Machine learning guided high first efficiency porous silicon-carbon synthesis method

Abstract

This invention relates to a machine learning-guided method for synthesizing high-efficiency porous silicon-carbon composites, comprising: obtaining physicochemical parameters of porous carbon precursors and process parameters for silane deposition and pyrolysis carbon coating as synthesis-related parameters; synthesizing porous silicon-carbon composite materials based on these parameters and fabricating them into batteries, testing their electrochemical performance to obtain performance data including initial efficiency; training a machine learning prediction model based on a random forest algorithm, using the synthesis-related parameters as feature values ​​and the performance data as target values; using the trained model to predict the target initial efficiency, and using the prediction results to guide the optimization of the selection of synthesis-related parameters; and synthesizing high-efficiency verification porous silicon-carbon composite materials based on the optimized parameters. Compared with existing technologies, this invention achieves accurate prediction from parameters to performance, solves the problems of long material development cycles and high costs under multi-parameter coupling, and can efficiently obtain high-performance silicon-carbon anode materials.

Description

Technical Field

[0001] This invention relates to the field of porous material synthesis, and in particular to a machine learning-guided method for synthesizing highly efficient porous silicon-carbon. Background Technology

[0002] Silicon-based anode materials, especially porous silicon-carbon composites, have become key materials for next-generation high-energy-density lithium-ion batteries due to their high theoretical specific capacity. However, the significant volume effect of silicon during charge and discharge leads to low initial efficiency and poor cycle performance. While constructing porous structures and optimizing carbon coating can effectively alleviate volume expansion, the final electrochemical performance (such as initial efficiency) of the material is profoundly affected by the coupling effect of multiple parameters, including the complex pore characteristics of the carbon precursor (such as pore volume, pore size distribution, and specific surface area) and deposition process parameters (such as silane flow rate and time). Traditional trial-and-error experiments struggle to efficiently explore the complex structure-property relationships between synthesis parameters and electrochemical performance in a high-dimensional parameter space, resulting in long development cycles, high costs, and difficulty in accurately obtaining silicon-carbon materials with high initial efficiency.

[0003] To improve the efficiency of materials research and development, the industry has attempted to introduce data-driven methods. For example, patent application CN121565325A discloses a method for predicting the performance and optimizing the formulation of porous ceramics based on machine learning technology. This method constructs a dataset by collecting the preparation parameters and performance data of porous ceramics, establishes a structure-process-performance correlation model using a machine learning model, and then optimizes the formulation in reverse. However, this technology targets porous ceramic systems used in waste gas treatment, and its core optimization objective is physical properties such as "flux," which is fundamentally different from the electrochemical performance such as first-efficiency that is the focus of lithium-ion battery anode materials. Its model construction and optimization paradigm does not involve how to model and reverse-design the synthesis process characteristics and specific electrochemical performance indicators of silicon-carbon materials.

[0004] Patent application CN118605364A discloses a computational monitoring and control system and method for the preparation process of porous silicon-carbon anode materials. This method focuses on monitoring real-time process parameters (such as ball milling time and sintering temperature) on the production line, identifying process patterns, and predicting and controlling quality based on historical data to achieve stable and optimized production. However, this technology focuses on real-time control of the preparation process and prediction of quality attributes. These quality attributes are mainly related to physicochemical parameters in the preparation process, without clearly and directly establishing an interpretable and predictable machine learning model connecting the initial carbon precursor pore structure characteristics and vapor deposition process parameters to the core electrochemical performance indicator of the first-efficiency electrode. Its method based on process pattern recognition and sub-model invocation lacks direct and efficient support for reverse engineering needs that aim to obtain high-efficiency new materials from scratch through proactive design of synthesis parameters.

[0005] In summary, existing technologies lack a method specifically for porous silicon-carbon anode materials that can efficiently and accurately establish the structure-property relationship between synthesis parameters such as the carbon precursor pore structure and silane deposition process and the final electrochemical performance such as the first-efficiency of the battery, and can directly guide the optimization of synthesis parameters based on this relationship model, thereby targeting the preparation of high-efficiency silicon-carbon materials. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of the existing technology and provide a machine learning-guided high-efficiency porous silicon-carbon synthesis method, which realizes accurate prediction from parameters to performance, solves the problems of long material development cycle and high cost under multi-parameter coupling, and can efficiently obtain high-performance silicon-carbon anode materials.

