Method for determining preparation parameters of biochar with high electron exchange capacity based on machine learning

Abstract

The invention provides a method for determining preparation parameters of biochar with high electron exchange capacity based on machine learning, and is suitable for the field of biomass energy and environmental engineering. The method comprises the following steps: collecting biomass raw material composition, carbonization conditions, biochar physicochemical structure parameters and electron exchange capability data, and preprocessing; a two-stage series model is constructed and optimized, and the electronic exchange capacity of the biochar is predicted; and determining an optimal raw material ratio and carbonization process parameters for maximizing the electronic exchange capacity of the biochar through a reverse optimization method. According to the method, the double-layer mapping relation between the electron exchange capacity of the biochar and the raw material ratio and carbonization process parameters is established, so that the model accuracy is effectively improved. According to the raw material ratio and carbonization process parameters of the method, the biochar with high electron exchange capability can be efficiently prepared.

Description

Technical Field

[0001] The present invention relates to the field of biomass energy and environmental engineering technology, and in particular to a method for determining preparation parameters of biochar with high electron exchange capacity based on machine learning. Background Art

[0002] Biochar can be produced by pyrolysis and carbonization of organic solid waste, and can be widely used in environmental remediation, soil improvement, and greenhouse gas emission reduction. In the process of environmental remediation, the electron exchange capacity of biochar is one of its important characteristics that affects the redox of pollutants and the energy metabolism of microorganisms. For example, as an electron acceptor, it promotes the acidification efficiency of organic matter in anaerobic fermentation processes, and as an electron donor, it promotes the reduction of hexavalent chromium in soil. Improving the electron exchange capacity of biochar can further enhance its environmental remediation capabilities and also contribute to the high-value utilization of the products of pyrolysis and carbonization of organic solid waste.

[0003] The electron exchange capacity of biochar is closely related to the raw materials and carbonization conditions. However, organic solid waste has complex components and a wide variety of types, and there are many combinations with different carbonization conditions. At present, it takes a huge amount of manpower and material resources to determine the preparation parameters of biochar (i.e. the best raw materials and carbonization conditions) through artificial experiments. Machine learning can assist in the rapid regulation of organic solid waste synergistic pyrolysis to prepare biochar with high electron exchange capacity. Summary of the invention

[0004] The present invention proposes a method for determining preparation parameters of high electron exchange capacity biochar based on machine learning, which can solve the above problems.

[0005] In order to achieve the above object, the present invention adopts the following technical solution:

[0006] In a first aspect, the present invention provides a method for determining parameters of high electron exchange capacity biochar preparation based on machine learning, comprising the following steps:

[0007] Establishing a mapping model; the mapping model includes a first model and a second model connected in series; the input of the first model is the preparation parameters of the biochar, the output of the first model is the physicochemical structure parameters of the biochar, the input of the second model is the physicochemical structure parameters of the biochar, the output of the second model is the electron exchange capacity value of the biochar, and the electron exchange capacity value of the biochar is used to indicate the level of the electron exchange capacity of the biochar;

[0008] Based on the mapping model, the preparation parameters corresponding to the biochar with the highest electron exchange capacity are determined.

[0009] Note: The above method can directly establish the mapping relationship between the preparation parameters of biochar and the physicochemical structure parameters of biochar through the first model, which can reveal the intrinsic connection between the preparation parameters (such as pyrolysis temperature, time conditions, etc.) and the physicochemical structure of biochar, and more accurately describe its reaction mechanism; at the same time, since the physicochemical structure of biochar directly affects its electron exchange capacity, the second model can establish the corresponding relationship between the physicochemical structure of biochar and the electron exchange capacity of biochar, and more accurately describe the behavior of biochar in related reactions; the establishment of the above mapping and corresponding relationship can prepare biochar with higher electron exchange capacity by setting the preparation parameters, and the prediction accuracy is higher.

[0010] Furthermore, the preparation parameters of the biochar include carbonization conditions and raw material composition. The carbonization conditions include carbonization temperature, heating rate, carbonization time, and carbonization atmosphere; the raw material composition includes the volatile matter content, fixed carbon content, and ash content of the biochar, the elemental composition of the biochar, and the lignin content, cellulose content, protein content, lipid content, and sugar content in the biochar.

