentiation technology to decouple data-driven and physical criteria optimization, forcing the neural network output to conform to the laws of mass and energy conservation. By introducing a weighted summation of the data-driven loss term and the physical regularization term, the prediction bias problem under sparse sensor measurement points or meteorological noise interference is effectively solved. This improves the physical consistency and prediction robustness of the generated water content distribution heatmap, providing a decision-making basis for precise evaporation and cooling scheduling under complex microclimates.

[0046] Further, the process of generating a moisture content distribution heatmap is as follows: a three-dimensional coordinate grid model of the building's exterior wall surface is established as the input coordinate set of the neural network prediction module; the preprocessed measurement point data is used as the true value label, and the neural network prediction module is used to perform spatiotemporal interpolation and nonlinear feature extraction to initially map and obtain the initial moisture content prediction values ​​for all grid nodes; the initial moisture content prediction values ​​are iteratively optimized using the loss function to ensure that the predicted values ​​for all grid nodes satisfy the conservation constraint of the moisture migration residual term in spatial distribution and the balance constraint of the heat conduction residual term and the boundary residual term in temporal evolution, thereby generating a continuous moisture content distribution heatmap with physical consistency.

[0047] The neural network prediction module is constructed as a water content mapping function. ,in These are the learnable weights of the network. The specific steps in the generation process include: Before inputting the input coordinate set into the network, the three-dimensional coordinates of the building's exterior walls are first determined. and time Perform normalization. The normalization formula is: ; ; in, , The geometric boundary span of the building facade. To predict the period. Normalization ensures that the neural network performs a high-sensitivity mapping within the linear region of the activation function.

[0048] The neural network prediction module adopts a fully connected deep neural network architecture, which includes one input layer (receiving normalized x, y, z, t coordinates), five hidden layers (each layer has 128, 128, 64, 64, and 32 neurons respectively, and uses the Tanh activation function to ensure the continuity of higher-order derivatives and meet the requirements of automatic differentiation), and one output layer (outputting two predicted values: water content and temperature).

[0049] The nonlinear feature extraction is achieved by spreading the residual information from the measured points to the entire grid. Initial values ​​for water content prediction at all grid nodes. The calculation method is as follows: ; in, For grid node coordinates, The coordinates of the measuring point are... The total number of measurement points. This is a distance-based Gaussian kernel function. This formula ensures that the initial heatmap values ​​are forced to revert to the true values ​​at the measurement points, while intelligent interpolation is performed in non-measurement point areas using the generalization ability of neural networks; based on the density of sensor measurement points, [the following is applied]. The preset value is 0.5 to 1.5 times the average distance between adjacent measurement points (for example, 1.0 times the average spacing) to ensure the spatial smoothness and continuity of the residual field without distortion.

[0050] Using the Adam optimizer to optimize network weights The update is performed, and the optimization criterion is to minimize the total loss function L. During the iteration process, the output gradients of all grid nodes are required to satisfy the following physical conservation consistency criterion: ; in, For the residual term of the aforementioned Richards equation, The preset convergence tolerance (range of values) ).

[0051] The initial learning rate of the Adam optimizer is set to... The training employs a full-batch training approach (inputting all grid nodes and measurement points as a single batch), with a maximum training epoch set at 20,000 epochs. A learning rate decay strategy is used: the learning rate is multiplied by a decay factor of 0.9 every 2,000 epochs. When the total loss function L decreases by less than 0.9 for 1,000 consecutive epochs... Early stopping is triggered, and the network weights corresponding to the lowest loss function are saved as the final model parameters.

[0052] This invention achieves intelligent fusion of discrete measurement point data and the full-grid continuous field by constructing a three-dimensional coordinate grid model and a moisture content mapping function, combined with a Gaussian kernel residual diffusion mechanism, ensuring accurate regression of initial heat map values ​​at the measurement points. Iterative optimization using the Adam optimizer and physical conservation consistency criteria forces the full-grid prediction values ​​to strictly adhere to the Richards equation and energy balance constraints in their spatiotemporal evolution. This method effectively eliminates the physical distortion of traditional interpolation algorithms, generating a moisture content distribution heat map with extremely high physical consistency and spatiotemporal resolution, providing reliable data support for accurate moisture prediction and efficient cooling scheduling of building facades.