[0007] The objective of this invention can be achieved through the following technical solutions: This invention provides a machine learning-guided method for synthesizing high-efficiency porous silicon-carbon, comprising the following steps: S1. Obtain the physicochemical parameters of the porous carbon precursor, the process parameters of silane deposition and pyrolysis carbon coating, as synthesis-related parameters; S2. Based on the synthesis-related parameters, the porous carbon precursor is subjected to silane deposition and carbon coating treatment to synthesize a porous silicon-carbon composite material. S3. Prepare the porous silicon-carbon composite material into a battery and test the electrochemical performance of the battery to obtain performance data including the first efficiency. S4. Using the synthesis-related parameters as feature values ​​and the performance data as target values, train a machine learning prediction model based on the random forest algorithm to establish the structure-property relationship between the synthesis parameters and the material properties. S5. The trained machine learning prediction model is used to predict the target performance, and the prediction results are used to guide the optimization of the selection of the synthesis-related parameters. Based on the optimized parameters, a high-efficiency porous silicon-carbon composite material for verification is synthesized.

[0008] Furthermore, in S1, the physicochemical parameters of the porous carbon precursor include pore volume characteristics, pore size distribution characteristics, and specific surface area characteristics. The volumetric characteristics of the pores include the total pore volume and its distributed volume within different pore diameter ranges. The pore size distribution characteristics are characterized by the proportion of pore volume in different pore size ranges to the total pore volume. The different pore size ranges include the micropore range, the mesopore range, and the macropore range.

[0009] Furthermore, in S1, among the physicochemical parameters, the pore volume ratio of the mesoporous range affects the initial efficiency of the final composite material by regulating the diffusion rate and deposition behavior of silane within the pores. The process parameters for silane deposition include the flow rate of silane gas and the duration of its introduction. These process parameters determine the loading and distribution of silicon in the porous carbon support.

[0010] Furthermore, in S1, the pyrolysis carbon coating process parameters include the flow rate of the carbon source gas and the duration of its introduction. The carbon source gas is preferably acetylene gas. The pyrolysis carbon coating process parameters determine the thickness and uniformity of the carbon coating layer.

[0011] Furthermore, in S2, the specific process of performing silane deposition and carbon coating on the porous carbon precursor based on the aforementioned synthesis-related parameters includes: The porous carbon precursor is placed in a reactor, and silane gas is introduced according to the silane gas flow rate and duration in the synthesis parameters to deposit silicon in the pores and surface of the porous carbon precursor. Then, acetylene gas is introduced to pyrolyze carbon coating, thereby obtaining the porous silicon-carbon composite material.

[0012] Furthermore, in S3, the specific process of preparing the porous silicon-carbon composite material into a battery includes: The porous silicon-carbon composite material is mixed with binder, conductive agent and dispersant in a preset mass ratio to prepare a uniform electrode slurry. The slurry is then coated onto a current collector and dried to obtain an electrode sheet. The electrode sheet is then used as the working electrode and lithium metal is used as the counter electrode to assemble a coin cell in an inert atmosphere glove box.

[0013] Furthermore, in S3, the specific process for testing the electrochemical performance of the battery to obtain performance data, including first-efficiency performance, includes: The assembled battery is subjected to a constant current charge-discharge test. First, it is discharged with a constant current to the first cutoff voltage. After being left to stand for a predetermined time, it is charged with the same constant current to the second cutoff voltage. The first discharge capacity and the first charge capacity are recorded. The first efficiency value is obtained by calculating the ratio of the first charge capacity to the first discharge capacity.

[0014] Furthermore, in S4, a machine learning prediction model is trained based on the random forest algorithm to establish the structure-property relationship between synthesis parameters and material properties. The specific process includes: The synthesis-related parameters obtained in S1 are organized into a feature value vector, and the performance data obtained in S3, including the first effect, are organized into a target value vector to form the initial dataset. The initial dataset is preprocessed to form a standardized training dataset; The standardized training dataset is divided into a model training subset for model training and a model test subset for evaluating model performance according to a preset ratio. The random forest algorithm was trained using the training subset of the model, and hyperparameter optimization was performed on the number of decision trees, the maximum depth of the trees, and the minimum number of samples required for node splitting to construct a preliminary machine learning prediction model. The performance of the preliminary machine learning prediction model is evaluated using the model test subset. The model accuracy is quantified by calculating the coefficient of determination and root mean square error between the predicted value and the actual target value. The model is then iteratively optimized based on the evaluation results until a final machine learning prediction model that meets the preset accuracy requirements is obtained, thereby completing the establishment of the structure-property relationship.

