Method and device for predicting pore structure of lignin-derived porous carbon

Abstract

The invention discloses a method and a device for predicting a pore structure of lignin-derived porous carbon. The invention provides a hybrid machine learning framework for solving the technical problems of scarcity of lignin-derived porous carbon samples and data missing. The pore structure is pre-trained by adopting an XGBoost model, feature latent variable extraction and sample generation are realized by utilizing transfer learning and a generative adversarial network model, and finally, an XGBoost regression device is trained in combination with generated data and original data to finish accurate prediction of the lignin-derived porous carbon pore structure. According to the method, a hybrid machine learning framework based on transfer learning and generative adversarial networks is adopted, and a new method is provided for prediction of the lignin-derived porous carbon pore structure.

Description

Technical Field

[0001] The present application relates to the fields of machine learning and biomass material technology, and in particular to a method and device for predicting the pore structure of lignin-derived porous carbon. Background Art Lignin is a widely occurring biomass in nature. Through catalytic pyrolysis, lignin can be converted into lignin-derived porous carbon. This material is widely used in supercapacitors, catalysis, adsorption, and other fields due to its high specific surface area, adjustable pore size distribution, and excellent electrochemical properties. The pore structure of lignin-derived porous carbon is one of the key factors determining its performance. Its porosity, specific surface area, pore size distribution, and other characteristics directly affect its application effect. However, the pore structure of lignin-derived porous carbon is affected by many factors, including the type of lignin raw material, pyrolysis temperature, pyrolysis time, atmospheric conditions, and heating rate. Porous carbon materials generated under different reaction conditions have significant differences in their pore structure and performance, and these differences are often difficult to quickly predict and optimize using traditional experimental methods.

[0002] Machine learning, as a powerful data analysis technology, can generate samples and extract features based on limited experimental data, mining the underlying patterns of a system to achieve accurate predictions of target variables. The invention patent (CN118551169 A) discloses a machine learning-based method and system for predicting biochar adsorption rates. This method uses an XGBoost model to rank biochar parameter information features and then employs an LSTM model to build a prediction model based on these ranked features, ultimately achieving accurate predictions of biochar adsorption rates.

[0003] However, compared with the prediction of physical properties of other biochars, the prediction of lignin porous carbon has the problems of small samples and serious missing values, and there is no method for predicting the physical properties of lignin-derived porous carbon. Summary of the Invention

[0004] This application provides a machine learning-based method and apparatus for predicting the pore structure of lignin-derived porous carbon. The prediction method described in this application achieves accurate prediction of the specific surface area and total pore volume of lignin-derived porous carbon by constructing a hybrid machine learning framework that combines a pre-trained model, an adversarial network, and a regressor model.

[0005] In a first aspect, the present application provides a method for predicting the pore structure of lignin-derived porous carbon, comprising: Obtaining parameter information of biomass-derived porous carbon and corresponding experimental parameters, and constructing a parameter database; the biomass includes lignin; Using a parameter database to train a machine learning model to obtain a prediction model; the machine learning model includes an initial prediction model, an adversarial model, and a regressor model; the adversarial model includes a generator and a discriminator; Evaluate and optimize the prediction model to obtain the optimal prediction model; Use the optimal prediction model to predict the material to be tested and output the prediction results.

[0006] In one possible implementation, the biomass-derived porous carbon is obtained by catalytic pyrolysis of biomass.

[0007] Furthermore, the parameter information of the biomass-derived porous carbon includes: the type of raw materials for catalytic pyrolysis, the raw material ratio, elemental analysis data, industrial analysis data and pore characteristic data; the corresponding experimental parameters include pyrolysis temperature, heating rate and reaction time.

