A machine learning-based multi-objective optimization method for MOFs synthesis routes

Abstract

The application discloses a MOFs synthesis route multi-objective optimization method based on machine learning. The method is to collect the synthesis conditions of prepared Ce-UiO-66, and evaluate the defect content and thermal stability thereof through a thermogravimetric analysis curve, as initial data set; the data set is randomly divided into a training set and a test set, eight algorithms are adopted to model each performance of Ce-UiO-66 and select the proxy model for performance prediction; the target achievement probability (PA) value of each performance in the synthesis space is calculated and expanded into a multi-objective evaluation factor; the Ce-UiO-66 material is prepared; the obtained material is subjected to characterization test, if the test data does not meet the requirement, the data set and the proxy model are updated. The application has the advantages of low cost, short cycle and the like in optimizing the catalytic and stable performances of MOFs based on reliable experimental data and machine learning, and can be popularized to the design of synthesis routes of other materials.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence and nanocomposite catalytic materials technology, and in particular to a multi-objective optimization method for MOF synthesis routes based on machine learning. Background Technology

[0002] Metal-organic frameworks (MOFs) are a new family of highly porous crystalline compounds, consisting of periodic network structures assembled from inorganic structural units (atoms, clusters, etc.) and organic ligands via coordination bonds. Due to their unique porous structure, tunable chemical functions, and diverse topologies, MOFs have attracted significant attention and active participation from researchers across multiple disciplines in the field of catalysis. Based on in-depth experimental and theoretical studies, the application of MOFs in catalytic reactions (e.g., Diels-Alder reactions, Knoevenagel condensation reactions, carbon dioxide hydrogenation reactions, dicyclopentadiene hydrogenation reactions, etc.) largely depends on their structural irregularities (i.e., defects). An effective method to improve catalytic performance is to adjust the electronic structure of the active center by introducing defects or dopants; unsaturated coordination sites (i.e., ligand defect sites) can serve as active sites for substrate activation. Therefore, intelligent control of structural defects in MOFs is crucial for achieving ideal performance.

[0003] On the other hand, while defect engineering in MOFs offers new opportunities for catalytic applications, it also reduces the crystal stability of MOFs, making their structures prone to collapse. The thermal stability of MOFs is particularly critical for high-temperature applications, such as those involving exothermic catalytic reactions. The thermal stability of MOF materials is a result of the binding strength between the metal center and the organic ligands, as well as the number of organic ligands connecting the metal ions or clusters. The absence of organic ligands between the metal centers can adversely affect thermal stability. Therefore, developing MOFs with good catalytic performance while maintaining a certain level of thermal stability is an urgent need for the efficient development of MOF catalytic materials.

[0004] Currently, although research on MOF synthesis is in-depth, systematic analysis of the synthesis process itself and the influence of various experimental variables is still lacking. Research on defect control mainly relies on traditional experimental trial-and-error methods, which suffer from typical problems such as long development cycles, high costs, and low efficiency. Therefore, accurately formulating synthesis routes in the unexplored experimental space, fundamentally changing the traditional trial-and-error experimental methods that rely on prior knowledge, is a scientific challenge that needs to be solved. Using machine learning to assist in the efficient planning of synthesis strategies is essential for achieving the precise preparation of target MOF materials. It can effectively avoid the waste of time and costs caused by blind experiments and drive the precise synthesis of MOF materials. Therefore, developing a multi-objective optimization method for MOF synthesis routes based on machine learning has broad practical prospects. Summary of the Invention

[0005] The technical problem this invention aims to solve is to provide a multi-objective optimization method for MOF synthesis routes based on machine learning. This method overcomes the drawbacks of long experimental cycles and high costs associated with simultaneously optimizing both defect content and thermal stability of MOF materials. Instead, it offers a simple, fast, low-cost, and labor-saving method based on machine learning to simultaneously optimize multiple properties. This invention employs a multi-objective optimization strategy using machine learning to fundamentally change the traditional trial-and-error experimental method that relies on prior knowledge, accurately and efficiently guiding the experimental synthesis of MOF materials and achieving a significant leap in research and development efficiency.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] A machine learning-based multi-objective optimization method for MOF synthesis routes includes the following steps:

[0008] S1. Collect the synthesis conditions such as reaction temperature, reaction time, and amount of regulator used to prepare Ce-UiO-66, and evaluate its defect content and thermal stability by thermogravimetric analysis curves. This data will be used as the Ce-UiO-66 synthesis condition-performance dataset.

[0009] S2. Randomly divide the dataset into training and test sets, use eight machine learning algorithms to model the performance of Ce-UiO-66, and select and determine the surrogate models used for performance prediction.