[0015] Furthermore, in S4, the preprocessing of the initial dataset includes: The initial dataset is cleaned to remove data samples containing outliers that are outside the reasonable range, and blank values ​​are filled or deleted. The cleaned data is standardized by transforming the feature values ​​and target values ​​of different dimensions to the same numerical scale through mathematical transformation, thus forming the standardized training dataset.

[0016] Furthermore, in S5, the trained machine learning prediction model is used to predict the target performance, and the prediction results are used to guide the optimization of the selection of the synthesis-related parameters. The specific process includes: Based on the preset high first-efficiency target value, the optimization algorithm is used to search within the value space of the synthesis-related parameters, and the trained machine learning prediction model is called to predict each candidate combination of synthesis-related parameters. From the search process, select one or more synthetic relevant parameter optimization combinations that make the model's predicted first-efficiency value closest to the high first-efficiency target value; Based on the optimized combination of the selected synthesis-related parameters, a precursor with matching physicochemical parameters is selected from the candidate porous carbon precursors, and the corresponding silane gas flow rate and inlet duration, acetylene gas flow rate and inlet duration are set as the optimized synthesis parameters to guide the synthesis of porous silicon-carbon composite materials.

[0017] Compared with the prior art, the present invention has the following beneficial effects: 1) This invention innovatively combines the physicochemical parameters of porous carbon precursors and silane deposition process parameters as machine learning features, using first-efficiency as the direct optimization target, to construct a proprietary synthesis-performance structure-property relationship prediction model. This method breaks free from the limitations of traditional trial-and-error methods and general optimization frameworks, achieving efficient and targeted design for the specific system of porous silicon-carbon composite materials.

[0018] 2) The prediction model based on the random forest algorithm described above can accurately reveal the influence of key parameters such as the proportion of micropores / mesopores / macropores in the carbon precursor and the silane flow rate on the final first-efficiency performance. This allows for performance prediction and parameter optimization before synthesis. It solves the core problem in existing technologies where complex parameter coupling makes it difficult to analyze structure-property relationships, transforming the previously experience-based, trial-and-error process into a data-driven, rational design. This significantly reduces the number of repetitive experiments required to find high-efficiency materials, greatly lowering R&D costs and time. The verification materials synthesized based on the model demonstrate significantly improved and guaranteed electrochemical performance, including first-efficiency performance. This provides an efficient and reliable technical means for obtaining high-performance silicon-carbon anode materials and promoting their application in high-end lithium-ion batteries. Attached Figure Description

[0019] Figure 1 This is a flowchart of machine learning prediction for the first efficiency of porous silicon-carbon half-cells in this invention.

[0020] Figure 2 These are the isothermal adsorption-desorption curves of porous carbon and nitrogen gases with different pore characteristics in this invention.

[0021] Figure 3 This is the XRD characterization of porous silicon carbon in this invention.

[0022] Figure 4 This is the charge-discharge curve and first-effect of the porous silicon carbon in this invention.

[0023] Figure 5 It is the Spearman / Pearson correlation coefficient between the synthesis parameters and the first efficiency of silicon-carbon in this invention.

[0024] Figure 6 This is a graph showing the relationship between the predicted first-efficiency and the actual first-efficiency of porous silicon carbon in this invention. Detailed Implementation

[0025] The following is in conjunction with the appendix Figure 1-6 The present invention will be described in detail with reference to specific embodiments. Component models, material names, connection structures, circuit structures, control methods, algorithms, and other features not explicitly described in this technical solution are considered common technical features disclosed in the prior art.

[0026] Example 1 This embodiment provides a machine learning-guided method for synthesizing high-efficiency porous silicon-carbon, referring to... Figure 1 This includes the following steps: S1, Nitrogen isothermal adsorption-desorption test (refer to...) Figure 2 The physicochemical parameters of the porous carbon precursor, as well as the process parameters for silane deposition and pyrolysis carbon coating, were obtained as relevant parameters for synthesis.