[0008] In one possible implementation, the method of using a parameter database to train a machine learning model to obtain a prediction model includes: Using the parameter database to train the initial prediction model, obtain a pre-trained initial model, and calculate the first loss function; The generator takes random noise as input and, after noise dimensionality reduction, outputs samples that conform to the characteristics of lignin-derived porous carbon. The discriminator optimizes the training of the generator by minimizing the Wasserstein distance between the generated samples and the real samples, and calculates a second loss function. The first and second loss functions are combined to optimize the training of the generator and discriminator. Fuse the samples generated by the generator with the experimental parameters in the parameter database to obtain a fused training dataset; The regressor model is trained using the fused training data set, the regressor model parameters are optimized, and a prediction model is obtained.

[0009] In one possible implementation, the initial prediction model includes an XGBoost regression model; the adversarial model is a Wasserstein GAN framework; and the regressor model includes an XGBoost regression model.

[0010] Furthermore, the adversarial model is an improved Wasserstein GAN framework. Specifically, based on the Wasserstein GAN framework, the generated samples obtained by the generator are input into the initial prediction model, and the root mean square error of the generated samples in the initial prediction model is calculated, and the root mean square error is used to assist in optimizing the generator.

[0011] In a second aspect, a device for predicting the pore structure of lignin-derived porous carbon is provided, comprising: A database construction module is used to obtain parameter information of biomass-derived porous carbon and corresponding experimental parameters to construct a parameter database; the biomass includes lignin; A training model for training a machine learning model using a parameter database to obtain a prediction model; the machine learning model includes an initial prediction model, an adversarial model, and a regressor model; the adversarial model includes a generator and a discriminator; Optimization module, used to evaluate and optimize the prediction model to obtain the optimal prediction model; The output module is used to predict the material to be tested using the optimal prediction model and output the prediction results.

[0012] In a third aspect, an electronic device is provided, comprising: a processor and a memory, wherein the memory is coupled to the processor, the memory is used to store computer program code, and the computer program code includes computer instructions. When the processor reads the computer instructions from the memory, the electronic device executes the method described in the first aspect.

[0013] In a fourth aspect, a non-transitory computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed, the method described in the first aspect is executed.

[0014] According to a fifth aspect, a computer program product is provided, wherein the computer program product comprises instructions, and when the instructions are executed by a processor, the method described in the first aspect is implemented.

[0015] Furthermore, the regressor model adopts an XGBoost model, uses lignin analysis information and lignin-derived porous carbon analysis information in the database, and lignin porous carbon sample data generated by a generative adversarial model to achieve accurate prediction of the specific surface area and total pore volume of lignin-derived porous carbon.

[0016] The beneficial effects of this application are as follows: (1) The method described in this application constructs a cross-material domain prediction model, which includes an initial prediction model, an adversarial model, and a regressor model. It can migrate the pore structure data of biochar to the prediction task of small sample lignin-derived porous carbon, and solve the data scarcity problem through latent space mapping.

[0017] (2) The prediction model in the method described in this application includes a dynamic collaborative training framework, in which the output of the pre-trained model serves as the regularization term of the generator, and the discriminator loss is combined to reversely optimize the feature extraction process, forming a closed-loop optimization framework of "pre-training-transfer generation-post-training", thereby improving the performance of the model.

[0018] (3) This application achieves efficient and accurate prediction of the specific surface area and total pore volume of lignin-derived porous carbon prepared by catalytic cracking, providing a feasible solution for the directional design of lignin-derived porous carbon. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 A schematic diagram of a process for predicting the pore structure of lignin-derived porous carbon based on machine learning provided in an embodiment of the present application; Figure 2 A schematic diagram of the prediction model training process provided in an embodiment of the present application; Figure 3 This is a schematic diagram of the structure of the adversarial model described in the embodiment of the present application; Figure 4 This is a schematic diagram of the structure of the device described in the embodiment of the present application; Figure 5 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0020] The pore structure of lignin-derived porous carbon is influenced by a variety of factors, including the type of lignin raw material, pyrolysis temperature, pyrolysis time, atmosphere conditions, and heating rate. Existing methods for predicting lignin-derived porous carbon suffer from a limited sample size and significant missing values. There are also no methods specifically designed to predict the physical properties of lignin-derived porous carbon.

