Edible radiation-cooled gas gel based on machine learning heterogeneous pore structure, preparation method and application

Abstract

The application belongs to the technical field of new materials, and discloses an edible radiation refrigeration aerogel based on machine learning heterogeneous hierarchical pore structure, a preparation method and application. In the preparation of the edible radiation refrigeration aerogel with a heterogeneous hierarchical pore structure, the machine learning is used to obtain the combined machine learning optimization pore diameter parameter of the preparation of the edible radiation refrigeration aerogel with a heterogeneous hierarchical pore structure, the finite difference time domain (FDTD) method is used to simulate the scattering behavior of light in the multi-stage pores, the freezing and thawing method is used to construct the heterogeneous hierarchical pore structure of the gelatin-based aerogel, and finally the edible radiation refrigeration aerogel is prepared. In the application, without using any chemical cross-linking agent, only by accurately controlling the thermodynamic process of freezing and thawing cycles, using water as a green pore-forming agent (ice template), and relying on the physical action of hydrogen bonds formed between the gelatin molecular chains, a stable three-dimensional network structure with micron-nanometer multi-stage pores is constructed. The structure is the physical basis for realizing efficient spectral scattering.

Description

Technical Field

[0001] This invention belongs to the field of new materials technology, and in particular to an edible radiation-cooled aerogel based on machine learning-based heterogeneous porous structure, its preparation method and application. Background Technology

[0002] Currently, radiation cooling material systems used for agricultural product preservation have the following limitations: First, many high-performance materials rely on synthetic polymer or inorganic nanoparticle composite systems (such as acrylamide / polyvinyl alcohol as a template for composite zirconium dioxide and polytetrafluoroethylene nanoparticles, or multilayer structures based on photonic crystals). These materials may introduce or retain toxic and harmful chemicals during preparation, their biocompatibility has not been fully verified, and there is a potential risk of migration, making them unable to meet the stringent safety standards for food contact materials. Second, traditional aerogel or thin film materials often employ complex and expensive preparation processes (such as supercritical drying and magnetron sputtering) to achieve excellent optical properties, resulting in high costs and making them difficult to promote in large-scale, cost-sensitive scenarios such as agricultural product logistics. Therefore, developing a new material that can directly contact agricultural products, is safe and non-toxic, possesses both high-efficiency radiation cooling performance and good mechanical properties, and can adapt to the logistics environment has become a key requirement for promoting the large-scale and safe application of this technology in the agricultural product supply chain.

[0003] In summary, existing radiation refrigeration materials still have significant shortcomings in terms of biosafety, manufacturing cost, mechanical durability, refrigeration efficiency, and environmental friendliness when applied to the practical application of agricultural product logistics preservation.

[0004] The search revealed the following patent publications related to this invention's patent application:

[0005] 1. Chinese Patent Publication CN118324103A discloses a micro / nano porous composite membrane aerogel, its preparation method, and its application. The method involves first preparing an aerogel by mixing hydroxyapatite as the main component with polyvinyl alcohol and other substances. Then, a sodium polyacrylate solution is coated onto the surface of the aerogel to form a composite film. The resulting material has a micron-scale porous structure and a diffuse white appearance, designed to achieve cooling through the scattering of visible light.

[0006] The disadvantages are:

[0007] (1) The material systems used (such as polyvinyl alcohol and sodium polyacrylate) are not natural edible substances. Aerogels prepared from these materials are not suitable for fields such as food packaging where there are strict requirements for biosafety and non-toxicity.

[0008] (2) The pore structure design is simple and the spectral performance may be poor. The patent literature focuses on the construction of micron-level porous structures and lacks the collaborative design of "heterogeneous pores" from nano to micro.

[0009] 2. Chinese Patent Publication CN119805640A discloses a colored radiation-cooled thin film based on Fano resonance and its preparation method. This patent publication utilizes the coupling of discrete states (such as metal-dielectric-metal structures) and continuous states (such as broadband absorbers) to generate sharp asymmetric spectral lines, thereby achieving high-purity structural colors. This technical solution involves a complex multilayer film system.

[0010] The disadvantages are:

[0011] (1) Complex structure, difficult and costly preparation: This method requires the deposition of a dozen or more precision thin films, and the material and thickness of each layer need to be precisely controlled. This relies on expensive vacuum coating equipment such as magnetron sputtering and electron beam evaporation, which is complex, has a long production cycle, and is costly, making it difficult to achieve large-scale application.

[0012] (2) Material system limitations: The scheme explicitly limits the use of specific metals (such as Au, Ag, Al) and dielectric materials (such as MgF2, SiO2, TiO2) for each layer, which limits the possibility of selecting a better or lower cost combination from a wider range of materials.

[0013] By comparison, the present invention patent application is fundamentally different from the aforementioned patent publications. Summary of the Invention

[0014] The purpose of this invention is to overcome the shortcomings of the prior art and provide an edible radiation-cooled aerogel based on a machine learning-based heterogeneous porous structure, its preparation method, and its application.

[0015] The technical solution adopted by this invention to solve its technical problem is:

[0016] Application of machine learning in the preparation of edible radiation-cooled aerogels with heterogeneous porous structures.

[0017] Furthermore, the edible radiation-cooling aerogel has a visible light reflectance of 0.93 and a mid-infrared emissivity of 0.96.

[0018] Furthermore, the application is as follows: when preparing edible radiation-cooled aerogels with hierarchical pore structures, the pore size parameters for preparing edible radiation-cooled aerogels with hierarchical pore structures are optimized by combining machine learning with machine learning, the scattering behavior of light in hierarchical pores is simulated by the finite-difference time-domain method (FDTD), and the hierarchical pore structure of gelatin-based aerogels is constructed by the freeze-thaw method, and finally edible radiation-cooled aerogels are prepared.