[0010] S3. Determine the candidate synthesis condition space as the search space to be optimized, calculate the target achievement probability (PA) value of each performance of Ce-UiO-66 in the search space and expand it into a multi-objective evaluation factor, and screen to obtain the synthesis scheme that is expected to achieve the target performance.

[0011] S4. Ce-UiO-66 material was prepared using a hydrothermal method according to the expected synthesis scheme;

[0012] S5. Characterize the Ce-UiO-66 material obtained by preparation with the target performance. If the test data does not meet the target requirements, update the dataset and surrogate model.

[0013] Preferably, in S1, the Ce-UiO-66 synthesis conditions-performance dataset specifically refers to: conducting a literature review on Ce-UiO-66 to establish a Ce-UiO-66 synthesis conditions-performance dataset containing seven experimental variables, including reaction temperature, reaction time, formic acid dosage, cerium ammonium nitrate dosage, terephthalic acid dosage, N,N-dimethylformamide (DMF) dosage, and water dosage, including synthesis conditions, ligand deficiency content, and thermal stability.

[0014] Preferably, in S1, the evaluation of defect content and thermal stability through thermogravimetric analysis (TGA) curves specifically involves: finding the second derivative extreme point of the TGA data curve as the region of material weight loss; drawing tangents at the horizontal level before the curve step in the weight loss region and at the curve inflection point; the intersection of the two tangents is the reference temperature point at which the weight loss process begins, used to characterize the material's thermal stability, and serving as one of the target performance parameters for machine learning; the weight percentage at the intersection of the two tangents is denoted as y1; the weight percentage after complete thermal decomposition of the material at 500℃ is denoted as y2; and Ce6O is calculated using theoretical formulas. 6+x (BDC) 6-x Defective content of ligand deficiency Used to characterize the catalytic activity of materials, as one of the target performance parameters for machine learning.

[0015] Preferably, in S2, during the performance modeling process, eight machine learning algorithms are selected, including: random forest, kernel ridge regression, support vector machine, K nearest neighbors, neural network, Xgboost (eXtreme Gradient Boost), Adaboost (Adaptive Boost), and LightGBM (Light Gradient Boosting Machine), and Bayesian optimization is used to tune the hyperparameters.

[0016] Preferably, in S2, the step of randomly dividing the dataset into a training set and a test set, and selecting surrogate models for each performance prediction, specifically involves: during the training process, randomly dividing the dataset into a training set and a test set in a ratio of 0.8:0.2, using the formula... Calculate the mean absolute error (MAE) of the model predictions, where x represents the entire training set data and y represents the entire test set data. i y represents the actual data. i The data predicted by the machine learning model is represented by i, which represents the index of the data sample in the training or test set, and n represents the total number of data samples in the training or test set. This process is repeated 100 times to calculate the mean of MAE on the training and test sets to analyze the accuracy of the model. The machine learning model with the lowest mean absolute error is the surrogate model used for predicting material defect content and thermal stability performance.

[0017] Preferably, in S2, the optimal model algorithm is the Adaboost algorithm.

[0018] Preferably, in S3, determining the candidate synthesis condition space as the search space to be optimized specifically involves: based on the synthesis condition categories for preparing Ce-UiO-66, determining the search space for Ce-UiO-66 synthesis conditions composed of seven experimental variables: reaction temperature, reaction time, formic acid dosage, cerium ammonium nitrate dosage, terephthalic acid dosage, N,N-dimethylformamide dosage, and water dosage; limiting the reaction temperature to be greater than or equal to 30℃ and less than or equal to 130℃, with a change step size of 10℃; and limiting the reaction time to be greater than or equal to 15 min and less than or equal to 1470 min. The dosage of formic acid is limited to 0 mmol / L and 110 mmol / L, with a step size of 10 mmol / L; the dosage of cerium ammonium nitrate is limited to 2 mmol / L; the dosage of terephthalic acid is limited to 1 mmol / L and 4 mmol / L, with a step size of 1 mmol / L; the dosage of water is limited to 400 mmol / L and 2200 mmol / L, with a step size of 100 mmol / L; to ensure complete dissolution of terephthalic acid, the dosage of N,N-dimethylformamide is limited to 80 times the dosage of terephthalic acid.

