Precise preparation method of functional group biochar based on machine learning

Abstract

The invention relates to a machine learning-based accurate preparation method of functional group biochar, which comprises the following steps: constructing a multi-dimensional fingerprint factor descriptor which affects the performance and active sites of biochar and comprises an internal impact factor and an external impact factor, and determining input characteristics and output characteristics; collecting a data set influencing the performance and active sites of the biochar by using a database; the data set is divided into a training set and a test set, RF and ANN model training is adopted, and a biochar performance prediction model is determined through cross validation and hyper-parameter optimization; and based on the biochar performance prediction model, accurately preparing the biochar material. According to the method, the optimal preparation condition of the material with the optimal performance is predicted through machine learning, the actual production of the biochar is guided, the test period is greatly shortened, the test cost is saved, and the method has important significance on precise preparation of the high-performance biochar and application of the high-performance biochar in the field of environmental restoration and has high application and popularization value.

Description

Technical Field

[0001] The present invention relates to the technical field of biochar preparation, and in particular to a method for accurately preparing functional-based biochar based on machine learning. Background Art

[0002] Biochar is a carbon-rich material made from agricultural waste, livestock and poultry manure and urban sludge. It is considered a potential adsorbent due to its high specific surface area, porous structure and rich oxygen-containing functional groups. It can remove pollutants such as ammonia nitrogen, phosphate, heavy metals, and organic dyes from water bodies by using its complex surface functional groups and developed pore structure. Among them, biochar prepared from agricultural waste has been widely studied in increasing crop income, and it also has significant benefits in sewage treatment, soil remediation and improvement.

[0003] The active sites of biochar are the key factors affecting the performance of biochar. The types and number of active sites are directly related to the composition of biomass raw materials, pyrolysis conditions, etc. However, current research mainly prepares biochar by simply changing the biomass raw materials and pyrolysis conditions. There are also problems such as long material preparation time and high trial and error costs, and the relationship between biomass raw materials and pyrolysis conditions and biochar active sites is still unclear. On the other hand, although there have been studies on the application of machine learning to the prediction of biochar performance, the input features considered are mainly external factors such as pyrolysis conditions and experimental conditions, which have limited effect on the prediction of biochar performance, and it is difficult to guide the precise preparation of high-performance biochar, which in turn limits the application of biochar. Summary of the invention

[0004] The technical problem to be solved by the present invention is to overcome the deficiencies in the prior art and provide a precise preparation method of functional-based biochar based on machine learning, which solves the problems of long preparation time of functional-based biochar and high cost caused by many trial and error times.

[0005] The present invention is achieved through the following technical solutions: A precise preparation method of functional biochar based on machine learning, comprising the following steps: S1. construct a multidimensional fingerprint factor descriptor that affects the performance and active sites of biochar, wherein the multidimensional fingerprint factor descriptor includes internal influencing factors and external influencing factors, and determine that the input feature is the multidimensional fingerprint factor descriptor, and the output feature is the content of active sites; S2. Using a database to collect data sets that affect biochar performance and active sites, including data on the input characteristics and data on the output characteristics; S3. Dividing the data set into a training set and a test set, using RF and ANN model training, cross-validating the data set to prevent overfitting, optimizing hyperparameters to improve the generalization ability of the model, and determining a biochar performance prediction model; S4. Based on the biochar performance prediction model, the biochar material is accurately prepared.

[0006] According to the above technical solution, preferably, in step S1, the internal influencing factor is the biomass composition, and the external influencing factor is the pyrolysis condition. The biomass composition includes the lignin, cellulose and hemicellulose content of the biomass, and the C, N, O and H element content of the biochar, and the pyrolysis condition includes the pyrolysis temperature, pyrolysis time and heating rate for preparing the biochar.

[0007] According to the above technical solution, preferably, in step S1, the output characteristics include the contents of four active sites: CO, C=C / C=O, C-OH, and OC=O.

[0008] According to the above technical solution, preferably, in step S2, data preprocessing is performed on the data set, wherein the data preprocessing includes one or more of unifying units, filling missing values, and normalization processing.

[0009] According to the above technical solution, preferably, in step S2, the data preprocessing includes: The data units in the biomass composition are unified as "%", the pyrolysis temperature unit is unified as "℃", the pyrolysis time unit is unified as "h", and the heating rate unit is unified as "℃ / min". For missing data, fill it with the average value of the existing data collected; The data is processed using the normalization method, namely:

[0010] Among them, X i * is a factor (X i ). max and X min X i After normalization, all data are mapped to the range [0,1].