[0011] Note: The above parameters are directly related to the performance of biochar

[0012] Furthermore, the oxygen-containing functional groups, nitrogen-containing functional groups and metal component contents of the biochar, the BET specific surface area, and the electron exchange capacity of the biochar, including the electron-donating capacity and the electron-accepting capacity, are the number of electrons that the unit mass of the biochar can provide and accept, in mol e - / g.

[0013] Furthermore, the first model / second model is one of an XGBoost model, a random forest regression model, a gradient boosting regression model, and a support vector regression model.

[0014] Note: The above models have high prediction accuracy, strong robustness, and high flexibility, and can be customized and optimized according to the needs of specific problems. This makes them adaptable to various complex data sets and machine learning tasks.

[0015] Furthermore, the first model / the second model is screened from the XGBoost model, the random forest regression model, the gradient boosting regression model, and the support vector regression model; the screening method is:

[0016] Acquire a data set; the data set includes true values ​​of preparation parameters of biochar, true values ​​of physicochemical structure parameters, and true values ​​of electron exchange capacity values ​​of biochar;

[0017] The true values ​​of the preparation parameters of the biochar in the data set are respectively input into the XGBoost model, the random forest regression model, the gradient boosting regression model, and the support vector regression model to obtain the predicted values ​​of the electron exchange capacity of the biochar of each model;

[0018] The predicted value and the true value are used to perform model evaluation, and the model with the best prediction performance among the XGBoost model, the random forest regression model, the gradient boosting regression model, and the support vector regression model is obtained, and the model is used as the first model / the second model.

[0019] Note: Through the above screening method, the model that is most suitable for the research and prediction of biochar preparation parameters can be selected from various models, and it can significantly improve the prediction accuracy, enhance the model generalization ability, promote model interpretation and feature selection, and adapt to the needs of this program.

[0020] Furthermore, the method further includes, after obtaining the data set, adjusting the hyperparameters of each model of the XGBoost model, the random forest regression model, the gradient boosting regression model, and the support vector regression model in turn, and the adjustment method includes:

[0021] The data set is divided into 5 subsets, and training and testing are performed 5 times, with 4 subsets used for training each time and the remaining 1 subset used for testing. The accuracy under various hyperparameters is obtained, and the hyperparameter with the highest accuracy is selected. The hyperparameters include one of the learning rate, batch size, number of iterations / number of training rounds, regularization parameter, activation function, optimizer, and dropout rate.

[0022] Note: Through the above training and testing methods, the adaptability between the model and the data set can be enhanced, thereby improving the accuracy of the prediction results of each model.

[0023] Furthermore, the model evaluation method is to calculate the determination coefficient R between the predicted value and the true value using the following formula (1) and formula (2) respectively: 2 And the root mean square error RMSE:

[0024]

[0025] In the formula, y i is the i-th true value; is the i-th predicted value; is the mean of all true values; n is the number of true values, the R 2 The closer it is to 1 and the smaller the RMSE, the smaller the error and the better the model prediction performance.

[0026] Note: In the above method, R 2Used to evaluate the fit of the model and compare the performance of different models, RMSE provides an intuitive measure of the size of the prediction error, so the prediction performance of the model can be evaluated based on these two indicators.

[0027] Furthermore, the method for determining the preparation parameters corresponding to the biochar with the highest electron exchange capacity based on the mapping model is:

[0028] Based on the mapping model, a fitness function is defined; the independent variable of the fitness function is the preparation parameter of the biochar, and the dependent variable of the fitness function is the electron exchange capacity value of the biochar;

[0029] Determine constraint conditions; determine the value range of the preparation parameters of the biochar and the value range of the electron exchange capacity value of the biochar in the fitness function;

[0030] According to the fitness function and the constraint conditions, the preparation parameters of the biochar are obtained with the goal of maximizing the electron exchange capacity value of the biochar.

[0031] Note: By defining the fitness function as above, the mapping model can be expressed in the form of a function, and the constraints can be determined, that is, the domain and value range of the fitness function can be determined according to the data range of the collected independent variables and dependent variables and the numerical range that conforms to common sense. Then, the preparation parameters of biochar can be obtained when the electronic exchange capacity of biochar within the achievable range is maximized, providing a reference with good accuracy for the actual preparation process.

