A method for preparing metal oxide nanoparticles with controllable particle size

Abstract

The present invention discloses a method for preparing metal oxide nanoparticles with controllable particle size, belonging to the field of nanomaterial preparation. Based on existing literature data and experimental synthesis, the method obtains a dataset of process parameters and nano-metal oxide particle size distribution during the preparation process. The method processes the variables and product particle size in the database through computational analysis methods such as one-hot encoding, data normalization, and model explanatory analysis to construct a prediction model. The prediction model is embedded in a Bayesian optimization algorithm to construct a reverse design platform for key process parameters. The method utilizes the reverse design platform to obtain synthesis process parameters based on the target particle size to prepare the metal oxide. The method experimentally verifies the size-controllable synthesis of three nanoparticles: titanium oxide, zirconium oxide, and cerium oxide.

Description

Technical Field

[0001] The present invention relates to the technical field of nanomaterial preparation; more specifically, the present invention relates to a method for preparing metal oxide nanoparticles with controllable particle size, wherein the method prepares nano metal oxides with a target particle size through a solvent thermal synthesis method under the guidance of a big data optimization algorithm. Background Art

[0002] The International Union of Pure and Applied Chemistry (IUPAC) defines nanoparticles (NPs) as particles with a size below 100 nanometers in one or more dimensions. Titanium oxide (TiO2) is a widely studied metal oxide. Due to its unique electronic structure at different sizes, it holds significant applications in optoelectronics, thermoelectrics, medicine, and energy storage. Furthermore, small (approximately 15 nm) TiO2 NPs offer irreplaceable advantages, such as exceptional dielectric constants, low Rayleigh scattering, and high stability, making them attractive candidates for inorganic modification of polymers with excellent optical properties. Precise size control depends crucially on the control of the TiO2 NP synthesis process. However, the synthesis of TiO2 NPs is influenced by multiple factors, such as temperature, reaction duration, pH, reactants, and mixing efficiency. In particular, different reactants, such as dopants, can influence NP nucleation and size, while the solvent, acting as a surface ligand, influences the monodispersity of the NPs. These factors in the reaction system are interdependent and mutually influential. Optimizing each reaction condition to achieve the target particle size consumes considerable experimental time and resources. The interrelationship of these factors makes the precise and controllable synthesis of TiO2 NPs challenging.

[0003] Complex TiO₂ NPs (TiO₂ NPs) composited with optical polymers have diverse applications, not limited to UV-shielding weathering coatings, photocatalytically mediated reversible color conversion, ultrahigh diffraction efficiency volume holographic gratings, and thermally tunable transparent coatings. In these applications, TiO₂ size significantly influences the performance and stability of the nanocomposites. Many size-controlled synthesis processes for complex TiO₂ NPs rely on manual grid searches (orthogonal experiments) to determine the optimal synthesis parameters. Each variable in the experimental process requires constant tuning to determine the optimal formulation, which not only consumes significant experimental time and resources but also results in a highly inefficient approach. Thoroughly identifying a suitable set of additive and dopant parameters to achieve the target particle size requires hundreds of experiments and characterizations. For more than one additive, the required experimental time and resources multiply, making them virtually impossible to achieve with laboratory manpower. In actual synthesis, in addition to the effects of additives and dopants, more than six variables, such as temperature, pH, and reaction time, influence the size distribution of the synthesized NPs. Screening these reaction conditions consumes significant experimental time and resources.

[0004] The field of artificial intelligence (AI) is advancing by leaps and bounds. Machine learning (ML), as a branch of artificial intelligence, plays a powerful role in data prediction and classification. When physical conditions are unknown, ML can break through physical limitations and obtain good results through established algorithms. Many papers have shown that ML has also been developed in the field of nanomaterial synthesis. For example, Zhang et al. applied ML to the synthesis process of Fe3O4 particles. They obtained the best model for predicting average particle size and crystallinity through multi-model comparison. Ling et al. established an ML model to study the purity, average particle size and shape of Ag NPs mixtures by extinction spectroscopy. However, most ML studies only focus on the synthesis parameters of NPs with a specific average particle size under a single condition, while the synthesis of NPs with precise particle size distribution and complex composition still requires a lot of experiments. Summary of the Invention

[0005] The purpose of the present invention is to overcome the deficiencies of the prior art and provide a method for preparing metal oxide nanoparticles with controllable particle size.