[0019] Preferably, in S3, the calculation of the target achievement probability (PA) values ​​of each performance of Ce-UiO-66 in the search space and its expansion into a multi-objective evaluation factor specifically involves: performing performance prediction on each synthesis scheme in the search space using the M performance prediction proxy models to obtain M predicted values, determining the mean μ and variance σ of the performance prediction values, and determining the target material defect content g. DC(defect content) =50%, thermal stability target g TS(thermal stability) =300℃, according to the formula Determine the target achievement probability (PA) value for each performance of each synthesis scheme in the search space, where Φ represents the cumulative distribution function, and μ(x), σ(x), and g are the predicted mean, predicted standard deviation, and expected target value, respectively, using the formula... Calculate the weighted sum of the probabilities of achieving each performance objective, expanding the evaluation factors to multiple objectives, where ω m PA m The weights, set ω m = (0.7, 0.3) determines the ratio of defect content to thermal stability.

[0020] The method also includes terminating the design if the test data meets the requirements, thereby obtaining a Ce-UiO-66 material with multi-performance synergistic optimization.

[0021] The beneficial effects of the above-described technical solution of the present invention are as follows:

[0022] 1) The efficient planning method of using machine learning-assisted synthesis strategy avoids the waste of resources caused by traditional experimental-intensive "trial and error" experiments, reduces R&D costs, and shortens the R&D cycle;

[0023] 2) The multi-objective optimization design method synergistically improves the catalytic activity and thermal stability of the material, and has good practical value;

[0024] 3) The method provided by this invention can realize the rapid search of the globally optimal synthesis scheme in the candidate synthesis space, overcoming the disadvantage that it is difficult to try each of the massive experimental schemes in a vast space. Attached Figure Description

[0025] Figure 1 This is a logical framework diagram of the present invention;

[0026] Figure 2 a is a performance evaluation diagram of the model for predicting the defect content of MOFs in this invention.

[0027] Figure 2 b is a performance evaluation diagram of the model for predicting the thermal stability performance of MOFs in this invention. Detailed Implementation

[0028] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0029] Example

[0030] In one embodiment, such as Figure 1 As shown, this embodiment of the invention provides a multi-objective optimization method for MOF synthesis routes based on machine learning, including the following steps:

[0031] S1. A literature review was conducted on Ce-UiO-66 to establish a Ce-UiO-66 synthesis condition-performance dataset containing seven experimental variables: reaction temperature, reaction time, formic acid dosage, cerium ammonium nitrate dosage, terephthalic acid dosage, N,N-dimethylformamide dosage, and water dosage. This dataset includes synthesis conditions, ligand deficiency content, and thermal stability. The second derivative extreme point of the thermogravimetric analysis (TGA) curve was determined as the region of material weight loss. Tangents were drawn at the horizontal level before the curve step and at the curve inflection point in the weight loss region. The intersection of these two tangents is the reference temperature at which the weight loss process begins, used to characterize the material's thermal stability and serving as one of the target performance parameters for machine learning. The weight percentage at the intersection of the two tangents is denoted as y1. The weight percentage after complete thermal decomposition at 500℃ is denoted as y2. Ce6O is calculated using theoretical formulas. 6+x (BDC) 6-x Defective content of ligand deficiency Used to characterize the catalytic activity of materials, as one of the target performance parameters for machine learning.

[0032] S2. For the defect content performance training set and the thermal stability performance training set, the datasets are randomly divided into training and test sets in a ratio of 0.8:0.2. Eight machine learning model algorithms are trained respectively: Random Forest, Kernel Ridge Regression, Support Vector Machine, K-Nearest Neighbors, Neural Network, XGBoost, AdaBoost, and LightGBM. The Ce-UiO-66 synthesis conditions are used as model inputs, and defect content and thermal stability are used as model outputs. Bayesian optimization is used to fine-tune the model hyperparameters, using the formula... Calculate the mean absolute error, where x represents the entire training set data and y represents the entire test set data. i y represents the actual data. i The data predicted by the machine learning model is represented by i, which represents the index of the data sample in the training or test set, and n represents the total number of data samples in the training or test set. This process is repeated 100 times to calculate the mean of MAE on the training and test sets to analyze the accuracy of the model. The determined optimal parameter model is then applied to the training set for training. Figure 2 a and Figure 2 Figure b shows the performance evaluation graphs of the Adaboost model for predicting defect content and thermal stability, respectively. The hyperparameters of the Adaboost model for predicting thermal stability are 'learning_rate': 0.80, 'loss': 'square', 'n_estimators': 85, 'random_state': 420; the hyperparameters of the Adaboost model for predicting defect content are 'learning_rate': 0.19, 'loss': 'exponential', 'n_estimators': 84, 'random_state': 420.