[0032] Furthermore, the definition of the fitness function based on the mapping model includes:

[0033] Convert the mapping model into the form of fitness function;

[0034] Determine the independent variables of the fitness function; use an explanatory model to first determine one or more biochar physicochemical structure parameters that have the greatest influence on the electron exchange capacity of the biochar as intermediate parameters; then screen out one or more biochar preparation parameters that have a greater influence on the intermediate parameters as the independent variables of the fitness function.

[0035] Note: By screening and determining the number of independent variables as described above, the amount of calculation can be reduced, the complexity of model calculation can be reduced, and the pertinence of the selection of preparation parameters can be improved, thereby making the obtained preparation parameters more practical.

[0036] Beneficial effects of the present invention:

[0037] The method of the present invention can directly establish a mapping relationship between the preparation parameters of biochar and the physicochemical structure parameters of biochar through the first model, which can reveal the intrinsic connection between the preparation parameters (such as pyrolysis temperature, time conditions, etc.) and the physicochemical structure of biochar, and more accurately describe its reaction mechanism; at the same time, since the physicochemical structure of biochar directly affects its electron exchange capacity, the corresponding relationship between the physicochemical structure of biochar and the electron exchange capacity of biochar can be established through the second model, which can more accurately describe the behavior of biochar in related reactions; the establishment of the above-mentioned mapping and corresponding relationship can prepare biochar with higher electron exchange capacity through the setting of preparation parameters, and the prediction accuracy is higher. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 This is one of the schematic flow diagrams of the method for preparing biochar with high electron exchange capacity based on machine learning provided in an embodiment of the present invention;

[0039] Figure 2 This is the second schematic flow chart of the method for preparing biochar with high electron exchange capacity based on machine learning provided in an embodiment of the present invention;

[0040] Figure 3 is a schematic diagram of the operation of the mapping model provided by an embodiment of the present invention;

[0041] Figure 4 It is a schematic diagram of the mapping relationship in the mapping model provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0042] The terms "first" and "second" etc. in the description and claims of the present invention are used to distinguish different objects rather than to describe a specific order of the objects.

[0043] In the embodiments of the present invention, words such as "exemplary" or "for example" are used to indicate examples, illustrations or explanations. Any embodiment or design described as "exemplary" or "for example" in the embodiments of the present invention should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary" or "for example" is intended to present related concepts in a specific way.

[0044] The method provided in the embodiment of the present invention relates to the technical field of biochar materials and can be used to optimize the preparation of biochar. Specifically, a mapping model is used to predict the electron exchange capacity of the biochar based on the preparation parameters of the biochar, and then the preparation parameters corresponding to the biochar with higher electron exchange capacity are selected to obtain the optimal raw materials and carbonization conditions, that is, the optimal preparation parameters.

[0045] In the prior art, the selection of biochar preparation parameters is mostly based on experimental selection, empirical determination or research data. These methods rely heavily on the experience of the experimenters and cannot form a complete and accurate determination method. The process of obtaining the parameters is relatively cumbersome and relies on manpower and experimental instruments, which causes huge costs. At the same time, in the current methods that rely on models, most of the research on predicting the function of biochar takes the raw materials, carbonization conditions and physical and chemical structure of biochar as input features at the same time, and directly uses the index values ​​to be obtained as output features for model calculation, and clarifies the accurate input variables, which limits the accuracy of the prediction.

[0046] At the same time, most current studies can only output recommended raw material types through feature importance analysis, but cannot provide a supporting solution for synergistic pyrolysis and carbonization for a given raw material, making it impossible to use it in practice. For example, when a certain organic solid waste to be treated is used as a resource, the organic solid waste is used as a raw material for preparation and treatment. The key point is to determine the preparation conditions such as biochemical composition and temperature required in addition to the raw materials. This function cannot be achieved in the current existing technology.

[0047] In order to solve the problems of high consumption and low efficiency in determining the preparation parameters of biochar by artificial experiments, and low electron exchange capacity of biochar prepared by the preparation parameters of the prior art, an embodiment of the present invention provides a biochar preparation method based on machine learning, including the establishment of a mapping model, and selecting the optimal preparation parameters according to the results obtained by the mapping model, so as to obtain biochar with higher electron exchange capacity.