[0019] Furthermore, the specific method for preparing edible radiation-cooled aerogels with heterogeneous porous structures is as follows: Based on machine learning training simulation to adjust freezing temperature, gelatin solution concentration, and freezing rate parameters, firstly, an electromagnetic scattering model of the aerogel porous structure is established using the finite-difference time-domain method (FDTD) or Monte Carlo method to calculate the spectral scattering efficiency factor under different pore size distributions; secondly, a neural network algorithm is used to train the nonlinear mapping relationship between pore size parameters and spectral characteristics to form a high-precision surrogate model; thirdly, a multi-objective optimization algorithm is used with high reflectivity in the solar band and high emissivity in the atmospheric window as objectives to inversely solve for the optimal pore size ratio, and outputs key process parameters guiding the gelatin prepolymer concentration, freeze-thaw temperature, and number of cycles, which can effectively control the nucleation and growth of ice crystals, and precisely regulate the micron- and nano-sized pore sizes and their distribution in the final aerogel formation, thus obtaining edible radiation-cooled aerogels.

[0020] A method for preparing edible radiation-cooled aerogel based on machine learning-based hierarchical porous structure is disclosed. The method is based on machine learning to obtain the pore size parameters of edible radiation-cooled aerogel with hierarchical porous structure, and uses the finite-difference time-domain method (FDTD) to simulate the scattering behavior of light in multi-level pores. The hierarchical porous structure of gelatin-based aerogel is constructed by freeze-thaw method, and finally edible radiation-cooled aerogel is prepared.

[0021] Furthermore, it includes the following steps:

[0022] (1) Based on finite-difference time-domain (FDTD) simulation and machine learning, pore size design: a porous model of gelatin aerogel is established and the pore size range is set; the spectral scattering efficiency under different pore size distributions is calculated by finite-difference time-domain and Monte Carlo method equations, and the pore size-spectrum relationship model is trained by machine learning to output the optimal pore size ratio with excellent spectral characteristics.

[0023] (2) Preparation of gelatin aerogel: Gelatin was dissolved in deionized water with a mass concentration of 10 wt%. The mixture was stirred at 90℃ for 1 h to form a sol. The sol was injected into a mold and pre-frozen at -20℃ for 0.5 h to form an ice template. Then it was transferred to -80℃ for cryogenic solidification for 0.5 h. The mixture was then cross-linked at room temperature (25℃) to induce the formation of heterogeneous pores. The frozen sample was freeze-dried to generate an edible radiation-cooled aerogel.

[0024] Further, the optimal pore size ratio in step (1) is: 60% for 50-500 nm nanopores and 40% for 1-20 µm micropores, or 80% for 50-200 nm nanopores and 20% for 1-5 μm micropores, or 35% for 5-10 μm micropores and 65% for 100-300 nm nanopores.

[0025] The edible radiation-cooled aerogel was prepared by the method described above.

[0026] A method for preparing an edible radiation-cooled aerogel based on a machine learning-based heterogeneous porous structure includes the following steps:

[0027] Gelatin powder was weighed and added to deionized water to achieve a final gelatin concentration of 10 wt%. The mixture was magnetically stirred at 300 rpm for 1 hour in a 90℃ water bath to obtain a clear and transparent gelatin sol. The gelatin sol was then injected into a polytetrafluoroethylene mold and pre-frozen at -20℃ for 0.5 h to form an initial ice template. It was then rapidly transferred to -80℃ for another 0.5 h, followed by 1 h at room temperature to induce the formation of a finer hierarchical porous structure. The frozen sample was then rapidly placed in a container pre-cooled to -50℃ and dried under a vacuum of less than 10 Pa for 48 hours to obtain an edible radiative-cooled aerogel.

[0028] Alternatively, it may include the following steps:

[0029] Gelatin powder was weighed and added to deionized water to achieve a final gelatin concentration of 10 wt%. The mixture was then magnetically stirred at 300 rpm for 1 hour in a 90℃ water bath to obtain a clear and transparent gelatin sol. Freezing process: The gelatin sol was injected into a polytetrafluoroethylene mold and rapidly frozen at -80℃ for 8 hours. Freeze-drying: The frozen sample was quickly placed in a container pre-cooled to -50℃ and dried under a vacuum of less than 10 Pa for 48 hours to obtain an edible radiative refrigerant aerogel.

[0030] The edible radiation-cooled aerogel was prepared by the method described above.

[0031] The advantages and positive effects of this invention are as follows:

[0032] 1. This invention relates to the controllable fabrication of a heterogeneous porous structure based on physical cross-linking: Without using any chemical cross-linking agents, a stable three-dimensional network structure with micron-nano hierarchical pores is constructed solely through precise control of the thermodynamic process of freeze-thaw cycles (e.g., pre-freezing at -20°C and -80°C, followed by cross-linking at 25°C room temperature). Water is used as a green pore-forming agent (ice template), and physical interactions such as hydrogen bonds formed between gelatin molecular chains are employed to achieve this structure. This structure forms the physical basis for achieving efficient spectral scattering.

[0033] 2. Intrinsic Safety and Environmental Friendliness of the Gel Material of this Invention: The method of this invention ensures the biosafety of the final product from the source. Its core features are: the raw materials are only edible gelatin and water; no toxic chemical cross-linking agents (such as glutaraldehyde) are introduced throughout the process; the process is mild, with no high pressure or harmful byproducts generated. A stable three-dimensional porous structure is constructed by relying on the physical cross-linking network (such as hydrogen bonds) formed by the gelatin molecular chains during freeze-thaw cycles, resulting in an aerogel that is both edible and biocompatible.