[0011] According to the above technical solution, preferably, in step S3, the RF model is used as the biochar performance prediction model.

[0012] According to the above technical solution, preferably, after step S3, the SHAP method is used to analyze the influence degree and importance ranking of the multidimensional fingerprint factors on the active sites of biochar.

[0013] According to the above technical solution, preferably, the internal influencing factors have a greater influence on the content of active sites than the external influencing factors, and the C content, O content, pyrolysis temperature and pyrolysis time have a significant influence on the predicted content of active sites.

[0014] The beneficial effects of the present invention are: The present invention provides a method for rapidly and targetedly preparing biochar with high performance and functional active sites based on multidimensional fingerprint descriptors and machine learning, and uses big data to comprehensively and comprehensively study the effects of factors such as pyrolysis conditions, biochar structural characteristics, and experimental conditions on the performance and active sites of prepared biochar, and clarifies the key parameters that affect the performance of prepared biochar; The present invention predicts the optimal preparation conditions of materials through machine learning and guides the actual production of biochar, which greatly shortens the test cycle and saves experimental costs. It is of great significance to the precise preparation of high-performance biochar and its application in the field of environmental remediation, and has high application and promotion value. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is the degree of influence of the multidimensional fingerprint factor on the active site (CO site) of biochar in the present invention.

[0016] Figure 2 It is the degree of influence of the multidimensional fingerprint factor on the active sites (C=C / C=O sites) of biochar in the present invention.

[0017] Figure 3 It is the degree of influence of the multidimensional fingerprint factor on the active site (C-OH site) of biochar in the present invention.

[0018] Figure 4 It is the degree of influence of the multidimensional fingerprint factor on the active site (OC=O site) of biochar in the present invention.

[0019] Figure 5 It is the interface diagram of the GUI toolbox in the present invention. DETAILED DESCRIPTION

[0020] In order to enable those skilled in the art to better understand the technical solution of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and the best embodiment. Based on the embodiments in the invention, all other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the invention.

[0021] As shown in the figure, the present invention comprises the following steps: S1. Construct a multidimensional fingerprint factor descriptor that affects the performance and active sites of biochar, wherein the multidimensional fingerprint factor descriptor includes internal influencing factors and external influencing factors, and determine that the input feature is the multidimensional fingerprint factor descriptor, and the output feature is the content of active sites.

[0022] Biomass composition and pyrolysis conditions are the key factors that determine the performance and active sites of biochar. Therefore, a multidimensional fingerprint factor descriptor covering the internal influencing factors of biomass composition and the external influencing factors of pyrolysis conditions is constructed. Among them, the internal influencing factors include the lignin, cellulose and hemicellulose content of biomass, and the C, N, O and H element content of biochar; the external influencing factors include the pyrolysis temperature, pyrolysis time and heating rate of biochar preparation. Among them, the multidimensional fingerprint factor descriptor constructed above is the input feature, and the contents of the four active sites of CO, C=C / C=O, C-OH and OC=O are selected as the output features that affect the performance of biochar.

[0023] S2. Using a database, collect data sets that affect biochar performance and active sites, including data on the input characteristics and data on the output characteristics.

[0024] The above biochar-related data sets were collected using a database, including input characteristic data: internal influencing factors (lignin, cellulose and hemicellulose content of biomass, C, N, O and H element content of biochar) and external influencing factors (pyrolysis temperature, pyrolysis time and heating rate); output characteristic data: CO, C=C / C=O, C-OH, OC=O of biochar.

[0025] At the same time, the data set is preprocessed, where the data preprocessing includes one or more of unifying units, filling missing values, and normalization processing, specifically including: (1) The units of the collected biomass lignin, cellulose and hemicellulose contents, the C, N, O and H element contents of biochar, and the biochar CO, C=C / C=O, C-OH, and OC=O data were unified into "%", the pyrolysis temperature was unified into "℃", the pyrolysis time was unified into "h", and the heating rate was unified into "℃ / min". (2) For missing data, fill it with the average value of the existing data collected; (3) Use normalization method to process data to speed up convergence, that is:

[0026] Among them, X i * is a factor (X i ). max and X min Xi After normalization, all data are mapped to the range [0,1].