[0027] In specific implementation, in S1, the physicochemical parameters of the porous carbon precursor include pore volume characteristics, pore size distribution characteristics, and specific surface area characteristics. The volumetric characteristics of the pores include the total pore volume (1.07 cm³). 3 / g) and its distribution volume within different pore size ranges, wherein the specific surface area of ​​the porous carbon is 1481.90 m². 2 / g, The pore size distribution characteristics are characterized by the proportion of pore volume in different pore size ranges to the total pore volume and the average pore size. The different pore size ranges include micropores (< 2 nm), mesopores (2-50 nm), and macropores (> 50 nm). The volume of micropores accounted for 17.44%, mesopores accounted for 78.95%, macropores accounted for 3.61%, and the average pore size was 3.044 nm.

[0028] In specific implementation, in S1, the pore volume ratio of the mesoporous range in the physicochemical parameters affects the initial efficiency of the final composite material by regulating the diffusion rate and deposition behavior of silane in the pores. The process parameters for silane deposition include the flow rate of silane gas and the duration of its introduction. These process parameters determine the loading and distribution of silicon in the porous carbon support.

[0029] In specific implementation, in S1, the pyrolysis carbon coating process parameters include the flow rate of the carbon source gas and the duration of its introduction. The carbon source gas is preferably acetylene gas. The pyrolysis carbon coating process parameters determine the thickness and uniformity of the carbon coating layer.

[0030] S2. Based on the synthesis parameters, the porous carbon precursor is subjected to silane deposition and carbon coating treatment to synthesize a porous silicon-carbon composite material.

[0031] In specific implementation, S2, the specific process of silane deposition and carbon coating of the porous carbon precursor based on the aforementioned synthesis-related parameters includes: The porous carbon precursor was placed in a reactor, and silane gas was introduced into the porous carbon at a gas flow rate of 3 L / min for 266 min, according to the synthesis parameters. Subsequently, silane was deposited in the pores and on the surface of the porous carbon precursor, forming silicon grains that continuously grew (see reference). Figure 3 XRD was introduced to characterize the size of silicon grains. Then, acetylene gas was introduced at a flow rate of 3 L / min for 100 min to complete the pyrolysis carbon coating and obtain the porous silicon-carbon composite material.

[0032] S3. Prepare a battery from the porous silicon-carbon composite material and test the electrochemical performance of the battery to obtain performance data, including the first-efficiency performance.

[0033] In specific implementation, S3, the process of preparing the porous silicon-carbon composite material into a battery includes: The porous silicon-carbon composite material Si / C, binder PAA, carbon nanotubes (CNTs) and carbon black SP are mixed in a preset mass ratio (90%:5%:1%:4%) to prepare a uniform electrode slurry. The slurry is then coated onto a current collector and dried to obtain an electrode sheet. The electrode sheet is then used as the working electrode and lithium metal is used as the counter electrode to assemble a coin cell in an inert atmosphere glove box.

[0034] In specific implementation, in S3, the process of testing the electrochemical performance of the battery to obtain performance data, including the first-efficiency result, includes: The assembled battery was subjected to constant current charge-discharge testing. First, it was discharged at a constant current of 0.1C to 0.05V to obtain the lithium insertion capacity. After resting for 30 seconds, it was charged at a constant current of 0.1C to 2V to obtain the lithium extraction capacity. The specific capacity and voltage data during the charge-discharge process were recorded. The specific capacity-voltage curve for the first lithium insertion / extraction of silicon-carbon was plotted using these data. Simultaneously, the first-stage efficiency value was obtained by calculating the ratio of the first-stage charge capacity to the first-stage discharge capacity (as shown in the charge-discharge curve and first-stage efficiency curve of the porous silicon-carbon sample). Figure 4 (As shown).

[0035] S4. Using the aforementioned synthesis-related parameters (pore characteristics, deposition parameters, etc.) as feature values ​​and the aforementioned performance data (first-effect value) as target values, a machine learning prediction model is trained based on the random forest algorithm to establish the structure-property relationship between synthesis parameters and the material's first-effect (refer to...). Figure 5 ).