[0034] 3. This invention is a novel aerogel material based on natural edible gelatin and designed with a heterogeneous porous structure to achieve efficient radiative cooling, in order to meet the urgent needs of the agricultural product supply chain for safe, green, and efficient cooling technologies.

[0035] 4. This invention overcomes the limitations of single-pore structure in spectral scattering by synergistically optimizing visible light reflection and infrared emission through hierarchical pores (micron-nano multi-level pores). The spectral performance of the gel is improved: the hierarchical pores scatter visible light through micron-sized pores and enhance infrared emission through nano-sized pores, achieving excellent spectral performance and promoting efficient radiative cooling in the atmospheric window band.

[0036] 5. This invention combines optical simulation and machine learning. By inputting feature parameters, it outputs the optimal target for aperture distribution and spectral performance. It uses a convolutional neural network to accurately predict the optimal aperture distribution, replacing the traditional trial-and-error preparation method.

[0037] 6. The gel of this invention is a highly environmentally friendly and biocompatible radiation cooling material. The prepared material is edible and does not contain any cross-linking agents. It relies entirely on the physical cross-linking network (such as hydrogen bonds) formed by the gelatin molecular chains in the freeze-thaw cycle to construct a stable three-dimensional porous structure. It uses all natural edible raw materials and avoids the use of chemical cross-linking agents, thus achieving safe and efficient radiation cooling.

[0038] 7. The spectral properties of the gel of this invention are as follows: visible light band (0.4-0.8 μm) reflectance 0.93 (… Figure 6 As shown in the figure, it is superior to traditional aerogels (reflectivity ≤ 0.9); simulation verification: FDTD predicted scattering efficiency and experimental error < 5%, and the machine learning model fit R is good. 2 >0.98.

[0039] 8. Replacement of machine learning network code training in this invention: In the modeling of porous structures of aerogels, if the data has spatial local correlation, convolutional layers can be used to extract local features, and recurrent neural networks (RNNs) can be used.

[0040] 9. In this invention, gelatin can be replaced with cellulose nanofibers (CNF) or polyvinyl alcohol, and good spectral effects can be achieved through hierarchical pore design.

[0041] In this invention, the prepolymer concentration gradient substitution can be achieved by using gelatin with a mass fraction of 5%-15%.

[0042] The fabrication process in this invention can be replaced: for applications that do not require extreme optical performance or are cost-sensitive, the machine learning step can be omitted, and a direct method based on FDTD simulation, involving parameterized scanning and iterative optimization, can be adopted. A set of well-defined and finite combinations of pore size distribution parameters is defined. For example, 3-5 gradients are selected for micron-sized pores in the range of 1-20 μm, and 3-5 gradients are selected for nanopore proportions in the range of 20%-80%, forming a discrete parameter space. Attached Figure Description

[0043] Figure 1 This is a schematic diagram of the intelligent closed-loop process flow based on "computational design-experimental verification" machine learning in this invention;

[0044] Figure 2 Establish a design algorithm diagram for the machine learning model in this invention;

[0045] Figure 3 This is a light field diagram of gelatin aerogel based on theoretical simulation in this invention;

[0046] Figure 4 This is the spectrum of the gelatin aerogel in the near-mid-infrared band (2.5 - 20 μm) of the present invention;

[0047] Figure 5 The images shown are scanning electron microscope (SEM) images of the gelatin aerogel in this invention, with cross-section (left) and longitudinal section (right) to illustrate its heterogeneous porous structure.

[0048] Figure 6 This is a reflectance diagram of the gelatin aerogel in the visible light band (0.4 - 0.8 μm) of the present invention;

[0049] Figure 7 This is an emissivity diagram of the gelatin aerogel in the mid-infrared band in this invention;

[0050] Figure 8 The above are average reflectance diagrams of aerogels in the visible light band for Examples 1, 2, Comparative Examples 1 and 2 of this invention.

[0051] Figure 9 The average emissivity of the aerogels in the mid-infrared band of Examples 1, 2, Comparative Example 1, and Comparative Example 2 of this invention is shown in the diagram.

[0052] Figure 10 This is the spectrum of nanocellulose aerogel in the mid-infrared band in Example 3 of this invention.

[0053] Figure 11This is the spectrum of polyvinyl alcohol aerogel in the mid-infrared band in Example 4 of this invention. Detailed Implementation

[0054] The present invention will be further described below with reference to the embodiments. The following embodiments are descriptive and not limiting, and should not be used to limit the scope of protection of the present invention.

[0055] The various experimental operations involved in the specific embodiments are all conventional techniques in the field. For parts not specifically annotated in this document, those skilled in the art can refer to various commonly used reference books, scientific and technological documents or related instructions and manuals prior to the filing date of this invention to carry out the operations.

[0056] Application of machine learning in the preparation of edible radiation-cooled aerogels with heterogeneous porous structures.

[0057] Furthermore, the edible radiation-cooling aerogel has a visible light reflectance of 0.93 and a mid-infrared emissivity of 0.96.

[0058] Furthermore, the application is as follows: when preparing edible radiation-cooled aerogels with hierarchical pore structures, the pore size parameters for preparing edible radiation-cooled aerogels with hierarchical pore structures are optimized by combining machine learning with machine learning, the scattering behavior of light in hierarchical pores is simulated by the finite-difference time-domain method (FDTD), and the hierarchical pore structure of gelatin-based aerogels is constructed by the freeze-thaw method, and finally edible radiation-cooled aerogels are prepared.