[0021] In view of this, this application proposes a machine learning-based method for predicting the pore structure of lignin-derived porous carbon. This method achieves accurate prediction of the specific surface area and total pore volume of lignin-derived porous carbon by constructing a hybrid machine learning framework, combining a pre-trained model, a generative adversarial model and a regressor.

[0022] See also Figure 1 , Figure 1 This is a schematic diagram of the process for predicting the pore structure of lignin-derived porous carbon provided in this application. The steps are as follows: S101. Obtain parameter information of biomass-derived porous carbon and corresponding experimental parameters, and construct a parameter database; the biomass includes lignin.

[0023] In one possible implementation, the parameter information of the biomass-derived porous carbon includes: the type of raw materials for catalytic pyrolysis, the raw material ratio, elemental analysis data, industrial analysis data and pore characteristic data; the corresponding experimental parameters include pyrolysis temperature, heating rate and reaction time.

[0024] Optionally, the elemental analysis data includes the percentage of element contents of the substance, for example, the content of carbon, hydrogen, and oxygen.

[0025] Optionally, the industrial analysis data includes the fixed carbon volatile matter and ash content of the lignin raw material.

[0026] Optionally, the pore characteristic data include specific surface area and total pore volume.

[0027] Optionally, the parameter database is divided into two parts: Parameter database of lignin and other biomass: used to store parameter information of lignin and other biomass, such as raw material type, proportion, pyrolysis temperature, etc.

[0028] Database of analytical information on lignin-derived porous carbon and other biochars: This database stores pore property data (specific surface area and total pore volume) of lignin-derived porous carbon and other biochars. This database will serve as the foundation for subsequent machine learning model training.

[0029] It should be noted that the parameter information in the parameter database was obtained through keyword searches of academic literature. Due to discrepancies in the presentation of the collected data, manual retrieval was used to organize the relevant data; inaccurate or missing data was deleted. All of this data was standardized to ensure consistency in the input model.

[0030] S102. Use the parameter database to train the machine learning model to obtain a prediction model; the machine learning model includes an initial prediction model, an adversarial model and a regressor model; the adversarial model includes a generator and a discriminator.

[0031] Alternatively, see Figure 2 , step S102 includes the following sub-steps: S102a: Train the initial prediction model using the parameter database to obtain a pre-trained initial model, and calculate a first loss function.

[0032] It should be noted that the task of the pre-training model is to make a preliminary prediction of the specific surface area and total pore volume of lignin-derived porous carbon, and provide a reference for the subsequent generative adversarial model.

[0033] In one possible implementation, the initial prediction model utilizes an XGBoost regression model. This model efficiently processes high-dimensional data and performs well in classification and regression tasks. Using the XGBoost model, a preliminary prediction of the specific surface area and total pore volume of lignin-derived porous carbon can be achieved based on lignin analysis information and lignin-derived porous carbon analysis information.

[0034] In one possible implementation, during training, the root mean square error (RMSE) of the predictions is calculated as a loss function. The model's training performance is optimized by comparing the error between the training and validation sets. This step can optimize the initial prediction model for lignin-derived porous carbon and guide the subsequent training of the generative adversarial model.

[0035] S102b, the generator takes random noise as input, and after noise dimensionality reduction, outputs sample data that conforms to the characteristics of lignin-derived porous carbon; the discriminator optimizes the training generator by minimizing the Wasserstein distance between the generated samples and the real samples, and calculates the second loss function; combined with the first loss function of S102a and the second loss function of S102b, the training generator and discriminator are optimized.

[0036] It should be noted that the adversarial network is used to extract characteristic latent variables from the common knowledge of biochar and lignin-derived porous carbon in the parameter database and generate sample data that conforms to the characteristics of lignin-derived porous carbon. The characteristic latent variables include the potential expression of the characteristic parameters of biomass porous carbon (such as the feedstock type and ratio, elemental analysis data, industrial analysis data, and pore characteristics data for lignin catalytic pyrolysis) and the corresponding experimental parameters (such as pyrolysis temperature, heating rate, and reaction time).