[0006] This invention uses machine learning to improve the efficiency of developing nanometal oxides with varying particle size distributions. By comprehensively considering the impact of different reaction variables on the product particle size distribution, the authors optimize the screening of precursors and the design of reaction conditions, ensuring that the process conditions meet the effective boundary conditions for nanometal oxide preparation and wide-ranging performance variations. The authors further develop a platform for reverse design of nanometal oxide synthesis parameters. By integrating a metal oxide particle size distribution prediction model with a Bayesian optimization algorithm, the authors develop a platform for reversely recommending metal oxide synthesis parameters with practical application value through hyperparameter optimization.

[0007] Based on this, the present invention mainly adopts the following technical solution, a method for preparing metal oxide nanoparticles with controllable particle size, which includes the following steps:

[0008] 1) extracting process parameters and corresponding nano-metal oxide particle size distributions from existing literature data on the solvothermal preparation of nano-metal oxides to obtain a literature dataset of process parameters and nano-metal oxide particle size distributions;

[0009] 2) preparing a precursor solution for preparing nano-metal oxides, synthesizing a nano-metal oxide wet aggregate precipitate by a solvothermal method, centrifuging and washing the wet aggregate precipitate, and obtaining the nano-metal oxide after post-processing; varying process parameters during the preparation process to obtain multiple sets of nano-metal oxides under different process parameters; testing the particle size distribution of each of the prepared nano-metal oxides to obtain an experimental data set of process parameter-nano-metal oxide particle size distribution; and merging the experimental data set with a literature data set into a final data set;

[0010] 3) Based on the final data set obtained in step 2), outliers in the particle size distribution are calculated and analyzed using the IQR method and the corresponding data are excluded. The process parameters are further screened to obtain key process parameters, and non-key process parameters are removed to obtain a preprocessed data set; a metal oxide target particle size prediction model based on the GBDT regression model is constructed, and the GBDT regression model is subjected to a ten-fold cross-validation using the preprocessed data set to obtain the optimal hyperparameters for preprocessing. The model is then trained using the preprocessed data set to obtain a practical nano-metal oxide target particle size prediction model;

[0011] 4) Embedding the above-mentioned nano-metal oxide target particle size prediction model into a Bayesian optimization algorithm to construct a key process parameter reverse design platform; inputting the target particle size distribution of the nano-metal oxide to be prepared, whether stirring is required, and the types of precursors, metal salts, hydrolysis inhibitors, additives and organic solvents in the precursor solution into the platform, searching the parameters through the Bayesian optimization algorithm to output the recommended remaining process parameters; designing a synthesis route according to the output process parameters, and synthesizing the nano-metal oxide according to the synthesis route.

[0012] Among them, the process parameters of the present invention include the type of precursor, the molar ratio of the precursor to water, the type of metal salt, the molar ratio of the metal salt to the precursor, the type of additive, the molar ratio of the additive to the precursor, the type of organic solvent, the molar ratio of the organic solvent to the precursor, the type of hydrolysis inhibitor, the molar ratio of the hydrolysis inhibitor to water, pH value, reaction temperature, reaction time, and whether stirring is performed.

[0013] Compared with the prior art, the present invention has at least the following beneficial effects:

[0014] 1) This paper uses a machine learning model to predict the particle size distribution of metal oxides. Through importance analysis and partial dependence analysis, it deeply explores the influencing conditions during the synthesis process of nano-metal oxides, contributing to the subsequent optimization of synthesis formulas and improvement of nano-metal oxide synthesis ideas.

[0015] 2) The method of the present invention builds a Bayesian optimization algorithm synthesis parameter reverse design platform, sets the target particle size area, and obtains recommended synthesis parameters through the Bayesian optimization method to guide the experiment, reducing the tedious formula optimization process to within a few minutes, thereby liberating productivity.