[0048] like Figure 1 , Figure 3 As shown, the method for determining parameters of high electron exchange capacity biochar preparation based on machine learning provided in an embodiment of the present invention includes S01-S03;

[0049] S01, establishing a mapping model; the mapping model includes a first model and a second model connected in series; the input of the first model is the preparation parameters of biochar, the output of the first model is the physicochemical structure parameters of the biochar, the input of the second model is the physicochemical structure parameters of the biochar, the output of the second model is the electron exchange capacity value of the biochar, and the electron exchange capacity value of the biochar is used to indicate the level of the electron exchange capacity of the biochar;

[0050] like Figure 4 As shown, the preparation parameters of the biochar include carbonization conditions and raw material composition. The carbonization conditions include carbonization temperature, heating rate, carbonization time, and carbonization atmosphere; the raw material composition includes the volatile matter content, fixed carbon content, and ash content of the biochar, the elemental composition of the biochar, and the lignin content, cellulose content, protein content, lipid content, and sugar content in the biochar.

[0051] The physical and chemical structural parameters of the biochar include: the content of oxygen-containing functional groups, nitrogen-containing functional groups and metal components, BET specific surface area, and the electron exchange capacity of the biochar, including the electron-donating capacity and the electron-accepting capacity, which is the number of electrons that the unit mass of the biochar can provide and accept, in units of mol e - / g.

[0052] It should be understood that the electron exchange capacity of biochar mainly depends on its physical and chemical structure. For example, oxidizing functional groups such as C=O and pyridine have electron-accepting ability, and reducing functional groups such as C-OH and pyrrole have electron-donating ability. The pore structure and specific surface area of ​​biochar will affect its electron exchange capacity by affecting the accessibility of the above redox-active functional groups. The physical and chemical structure of biochar depends on the biochemical composition of the raw materials and the carbonization conditions. For example, the biochar produced by pyrolysis of biomass with high lignin content has a higher fixed carbon content. Compared with woody biomass, biomass such as agricultural waste, livestock manure, and urban sludge contains more water and ash. Water will reduce the yield of biochar and affect its surface functional group structure, while ash will reduce the C content and porosity of biochar. As the pyrolysis temperature increases, the H and O content in the biochar decreases, and the C and ash content increases.

[0053] The study found that the preparation parameters of biochar have a relatively accurate functional mapping relationship with the physicochemical structure parameters, and the physicochemical structure parameters of biochar have a relatively accurate mapping relationship with the electron exchange capacity of biochar. However, due to the complex intermediate process from the preparation parameters of biochar to the electron exchange capacity, including pyrolysis, carbonization, activation and other steps, the changes in these processes are difficult to describe with a simple functional relationship; and the relationship between the preparation parameters of biochar and the electron exchange capacity may be nonlinear, that is, a small change in the preparation parameters may lead to a significant change in the electron exchange capacity. If the preparation parameters of biochar are directly related to the electron exchange capacity of biochar, the complex physical and chemical processes need to be simplified, which may lead to a reduction in the accuracy of the mapping model;

[0054] Therefore, the embodiment of the present invention uses two models connected in series for calculation, which can achieve convenient calculation and ensure the accuracy of the mapping model.

[0055] The mapping model is one of an XGBoost model, a random forest regression model, a gradient boosting regression model, and a support vector regression model, for example, an XGBoost model is used.

[0056] The first model / the second model is screened from the XGBoost model, the random forest regression model, the gradient boosting regression model, and the support vector regression model; the screening method is:

[0057] like Figure 2 As shown, including A1~A3;

[0058] A1. Obtaining a data set; the data set includes the true values ​​of the preparation parameters of the biochar, the true values ​​of the physical and chemical structure parameters, and the true value of the electron exchange capacity value of the biochar;

[0059] In the embodiments of the present invention, the data in the data set are derived from Web of Science literature and public biochar databases.