[0037] The more specific steps are as follows: (1) Generator: The generator is used to take random noise as input and output a sample that conforms to the characteristics of lignin-derived porous carbon after noise dimensionality reduction. The goal of the generator is to maximize the authenticity of the sample and make the generated sample as similar as possible to the actual sample.

[0038] It should be noted that the samples that meet the characteristics of lignin-derived porous carbon refer to samples generated by the generator that have similar characteristic distributions to the real samples in the lignin parameter database.

[0039] (2) Discriminator: The task of the discriminator is to distinguish the difference between real samples and generated samples. By minimizing the Wasserstein distance between generated samples and real samples, the discriminator continuously improves the accuracy of judging generated samples and calculates the classification loss function.

[0040] (3) Joint Optimization: During the training process, the training of the generator and the discriminator is optimized by combining the first loss function (the loss function of the initial prediction model) and the second loss function (the feedback from the discriminator) in S102a. This ensures the effective extraction of feature latent variables and maintains high-quality sample generation.

[0041] Furthermore, the adversarial model is an improved Wasserstein GAN framework, see Figure 3Specifically, based on the Wasserstein GAN framework, the generated samples obtained by the generator are input into the initial prediction model, the prediction results are output, and the root mean square error of the generated samples in the initial prediction model is calculated, and the root mean square error is used to assist in optimizing the generator.

[0042] S102c: Fuse the sample data generated by the generator with the experimental parameters in the parameter database to obtain a fused training data set.

[0043] S102d. Use the fusion training data set to train the regressor model, optimize the regressor model parameters, and obtain a prediction model.

[0044] Understandably, the dataset contains both original samples and generated high-quality samples, thus ensuring the diversity and richness of the training data.

[0045] The regressor model is used to accurately predict the specific surface area and total pore volume of lignin-derived porous carbon.

[0046] In one possible implementation, the regressor model uses an XGBoost regression model and is trained based on a fusion training dataset. By training the model, the parameters of the regressor model are optimized, ultimately achieving accurate prediction of the pore structure of lignin-derived porous carbon.

[0047] S103: Evaluate and optimize the prediction model to obtain the optimal prediction model.

[0048] Optionally, the performance of the regressor model is evaluated by using the coefficient of determination and root mean square error to ensure high accuracy and generalization ability of the prediction results.

[0049] Optionally, the method of optimizing the prediction model includes increasing the number of model training times.

[0050] S104: Use the optimal prediction model to predict the material to be tested and output the prediction result.

[0051] In one possible implementation, S104 includes: The elemental analysis and industrial analysis results of the lignin raw materials of the test material, the type of catalyst used in the catalytic pyrolysis preparation process, and the pyrolysis temperature and time are obtained as input; the specific surface area and total pore volume are output through the optimal prediction model.

[0052] The following describes the lignin-derived porous carbon pore structure prediction device provided by the present invention. The lignin-derived porous carbon pore structure prediction device described below and the lignin-derived porous carbon pore structure prediction method described above can be referenced to each other.

[0053] Figure 4Schematic diagram of the structure of the lignin-derived porous carbon pore structure prediction device provided by an embodiment of the present invention. Figure 4 As shown, it includes: a database construction module 41, a training model 42, an optimization module 43 and an output module 45, wherein: A database construction module 41 is used to obtain parameter information of biomass-derived porous carbon and corresponding experimental parameters to construct a parameter database; the biomass includes lignin; A training model 42 is used to train a machine learning model using a parameter database to obtain a prediction model; the machine learning model includes an initial prediction model, an adversarial model, and a regressor model; the adversarial model includes a generator and a discriminator; The optimization module 43 is used to evaluate and optimize the prediction model to obtain the optimal prediction model; The output module 44 is used to predict the material to be tested using the optimal prediction model and output the prediction result.

[0054] Figure 5 An example of a physical structure diagram of an electronic device is shown below. Figure 5 As shown, the electronic device may include: a processor 510, a communication interface 520, a memory 530, and a communication bus 540, wherein the processor 510, the communication interface 520, and the memory 530 communicate with each other via the communication bus 540. The processor 510 may call the logic instructions in the memory 530 to execute the lignin-derived porous carbon pore structure prediction method.