[0016] 3) Improve the development efficiency of size-adjustable nano-metal oxides through machine learning methods, comprehensively consider the impact of different reaction variables on the various properties of the products, and thus optimize the screening of precursors and the design of reaction conditions. This can save a lot of manpower and material resources and has inspiring significance for the design of other nano-functional materials. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 This is a flow chart of the method for preparing metal oxide nanoparticles with controllable particle size according to an embodiment of the present invention;

[0018] Figure 2 This is a heat map of the correlation coefficients of the response variables in Example 1;

[0019] Figure 3 This is a ranking diagram of the importance of reaction variables in Example 1;

[0020] Figure 4 This is a partial dependence analysis diagram of the reaction variables in Example 1;

[0021] Figure 5 This is a graph showing the change in relative error of the reverse recommendation platform with respect to the number of iterations in Example 1;

[0022] Figure 6 Figure 2 shows the dispersions of titanium oxide, zirconium oxide, and cerium oxide obtained by optimizing the platform in the examples;

[0023] Figure 7 is the titanium oxide particle size distribution diagram obtained by dynamic light scattering in Example 1;

[0024] Figure 8 is the X-ray diffraction pattern of titanium oxide in Example 1;

[0025] Figure 9 This is a high-resolution transmission electron micrograph of titanium oxide in Example 1;

[0026] Figure 10 is the zirconium oxide particle size distribution diagram obtained by dynamic light scattering in Example 3;

[0027] Figure 11 is the X-ray diffraction pattern of zirconium oxide in Example 3;

[0028] Figure 12 This is a high-resolution transmission electron micrograph of zirconium oxide in Example 3;

[0029] Figure 13 is a graph showing the particle size distribution of cerium oxide obtained by dynamic light scattering in Example 4;

[0030] Figure 14 is the X-ray diffraction pattern of cerium oxide in Example 4;

[0031] Figure 15 This is a high-resolution transmission electron micrograph of cerium oxide in Example 4. DETAILED DESCRIPTION

[0032] The present invention will be further described and illustrated below in conjunction with specific embodiments. The embodiments are merely illustrative of the present disclosure and do not limit its scope. The technical features of the various embodiments of the present invention may be combined accordingly, provided that there is no conflict between them.

[0033] In terms of the controlled synthesis of nano-metal oxide particle size, the present invention introduces a machine learning method to construct a prediction model based on the gradient boosting tree algorithm, trains the model with processed and screened preliminary experimental data, embeds the above-mentioned nano-metal oxide target particle size prediction model into the Bayesian optimization algorithm to construct a key process parameter reverse design platform, and uses the Bayesian optimization algorithm to reversely recommend synthesis parameter strategies to discover target products that meet the expected product performance requirements, thereby optimizing the preparation and analysis of materials, greatly improving efficiency and the comprehensive understanding of the overall experimental variables.

[0034] In the application phase of the present invention, the target particle size distribution of the nano-metal oxide (titanium oxide, zirconium oxide or cerium oxide) to be prepared, whether to stir (stirring conditions) and the types of precursor, metal salt, hydrolysis inhibitor, additive and organic solvent in the precursor solution are input into the constructed reverse design platform, and the Bayesian optimization algorithm is used to search the parameters and output the recommended remaining key process parameters (i.e., the molar ratio of precursor to water, the molar ratio of metal salt to precursor, the molar ratio of additive to precursor, the molar ratio of organic solvent to precursor, the molar ratio of hydrolysis inhibitor to water, pH value, reaction temperature, reaction time); according to the output key process parameters combined with the stirring conditions and the pre-selected types of precursor, metal salt, hydrolysis inhibitor, additive and organic solvent in the precursor solution, a statistical thermal synthesis route is designed, and nano-metal oxides with the desired target particle size distribution are synthesized according to the synthesis route.

[0035] The present invention uses a platform to obtain process parameters before synthesizing nanometal oxides using a solvothermal method. This method is a common method for preparing nanometal oxides. The synthesis route is as follows: preparing a precursor solution for preparing the nanometal oxide, synthesizing a wet aggregate precipitate of the nanometal oxide using a solvothermal method, centrifuging and washing the wet aggregate precipitate, and then post-processing the nanometal oxide. The solvothermal reaction temperature is 100-250°C, and the solvothermal reaction time is 1-72 hours. Preferably, the solvothermal reaction temperature is 110-200°C, and the solvothermal reaction time is 5-48 hours. More preferably, the solvothermal reaction temperature is 150-200°C, and the solvothermal reaction time is 8-24 hours.