[0036] In specific implementation, in S4, a machine learning prediction model is trained based on the random forest algorithm to establish the structure-property relationship between synthesis parameters and material properties. The specific process includes: The synthesis-related parameters obtained in S1 are organized into a feature value vector, and the performance data obtained in S3, including the first effect, are organized into a target value vector to form the initial dataset. The initial dataset is preprocessed to form a standardized training dataset; The standardized training dataset is divided into a model training subset for model training and a model test subset for evaluating model performance according to a preset ratio (7:3). The random forest algorithm was trained using the training subset of the model, and hyperparameter optimization was performed on the number of decision trees, the maximum depth of the trees, and the minimum number of samples required for node splitting to construct a preliminary machine learning prediction model. The performance of the preliminary machine learning prediction model is evaluated using the model test subset. The model accuracy is quantified by calculating the coefficient of determination and root mean square error between the predicted value and the actual target value. The model is then iteratively optimized based on the evaluation results until a final machine learning prediction model that meets the preset accuracy requirements is obtained, thereby completing the establishment of the structure-property relationship.

[0037] In specific implementation, the preprocessing of the initial dataset in S4 includes: The initial dataset is cleaned to remove data samples containing outliers that are outside the reasonable range, and blank values ​​are filled or deleted. Correlation analysis was performed on the feature values ​​of the initial dataset and the target dataset. The features were then ranked, and irrelevant features were removed to improve training efficiency (see reference). Figure 5 , Figure 5 The figure shows the composite parameters with high relevance ranked first. The cleaned data is standardized by transforming the feature values ​​and target values ​​of different dimensions to the same numerical scale through mathematical transformation, thus forming the standardized training dataset.

[0038] S5. The trained machine learning prediction model is used to predict the target performance, and the prediction results are used to guide the optimization of the selection of the synthesis-related parameters. Based on the optimized parameters, a high-efficiency porous silicon-carbon composite material for verification is synthesized.

[0039] In specific implementation, in S5, the trained random forest prediction model is used to predict the target performance, and the prediction results are used to guide the optimization of the selection of the synthetic relevant parameters. The specific process includes: Based on the preset high first-efficiency target value, the parameter scanning (or genetic algorithm or Bayesian optimization) optimization algorithm is used to search in the value space of the synthetic relevant parameters, and the trained random forest prediction model is called to predict each candidate combination of synthetic relevant parameters. From the search process, select one or more synthetic relevant parameter optimization combinations that make the model's predicted first-efficiency value closest to the high first-efficiency target value; Based on the optimized combination of the selected synthesis-related parameters, a precursor with matching physicochemical parameters is selected from the candidate porous carbon precursors, and the corresponding silane gas flow rate and inlet duration, acetylene gas flow rate and inlet duration are set. These are used as the optimized synthesis parameters to guide the subsequent synthesis of porous silicon-carbon composite materials.

[0040] Reference Figure 6 This method compares the predicted first efficiency of different porous silicon carbons with the tested first efficiency after they are made into half-cells. The results show that the error is small, which proves that the random forest prediction model in this method has been successfully built and can be used to guide the synthesis of porous silicon carbon with high first efficiency.

[0041] Based on the trained random forest model constructed and verified in this embodiment, when used to guide synthesis: First, a clear target value for high first-efficiency is set; then, an optimization algorithm (such as parameter scanning, genetic algorithm, Bayesian optimization, etc.) is used to systematically search and traverse the multidimensional synthesis parameter space. For each candidate parameter combination (i.e., a possible precursor and process scheme) generated by the optimization algorithm, the trained random forest model is called to quickly predict the positive performance, obtaining a predicted first-efficiency value; the optimization algorithm continuously compares the closeness between the predicted value and the target value, and adjusts the search direction accordingly, ultimately selecting one or more optimal synthesis parameter combinations from a massive number of possibilities that make the predicted first-efficiency closest to or even exceed the preset target. Based on the recommended parameter combination calculated by this model and the optimization algorithm, researchers can accurately select porous carbon precursors with matching pore characteristics from the candidate material library, and precisely set the flow rate and introduction time of silane and acetylene gases, thereby achieving targeted synthesis of high-performance targets. This transforms traditional trial-and-error experiments into data-driven rational design, replacing a large number of time-consuming and labor-intensive repetitive experiments with model calculations, thus greatly improving the efficiency and success rate of developing high-efficiency porous silicon-carbon composite materials.

[0042] Verification Example 1 Based on the method in Example 1, the method for producing porous silicon-carbon composite materials provided in this embodiment includes the following steps: Step 1: Select 1 kg of porous carbon A1 sample, which has the following characteristics: specific surface area 1481.90 m². 2 / g, with a total pore volume of 1.07 cm³. 3 / g, micropore ratio: mesopore ratio: macropore ratio = 17.44%: 78.95%: 3.61%, average pore size is 3.04 nm; 1 kg of porous carbon A was put into the furnace, silane was introduced at a flow rate of 3 L / min for 266 min, followed by acetylene at a flow rate of 3 L / min for 100 min; finally, 2.13 kg of porous silicon-carbon sample B1 was obtained.