[0027] S3. 80% of the data set was used as the machine learning model training set, and the remaining 20% ​​was used as the model test set. The RF and ANN models were used for training, and the data set was cross-validated 5 times. At the same time, the hyperparameters were optimized to determine the biochar performance prediction model. The model performance was determined by the coefficient of determination (R 2 ), root mean square error (RMSE), mean absolute error (MAE), absolute correction factor (Adjusted R 2 ) indicator evaluation. The model training results are shown in Table 1. The ANN model training determination coefficient (R 2 ) is between 0.72 and 0.84, the root mean square error (RMSE) is between 10 and 18, the mean absolute error (MAE) is between 6 and 12, and the absolute correction coefficient (Adjusted R 2 ) is between 0.72 and 0.82. The RF model training determination coefficient (R 2 )>0.90, RMSE<0.8, MAE<2, Adjusted R 2 )>0.90. Therefore, the RF model was determined to be the preferred prediction model.

[0028] Table 1

[0029] At the same time, the SHAP method was used to analyze the influence of multidimensional fingerprint factors on biochar active sites and their importance ranking. The higher the SHAP value of the input feature, the greater the influence of the feature on the model. Figure 1-4 As shown in the figure, all input features have a significant impact on the four active sites, proving the correctness of the input features constructed based on the multidimensional fingerprint factor descriptor. In the CO, C=C / C=O, C-OH and OC=O active sites, the contribution rates of internal influencing factors are 69%, 63%, 59% and 60%, respectively, while the contribution rates of external influencing factors are 31%, 37%, 41% and 40%, respectively. Therefore, the influence of internal influencing factors on the CO, C=C / C=O, C-OH and OC=O active sites is greater than that of external influencing factors. According to the SHAP feature importance analysis, C content, O content, pyrolysis temperature and pyrolysis time have the most significant effects on the prediction of CO, C=C / C=O, C-OH and OC=O active sites, and are the most important factors affecting the active sites on the biochar surface.

[0030] S4. Based on the biochar performance prediction model, the biochar material is accurately prepared. In this example, based on the SHAP feature importance analysis, the optimal preparation conditions of high-performance biochar are predicted using the best-performing RF model, which are: pyrolysis temperature 400℃~550℃, pyrolysis time 1.5~2.5h, pyrolysis rate 5~10℃ / min, C content 50~60%, O content 30~40%, H content 1~5%, and N content 0.1~1%.

[0031] In this example, environmental waste was used as raw material to prepare biochar under different conditions. The C, N, O, and H contents of the samples were detected by elemental analyzer; the active site content of the samples was detected by X-ray photoelectron spectrometer (XPS), where CO and C=C / C=O contents were obtained through C1s peak, and C-OH and OC=O contents were obtained through O1s peak.

[0032] Wheat straw was used as biomass raw material, and the contents of cellulose, hemicellulose, and lignin were 34.82%, 27.15%, and 18.27%, respectively. Three types of biochar were prepared under three pyrolysis conditions. The preparation conditions and biochars were named as follows: (1) Biochar was prepared at a pyrolysis temperature of 300 °C, a pyrolysis time of 2 h, and a pyrolysis rate of 5 °C / min and was named BC1. (2) Biochar was prepared at a pyrolysis temperature of 500 °C, a pyrolysis time of 2 h, and a pyrolysis rate of 5 °C / min and was named BC2. (3) Biochar was prepared at 700 °C, 2 h, and 5 °C / min, and was named BC3. (4) Biochar was prepared at a pyrolysis temperature of 500 °C, a pyrolysis time of 4 h, and a pyrolysis rate of 5 °C / min and was named BC4. The prepared biochar materials were tested, and the data of each index are shown in Table 2. Compared with the four biochar materials, the biochar (BC2) material prepared by pyrolysis at 500℃ for 2h at a pyrolysis rate of 5℃ / min had CO, C=C / C=O, C-OH and OC=O contents of 47.06%, 59.44%, 43.91% and 46.89%, respectively, which were significantly higher than the other three materials. Therefore, compared with other biochars, biochar (BC2) contains the most surface active sites and has the best performance.