[0036] The precursor of the present invention is one of a titanium source, a zirconium source, and a cerium source. The resulting nano-metal oxide is titanium oxide, zirconium oxide, or cerium oxide. The target particle size distribution parameters of the nano-metal oxide are one or more of D3, D10, D25, D50, D75, D90, and D99, preferably one or more of D3, D50, and D90. The D3 input range is 1-100 nm, the D50 input range is 1-100 nm, and the D90 input range is 10-1000 nm.

[0037] In the solvothermal method, the precursor solution includes the precursor, metal salt, water, hydrolysis inhibitor, additives and organic solvent.

[0038] The titanium source includes one or more of tetrabutyl titanate, tetraethyl titanate, isopropyl titanate, titanium tetrachloride, titanyl sulfate and titanium sulfate; the zirconium source includes one or more of zirconium oxychloride octahydrate, zirconium sulfate, zirconium nitrate, basic zirconium carbonate, tetrabutyl zirconate and tetraisopropyl zirconate; the cerium source includes one or more of cerium nitrate, cerium chloride and cerium sulfate; the hydrolysis inhibitor includes one or more of octanoic acid, butyric acid, nitric acid, hydrochloric acid and sulfuric acid; the additive includes phenyltrimethoxysilane, γ-aminopropyl The organic solvent comprises one or more of triethoxysilane, γ-glycidyloxypropyltrimethoxysilane, 3-(methacryloyloxy)propyltrimethoxysilane, 3-(acryloyloxy)propyltrimethoxysilane, 3-mercaptopropyltrimethoxysilane, vinyltrimethoxysilane, and vinyltriethoxysilane; and the organic solvent comprises one or more of methanol, ethanol, isopropanol, n-butanol, tert-butanol, octanol, isomeric hexadecanol, propylene glycol, benzyl alcohol, acetone, butanone, propylene glycol methyl ether acetate, and tetrahydrofuran.

[0039] An embodiment of the present invention provides an exemplary method for preparing metal oxide nanoparticles with controllable particle size, comprising the following steps:

[0040] 1) extracting process parameters and corresponding nano-metal oxide particle size distributions from existing literature data on the solvothermal preparation of nano-metal oxides to obtain a literature dataset of process parameters and nano-metal oxide particle size distributions;

[0041] The process parameters include the type of precursor, the molar ratio of the precursor to water, the type of metal salt, the molar ratio of the metal salt to the precursor, the type of additive, the molar ratio of the additive to the precursor, the type of organic solvent, the molar ratio of the organic solvent to the precursor, the type of hydrolysis inhibitor, the molar ratio of the hydrolysis inhibitor to water, pH value, reaction temperature, reaction time, and whether stirring is performed.

[0042] In a specific embodiment of the present invention, a literature data automatic processing script can be used to obtain literature data, and optical-grade image recognition technology can be used to search the literature data for relevant images related to the size of nanometal oxides. A multivariate bag-of-words method is used to identify specific keywords contained in the images and exclude irrelevant images. The multivariate bag-of-words method is as follows: first, images containing keywords such as "size", "diameter" or "dimension" are identified from the images as candidate images, and then images containing keywords such as "pore" are removed from the candidate images. The image recognition effect obtained by this constraint is the best.

[0043] The obtained images were digitized using Origin and the particle size distribution data was extracted. The process parameters for synthesizing the nano-metal oxide were further obtained from literature data to obtain a literature data set of process parameters-nano-metal oxide particle size distribution.

[0044] This example collects a large amount of nano-metal oxide synthesis data published in international journals using the method described above.

[0045] 2) preparing a precursor solution for preparing nano-metal oxides, synthesizing a nano-metal oxide wet aggregate precipitate by a solvothermal method, centrifuging and washing the wet aggregate precipitate, and obtaining the nano-metal oxide after post-processing; varying process parameters during the preparation process to obtain multiple sets of nano-metal oxides under different process parameters; and testing the particle size distribution of each of the prepared nano-metal oxides to obtain an experimental data set of process parameter-nano-metal oxide particle size distribution;

[0046] The precursor solution used in the solvothermal preparation process is described above. The post-treatment includes drying, vacuum drying, or freeze drying. The particle size distribution of the nanometal oxide is determined by characterization methods such as dynamic light scattering and high-resolution transmission electron microscopy.

[0047] In this embodiment, in order to quantify the process parameters and particle size distribution data, the particle size distribution data obtained in step 1) and step 2) and the corresponding process parameters are quantified, and after quantification, the experimental data set and the literature data set are merged into the final data set.