[0059] Furthermore, the specific method for preparing edible radiation-cooled aerogels with heterogeneous porous structures is as follows: Based on machine learning training simulation to adjust freezing temperature, gelatin solution concentration, and freezing rate parameters, firstly, an electromagnetic scattering model of the aerogel porous structure is established using the finite-difference time-domain method (FDTD) or Monte Carlo method to calculate the spectral scattering efficiency factor under different pore size distributions; secondly, a neural network algorithm is used to train the nonlinear mapping relationship between pore size parameters and spectral characteristics to form a high-precision surrogate model; thirdly, a multi-objective optimization algorithm is used with high reflectivity in the solar band and high emissivity in the atmospheric window as objectives to inversely solve for the optimal pore size ratio, and outputs key process parameters guiding the gelatin prepolymer concentration, freeze-thaw temperature, and number of cycles, which can effectively control the nucleation and growth of ice crystals, and precisely regulate the micron- and nano-sized pore sizes and their distribution in the final aerogel formation, thus obtaining edible radiation-cooled aerogels.

[0060] A method for preparing edible radiation-cooled aerogel based on machine learning-based hierarchical porous structure is disclosed. The method is based on machine learning to obtain the pore size parameters of edible radiation-cooled aerogel with hierarchical porous structure, and uses the finite-difference time-domain method (FDTD) to simulate the scattering behavior of light in multi-level pores. The hierarchical porous structure of gelatin-based aerogel is constructed by freeze-thaw method, and finally edible radiation-cooled aerogel is prepared.

[0061] Furthermore, it includes the following steps:

[0062] (1) Based on finite-difference time-domain (FDTD) simulation and machine learning, pore size design: a porous model of gelatin aerogel is established and the pore size range is set; the spectral scattering efficiency under different pore size distributions is calculated by finite-difference time-domain and Monte Carlo method equations, and the pore size-spectrum relationship model is trained by machine learning to output the optimal pore size ratio with excellent spectral characteristics.

[0063] (2) Preparation of gelatin aerogel: Gelatin was dissolved in deionized water with a mass concentration of 10 wt%. The mixture was stirred at 90℃ for 1 h to form a sol. The sol was injected into a mold and pre-frozen at -20℃ for 0.5 h to form an ice template. Then it was transferred to -80℃ for cryogenic solidification for 0.5 h. The mixture was then cross-linked at room temperature (25℃) to induce the formation of heterogeneous pores. The frozen sample was freeze-dried to generate an edible radiation-cooled aerogel.

[0064] Further, the optimal pore size ratio in step (1) is: 60% for 50-500 nm nanopores and 40% for 1-20 µm micropores, or 80% for 50-200 nm nanopores and 20% for 1-5 μm micropores, or 35% for 5-10 μm micropores and 65% for 100-300 nm nanopores.

[0065] The edible radiation-cooled aerogel was prepared by the method described above.

[0066] A method for preparing an edible radiation-cooled aerogel based on a machine learning-based heterogeneous porous structure includes the following steps:

[0067] Gelatin powder was weighed and added to deionized water to achieve a final gelatin concentration of 10 wt%. The mixture was magnetically stirred at 300 rpm for 1 hour in a 90℃ water bath to obtain a clear and transparent gelatin sol. The gelatin sol was then injected into a polytetrafluoroethylene mold and pre-frozen at -20℃ for 0.5 h to form an initial ice template. It was then rapidly transferred to -80℃ for another 0.5 h, followed by 1 h at room temperature to induce the formation of a finer hierarchical porous structure. The frozen sample was then rapidly placed in a container pre-cooled to -50℃ and dried under a vacuum of less than 10 Pa for 48 hours to obtain an edible radiative-cooled aerogel.

[0068] Alternatively, it may include the following steps:

[0069] Gelatin powder was weighed and added to deionized water to achieve a final gelatin concentration of 10 wt%. The mixture was then magnetically stirred at 300 rpm for 1 hour in a 90℃ water bath to obtain a clear and transparent gelatin sol. Freezing process: The gelatin sol was injected into a polytetrafluoroethylene mold and rapidly frozen at -80℃ for 8 hours. Freeze-drying: The frozen sample was quickly placed in a container pre-cooled to -50℃ and dried under a vacuum of less than 10 Pa for 48 hours to obtain an edible radiative refrigerant aerogel.

[0070] The edible radiation-cooled aerogel was prepared by the method described above.

[0071] Specifically, the relevant preparation and testing methods are as follows:

[0072] Based on machine learning training and simulation to adjust parameters such as freezing temperature, gelatin solution concentration, and freezing rate, this system first establishes an electromagnetic scattering model of the aerogel porous structure using the finite-difference time-domain (FDTD) method or Monte Carlo method, calculating the spectral scattering efficiency factor under different pore size distributions. Then, it uses a neural network algorithm to train the nonlinear mapping relationship between pore size parameters and spectral characteristics, forming a high-precision surrogate model. Finally, a multi-objective optimization algorithm is employed, targeting high reflectivity in the solar band and high emissivity in the atmospheric window, to inversely solve for the optimal pore size ratio, outputting guidance for key process parameters such as gelatin prepolymer concentration, freeze-thaw temperature, and number of cycles. This effectively controls the nucleation and growth of ice crystals, thereby precisely regulating the final micron- and nanometer-scale pore sizes and their distribution in the aerogel.