[0033] S3. Based on the categories of synthetic conditions for preparing Ce-UiO-66, determine the search space for Ce-UiO-66 synthetic conditions, consisting of seven experimental variables: reaction temperature, reaction time, formic acid dosage, cerium ammonium nitrate dosage, terephthalic acid dosage, N,N-dimethylformamide dosage, and water dosage. Limit the reaction temperature to ≥30℃ and ≤130℃, with a step size of 10℃; limit the reaction time to ≥15min and ≤1470min, with a step size of 30min; limit the formic acid dosage... The amount of acid used is greater than or equal to 0 mmol and less than or equal to 110 mmol, with a change step of 10 mmol; the amount of cerium ammonium nitrate is limited to 2 mmol; the amount of terephthalic acid is limited to greater than or equal to 1 mmol and less than or equal to 4 mmol, with a change step of 1 mmol; the amount of water is limited to greater than or equal to 400 mmol and less than or equal to 2200 mmol, with a change step of 100 mmol; to ensure complete dissolution of terephthalic acid, the amount of N,N-dimethylformamide is limited to 80 times the amount of terephthalic acid.

[0034] Furthermore, the target achievement probability (PA) values ​​for each performance of Ce-UiO-66 in the search space are calculated and expanded into multi-objective evaluation factors. Specifically, the performance of each synthesis scheme in the search space is predicted using the M performance prediction surrogate models to obtain M predicted values. The mean μ and variance σ of the predicted performance values ​​are determined, and the target material defect content g is determined. DC(defect content) =50%, thermal stability target g TS(thermal stability) =300℃, according to the formula Determine the target achievement probability (PA) value for each performance of each synthesis scheme in the search space, where Φ represents the cumulative distribution function, and μ(x), σ(x), and g are the predicted mean, predicted standard deviation, and expected target value, respectively, using the formula... Calculate the weighted sum of the probabilities of achieving each performance objective, expanding the evaluation factors to multiple objectives, where ω m PA m The weight, set ω m = (0.7, 0.3) represents the weight ratio of defect content and thermal stability in the multi-objective optimization process.

[0035] S4. Ce-UiO-66 material was prepared using a hydrothermal method according to the candidate synthesis scheme, as follows:

[0036] Cerium ammonium nitrate was dissolved in deionized water to obtain solution A, and terephthalic acid was dissolved in N,N-dimethylformamide solution to obtain solution B. Both solutions were sonicated for 5 min, and then solution B was poured into solution A and sonicated for another 1 min to mix them evenly. The solutions were then poured into a round-bottom flask, sealed, and reacted at the expected temperature for the expected time. After cooling to room temperature, the mixture was washed twice by centrifugation in DMF and four times by centrifugation in acetone. The resulting solid was dried in air at 70°C for 24 h to obtain Ce-UiO-66 material.

[0037] S5. For the Ce-UiO-66 material prepared in S4, the thermal stability of the sample was determined using a thermogravimetric analyzer (TGA) under air atmosphere at a heating rate of 5℃·min. -1 The temperature range is from 30℃ to 800℃ to obtain the synthesis conditions and target performance data of the newly prepared Ce-UiO-66 material, which are then fed back to the initial dataset to perform iterative collaborative optimization of each target performance.

[0038] Based on the above implementation steps, through three rounds of experimental iterations, with 10 experimental schemes selected and implemented in each round, a Ce-UiO-66 material meeting the target requirements was obtained, with a thermal decomposition temperature of 303℃ and a defect content of 45.03%, which is consistent with the material thermal stability target g set in the multi-objective optimization. TS =300℃, target defect content (g) DC =50%, basically consistent.