[0048] The quantification steps include: first, one-hot encoding the process parameters such as "precursor type", "metal salt type", "additive type", "solvent type", "hydrolysis inhibitor type", and "whether to stir"; then, standardizing, scaling, and reducing the dimension of the particle size distribution data and its corresponding process parameter data.

[0049] In one embodiment of the present invention, one-hot encoding is as follows: a sparse matrix is ​​established according to these category variables, and the sparse matrix is ​​converted into {0, 1} related features as input.

[0050] Based on the above data processing, through normalization, using the expectation and variance of variable data, data of different orders of magnitude are uniformly scaled to data with an expectation of 0 and a variance of 1.

[0051] Use heatmap to analyze the correlation between input features. Heatmap can intuitively display the correlation coefficients between various features. The heatmaps of experimental variables are as follows: Figure 2 As shown,

[0052] 3) If Figure 1 As shown, according to the final data set obtained in step 2), the outliers of the particle size distribution are calculated and analyzed by the IQR method and the corresponding data are excluded (the range after excluding the outliers here is consistent with the target particle size distribution input range of the reverse design platform constructed subsequently. Exemplarily, taking D3, D50, and D90 as examples, this step can exclude particle size distributions outside the range of 1-100 nm for D3, 1-100 nm for D50, and 10-1000 nm for D90 as outliers), further screen the process parameters to obtain key process parameters, and remove non-key process parameters to obtain a preprocessed data set; construct a metal oxide target particle size prediction model based on the GBDT regression model, perform a ten-fold cross validation on the GBDT regression model through the preprocessed data set to obtain the optimal hyperparameters for preprocessing, and then train the model with the preprocessed data set to obtain a nano-metal oxide target particle size prediction model that can be practically applied;

[0053] The optimal hyperparameters for preprocessing are obtained by performing a ten-fold cross validation on the GBDT regression model using the preprocessed dataset, specifically including:

[0054] a) The preprocessed data set is randomly divided into a training set and a test set. The metal oxide target particle size prediction model is a gradient boosting tree model. The model parameters of the gradient boosting tree model are adjusted using the training set. The model parameters include optimization criteria, number of decision trees, and maximum depth of decision trees. The prediction results of the metal oxide target particle size prediction model after parameter adjustment are evaluated and verified using the test set. The evaluation indicators include mean square error (RMSE) and determination coefficient (R). 2 ;

[0055] b) Re-divide the training set and test set, and repeat step a) until the set number of cycles is reached.

[0056] 4) Embedding the above-mentioned nano-metal oxide target particle size prediction model into a Bayesian optimization algorithm to construct a key process parameter reverse design platform; inputting the target particle size distribution of the nano-metal oxide to be prepared into the platform, searching the parameters through the Bayesian optimization algorithm to output recommended key process parameters; designing a synthesis route according to the output key process parameters, and synthesizing the nano-metal oxide according to the synthesis route.

[0057] The synthesis parameter reverse design platform described in step 4) is obtained by embedding the preprocessing database obtained in step 2) and the nano-metal oxide target particle size prediction model obtained in step 3) into a Bayesian optimization algorithm; through the original uncertainty search capability of the Bayesian algorithm, the nano-metal oxide target particle size prediction model obtained in step 3) is used as a nano-metal oxide characterization proxy model, and the key process parameters corresponding to the input target particle size are obtained by continuous iterative optimization.

[0058] In this embodiment, for step 2), a specific experimental scheme and operating instructions are given as follows:

[0059] Dissolve 0.75 mL of deionized water and 0.13 g of aluminum nitrate nonahydrate in 20 mL of benzyl alcohol and stir thoroughly until dissolved. Add 5.7 mL of tetrabutyl titanate to the mixture and stir under high-speed magnetic stirring to form a homogeneous, transparent precursor solution. Transfer the precursor solution to a 50 mL hydrothermal autoclave and set the reaction temperature at 200°C for 1 hour. Centrifuge the wet solid suspension at 4500 rpm, wash with benzyl alcohol, and repeat the centrifugation three times.

[0060] The particle size distribution of the nano titanium dioxide was tested.