[0043] Step 2: Porous silicon-carbon B1 was used to prepare a lithium-ion half-cell with a Si / C:PAA:CNT:SP ratio of 90%:5%:1%:4%. The half-cell was then subjected to charge-discharge testing. The specific steps were as follows: A constant current discharge of 0.1C was applied to 0.05V to obtain the lithium insertion capacity. After resting for 30 seconds, the cell was charged at 0.1C to 2V to obtain the lithium extraction capacity, and then rested for 30 seconds. The specific capacity and voltage data during the charge-discharge process were recorded, ultimately obtaining the specific capacity, voltage, and initial efficiency data during the charge-discharge process.

[0044] Step 3: Input the porous carbon Al characteristics and deposition process parameters into the model to obtain the predicted first-effect data. Compare the predicted first-effect data with the first-effect data in Step 2. The predicted first-effect was 91.13%, the actual first-effect was 91.78%, and the error was 0.71%.

[0045] Verification Example 2 Based on the method in Example 1, the method for producing porous silicon-carbon composite materials provided in this embodiment includes the following steps: Step 1: Select 1 kg of porous carbon A2 sample, which has the following characteristics: specific surface area 368.99 m². 2 / g, with a total pore volume of 0.62 cm³. 3 / g, micropore ratio: mesopore ratio: macropore ratio = 2.36%: 82.10%: 15.54%, average pore size is 7.47 nm; 1 kg of porous carbon A was put into the furnace, silane was introduced at a flow rate of 3 L / min for 266 min, followed by acetylene at a flow rate of 3 L / min for 100 min; finally, 2.03 kg of porous silicon-carbon sample B2 was obtained.

[0046] Step 2: A lithium-ion half-cell was prepared using porous silicon-carbon B2 in a ratio of Si / C:PAA:CNT:SP = 90%:5%:1%:4%. Charge-discharge tests were then performed on the half-cell. The specific steps were as follows: A constant current discharge of 0.1C was applied to 0.05V to obtain the lithium insertion capacity. After resting for 30 seconds, the cell was charged at 0.1C to 2V to obtain the lithium extraction capacity. This process was then rested for 30 seconds. The specific capacity and voltage data during the charge-discharge process were recorded, ultimately yielding the specific capacity, voltage, and initial efficiency data.

[0047] Step 3: Input the porous carbon A2 characteristics and deposition process parameters into the model to obtain the predicted first-effect data. Compare the predicted first-effect data with the first-effect data in Step 2. The predicted first-effect is 91.72%, the actual first-effect is 90.87%, and the error is 0.94%.

[0048] Verification Example 3 Based on the method in Example 1, the method for producing porous silicon-carbon composite materials provided in this embodiment includes the following steps: Step 1: Select 1 kg of porous carbon A3 sample, which has the following characteristics: specific surface area 2114.96 m². 2 / g, with a total pore volume of 0.99 cm³. 3 / g, micropore ratio: mesopore ratio: macropore ratio = 75.11%: 22.64%: 2.26%, average pore size is 1.84 nm; 1 kg of porous carbon A was put into the furnace, silane was introduced at a flow rate of 3 L / min for 266 min, followed by acetylene at a flow rate of 3 L / min for 100 min; finally, 2.1 kg of porous silicon-carbon sample B3 was obtained.

[0049] Step 2: Porous silicon-carbon B3 was used to prepare a lithium-ion half-cell with a Si / C:PAA:CNT:SP ratio of 90%:5%:1%:4%. The half-cell was then subjected to charge-discharge testing. The specific steps were as follows: A constant current discharge of 0.1C was applied to 0.05V to obtain the lithium insertion capacity. After resting for 30 seconds, the cell was charged at 0.1C to 2V to obtain the lithium extraction capacity, and then rested for 30 seconds. The specific capacity and voltage data during the charge-discharge process were recorded, ultimately obtaining the specific capacity, voltage, and initial efficiency data during the charge-discharge process.