[0073] The intelligent closed-loop process of "computational design-experimental verification" has the following flowchart: Figure 1As shown, the primary key point of this invention lies in constructing a complete intelligent material development closed loop. Based on joint optimization of spectral simulation and machine learning, the scattering efficiency factor is calculated using the finite-difference time-domain method and the Monte Carlo method. A neural network is used to train the coupling effect between gelatin pore size and spectral properties, establishing an pore size-spectrum mapping model. A multi-objective optimization algorithm is then used to inversely solve for the optimal pore size distribution, accurately predicting and optimizing the relationship between pore size distribution and spectral performance. This is achieved through rigorous "optical objectives." Theoretical Structure The reverse reasoning process of the preparation process, supplemented by experimental verification and feedback, was used to construct a gelatin prepolymer system. This enabled precise and theoretically grounded optimization of parameters such as freeze-thaw temperature, freeze-thaw cycle number, crosslinking temperature, and prepolymer concentration. Scanning electron microscopy was used to observe the pore size and distribution in the aerogel's porous structure, and a spectrometer was used to measure its reflectance in the ultraviolet-visible-near-infrared band and its emissivity in the mid-infrared band. The spectral performance was transmitted to the model for feedback, allowing for the assessment of the accuracy of the predicted performance and further optimization of the model.

[0074] Example 1: Balanced heterogeneous pore gelatin aerogel

[0075] This embodiment aims to prepare a uniformly high-performance radiocooled aerogel that exhibits performance in both the visible light and atmospheric window bands. The machine learning optimization process is as follows: Figure 2 As shown, based on Mie theory, the scattering efficiency factor of particles with different sizes is calculated, and an optical response library is established. By solving Maxwell's equations, the scattering and absorption efficiencies of monodisperse particle systems at different wavelengths are calculated. Secondly, the complex particle size distribution of actual aerogels is mathematically characterized. A Gaussian mixture model or a log-normal distribution is used to fit the multi-peak pore size distribution, and distribution moment features such as mean, standard deviation, skewness, and kurtosis are extracted. Thirdly, a feature engineering example is provided, using principal component analysis to reduce the dimensionality of the pore size distribution features, or constructing a cross-feature between the number of particle size bands and the average scattering efficiency. Next, model training and cross-validation are performed, using support vector machines or random forest regression algorithms, and k-fold cross-validation is used to evaluate the model's generalization ability and prevent overfitting. Finally, the desired improvement EI is maximized based on the acquisition function of a Gaussian process through a Bayesian optimization equation. The core equation is: EI(x) = (μ(x) - f(x)). + ) - ξ)Φ(Z) + σ(x) (Z), where Z = (μ(x) - f(x) + In this scheme, x represents the target aperture distribution parameter; μ(x) represents the weighted spectral performance index; f(x) = 0.05 - ξ) / σ(x). +The ) represents the value that optimizes the weighted objective number through FDTD simulation verification, serving as the benchmark for determining whether "improvement" is needed; σ(x) represents the confidence level of the model's performance prediction at point x, with a larger value indicating higher uncertainty; ξ is a hyperparameter, setting ξ>0 will make it more inclined to explore regions with higher uncertainty when selecting the next evaluation point, helping to escape possible local optima; Z is the weighted average of [μ(x) - f(x)]. + The standardization is performed on the uncertainty [σ(x)] and Φ(Z); Φ(Z) represents the probability that the standard normal random variable takes a value less than or equal to Z; (Z) It represents the probability density of values ​​within a unit interval around the Z value. Optimal particle size distribution parameters are automatically searched to achieve directional design of aerogel optical properties. This method significantly improves the efficiency and accuracy of aerogel pore structure design by combining a physical model with a data-driven approach. FDTD simulation and machine learning for pore design. Model establishment: Using the two-dimensional finite-difference time-domain (FDTD) method, cylindrical or spherical air pores containing randomly distributed nanopores (50-500 nm) and micropores (1-20 μm) are established and embedded in a polymer gelatin substrate with a refractive index of 1.53. Periodic boundary conditions are set in the x and y directions of the model to simulate an infinitely large aerogel plane. A perfectly matched layer (PML) is set in the light propagation direction (z-direction) to absorb transmitted light and avoid non-physical reflections. The computational region covers 0.3-20 μm, simulating a plane wave incident perpendicularly. The reflection and transmission fields of the structure in the target wavelength band are calculated using a monitor. Machine learning optimization: 5000 different combinations of aperture distributions (including the ratio of nanopores to micropores, average size, and randomness of position) are generated as a training set. A neural network model (structure: input layer - 3 hidden layers - output layer) is used. The number of nodes in the input layer corresponds to the number of feature parameters of the aperture distribution. The hidden layers typically have 3 or more fully connected layers, each containing hundreds of neurons, using activation functions such as ReLU (Rectified Linear Unit) to introduce nonlinearity. The output layer has 2 nodes, corresponding to the predicted solar band reflectivity and atmospheric window emissivity, respectively. The loss function is typically the mean squared error (MSE), which is the sum of squares of the differences between the predicted value and the true FDTD value. Optimizers such as Adam are used to train the network weights by minimizing the loss function through backpropagation. The model was trained by minimizing a weighted objective function (0.6 for solar band reflectivity and 0.4 for atmospheric window emissivity). The optimized output showed the best predictive spectral performance when nanoscale pores (100-300 nm) comprised approximately 65% ​​and microscale pores (5-10 μm) comprised approximately 35%. The light field intensity is as follows: Figure 3 As shown, the heterogeneous porous structure exhibits strong scattering of visible light. The reflectance and transmittance of gelatin aerogel in the mid-infrared band are as follows: Figure 4 As shown, the emissivity of gelatin aerogel can reach 0.9 in the atmospheric window band (8-13 μm).