[0061] The importance of the process parameters concerned in the present invention is analyzed, such as Figure 3 As shown, f1 represents the type of precursor, f2 represents the molar ratio of water to precursor, f3 represents the type of additive, f4 represents the molar ratio of additive to precursor, f5 represents pH value, f6 represents the type of dopant, f7 represents the molar ratio of dopant to precursor, f8 represents temperature, f9 represents reaction duration, f10 represents the type of solvent, f11 represents the molar ratio of solvent to precursor, f12 represents the presence of stirring, and f13 represents the absence of stirring. The experiment found that the type of dopant, the type of precursor, and the type of solvent have the strongest influence on the particle size distribution of nano-metal oxides. The partial dependence analysis technique was used to analyze the numerical variables, such as Figure 4 , solvents, dopants and other substances have a certain regional effect on the particle size distribution of nano-metal oxides. When exceeding this region, the influence of these variables on the particle size distribution will become less obvious. Based on this, preferably, the process parameters can be screened according to importance according to demand to obtain key process parameters, thereby reducing the amount of data and the training cost of subsequent models. When screening, process parameters with low importance are preferentially removed, as their impact on the target particle size distribution is usually low. When ignoring the cost, more process parameters can be considered to be retained according to importance.

[0062] In one embodiment, the above-mentioned model of the present invention is embedded in the Bayesian optimization algorithm to develop a reverse design platform for synthesis parameters. When applied, the target particle size distribution of the nano-metal oxide, stirring conditions (whether to stir), and the types of precursors, metal salts, hydrolysis inhibitors, additives, and organic solvents in the precursor solution are set. The platform is guided by the particle size distribution of the nano-metal oxide and reversely recommends synthesis parameters through the platform, such as Figure 5 As shown in the figure, only 45 iterations of optimization are needed to obtain better results.

[0063] Example 1

[0064] For titanium oxide nanoparticles, the D50 range is set to 8-15nm, stirring is selected, the types of precursors, metal salts, hydrolysis inhibitors, additives and organic solvents in the required precursor solution are selected, and the synthesis parameter reverse design platform is input. The optimized synthesis process parameters are obtained using the platform. According to the process parameters, the solvent thermal synthesis route is designed according to the solvent thermal method described above: 22mL of butanol, 0.39mL of butyric acid and 0.75mL of deionized water are mixed evenly with a high-speed magnetic stirrer, and 5.72mL of tetrabutyl titanate is added to the solution and mixed evenly. The solution is placed in a reactor and kept warm at 180℃ for 20h. The turbid liquid after the reaction is taken out and centrifuged at 4500rpm for 5min to obtain a white wet cake at the bottom. An appropriate amount of acetone is added to wash and centrifuge, and repeated 2-3 times. It is prepared into a dispersion and subjected to dynamic light scattering ( Figure 7 )、XRD( Figure 8 )、TEM( Figure 9 ) analysis. Dynamic light scattering measured a D50 of approximately 10 nm, which falls within the target particle size range. XRD analysis confirmed that the synthesized titanium oxide nanoparticles were anatase-type. TEM analysis further confirmed that the measured particle size fell within the target range.

[0065] Example 2

[0066] Set the zirconium oxide D50 range to 8-15nm, choose not to stir, select the types of precursors, metal salts, hydrolysis inhibitors, additives and organic solvents in the required precursor solution, and input them into the platform. Use the platform to obtain optimized synthesis process parameters and design the solvent thermal synthesis route as follows: Mix 10g of basic zirconium carbonate, 10g of sodium hydroxide, and 100g of water, and place the solution in a reactor and keep it at 110°C for 6h. Take out the turbid liquid after the reaction and centrifuge it at 4500rpm for 5min to obtain a white wet cake at the bottom. Add appropriate amount of water to wash and centrifuge, repeat 2-3 times. Prepare it into a dispersion and perform dynamic light scattering ( Figure 10 )、TEM( Figure 11 )、XRD( Figure 12) analysis. Dynamic light scattering measured a D50 of approximately 9 nm, which falls within the target particle size range. XRD analysis confirmed that the synthesized zirconia nanoparticles were tetragonal. TEM analysis further confirmed that the measured particle size fell within the target range.