[0053] Furthermore, the specific steps for dynamically correcting the preset water supply duration include: extracting the predicted values ​​of each grid node in the moisture content distribution heatmap, calculating the spatial integral of the deviation between the predicted values ​​and the preset saturated moisture content threshold to obtain the global moisture deficit of the porous absorbent material layer at the current moment; calculating the water consumption rate under the current evaporation condition based on the interface drying time output by the physical state perception engine, and determining the theoretical water replenishment required to maintain a constant evaporation rate in combination with the global moisture deficit; mapping the theoretical water replenishment to the single pulse water supply time of the micro water pump, and adjusting the water supply frequency of the micro water pump according to the spatial non-uniformity of the moisture content distribution heatmap to update the preset water supply duration.

[0054] The deviation spatial integral is achieved by performing volume integration on the building facade mesh model, and the calculation formula is as follows: ; in, The preset saturated moisture content threshold; The predicted value of the grid node in the moisture content distribution heatmap; This represents the total volume of the porous absorbent material layer. It also represents the volume of water required to reach full saturation of the current system.

[0055] Combined with the interface drying time Theoretical water replenishment The calculation formula is as follows: ; in, The water consumption rate under the current evaporation conditions is given by the formula. get; This is the preset duration of the intermittent water supply cycle.

[0056] The single pulse water supply time of the micro water pump Determined according to the following formula: ; in, The number of pulses within a water supply cycle; This is the rated flow rate of the water pump.

[0057] The water supply frequency Correction coefficient According to the spatial standard deviation of the moisture content distribution heatmap Perform dynamic compensation: ; in, The reference water supply frequency; The average moisture content across the entire grid; The non-uniformity sensitivity coefficient (range of values) ).

[0058] Under cloudy or low-wind conditions, the evaporation process is relatively uniform in space, with small local differences in the moisture content heatmap. Therefore, frequent adjustments to the water supply frequency are unnecessary. It can take the value 0.1.

[0059] Typical summer weather conditions, such as sunny skies or strong crosswinds, present significant differences in evaporation intensity between sunlit and shaded surfaces, with a large spatial standard deviation. The system needs to be highly sensitive to significantly increase the size of the system. The value is used to amplify the frequency correction signal to prevent local areas from entering the dry period prematurely. It can take the value 0.3.

[0060] This invention quantifies the global water deficit of porous materials by spatially integrating a heat map of moisture content distribution, and achieves accurate feedforward calculation of theoretical water replenishment by combining the interface drying time. Utilizing a pulsed water supply mapping model and a frequency correction mechanism based on spatial standard deviation, the system can dynamically adjust pump operating parameters according to the evaporation intensity under different meteorological conditions. Compared to traditional timed spraying methods, this solution improves water use efficiency, effectively maintaining a constant rate of high-efficiency evaporation on the facade for extended periods, and preventing water waste caused by localized drying leading to cooling failure or excessive water supply.

[0061] Furthermore, the water supply scheduling module also includes an energy balance correction unit: The energy balance correction unit acquires the real-time remaining power of solar power and the predicted value of future light intensity, and outputs the predicted building heat load distribution for the next 24 hours based on historical meteorological sequences and wall thermal response characteristics. If a high-intensity heat load is predicted in the future and the current remaining power is sufficient, the energy balance correction unit instructs the water supply scheduling module to pre-wet the facade capillary network and use the thermal and humid inertia of the porous water-absorbing material layer for cold storage; if the current remaining power is lower than a preset threshold, the energy balance correction unit reduces the value of the deviation penalty weight of the water supply scheduling to reduce the single pulse water supply time to a minimum value that maintains basic evaporation requirements.