[0076] Preparation of Gelatin Aerogel (Graded Freeze-Thaw Method): Solution Preparation: Weigh 10.0 g of gelatin powder (Bloom strength ~250) and add it to 90 mL of deionized water. Stir magnetically (300 rpm) in a 90℃ water bath for 1 hour to obtain a clear and transparent gelatin sol. No chemical cross-linking agents were added throughout the process. Freezing Process: Inject the gelatin sol into a polytetrafluoroethylene mold. First, pre-freeze in a -20℃ freezer for 0.5 h to form an initial ice template; then quickly transfer it to a -80℃ cryogenic freezer for another 0.5 h, followed by 1 h at room temperature to induce the formation of a finer hierarchical porous structure. Freeze-Drying: Quickly place the frozen sample into a freeze dryer pre-cooled to -50℃ and dry under a vacuum of less than 10 Pa for 48 hours to obtain the final porous gelatin aerogel.

[0077] The gelatin aerogel obtained in Implementation Case 1 was characterized by scanning electron microscopy (SEM), and the results are as follows: Figure 5 As shown in the low-magnification image, the material exhibits a complete three-dimensional porous network structure. High-magnification images clearly show two pore sizes: one is a micrometer-scale macroporous framework with a size of approximately 5-10 μm and a regular morphology; the other is nanometer-scale pores with a size of approximately 100-300 nm distributed on the walls of the macroporous framework. Statistical results from multiple fields of view using image analysis software indicate that nanometer-scale pores account for approximately 65%, and micrometer-scale pores account for approximately 35%. This is highly consistent with the optimal pore size distribution prediction obtained from the machine learning model optimization, proving the effectiveness and controllability of the fabrication process parameters. Tests were performed using a UV-Vis-NIR spectrophotometer (equipped with an integrating sphere) and a Fourier transform infrared spectrometer. The results show that the aerogel achieves an average reflectance of 0.93 in the 0.3-2.5 μm solar band (e.g., ...). Figure 6 As shown), the average emissivity reaches 0.96 in the 8-13 μm atmospheric window band (as shown). Figure 7 As shown in the figure, the smooth spectral curves demonstrate that the aerogel exhibits excellent and balanced optical response across a wide spectral range. This is attributed to the efficient scattering of visible light and the enhancement of infrared emission through its optimized hierarchical pore structure, resulting in strong radiative cooling capabilities.

[0078] Example 2: High-emission heterogeneous pore gelatin aerogel

[0079] This embodiment focuses on improving the infrared emission performance of the material in the atmospheric window band, and is suitable for scenarios with higher heat dissipation requirements.

[0080] FDTD Simulation and Machine Learning Aperture Design: The simulation method in this embodiment is the same as in Embodiment 1, but the objective function of the machine learning model is adjusted to significantly increase the weight of emissivity in the atmospheric window band (reflectivity weight 0.4, emissivity weight 0.6). The optimized output leads to a distribution with a significantly increased proportion of nanopores: nano-sized pores (50-200 nm) account for approximately 80%, and micron-sized pores (1-5 μm) account for approximately 20%. Theoretical predictions indicate that this type of structure can more effectively excite photon localization and emission effects in the infrared band.

[0081] Preparation of gelatin aerogel (freeze-thaw method): The preparation process is basically the same as in Example 1, with the key difference being the control of the freezing rate to induce more nanopores: After injecting the gelatin sol (10 wt%, mass concentration) into the mold, it was directly placed in a -80°C cryogenic freezer for rapid freezing (skipping the -20°C pre-freezing step) and maintained at -80°C for 8 hours. A faster freezing rate helps to form finer ice crystals, thereby obtaining a higher proportion of nanopores after drying.

[0082] Performance Characterization: Microstructure (SEM): SEM images show that the aerogel's pore structure is significantly denser than that of Example 1. The number and size of micron-sized pores are reduced (approximately 1-5 μm), while the nanoscale network structure is extremely well-developed, with pore sizes mainly concentrated in the 50-200 nm range. Figure 8 He Ru Figure 9 As shown, the spectral performance is as follows: Test results show that, due to the relative reduction in micron-sized scattering centers, its average reflectance in the solar band is 0.89, slightly lower than that of Example 1. However, thanks to the high proportion of nanopores that greatly enhance infrared emission capabilities, its average emissivity in the atmospheric window band reaches as high as 0.95, achieving the design goal of this scheme.

[0083] Example 3: Heterogeneous porous nanocellulose aerogel

[0084] This embodiment successfully prepared an aerogel based on biomass-derived cellulose nanofibers (CNF). The process included: first, preparing a 2 wt% CNF aqueous dispersion, then adding a crosslinking agent, 1,4-butanediol diglycidyl ether (BDGE), at 25% of the CNF aqueous dispersion mass. The reaction temperature was 50-70℃, and the reaction time was 4-6 h. The hydrogel network was formed by the crosslinking reaction between the hydroxyl groups on the CNF surface and the epoxy groups of BDGE. Subsequently, a unidirectional freezing technique was used to directionally freeze a copper block cooled by liquid nitrogen at the bottom of the hydrogel mold, inducing the directional growth of ice crystals and forming micron-sized channels arranged along the temperature gradient. The recommended freezing rate was 1-10℃ / min. For hydrogels with a thickness of 1-2 cm, it may take 10-30 minutes for the interior to freeze completely after contact with the pre-cooled copper block. The actual time is based on the complete solidification of the sample; finally, the sample is freeze-dried at -80℃ for 20-48 hours to obtain CNF aerogel with an anisotropic hierarchical porous structure.

[0085] like Figure 10 As shown, the CNF aerogel has an average emissivity of 0.94 in the atmospheric window (8-13 μm). Its excellent performance is due to the effective scattering of visible light by the micron-sized channels (~10-50 μm) formed by directional freezing, and the enhanced infrared vibration absorption of polar bonds such as OH in CNF molecules by the CNF nanofiber network and the nano-sized pores (<100 nm) generated by cross-linking.