[0060] In the embodiment of the present invention, after the data is obtained, normalization processing and data filling processing are performed on the data respectively;

[0061] Exemplarily, the normalization process is: using the StandardScaler method to perform data normalization processing. Specifically, StandardScaler standardizes the data by subtracting the mean of each eigenvalue (i.e., the value in the data set) from the mean of the eigenvalue, and then dividing it by the standard deviation of the eigenvalue (i.e., normalized data = (original data - mean of the feature) / standard deviation) to ensure that the mean of each feature is 0 and the standard deviation is 1, and then removes the eigenvalues ​​whose absolute value of the normalized eigenvalue and its mean is greater than 3, thereby completing the data normalization process.

[0062] Exemplarily, the filling data processing is as follows: for missing data, the K-nearest neighbors regression (KNN regression) method is used for filling. It can be understood that K-nearest neighbors regression is a non-parametric regression method that can be used to estimate missing data. The basic principle is to find the K nearest neighboring values ​​​​closest to the missing data point, and then estimate the missing value based on these neighboring values. Specifically, the data is divided into parts with and without missing values, and the Euclidean distance between the data points with missing values ​​and the data points without missing values ​​is calculated. The nearest K neighboring values ​​are found for each data point with missing values, and then the optimal K value is determined by cross-validation, and the average of the nearest neighboring values ​​is used to fill the missing value.

[0063] In an embodiment of the present invention, the method further includes, after obtaining the data set, adjusting the hyperparameters of each model of the XGBoost model, the random forest regression model, the gradient boosting regression model, and the support vector regression model in sequence, and the adjustment method includes:

[0064] The data set is divided into 5 subsets, and training and testing are performed 5 times, with 4 subsets used for training each time and the remaining 1 subset used for testing. The accuracy under various hyperparameters is obtained, and the hyperparameter with the highest accuracy is selected. The hyperparameters include one of the learning rate, batch size, number of iterations / number of training rounds, regularization parameter, activation function, optimizer, and dropout rate.

[0065] A2, inputting the true values ​​of the preparation parameters of the biochar in the data set into the XGBoost model, the random forest regression model, the gradient boosting regression model, and the support vector regression model, respectively, to obtain the predicted value of the electron exchange capacity value of the biochar;

[0066] A3. Perform model evaluation on the predicted value and the true value to obtain the model with the best prediction performance among the XGBoost model, random forest regression model, gradient boosting regression model, and support vector regression model, and use it as the first model / second model.

[0067] In the embodiment of the present invention, the training process is a process of continuously updating the initialized first model / second model using the training set, specifically, continuously updating the parameters of the first model / second model; the above data set is divided into a training set, a validation set and a test set, and model training is started at the same time. The model training and evaluation process includes:

[0068] The first step is model training.

[0069] The training samples in the training set are input into the first model / second model, the predicted value is calculated, and the loss function value between the predicted value and the true value is calculated, and gradient backpropagation is performed, and the parameters of the model are updated according to the set optimization algorithm.

[0070] It should be understood that model training is a process of multiple rounds of iterations. Each round of iteration traverses the training set once, and each time a small batch of samples is obtained from it and sent to the model to perform forward calculations to obtain predicted values. The effect of model training can be judged by observing the decreasing trend of the loss value in each round of iteration. For example, when the decreasing trend of the loss value remains unchanged, it means that the model tends to converge.

[0071] The second step is model evaluation.

[0072] The test set is input into the trained model for evaluation to obtain the predicted value, calculate the loss function value between the predicted value and the true value, and calculate the evaluation index value to facilitate the evaluation of the model effect.

[0073] The third step is model reasoning.

[0074] Input the preparation data of the biochar to be verified into the trained model to perform reasoning, and observe and verify whether the reasoning results meet expectations.

[0075] The method for evaluating the model is as follows: The method for evaluating the model is to calculate the determination coefficient R between the predicted value and the true value using the following formula (1) and formula (2) respectively: 2 And the root mean square error RMSE:

[0076]

[0077] In the formula, y i is the i-th true value; is the i-th predicted value; is the mean of all true values; n is the number of true values, the R 2 The closer it is to 1 and the smaller the RMSE, the smaller the error and the better the model prediction performance.

[0078] S02. Based on the mapping model, determining the preparation parameters corresponding to the biochar with the highest electron exchange capacity.