[0062] The neural network prediction module integrates a time-series feature processing unit, which employs a two-layer stacked LSTM network, with each layer containing 64 hidden units, based on hourly outdoor temperatures over the past 72 hours. With solar radiation intensity The input sequence is the heat load prediction sequence for the next 24 hours. The output of the LSTM is mapped through a fully connected layer to a backbone network shared with the moisture content mapping network, and jointly trained to establish the heat load prediction function. ; in, This is a prediction model with learned weights; and These are the time-series forecasts of outdoor temperature and light intensity for the next 24 hours; Set a baseline temperature for the indoor environment; This is the characteristic constant of the wall's thermal response (determined by the material's thermal diffusivity).

[0063] The pre-wetting and cold storage strategy must be triggered simultaneously to meet the following logical conditions: (1) Heat load criterion: , .in, The threshold for high-intensity heat load is set to 80% of the building's design cooling load.

[0064] (2) .in, To show the real-time remaining battery power, This represents the total capacity of the system's energy storage devices. This threshold ensures that cold storage operations do not affect the subsequent basic operation of the system.

[0065] when When the water supply is below a preset threshold (e.g., 20%), the water supply scheduling module regenerates the water supply duration by minimizing the following objective function J: ; in, The deviation penalty weight; This is the corrected actual water supply. This refers to the theoretical water replenishment amount; This refers to the power consumption of the water pump.

[0066] The specific numerical representation of the reduced deviation penalty weight is as follows: when the power is low, let The attenuation slope coefficient This is used to control the smoothness of water supply time reduction when power consumption decreases, with a value ranging from 5 to 15. At this time, the single pulse water supply time... Reduced to: ; The starting threshold time for a micro water pump refers to the minimum energizing time required to maintain a stable flow rate. It is experimentally determined to be between 0.5s and 1.2s. The energy warning trigger threshold is usually taken as the total system capacity. 20%.

[0067] This invention achieves integrated coordinated scheduling of water, energy, and heat through an energy balance correction unit. Utilizing a neural network for forward-looking prediction of future heat load, the system can pre-wet the facade capillary network during periods of abundant power supply, leveraging the thermal and hygroscopic inertia of porous materials to achieve spatiotemporal shift of cooling capacity, effectively reducing the cooling load during peak afternoon electricity consumption. When power is limited, a dynamic correction mechanism with deviation penalty weights allows the system to autonomously switch to a simplified water supply mode, minimizing operating energy consumption while ensuring system continuity. This strategy enhances the system's self-sufficiency and operational flexibility under off-grid or fluctuating energy conditions.

[0068] This system achieves high-fidelity perception and precise scheduling of building facade evaporation status through deep coupling of physical mechanisms and neural networks. By utilizing the Richards equation and critical point identification during constant evaporation rates, it overcomes the shortcomings of traditional systems that cannot determine the timing of interface drying, ensuring that materials are always in the efficient evaporation phase. Simultaneously, the regularized constrained neural network improves the prediction accuracy of the moisture content distribution heatmap, solving the problem of nonlinear mapping under sparse sensor data. Finally, by dynamically adjusting the water supply duration, it achieves an optimal match between cooling performance and resource consumption, reducing system operating energy consumption.

[0069] Example 2 In a typical sunny summer weather scenario, the specific workflow of this system is as follows: During the early morning hours, the data acquisition module uses sensors deployed on the west-facing facade of the building to acquire the initial temperature of the exterior wall base and the humidity of the porous absorbent material layer in real time. The preprocessing unit activates the extended Kalman filter algorithm, which automatically filters out microclimate parameter fluctuations caused by early morning gusts by introducing a weighted attenuation memory exponent, providing smooth measurement data for subsequent calculations.