[0067] Example 3

[0068] Set the D50 range of cerium oxide to 10-20nm, choose stirring, select the types of precursors, metal salts, hydrolysis inhibitors, additives and organic solvents in the required precursor solution, and input them into the platform. Use the platform to obtain optimized synthesis process parameters and design a solvothermal synthesis route as follows: In a 250ml beaker, dissolve 2.52g of cerium nitrate hexahydrate in 100ml of a 1:1 water-ethylene glycol mixture. Heat the mixed solution to 60°C and add 16ml of 25% ammonia water under strong stirring. Stir for three hours, cool to room temperature, wash with deionized water, and repeat 2-3 times. Prepare it into a dispersion and perform dynamic light scattering ( Figure 13 )、TEM( Figure 14 )、XRD( Figure 15 ) analysis. Dynamic light scattering measured a D50 of approximately 20 nm, which falls within the target particle size range. XRD analysis confirmed that the synthesized cerium oxide nanoparticles were cubic. TEM analysis further confirmed that the measured particle size fell within the target range.

[0069] The titanium oxide, zirconium oxide and cerium oxide dispersions obtained by the solvent thermal method through platform optimization in the above examples 1-3 are as follows: Figure 6 As shown above, the present invention uses a machine learning model to predict the particle size distribution of metal oxides. Through importance analysis and partial dependence analysis, it deeply explores the influencing conditions of nano-metal oxides in the synthesis process, which contributes to the subsequent optimization of synthesis formula and improvement of nano-metal oxide synthesis ideas.

[0070] The above-described embodiments merely represent several implementation methods of the present invention. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art would be able to make various modifications and improvements without departing from the spirit of the present invention, all of which fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be determined by the appended claims.

Claims

1. A method for preparing metal oxide nanoparticles with controllable particle size, characterized in that: The steps include: 1) extracting process parameters and corresponding nano-metal oxide particle size distributions from existing literature data on the solvothermal preparation of nano-metal oxides to obtain a literature dataset of process parameters and nano-metal oxide particle size distributions; 2) preparing a precursor solution for preparing nano-metal oxides, synthesizing a wet aggregate precipitate of the nano-metal oxides by a solvothermal method, centrifuging and washing the wet aggregate precipitate, and obtaining the nano-metal oxides after post-treatment; Changing the process parameters during the preparation process to obtain multiple groups of nano-metal oxides under different process parameters; performing a particle size distribution test on each of the prepared nano-metal oxides to obtain an experimental data set of process parameter-nano-metal oxide particle size distribution; Merge the experimental dataset and the literature dataset into the final dataset; 3) Based on the final data set obtained in step 2), outliers in the particle size distribution are calculated and analyzed using the IQR method and the corresponding data are excluded. The process parameters are further screened to obtain key process parameters, and non-key process parameters are removed to obtain a preprocessed data set; a metal oxide target particle size prediction model based on the GBDT regression model is constructed, and the GBDT regression model is subjected to a ten-fold cross-validation using the preprocessed data set to obtain the optimal hyperparameters for preprocessing. The model is then trained using the preprocessed data set to obtain a practical nano-metal oxide target particle size prediction model; 4) Embedding the above-mentioned nano-metal oxide target particle size prediction model into the Bayesian optimization algorithm to construct a reverse design platform for key process parameters; The platform inputs the target particle size distribution of the nanometal oxide to be prepared, whether stirring is required, and the types of precursors, metal salts, hydrolysis inhibitors, additives, and organic solvents in the precursor solution. The platform then searches for parameters using a Bayesian optimization algorithm and outputs the recommended remaining process parameters. The synthesis route is designed according to the output process parameters, and the nanometal oxide is synthesized according to the synthesis route.

2. The preparation method according to claim 1, wherein: In step 1), the process parameters and corresponding nano-metal oxide particle size distribution extracted from literature data include: Literature data was obtained, and relevant images about the size of nanometal oxides were searched in the literature data using image recognition technology. The specific keywords contained in the images were identified using a multivariate bag-of-words method, and irrelevant images were excluded. The obtained images were digitized using Origin and the particle size distribution data was extracted. The process parameters for synthesizing the nanometal oxides were further obtained from the literature data to obtain a literature dataset of process parameters-nanometal oxide particle size distribution.

3. The preparation method according to claim 2, wherein : The multivariate bag-of-words method is used to identify specific keywords contained in pictures and exclude irrelevant pictures. Specifically, the pictures containing the keyword "size", "diameter" or "dimension" are first identified as candidate pictures, and then the pictures containing the keyword "pore" are removed from the candidate pictures.