[0033] Table 2

[0034] The measured values ​​of the four surface active sites of the prepared biochars BC1, BC2, BC3 and BC4 were compared with the predicted values ​​of the RF model, and the results are shown in Table 3. The predicted values ​​of CO, C=C / C=O, C-OH and OC=O contents of the four biochar materials were close to the measured values, and the prediction error rate was less than 5%. This result shows that the RF model has good prediction performance.

[0035] Table 3

[0036] In addition, this application can also develop a GUI toolbox based on the optimal prediction model. The toolbox interface parameters are as follows: Figure 5 As shown, including the lignin, cellulose and hemicellulose contents of biomass, the C, N, O and H element contents of biochar, the pyrolysis temperature, pyrolysis time and heating rate of biochar, the prediction of the precise preparation method of biochar can be visualized.

[0037] The present invention provides a method for rapidly and targetedly preparing biochar with high performance and functional active sites based on multidimensional fingerprint descriptors and machine learning. Big data is used to comprehensively and integratedly study the influence of factors such as pyrolysis conditions, biochar structural characteristics and experimental conditions on the performance and active sites of prepared biochar, and key parameters affecting the performance of prepared biochar are clarified. The present invention predicts the optimal preparation conditions of materials through machine learning, guides the actual production of biochar, greatly shortens the test cycle, saves experimental costs, is of great significance to the precise preparation of high-performance biochar and its application in the field of environmental remediation, and has high application and promotion value.

[0038] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A precise preparation method of functional biochar based on machine learning, characterized in that: The steps include: S1. construct a multidimensional fingerprint factor descriptor that affects the performance and active sites of biochar, wherein the multidimensional fingerprint factor descriptor includes internal influencing factors and external influencing factors, and determine that the input feature is the multidimensional fingerprint factor descriptor, and the output feature is the content of active sites; S2. Using a database to collect data sets that affect biochar performance and active sites, including data on the input characteristics and data on the output characteristics; S3. Dividing the data set into a training set and a test set, using RF and ANN model training, cross-validating and optimizing hyperparameters on the data set, and determining a biochar performance prediction model; S4. Based on the biochar performance prediction model, the biochar material is accurately prepared.

2. The method for accurately preparing functional biochar based on machine learning according to claim 1, characterized in that: In step S1, the internal influencing factor is the biomass composition, and the external influencing factor is the pyrolysis condition.

3. The method for accurately preparing functional biochar based on machine learning according to claim 2, characterized in that: In step S1, the biomass composition includes the lignin, cellulose and hemicellulose contents of the biomass, and the C, N, O and H element contents of the biochar, and the pyrolysis conditions include the pyrolysis temperature, pyrolysis time and heating rate for preparing the biochar.

4. The method for accurately preparing functional biochar based on machine learning according to any one of claims 1 to 3, characterized in that: In step S1, the output characteristics include the contents of four active sites: CO, C=C / C=O, C-OH, and OC=O.

5. The method for accurately preparing functional biochar based on machine learning according to claim 3, characterized in that: In step S2, data preprocessing is performed on the data set.

6. The method for accurately preparing functional biochar based on machine learning according to claim 5, characterized in that: In step S2, the data preprocessing includes one or more of unifying units, filling missing values, and normalization processing.

7. The method for accurately preparing functional biochar based on machine learning according to claim 6, characterized in that: In step S2, the data preprocessing includes: The data units in the biomass composition are unified as "%", the pyrolysis temperature unit is unified as "℃", the pyrolysis time unit is unified as "h", and the heating rate unit is unified as "℃ / min"; For missing data, fill it with the average value of the existing data collected; The data is processed using the normalization method, namely: , Among them, X i * is a factor (X i ), X max and X min X i After normalization, all data are mapped to the range [0,1].

8. A method for accurately preparing functional biochar based on machine learning according to claim 5 or 6, characterized in that: In step S3, the RF model is used as the biochar performance prediction model.

9. The method for accurately preparing functional biochar based on machine learning according to claim 8, characterized in that: After step S3, the SHAP method was used to analyze the influence degree and importance ranking of multidimensional fingerprint factors on biochar active sites.

10. The method for accurately preparing functional biochar based on machine learning according to claim 9, characterized in that: The internal influencing factors have a greater influence on the content of active sites than the external influencing factors, and the C content, O content, pyrolysis temperature and pyrolysis time have a significant influence on the predicted active site content.

Images