[0086] Example 4: Heterogeneous porous polyvinyl alcohol aerogel

[0087] This embodiment details the preparation process of a composite aerogel constructed using polyvinyl alcohol (PVA) and sodium alginate (SA): First, prepare a 10 wt% PVA aqueous solution (dissolved by heating and stirring at 95°C for 3 hours) and a 1 wt% SA aqueous solution (dissolved by stirring at 400 rpm at room temperature). Then, mix the two solutions at a 1:1 mass ratio. While stirring at 300 rpm, add the crosslinking agent boric acid (10 wt% - 12 wt% of the total mass of the mixed solution) and the foaming agent sodium bicarbonate (10 wt% - 20 wt% of the total mass of the mixed solution) sequentially to the mixture, and continue stirring for 10-15 minutes to introduce microbubbles. Next, pour the mixture into a mold and freeze at -20°C for 24 hours to allow the PVA and boric acid to form the first crosslinking network. Then, immerse the gel in a 2 wt% - 15 wt% calcium chloride solution, and pass the calcium chloride through a precipitate solution. 2+A second heavy ion crosslinking network is formed by combining SA in an "egg-box" manner; finally, PVA / SA composite aerogel is obtained by freeze-drying at -60℃. Performance characterization shows that the aerogel has an average reflectance of 0.91 in the solar band (0.3-2.5 μm), and its spectral characteristics in the mid-infrared band are as follows: Figure 11 As shown, the average emissivity in the atmospheric window band (8-13 μm) is 0.93, and its micron pores are mainly formed by foaming and ice templates, while the nanopores originate from cross-linked networks.

[0088] Comparative Example 1: Gelatin Aerogel with Single Micron-Porous Structure

[0089] This comparative example aims to illustrate the limitations of traditional aerogels without a hierarchical porous structure in terms of radiative cooling performance.

[0090] Preparation process: The gelatin sol was prepared exactly the same as in Example 1 (10 wt%, mass concentration). The difference lies in the freezing process: after the sol was injected into the mold, it was slowly frozen at -20°C for 24 hours. This condition tends to form larger and more uniform ice crystals, resulting in a structure dominated by micron-sized pores. The subsequent freeze-drying process was the same.

[0091] Performance Characterization: Microstructure (SEM): SEM images show that the aerogel has large pores (10-20 μm), but a simple structure with smooth pore walls, and almost no nanoscale fine pores are observed. This contrasts sharply with the complex hierarchical pore structures in Examples 1 and 2. Spectroscopic Performance: Spectroscopic results clearly demonstrate the disadvantages of a single pore size. Figure 8 As shown, in the solar band, due to the lack of cooperative scattering by nanoscale structures, its average reflectivity is only 0.75. For example... Figure 9 As shown, in the infrared atmospheric window band, the average emissivity is also low, only 0.85, due to the inability to effectively control the emission characteristics. This directly proves that machine learning-based heterogeneous aperture design is necessary and crucial for achieving efficient radiative cooling.

[0092] Comparative Example 2:

[0093] Preparation process: The preparation of gelatin sol is exactly the same as in Example 1, except that the mass concentration of gelatin in the clear and transparent gelatin sol is 20 wt%.

[0094] Test results as follows Figure 8 and Figure 9As shown, its performance is significantly inferior to that of Example 1. The dense structure leads to a decrease in solar reflectivity to 0.82 and infrared emissivity to 0.86, and the material becomes much more brittle (although the compressive strength increases to 280 kPa, it breaks at 40% strain). The fundamental reason is that the excessively high concentration inhibits the ice crystal template effect, resulting in the absence of micron-level pores and insufficient nano-level pores, making it impossible to form the heterogeneous pore structure of "micron-nano synergy" in the example.

[0095] Meanwhile, by comparing Example 1, Comparative Example 1, and Comparative Example 2, it can be seen that the steps "the mass concentration of gelatin in the gelatin sol is 10 wt%" and "then quickly transfer it to a -80℃ cryogenic freezer for 0.5 h, and then place it at room temperature for 1 h to induce the formation of a finer hierarchical pore structure; freeze-drying: quickly place the frozen sample into a freeze dryer pre-cooled to -50℃ and dry it for 48 hours under a vacuum of less than 10 Pa" have a synergistic effect, which can synergistically improve the relevant properties of the prepared porous gelatin aerogel.

[0096] Although embodiments of the invention have been disclosed for illustrative purposes, those skilled in the art will understand that various substitutions, variations, and modifications are possible without departing from the spirit and scope of the invention and the appended claims. Therefore, the scope of the invention is not limited to the contents disclosed in the embodiments.

Claims

1. The application of machine learning in the preparation of edible radiation-cooled aerogels with hierarchical pore structures, wherein the application is as follows: when preparing edible radiation-cooled aerogels with hierarchical pore structures, the pore size parameters of the edible radiation-cooled aerogels with hierarchical pore structures are obtained based on machine learning, the scattering behavior of light in hierarchical pores is simulated by the finite-difference time-domain method (FDTD), the hierarchical pore structure of gelatin-based aerogels is constructed by the freeze-thaw method, and finally the edible radiation-cooled aerogels are prepared; The specific method for preparing edible radiation-cooled aerogels with heterogeneous porous structures is as follows: Based on machine learning training simulation, the freezing temperature, gelatin solution concentration, and freezing rate parameters are adjusted. First, an electromagnetic scattering model of the aerogel porous structure is established using the finite-difference time-domain method (FDTD) or Monte Carlo method to calculate the spectral scattering efficiency factor under different pore size distributions. Then, a neural network algorithm is used to train the nonlinear mapping relationship between pore size parameters and spectral characteristics to form a high-precision surrogate model. A multi-objective optimization algorithm is used with high reflectivity in the solar band and high emissivity in the atmospheric window as objectives to solve for the optimal pore size ratio in reverse, and outputs key process parameters to guide the concentration of gelatin prepolymer, freeze-thaw temperature, and number of cycles. This can effectively control the nucleation and growth of ice crystals and precisely regulate the micron- and nano-sized pore sizes and their distribution in the final aerogel formation.