[0079] 1) Based on the mapping model, a fitness function is defined; the independent variable of the fitness function is the preparation parameter of the biochar, and the dependent variable of the fitness function is the electron exchange capacity value of the biochar;

[0080] First, the mapping model is converted into the form of a fitness function, such as F(x);

[0081] Then the independent variable of the fitness function is determined; an explanatory model (SHAP, Shapley AdditiveExplanations) is used to first determine one or more biochar physicochemical structure parameters that have the greatest impact on the electron exchange capacity of the biochar as intermediate parameters; then one or more biochar preparation parameters that have a greater impact on the intermediate parameters are screened out as the independent variables of the fitness function.

[0082] The explanation model SHAP is an explanation model based on the Shapley value of game theory, which is used to explain the prediction results of the model. It quantifies the contribution of each feature to the model prediction as a numerical value. The calculation process can be summarized as the following steps: ① Feature contribution calculation: For a given prediction result, the SHAP model calculates the contribution of each feature, that is, the impact of the feature on the prediction result in all possible feature combinations; ② Marginal contribution summation: Calculate the marginal contribution of each feature and perform a weighted summation to obtain the SHAP value of the feature; ③ Explanation model construction: Use the calculated SHAP value to build an explanation model that can explain the prediction results of the original model;

[0083] The calculation of the SHAP model can be performed in a variety of ways, including enumeration, approximation, and model-based specific algorithms. For example, for a linear model, the coefficients of the model can be directly used as SHAP values; for a deep learning model, the DeepSHAP algorithm can be used.

[0084] In the embodiment of the present invention, a fitness function F(x) is defined, where there are n input features (such as biomass type, volatile matter, biomass components, etc.), that is, F(x) = F(x1, x2, ... x n ); the dependent variables of F(x) include electron donating capacities (EDC) and electron accepting capacities (EAC), which can measure the current response of biochar at a specific potential and quantify EDC and EAC by calculating the integrated area of ​​current-time.

[0085] 2) Determine the constraint conditions; determine the value range of the preparation parameters of the biochar and the value range of the electron exchange capacity of the biochar in the fitness function; illustratively, set the range value of each parameter value, for example, the range value of the temperature parameter is 10 to 2000°C, and the range value of the content of the biochemical component / element / functional group is 0 to 100%.

[0086] 3) According to the fitness function and the constraint conditions, the preparation parameters of the biochar are obtained with the goal of maximizing the electron exchange capacity of the biochar.

[0087] Exemplarily, a function F(x)=E(x) is defined; F(x) represents the electron exchange capacity value of the biochar, and E(x) is the electron exchange capacity value of the biochar predicted based on given preparation parameters; an optimization calculation method is used to find a set of preparation parameters (x1, x2, ...x n ), so that the function F(x) is maximized;

[0088] The above-mentioned optimization calculation method is specifically one of gradient descent, genetic algorithm, particle swarm optimization, etc.

[0089] In an embodiment of the present invention, a method for obtaining the preparation parameters of the biochar with the goal of maximizing the electron exchange capacity value of the biochar is implemented in a given data set using an optimization algorithm, namely, particle swarm optimization (PSO), which identifies and updates the position of a particle swarm to explore the optimal solution in a restricted domain space.

[0090] It should be understood that the particle swarm optimization algorithm is an evolutionary computing method that searches for the optimal solution through collaboration and information sharing between individuals in a group. The particle swarm algorithm designs a massless particle that has only two properties: speed and position. The speed represents the speed of movement, and the position represents the direction of movement. Each particle searches for the optimal solution in the search space individually, and records it as the current individual extreme value, and shares the individual extreme value with other particles in the entire particle swarm. The best individual extreme value is found as the current global optimal solution of the entire particle swarm. All particles in the particle swarm adjust their speed and position according to the current individual extreme value they have found and the current global optimal solution shared by the entire particle swarm.

[0091] Experimental example:

[0092] Based on the above embodiments, this solution provides a specific experimental example including the following S1-S8:

[0093] S1. Retrieve data and construct a data set, including biomass type (biomass), volatile matter (VM), fixed carbon (FC), ash (Ash), biomass components (lignin, cellulose, protein, lipid, sugar content), elemental composition (C, H, O, N, S, Fe, Zn), carbonization temperature (T), carbonization time (t), heating rate (v), biochar structural parameters - oxygen-containing functional group content (quinone, C=O, C-OH, -COOH), nitrogen-containing functional group content (pyridinic-N, pyrrolic-N, protein-N, quaternary-N), the ratio of D peak to G peak in Raman spectrum (ID / IG), specific surface area (SA), electron exchange capacity (electron donation capacity; electron acceptance capacity).