[0070] As solar radiation intensity gradually increases, the physical state perception engine begins high-frequency calculations based on the Richards equation. The capillary suction calculation unit determines that liquid water is migrating to the outer surface at a stable capillary flux by analyzing the matrix potential gradient inside the material. At this time, the evaporation phase change coupling unit detects that the gas-liquid interface is in a constant evaporation rate period. By identifying the water content evolution trend, the physical engine predicts that the interface will reach the preset drying threshold at 2 PM and outputs the specific interface drying time.

[0071] Meanwhile, the neural network prediction module combines the acquired temperature data with the equivalent thermal conductivity feedback from the physics engine. Through automatic differentiation, the moisture migration residual term and the heat conduction residual term in the loss function work together to ensure that the moisture content distribution heatmap output by the neural network strictly adheres to the conservation law in space. The energy balance correction unit simultaneously acquires the power information of the solar power supply unit, detects that the current remaining power is sufficient and that a high-intensity heat load will occur in the next 4 hours, and then issues a "pre-wetting" command.

[0072] Upon receiving the instruction, the water supply scheduling module extracts the full-grid predicted value from the moisture content distribution heatmap and calculates the current global moisture deficit through spatial integration. Combined with the predicted interface drying time, the scheduling module calculates the theoretical water replenishment required to maintain efficient cooling and maps it to instructions from the execution control module. The execution control module drives micro-pumps to inject water into the facade capillary network at specific pulse water supply intervals. The porous absorbent material layer fully utilizes its thermal and hygroscopic inertia to achieve cold storage, enabling the building facade to maintain excellent interface evaporative cooling even during periods of highest temperature, effectively resisting external thermal shock.

[0073] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A passive evaporative cooling system for buildings based on capillary water transport and interfacial evaporation, characterized in that, include: Data acquisition module: Real-time acquisition of building exterior wall surface temperature, humidity of porous absorbent material layer, and building microclimate environmental parameters composed of solar radiation intensity and wind speed; Physical state perception engine: Based on the Richards equation, a moisture migration model for unsaturated porous media is constructed. The humidity is used to calculate the capillary flux of liquid water permeating from the vertical capillary network to the surface of the porous absorbent material layer. By identifying the critical physical point of transition from the constant evaporation rate period to the decreasing evaporation rate period, the interface drying time is predicted. Neural network prediction module: The Richards equation, Fourier's law of heat conduction, and the convective heat transfer boundary conditions determined by the building microclimate environmental parameters are used as regularization constraints to embed the loss function. Combined with the temperature, the measurement point data obtained by the data acquisition module are nonlinearly mapped to generate a heat map of moisture content distribution covering the surface of the building exterior wall. Water supply scheduling module: connects the physical state perception engine and the neural network prediction module, and dynamically adjusts the preset water supply duration based on the water content distribution heat map and the interface drying time. Execution control module: Based on the water supply scheduling module, it drives the micro water pumps in the water supply unit to intermittently supply water to the capillary network on the facade.

2. The passive evaporative cooling system for buildings based on capillary water transport and interfacial evaporation according to claim 1, characterized in that, The data acquisition module includes a preprocessing unit, which uses an extended Kalman filter algorithm to smooth the acquired measurement point data. The preprocessing unit introduces multiple decay factors with a weighted decay memory exponent to increase the confidence weight of the measurement point data in the current sampling period and suppress nonlinear measurement noise caused by fluctuations in building microclimate environmental parameters.

3. The passive evaporative cooling system for buildings based on capillary water transport and interfacial evaporation according to claim 1, characterized in that, The unsaturated porous medium moisture migration model includes a capillary suction calculation unit, an evaporation phase change coupling unit, and a hydrothermal coupling evolution unit: the capillary suction calculation unit calculates the matrix potential gradient inside the porous water-absorbing material layer according to the Richards equation, determines the spontaneous migration flux of liquid water driven by capillary force, and obtains the capillary flux; the evaporation phase change coupling unit simulates the mass exchange at the gas-liquid interface, takes the surface moisture evaporation rate determined by the building microclimate environment parameters as a negative mass source term, and performs mass conservation superposition with the capillary flux to predict the water content evolution trend of the gas-liquid interface; Determine the moment when the moisture content evolution trend decreases to a preset drying threshold, and predict the interface drying time; The hydrothermal coupling evolution unit simultaneously calculates the equivalent thermal conductivity of the porous water-absorbing material layer based on the water content evolution trend, and feeds back the equivalent thermal conductivity and the gas-liquid interface state determined by the evaporation phase change coupling unit to the neural network prediction module.