[0055] Furthermore, the logic instructions in the aforementioned memory 530 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0056] On the other hand, the present invention also provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the lignin-derived porous carbon pore structure prediction method provided by the above methods.

[0057] In another aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the lignin-derived porous carbon pore structure prediction method provided by the above methods.

[0058] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0059] Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus a necessary general-purpose hardware platform, or of course, hardware. Based on this understanding, the essence of the above technical solution, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, or an optical disk, and includes a number of instructions for causing a computer device (such as a personal computer, server, or network device) to execute the methods described in each embodiment or certain portions of the embodiments.

[0060] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for predicting the pore structure of lignin-derived porous carbon, characterized in that: include: Obtain parameter information of biomass-derived porous carbon and corresponding experimental parameters, and construct a parameter database; The biomass includes lignin; Using a parameter database to train a machine learning model to obtain a prediction model; the machine learning model includes an initial prediction model, an adversarial model, and a regressor model; the adversarial model includes a generator and a discriminator; Evaluate and optimize the prediction model to obtain the optimal prediction model; Use the optimal prediction model to predict the material to be tested and output the prediction results.

2. The prediction method according to claim 1, characterized in that The biomass-derived porous carbon is obtained by catalytic pyrolysis of biomass.

3. The prediction method according to claim 2, characterized in that The parameter information of the biomass-derived porous carbon includes: the type of raw materials for catalytic pyrolysis, the raw material ratio, elemental analysis data, industrial analysis data and pore characteristic data; the corresponding experimental parameters include pyrolysis temperature, heating rate and reaction time.

4. The prediction method according to claim 1, characterized in that The method of using a parameter database to train a machine learning model to obtain a prediction model includes: Using the parameter database to train the initial prediction model, obtain a pre-trained initial model, and calculate the first loss function; The generator takes random noise as input and, after noise dimensionality reduction, outputs samples that conform to the characteristics of lignin-derived porous carbon. The discriminator optimizes the training of the generator by minimizing the Wasserstein distance between the generated samples and the real samples, and calculates a second loss function. The first and second loss functions are combined to optimize the training of the generator and discriminator. Fuse the samples generated by the generator with the experimental parameters in the parameter database to obtain a fused training dataset; The fusion training time is used to train the regressor model, optimize the regressor model parameters, and obtain the prediction model.

5. The prediction method according to claim 1, wherein The initial prediction model includes an XGBoost regression model; the adversarial model is a Wasserstein GAN framework; and the regressor model includes an XGBoost regression model.

6. The prediction method according to claim 5, wherein: The adversarial model is an improved WassersteinGAN framework. Specifically, based on the Wasserstein GAN framework, the generated samples obtained by the generator are input into the initial prediction model, and the root mean square error of the generated samples in the initial prediction model is calculated, and the root mean square error is used to assist in optimizing the generator.

7. A device for predicting the pore structure of lignin-derived porous carbon, characterized in that: include: A database construction module is used to obtain parameter information of biomass-derived porous carbon and corresponding experimental parameters and construct a parameter database; The biomass includes lignin; A training model for training a machine learning model using a parameter database to obtain a prediction model; the machine learning model includes an initial prediction model, an adversarial model, and a regressor model; the adversarial model includes a generator and a discriminator; Optimization module, used to evaluate and optimize the prediction model to obtain the optimal prediction model; The output module is used to predict the material to be tested using the optimal prediction model and output the prediction results.

8. An electronic device, characterized in that: include: A processor and a memory, the memory being coupled to the processor, the memory being used to store computer program code, the computer program code comprising computer instructions, and when the processor reads the computer instructions from the memory, the electronic device executes the method according to any one of claims 1 to 6.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed, the method according to any one of claims 1 to 6 is performed.

10. A computer program product, characterized in that The computer program product comprises instructions, and when the instructions are executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images