[0094] S2. StandardScaler is used to normalize the data, outliers with absolute values ​​greater than 3 are removed, and the K nearest neighbor regression (KNN) method is used to fill in missing data. The data set is constructed, and 80% of the samples are randomly selected as the training set, and the rest are used as the test set.

[0095] S3. The preparation parameters of the first model input are biomass volatile matter (VM), fixed carbon (FC), ash (Ash), biomass components (lignin, cellulose, protein, lipid, sugar content), elemental composition (C, H, O, N, S, Fe, Zn), carbonization temperature (T), carbonization time (t), and heating rate (v). The output is biochar structural parameters - oxygen-containing functional group content (quinone, C=O, C-OH, -COOH), nitrogen-containing functional group content (pyridinic-N, pyrrolic-N, protein-N, quaternary-N), and the ratio of D peak to G peak in Raman spectrum (ID / IG).

[0096] S4. The input of the second model is the biochar structural parameters - oxygen-containing functional group content (quinone, C=O, C-OH, -COOH), nitrogen-containing functional group content (pyridinic-N, pyrrolic-N, protein-N, quaternary-N), the ratio of the D peak to the G peak in the Raman spectrum (ID / IG), and the output is the electron exchange capacity (electron donating capacities, EDC; electron accepting capacities, EAC).

[0097] S5. Use the training set to train XGBoost, random forest regression, gradient boosting regression, and support vector regression models respectively, and use the five-fold cross-validation method to select the best hyperparameters of each model according to the prediction accuracy (accuracy>90%).

[0098] S6, through the coefficient of determination (R 2 ) and root mean square error (RMSE) were used to evaluate the model prediction performance, and XGBoost was selected as the most suitable model.

[0099] S7. Shapley Additive Explanations (SHAP) was used to explain the influence weights of biomass composition and carbonization conditions on the physicochemical structure parameters of biochar, as well as the influence weights of biochar physicochemical structure parameters on electron exchange capacity, and the biomass composition and carbonization conditions with larger weights were used as the input of the subsequent fitness function. The results showed that the biochemical composition of biomass and carbonization temperature were the most important characteristics affecting the physicochemical structure parameters of biochar, and the specific surface area, C=O, CO, pyridinic-N, and pyrrolic-N of biochar were important characteristics affecting the electron exchange capacity.

[0100] S8. Define the fitness function, encode the carbonization conditions and raw material composition parameters as individuals in the particle swarm, and use the Particle Swarm Optimization (PSO) algorithm to solve the fitness function under the data set boundary (carbonization temperature 300-800℃, biochemical component / element / functional group content 0-100%) to obtain the optimal carbonization conditions (temperature 500℃) and raw material composition (lignin and cellulose contents are 23% dw and 52% dw, respectively).

[0101] S9. For a given organic solid waste that needs to be carbonized, municipal sludge, the test shows that it contains lignin and cellulose contents of approximately 15% dw and 45% dw, respectively. It is necessary to supplement sawdust and herbaceous plants with high lignin and cellulose contents to achieve the optimal raw material composition, thereby directionally preparing biochar with high electron exchange capacity.

[0102] In summary, through the method of the embodiment of the present invention, the preparation process of biochar can be systematically guided and optimized, and its electron exchange capacity can be accurately predicted and efficiently regulated, thereby promoting the widespread application of biochar in fields related to environmental remediation.

[0103] 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 embodiments of the present invention.

Claims

1. A method for determining parameters for preparing biochar with high electron exchange capacity based on machine learning, characterized in that: include: Establishing a mapping model; the mapping model includes a first model and a second model connected in series; the input of the first model is the preparation parameters of the biochar, the output of the first model is the physicochemical structure parameters of the biochar, the input of the second model is the physicochemical structure parameters of the biochar, the output of the second model is the electron exchange capacity value of the biochar, and the electron exchange capacity value of the biochar is used to indicate the level of the electron exchange capacity of the biochar; Based on the mapping model, the preparation parameters corresponding to the biochar with the highest electron exchange capacity are determined.