4. A passive evaporative cooling system for buildings based on capillary water transport and interfacial evaporation according to claim 1, characterized in that, The steps of embedding the Richards equation, Fourier's law of heat conduction, and the convective heat transfer boundary conditions determined by the building microclimate environmental parameters as regularization constraints into the loss function include: using an automatic differentiation method to calculate the derivatives of the water content output by the neural network prediction module with respect to time and spatial coordinates, constructing a water migration residual term that conforms to the Richards equation, and constructing a heat conduction residual term according to the Fourier law of heat conduction; transforming the convective heat transfer boundary conditions into an interface energy balance equation, calculating the deviation between the predicted temperature gradient and the environmental heat transfer determined by wind speed and solar radiation intensity, forming a boundary residual term; and using the water migration residual term, the heat conduction residual term, and the boundary residual term as regularization terms, weighted and summed with the mean square error term of the measurement point data to obtain the loss function used to update the neural network weights.

5. A passive evaporative cooling system for buildings based on capillary water transport and interfacial evaporation according to claim 1, characterized in that, The process of generating a moisture content distribution heatmap is as follows: a three-dimensional coordinate grid model of the building's exterior wall surface is established as the input coordinate set of the neural network prediction module; the preprocessed measurement point data is used as the true value label, and the neural network prediction module is used to perform spatiotemporal interpolation and nonlinear feature extraction to initially map and obtain the initial moisture content prediction values ​​for all grid nodes; the initial moisture content prediction values ​​are iteratively optimized using the loss function to ensure that the predicted values ​​for all grid nodes satisfy the conservation constraint of the moisture migration residual term in spatial distribution and the balance constraint of the heat conduction residual term and the boundary residual term in temporal evolution, thereby generating a continuous moisture content distribution heatmap with physical consistency.

6. A passive evaporative cooling system for buildings based on capillary water transport and interfacial evaporation according to claim 1, characterized in that, The specific steps for dynamically correcting the preset water supply duration include: extracting the predicted values ​​of each grid node in the moisture content distribution heatmap, calculating the spatial integral of the deviation between the predicted values ​​and the preset saturated moisture content threshold to obtain the global moisture deficit of the porous absorbent material layer at the current moment; calculating the water consumption rate under the current evaporation condition based on the interface drying time output by the physical state perception engine, and determining the theoretical water replenishment required to maintain a constant evaporation rate based on the global moisture deficit; mapping the theoretical water replenishment to the single pulse water supply time of the micro water pump, and adjusting the water supply frequency of the micro water pump according to the spatial non-uniformity of the moisture content distribution heatmap to update the preset water supply duration.

7. A passive evaporative cooling system for buildings based on capillary water transport and interfacial evaporation according to claim 1, characterized in that, The water supply scheduling module also includes an energy balance correction unit: The energy balance correction unit acquires the real-time remaining power of solar power and the predicted value of future light intensity, and outputs the predicted building heat load distribution for the next 24 hours based on historical meteorological sequences and wall thermal response characteristics. If a high-intensity heat load is predicted in the future and the current remaining power is sufficient, the energy balance correction unit instructs the water supply scheduling module to pre-wet the facade capillary network and use the thermal and humid inertia of the porous water-absorbing material layer for cold storage; if the current remaining power is lower than a preset threshold, the energy balance correction unit reduces the value of the deviation penalty weight of the water supply scheduling to reduce the single pulse water supply time to a minimum value that maintains basic evaporation requirements.

Images