2. The application according to claim 1, characterized in that: The edible radiative cooling aerogel has a visible light reflectance of 0.93 and a mid-infrared emissivity of 0.

96.

3. A method for preparing an edible radiation-cooled aerogel based on a machine learning-based heterogeneous porous structure, characterized in that: The method is based on machine learning to obtain edible radio-cooled aerogels with hierarchical pore structures, combined with machine learning to optimize pore size parameters, using the finite-difference time-domain method (FDTD) to simulate the scattering behavior of light in hierarchical pores, and constructing the hierarchical pore structure of gelatin-based aerogels through freeze-thaw methods, ultimately preparing edible radio-cooled aerogels. The specific method for preparing edible radiation-cooled aerogels with heterogeneous porous structures is as follows: Based on machine learning training simulation, the freezing temperature, gelatin solution concentration, and freezing rate parameters are adjusted. First, an electromagnetic scattering model of the porous structure of the aerogel is established using the finite-difference time-domain method (FDTD), and the spectral scattering efficiency factor under different pore size distributions is calculated. The nonlinear mapping relationship between pore size parameters and spectral characteristics is trained using a neural network algorithm to form a high-precision surrogate model. A multi-objective optimization algorithm is used with high reflectivity in the solar band and high emissivity in the atmospheric window as objectives to solve for the optimal pore size ratio in reverse, and outputs key process parameters to guide the concentration of gelatin prepolymer, freeze-thaw temperature, and number of cycles. This can effectively control the nucleation and growth of ice crystals and precisely regulate the micron- and nano-sized pore sizes and their distribution in the final aerogel formation.

4. The preparation method according to claim 3, characterized in that: Includes the following steps: (1) Based on finite-difference time-domain (FDTD) simulation and machine learning, pore size design: establish a porous model of gelatin aerogel and set the pore size range; calculate the spectral scattering efficiency under different pore size distributions by using the finite-difference time-domain equation, train the pore size-spectrum relationship model using machine learning, and output the optimal pore size ratio with excellent spectral characteristics. (2) Preparation of gelatin aerogel: Gelatin was dissolved in deionized water with a mass concentration of 10 wt%. The mixture was stirred at 90℃ for 1 h to form a sol. The sol was injected into a mold and pre-frozen at -20℃ for 0.5 h to form an ice template. Then it was transferred to -80℃ for cryogenic solidification for 0.5 h. The mixture was then cross-linked at room temperature (25℃) to induce the formation of heterogeneous pores. The frozen sample was freeze-dried to generate an edible radiation-cooled aerogel.

5. The preparation method according to claim 4, characterized in that: The optimal pore size ratio in step (1) is: 60% for 50-500 nm nanopores and 40% for 1-20 µm micropores, or 80% for 50-200 nm nanopores and 20% for 1-5 μm micropores, or 35% for 5-10 μm micropores and 65% for 100-300 nm nanopores.

6. A method for preparing an edible radiation-cooled aerogel based on a machine learning-based heterogeneous porous structure, characterized in that: Includes the following steps: (1) Based on finite-difference time-domain (FDTD) simulation and machine learning, the freezing temperature, gelatin solution concentration and freezing rate parameters are adjusted by training simulation based on machine learning. First, the electromagnetic scattering model of the porous structure of aerogel is established by finite-difference time-domain (FDTD) and the spectral scattering efficiency factor under different pore size distributions is calculated. The nonlinear mapping relationship between pore size parameters and spectral characteristics is trained by using neural network algorithm to form a high-precision surrogate model. The multi-objective optimization algorithm is used with high reflectivity in the solar band and high emissivity in the atmospheric window as the target to solve the optimal pore size ratio in reverse and output the key process parameters of gelatin prepolymer concentration, freeze-thaw temperature and number of cycles. It can effectively control the nucleation and growth of ice crystals and accurately regulate the micron and nano-sized pore size and its distribution in the final formation of aerogel. (2) Preparation of edible radiation-cooled aerogel: Gelatin powder was weighed and added to deionized water. The final mass concentration of gelatin was 10 wt%. The gelatin was magnetically stirred at 300 rpm for 1 hour in a 90℃ water bath to obtain a clear and transparent gelatin sol. Freezing process: The gelatin sol was injected into a polytetrafluoroethylene mold and first placed in a -20℃ freezer for 0.5 h to form an initial ice template. Then it was quickly transferred to -80℃ for 0.5 h and then placed at room temperature for 1 h to induce the formation of a finer heterogeneous pore structure. Freeze-drying: The frozen sample was quickly placed in a container pre-cooled to -50℃ and dried for 48 hours under a vacuum of less than 10 Pa to obtain edible radiation-cooled aerogel. Alternatively, weigh gelatin powder and add it to deionized water to achieve a final gelatin concentration of 10 wt%. Stir magnetically at 300 rpm for 1 hour in a 90℃ water bath to obtain a clear and transparent gelatin sol. Freezing process: Inject the gelatin sol into a polytetrafluoroethylene mold and directly place it in -80℃ for rapid freezing, and maintain it at -80℃ for 8 hours. Freeze-drying: Quickly place the frozen sample into a container pre-cooled to -50℃ and dry it under a vacuum of less than 10 Pa for 48 hours to obtain an edible radiative cooling aerogel.

Images