2. The method for determining parameters of high electron exchange capacity biochar preparation based on machine learning according to claim 1, characterized in that: The preparation parameters of the biochar include carbonization conditions and raw material composition. The carbonization conditions include carbonization temperature, heating rate, carbonization time, and carbonization atmosphere; the raw material composition includes the volatile matter content, fixed carbon content, ash content of the biochar, the elemental composition of the biochar, and the lignin content, cellulose content, protein content, lipid content, and sugar content in the biochar.

3. The method for determining parameters of high electron exchange capacity biochar preparation based on machine learning according to claim 1, characterized in that: The physical and chemical structural parameters of the biochar include: oxygen-containing functional groups, nitrogen-containing functional groups and metal component content, BET specific surface area, and the electron exchange capacity of the biochar is the number of electrons that the unit mass of the biochar can provide and accept, in mol e - / g.

4. The method for determining parameters of high electron exchange capacity biochar preparation based on machine learning according to claim 1, characterized in that: The first model / second model is one of an XGBoost model, a random forest regression model, a gradient boosting regression model, and a support vector regression model.

5. The method for determining parameters of high electron exchange capacity biochar preparation based on machine learning according to claim 1, characterized in that: The first model / the second model is screened from an XGBoost model, a random forest regression model, a gradient boosting regression model, and a support vector regression model; the screening method is: Acquire a data set; the data set includes true values ​​of preparation parameters of biochar, true values ​​of physicochemical structure parameters, and true values ​​of electron exchange capacity values ​​of biochar; The true values ​​of the preparation parameters of the biochar in the data set are respectively input into the XGBoost model, the random forest regression model, the gradient boosting regression model, and the support vector regression model to obtain the predicted values ​​of the electron exchange capacity of the biochar of each model; The predicted value and the true value are used to perform model evaluation, and the model with the best prediction performance among the XGBoost model, the random forest regression model, the gradient boosting regression model, and the support vector regression model is obtained, and the model is used as the first model / the second model.

6. The method for determining parameters of high electron exchange capacity biochar preparation based on machine learning according to claim 5, characterized in that: The method further includes performing hyperparameter tuning on the XGBoost model, the random forest regression model, the gradient boosting regression model, and the support vector regression model, by: The data set was divided into 5 subsets, and training and testing were performed 5 times. Four subsets were used for training each time, and the remaining 1 subset was used for testing. The accuracy under various hyperparameters was obtained, and the hyperparameters with the highest accuracy were selected. The hyperparameters included learning rate, batch size, number of iterations / training rounds, and regularization parameters.

7. The method for determining parameters of high electron exchange capacity biochar preparation based on machine learning according to claim 5, characterized in that: The method of model evaluation is to calculate the determination coefficient R between the predicted value and the true value using the following formula (1) and formula (2) respectively: 2 And the root mean square error RMSE: In the formula, y i is the i-th true value; is the i-th predicted value; is the mean of all true values; n is the number of true values, the R 2 The closer it is to 1 and the smaller the RMSE, the smaller the error and the better the model prediction performance.

8. The method for determining parameters of high electron exchange capacity biochar preparation based on machine learning according to claim 1, characterized in that: The method for determining the preparation parameters corresponding to the biochar with the highest electron exchange capacity based on the mapping model is: Based on the mapping model, a fitness function is defined; the independent variable of the fitness function is the preparation parameter of the biochar, and the dependent variable of the fitness function is the electron exchange capacity value of the biochar; Identify constraints; Determining, in the fitness function, a value range of the preparation parameters of the biochar and a value range of the electron exchange capacity value of the biochar; According to the fitness function and the constraint conditions, the preparation parameters of the biochar are obtained with the goal of maximizing the electron exchange capacity value of the biochar.

9. The method for determining parameters of high electron exchange capacity biochar preparation based on machine learning according to claim 8, characterized in that: The independent variables of the fitness function are screened; the screening process is: using an explanatory model, first determining one or more biochar physical and chemical structure parameters with the greatest influence on the electron exchange capacity of the biochar as intermediate parameters; then screening out one or more biochar preparation parameters with a greater influence on the intermediate parameters as the independent variables of the fitness function.

Images