4. The preparation method according to claim 2, characterized in that : After obtaining the particle size distribution data and the corresponding process parameters in step 1) and step 2), the method further includes quantifying the particle size distribution and the process parameters to obtain a corresponding data set; The quantification steps include: firstly performing one-hot encoding on the process parameters such as "precursor type", "metal salt type", "additive type", "solvent type", "hydrolysis inhibitor type", and "whether stirring is used"; and then performing standardization, scaling, and dimensionality reduction on the particle size distribution data and its corresponding process parameter data.

5. The preparation method according to claim 1, wherein: The precursor solution in step 2) comprises a precursor, a metal salt, water, a hydrolysis inhibitor, an additive, and an organic solvent; wherein the precursor is one of a titanium source, a zirconium source, and a cerium source; The titanium source includes one or more of tetrabutyl titanate, tetraethyl titanate, isopropyl titanate, titanium tetrachloride, titanyl sulfate and titanium sulfate; The zirconium source includes one or more of zirconium oxychloride octahydrate, zirconium sulfate, zirconium nitrate, basic zirconium carbonate, tetrabutyl zirconate, and tetraisopropyl zirconate; The cerium source includes one or more of cerium nitrate, cerium chloride, and cerium sulfate; the hydrolysis inhibitor includes one or more of octanoic acid, butyric acid, nitric acid, hydrochloric acid, and sulfuric acid; The additive includes one or more of phenyltrimethoxysilane, γ-aminopropyltriethoxysilane, γ-glycidoxypropyltrimethoxysilane, 3-(methacryloyloxy)propyltrimethoxysilane, 3-(acryloyloxy)propyltrimethoxysilane, 3-mercaptopropyltrimethoxysilane, vinyltrimethoxysilane, and vinyltriethoxysilane; The organic solvent includes one or more of methanol, ethanol, isopropanol, n-butanol, tert-butanol, octanol, isomeric hexadecanol, propylene glycol, benzyl alcohol, acetone, butanone, propylene glycol methyl ether acetate, and tetrahydrofuran.

6. The preparation method according to claim 1, characterized in that The process parameters include the type of precursor, the molar ratio of the precursor to water, the type of metal salt, the molar ratio of the metal salt to the precursor, the type of additive, the molar ratio of the additive to the precursor, the type of organic solvent, the molar ratio of the organic solvent to the precursor, the type of hydrolysis inhibitor, the molar ratio of the hydrolysis inhibitor to water, the pH value, the reaction temperature, the reaction time, and whether stirring is used; the reaction temperature described in step 2) is 100-250°C, and the reaction time is 1-72h.

7. The preparation method according to claim 1, characterized in that : In step 2), the particle size distribution of the nano-metal oxide is measured by dynamic light scattering and high-resolution transmission electron microscopy.

8. The preparation method according to claim 1, characterized in that : The optimal hyperparameters for preprocessing are obtained by performing a ten-fold cross validation on the GBDT regression model using the preprocessed dataset, specifically including: a) The preprocessed data set is randomly divided into a training set and a test set. The metal oxide target particle size prediction model is a gradient boosting tree model. The model parameters of the gradient boosting tree model are adjusted using the training set. The model parameters include optimization criteria, number of decision trees, and maximum depth of decision trees. The prediction results of the metal oxide target particle size prediction model after parameter adjustment are evaluated and verified using the test set. The evaluation indicators include mean square error (RMSE) and determination coefficient (R). 2 ; b) Re-divide the training set and test set, and repeat step a) until the set number of cycles is reached.

9. The preparation method according to claim 1, wherein: The synthesis parameter reverse design platform described in step 4) is obtained by embedding the preprocessing database obtained in step 2) and the nano-metal oxide target particle size prediction model obtained in step 3) into a Bayesian optimization algorithm; through the original uncertainty search capability of the Bayesian algorithm, the nano-metal oxide target particle size prediction model obtained in step 3) is used as a nano-metal oxide characterization proxy model, and the key process parameters corresponding to the input target particle size are obtained by continuous iterative optimization.

10. The preparation method according to claim 1, characterized in that: The nano metal oxide is titanium oxide, zirconium oxide or cerium oxide; the target particle size distribution of the nano metal oxide input into the key process parameter reverse design platform in step 4) is one or more of D3, D10, D25, D50, D75, D90, and D99.

Images