[0039] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A multi-objective optimization method for MOF synthesis routes based on machine learning, characterized in that, Includes the following steps: S1. The synthesis conditions for Ce-UiO-66 were collected, and its defect content and thermal stability were evaluated using thermogravimetric analysis (TGA) curves. This data was used as the Ce-UiO-66 synthesis condition-performance dataset. The defect content and thermal stability were evaluated using TGA curves. Specifically, the second derivative extremum of the TGA curve was identified as the region of material weight loss. Tangents were drawn at the horizontal point before the curve step and at the curve inflection point in the weight loss region. The intersection of these two tangents is the reference temperature at which the weight loss process begins, used to characterize the material's thermal stability and serving as one of the target performance parameters for machine learning. The weight percentage at the intersection of the two tangents is denoted as y1. The weight percentage after complete thermal decomposition at 500 °C is denoted as y2. Ce6O was calculated using theoretical formulas. 6+x (BDC) 6-x Defective content of ligand deficiency This is used to characterize the catalytic activity of materials, serving as one of the target performance metrics for machine learning. S2. Randomly divide the dataset into training and test sets, use eight machine learning algorithms to model the performance of Ce-UiO-66, and select and determine the surrogate models used for performance prediction. S3. Determine the candidate synthesis condition space as the search space to be optimized, calculate the target achievement probability PA value of Ce-UiO-66 in the search space and expand it into a multi-objective evaluation factor, and screen to obtain the synthesis scheme that is expected to achieve the target performance. The determination of the candidate synthesis condition space as the search space to be optimized specifically involves: based on the synthesis conditions for preparing Ce-UiO-66, determining the search space for Ce-UiO-66 synthesis conditions composed of seven experimental variables: reaction temperature, reaction time, formic acid dosage, cerium ammonium nitrate dosage, terephthalic acid dosage, N,N-dimethylformamide dosage, and water dosage. The reaction temperature is limited to ≥30 °C and ≤130 °C, with a step size of 10 °C; the reaction time is limited to ≥15 min and ≤1470 min, with a step size of 30 min; the formic acid dosage is limited to ≥0 mmol and ≤110 mmol, with a step size of 10 mmol; the cerium ammonium nitrate dosage is limited to 2 mmol; the terephthalic acid dosage is limited to ≥1 mmol and ≤4 mmol, with a step size of 1 mmol; and the water dosage is limited to ≥400 mmol and ≤2200 mmol, with a step size of 100 mmol. mmol; To ensure complete dissolution of terephthalic acid, the amount of N,N-dimethylformamide used is limited to 80 times the amount of terephthalic acid. The target achievement probability PA value of each performance of Ce-UiO-66 in the calculation search space is expanded into a multi-objective evaluation factor. Specifically, the performance of each synthesis scheme in the search space is predicted by the M performance prediction surrogate models to obtain M predicted values. The mean μ and variance σ of the performance predicted values ​​are determined, and the target material defect content g is determined. DC =50%, thermal stability target g TS =300 ℃, according to the formula Determine the target achievement probability PA value for each performance of each synthesis scheme in the search space, where, Let μ(x), σ(x), and g represent the cumulative distribution function, respectively, and let μ(x), σ(x), and g be the predicted mean, predicted standard deviation, and expected target value. This can be expressed using the formula... Calculate the weighted sum of the probabilities of achieving each performance objective, expanding the evaluation factors to multiple objectives, where ω m PA m The weights are set by the ratio ω of defect content and thermal stability. m =0.7:0.3; S4. Ce-UiO-66 material was prepared using a hydrothermal method according to the expected synthesis scheme; S5. Characterize the target performance of the prepared Ce-UiO-66 material. If the test data does not meet the target requirements, update the dataset and surrogate model.

2. The multi-objective optimization method for MOF synthesis routes based on machine learning according to claim 1, characterized in that, In S1, the Ce-UiO-66 synthesis conditions-performance dataset specifically refers to: conducting a literature review on Ce-UiO-66 and establishing a Ce-UiO-66 synthesis conditions-performance dataset containing seven experimental variables, including reaction temperature, reaction time, formic acid dosage, cerium ammonium nitrate dosage, terephthalic acid dosage, N,N-dimethylformamide dosage, and water dosage, including synthesis conditions, ligand deficiency content, and thermal stability.

3. The multi-objective optimization method for MOF synthesis routes based on machine learning according to claim 1, characterized in that, In S2, the eight machine learning algorithms used include: Random Forest, Kernel Ridge Regression, Support Vector Machine, K Nearest Neighbors, Neural Network, XGBoost, AdaBoost, and LightGBM, and Bayesian optimization is used to tune the hyperparameters.

4. The multi-objective optimization method for MOF synthesis routes based on machine learning according to claim 1, characterized in that, In S2, the step of randomly dividing the dataset into training and test sets and selecting surrogate models for each performance prediction specifically involves: during training, randomly dividing the dataset into training and test sets in a ratio of 0.8:0.2, using the formula... Calculate the mean absolute error (MAE) of the model predictions, where x represents the entire training set data and y represents the entire test set data. i y represents the actual data. i The data predicted by the machine learning model is represented by i, which represents the index of the data sample in the training or test set, and n represents the total number of data samples in the training or test set. This process is repeated 100 times to calculate the mean of MAE on the training and test sets to analyze the accuracy of the model. The machine learning model with the lowest mean absolute error is the surrogate model used for predicting material defect content and thermal stability performance.

5. The multi-objective optimization method for MOF synthesis routes based on machine learning according to claim 1, characterized in that, In S2, the Adaboost algorithm is used to select the proxy model for each performance prediction.

6. The multi-objective optimization method for MOF synthesis routes based on machine learning according to claim 1, characterized in that, In S5, if the characterization test data meets the requirements, the design is terminated, and the Ce-UiO-66 material with multi-performance synergistic optimization is determined.

Images