[0050] Step 3: Input the porous carbon A3 characteristics and deposition process parameters into the model to obtain the predicted first-effect data. Compare the predicted first-effect data with the first-effect data in Step 2. The predicted first-effect was 87.71%, the actual first-effect was 87.34%, and the error was 0.42%.

[0051] Verification Example 4 Based on the method in Example 1, the method for producing porous silicon-carbon composite materials provided in this embodiment includes the following steps: Step 1: Select 1 kg of porous carbon A4 sample, which has the following characteristics: specific surface area 258.52 m². 2 / g, with a total pore volume of 0.49 cm³. 3 / g, micropore ratio: mesopore ratio: macropore ratio = 0%: 87.26%: 12.73%, average pore size is 8.02nm; 1kg of porous carbon A was put into the furnace, silane was introduced at a flow rate of 3 L / min for 230 min, followed by acetylene at a flow rate of 3 L / min for 100 min; finally, 1.88 kg of porous silicon-carbon sample B4 was obtained.

[0052] Step 2: A lithium-ion half-cell was prepared using porous silicon-carbon B4 in a ratio of Si / C:PAA:CNT:SP = 90%:5%:1%:4%. Charge-discharge tests were then performed on the half-cell. The specific steps were as follows: A constant current discharge of 0.1C was applied to 0.05V to obtain the lithium insertion capacity. After resting for 30 seconds, the cell was charged at 0.1C to 2V to obtain the lithium extraction capacity. This process was then rested for 30 seconds. The specific capacity and voltage data during the charge-discharge process were recorded, ultimately obtaining the specific capacity, voltage, and initial efficiency data.

[0053] Step 3: Input the porous carbon A4 characteristics and deposition process parameters into the model to obtain the predicted first-effect data. Compare the predicted first-effect data with the first-effect data in Step 2. The predicted first-effect was 92.10%, the actual first-effect was 90.85%, and the error was 1.38%.

[0054] The above description of the embodiments is provided to enable those skilled in the art to understand and use the invention. It will be apparent to those skilled in the art that various modifications can be made to these embodiments, and the general principles described herein can be applied to other embodiments without inventive effort. Therefore, the present invention is not limited to the above embodiments, and any improvements and modifications made by those skilled in the art based on the disclosure of the present invention without departing from the scope of the invention should be within the protection scope of the present invention.

Claims

1. A machine learning-guided method for synthesizing high-efficiency porous silicon-carbon, characterized in that, Includes the following steps: S1. Obtain the physicochemical parameters of the porous carbon precursor, the process parameters of silane deposition and pyrolysis carbon coating, as synthesis-related parameters; S2. Based on the synthesis-related parameters, the porous carbon precursor is subjected to silane deposition and carbon coating treatment to synthesize a porous silicon-carbon composite material. S3. Prepare the porous silicon-carbon composite material into a battery and test the electrochemical performance of the battery to obtain performance data including the first efficiency. S4. Using the synthesis-related parameters as feature values ​​and the performance data as target values, train a machine learning prediction model based on the random forest algorithm to establish the structure-property relationship between the synthesis parameters and the material properties. S5. The trained machine learning prediction model is used to predict the target performance, and the prediction results are used to guide the optimization of the selection of the synthesis-related parameters. Based on the optimized parameters, a high-efficiency porous silicon-carbon composite material for verification is synthesized.

2. The machine learning-guided high-efficiency porous silicon-carbon synthesis method according to claim 1, characterized in that, In S1, the physicochemical parameters of the porous carbon precursor include pore volume characteristics, pore size distribution characteristics, and specific surface area characteristics. The volumetric characteristics of the pores include the total pore volume and its distributed volume within different pore diameter ranges. The pore size distribution characteristics are characterized by the proportion of pore volume in different pore size ranges to the total pore volume. The different pore size ranges include the micropore range, the mesopore range, and the macropore range.

3. The machine learning-guided high-efficiency porous silicon-carbon synthesis method according to claim 2, characterized in that, In S1, among the physicochemical parameters, the pore volume ratio of micropores, mesopores, and macropores affects the initial efficiency of the final composite material by regulating the diffusion rate and deposition behavior of silane in the pores. The process parameters for silane deposition include the flow rate of silane gas and the duration of its introduction. These process parameters determine the loading and distribution of silicon in the porous carbon support.

4. The machine learning-guided high-efficiency porous silicon-carbon synthesis method according to claim 1, characterized in that, In S1, the pyrolysis carbon coating process parameters include the flow rate of the carbon source gas and the duration of its introduction. The carbon source gas is preferably acetylene gas. The pyrolysis carbon coating process parameters determine the thickness and uniformity of the carbon coating layer.

5. The machine learning-guided high-efficiency porous silicon-carbon synthesis method according to claim 1, characterized in that, In S2, the specific process of performing silane deposition and carbon coating on the porous carbon precursor based on the aforementioned synthesis-related parameters includes: The porous carbon precursor is placed in a reactor, and silane gas is introduced according to the silane gas flow rate and duration in the synthesis parameters to deposit silicon in the pores and surface of the porous carbon precursor. Then, acetylene gas is introduced to pyrolyze carbon coating, thereby obtaining the porous silicon-carbon composite material.

6. The machine learning-guided high-efficiency porous silicon-carbon synthesis method according to claim 1, characterized in that, In S3, the specific process of preparing the porous silicon-carbon composite material into a battery includes: The porous silicon-carbon composite material is mixed with binder, conductive agent and dispersant in a preset mass ratio to prepare a uniform electrode slurry. The slurry is then coated onto a current collector and dried to obtain an electrode sheet. The electrode sheet is then used as the working electrode and lithium metal is used as the counter electrode to assemble a coin cell in an inert atmosphere glove box.

7. The machine learning-guided high-efficiency porous silicon-carbon synthesis method according to claim 1, characterized in that, In S3, the specific process for testing the electrochemical performance of the battery to obtain performance data, including first-efficiency performance, includes: The assembled battery is subjected to a constant current charge-discharge test. First, it is discharged with a constant current to the first cutoff voltage. After being left to stand for a predetermined time, it is charged with the same constant current to the second cutoff voltage. The first discharge capacity and the first charge capacity are recorded. The first efficiency value is obtained by calculating the ratio of the first charge capacity to the first discharge capacity.

8. The machine learning-guided high-efficiency porous silicon-carbon synthesis method according to claim 1, characterized in that, In S4, a machine learning prediction model is trained based on the random forest algorithm to establish the structure-property relationship between synthesis parameters and material properties. The specific process includes: The synthesis-related parameters obtained in S1 are organized into a feature value vector, and the performance data obtained in S3, including the first effect, are organized into a target value vector to form the initial dataset. The initial dataset is preprocessed to form a standardized training dataset; The standardized training dataset is divided into a model training subset for model training and a model test subset for evaluating model performance according to a preset ratio. The random forest algorithm was trained using the training subset of the model, and hyperparameter optimization was performed on the number of decision trees, the maximum depth of the trees, and the minimum number of samples required for node splitting to construct a preliminary machine learning prediction model. The performance of the preliminary machine learning prediction model is evaluated using the model test subset. The model accuracy is quantified by calculating the coefficient of determination and root mean square error between the predicted value and the actual target value. The model is then iteratively optimized based on the evaluation results until a final machine learning prediction model that meets the preset accuracy requirements is obtained, thereby completing the establishment of the structure-property relationship.

9. The machine learning-guided high-efficiency porous silicon-carbon synthesis method according to claim 8, characterized in that, In S4, the preprocessing of the initial dataset includes: The initial dataset is cleaned to remove data samples containing outliers that are outside the reasonable range, and blank values ​​are filled or deleted. The cleaned data is standardized by transforming the feature values ​​and target values ​​of different dimensions to the same numerical scale through mathematical transformation, thus forming the standardized training dataset.

10. The machine learning-guided high-efficiency porous silicon-carbon synthesis method according to claim 1, characterized in that, In S5, the trained machine learning prediction model is used to predict the target performance, and the prediction results are used to guide the optimization of the selection of the synthesis-related parameters. The specific process includes: Based on the preset high first-efficiency target value, the optimization algorithm is used to search within the value space of the synthesis-related parameters, and the trained machine learning prediction model is called to predict each candidate combination of synthesis-related parameters. From the search process, select one or more synthetic relevant parameter optimization combinations that make the model's predicted first-efficiency value closest to the high first-efficiency target value; Based on the optimized combination of the selected synthesis-related parameters, a precursor with matching physicochemical parameters is selected from the candidate porous carbon precursors, and the corresponding silane gas flow rate and inlet duration, acetylene gas flow rate and inlet duration are set as the optimized synthesis parameters to guide the synthesis of porous silicon-carbon composite materials.

Images