Biomass pyrolysis activation energy analysis method based on stacking model

Abstract

The invention discloses a biomass pyrolysis activation energy analysis method based on a stacking model, and belongs to the field of chemical or physical analysis. The analysis method comprises the following steps: acquiring an approximate analysis result, an element analysis result, raw material type classification and activation energy data of biomass through open source data, and performing preprocessing operation on the acquired data to obtain a training set and a test set in a preset proportion; inputting the training set into the constructed stacking model to obtain a biomass pyrolysis activation energy prediction result, and completing analysis; the model is validated using the test set. According to the method, the sample breadth and the feature dimension of the current pyrolysis activation energy are remarkably expanded by considering the input variables such as the similar analysis result, the element analysis result, the raw material type classification and the activation energy data of the near biomass; by constructing the stacking model, the precision and generalization ability of predicting the pyrolysis activation energy are enhanced.

Description

Technical Field

[0001] The present invention relates to the field of chemical or physical analysis, and in particular to a biomass pyrolysis activation energy analysis method based on a stacking model. Background Art

[0002] As a key technology in the field of renewable energy, biomass pyrolysis has attracted much attention due to its potential in carbon neutrality and resource recycling. Existing research mainly obtains the pyrolysis activation energy ( ), and using traditional statistical models or single machine learning algorithms (such as random forest, support vector regression) to calculate the thermal activation energy ( ) and biomass characteristics (e.g. elemental composition, ash content).

[0003] However, existing research still has many limitations. First, most models are built based on data sets with limited sample sizes, lacking extensive coverage of diverse biomass feedstocks and experimental conditions, making it difficult to achieve high-quality generalized predictions. Second, the selection of input variables is relatively simple, mainly focusing on elemental analysis or approximate analysis indicators (such as C, H, N, O content, ash content, moisture, etc.), while ignoring important process variables that change dynamically during the pyrolysis process. In particular, the conversion rate (α), as a direct reflection of the pyrolysis reaction process, has a significant impact on the activation energy, but has not been systematically incorporated into the modeling framework in most studies. These problems restrict the rapid evaluation and optimization of the pyrolysis process, and urgently need to be broken through through large-scale data construction and advanced algorithm design. Summary of the Invention

[0004] In order to solve the above technical problems, the present invention provides a biomass pyrolysis activation energy analysis method based on a stacking model.

[0005] To implement the above technology, the specific steps include: S1. Collect biomass approximate analysis results, elemental analysis results, raw material type classification, and activation energy data through open source data, and perform preprocessing operations on the collected data to obtain training and test sets with preset proportions; Proximate analytical results include: ash content, expressed as mass percent; Elemental analysis results include: carbon (C), hydrogen (H), nitrogen (N) and oxygen (O) content, in terms of mass The mass percentages are expressed and the H / C and O / C mass ratios are calculated accordingly; The raw material types include: lignin biomass, herbaceous biomass, animal biomass, agricultural waste and industrial / municipal organic waste, and fruit shells; The activation energy data includes pyrolysis activation energy values calculated by FWO, KAS and Friedman method ); The preprocessing operation includes missing value processing, non-numeric feature encoding and data division operation. The missing value processing is to delete samples containing missing values. The non-numeric feature encoding is to encode lignin biomass, herbaceous biomass, animal-derived biomass, agricultural waste and industrial / urban organic waste, and fruit shell as [0, 1, 2, 3] by using Label Encoding method, and to encode the activation energy values calculated by FWO, KAS and Friedman method as [0, 1, 2]. ); The data division operation is to divide the data into training set and test set according to the preset proportion after randomizing the data, wherein the training set is used for model training and optimal hyperparameter selection, and the test set is used for evaluating the model performance.

[0006] S2, input the training set into the constructed stacked model to obtain the biomass pyrolysis activation energy prediction result, and complete the analysis. The step of constructing the stacked model includes: S2.1, selecting a preset number of models as primary models of the stacked model; In the present application, the preset number of models is 2, and the primary models include random forest model (RF) and decision tree model (DT). S2.2, performing hyperparameter optimization operation on the primary models of the stacked model. The hyperparameter optimization operation is to use grid search method (Grid SearchCV) to traverse all combinations in the given hyperparameter range, and to select the optimal parameters through cross-validation. The given hyperparameter range includes the number of decision trees (n_estimators) in the random forest model (RF) for obtaining the stability and prediction accuracy of the model, the maximum number of features (max_features) in each tree in the random forest model for considering the maximum number of input features, the minimum number of samples (min_samples_split) for node splitting in the decision tree model (DT), and the maximum depth (max_depth) of the tree in the decision tree model (DT) for maximizing the fitting ability of the model. S2.3, taking the approximate analysis result, the element analysis result and the raw material type classification in the training set as the input features of the primary model of the stacked model after the optimal hyperparameter optimization, and taking the activation energy data as the true label to be predicted respectively to obtain the prediction result of the primary model; S2.4, taking the prediction result of the primary model as the input feature of the secondary model, and taking the activation energy data as the true label of the secondary model to be predicted to obtain the final prediction result; The secondary model is a linear regression model.

[0007] S3, verifying the model using the test set; The verification method of the model is: The test set is used to verify the single model and the stacked model respectively, and the verification indexes include the determination coefficient (R²) and the root mean square error (RMSE).

[0008] Advantages of the present application The present application significantly expands the sample breadth and feature dimension of the current pyrolysis activation energy by considering the input variables of the approximate analysis result, the element analysis result, the raw material type classification and the activation energy data.

[0009] The present application enhances the prediction accuracy and generalization ability of the pyrolysis activation energy by constructing a stacked model. BRIEF DESCRIPTION OF DRAWINGS

[0010] Figure 1 The step flowchart of the present application. DETAILED DESCRIPTION

[0011] The present application will be further described in detail below in combination with specific embodiments.

[0012] S1, collecting the approximate analysis result, the element analysis result, the raw material type classification and the activation energy data of the biomass through open source data, and performing a pretreatment operation on the collected data to obtain a preset proportion of the training set and the test set; In this embodiment, the relevant literatures are retrieved from the Web of Science database, and the keywords are biomass, pyrolysis and isoconversional method; The approximate analysis result includes the ash content, which is expressed in mass percentage; The element analysis result includes the content of carbon (C), hydrogen (H), nitrogen (N) and oxygen (O) expressed in mass percentage, and the H / C and O / C mass ratios are calculated therefrom; wherein the calculation method is: ; ; The raw material type classification includes: lignin biomass, herbaceous biomass, animal-derived biomass, agricultural waste and industrial / urban organic waste, and fruit shell; The activation energy data includes: pyrolysis activation energy values calculated by FWO, KAS and Friedman methods; The preprocessing operation includes: missing value processing, non-numeric feature encoding and data division operation; The missing value processing method is to delete samples containing missing values; The non-numeric feature encoding method is: since part of the features are non-numeric, most machine learning models require input to be numeric variables, and the present application uses Label Encoding method to perform non-numeric feature encoding operation, wherein the non-numeric feature encoding includes: encoding lignin biomass, herbaceous biomass, animal-derived biomass, agricultural waste and industrial / urban organic waste, and fruit shell as [0, 1, 2, 3], and encoding activation energy values calculated by FWO, KAS and Friedman methods as [0, 1, 2]; The data division operation method is: after randomly shuffling the data, the data is divided into a training set and a test set according to a preset ratio; wherein the training set is used for model training and optimal hyperparameter selection, and the test set is used for evaluating the performance of the model; in the present embodiment, the preset ratio is 9:1.

[0013] S2, input the training set into the constructed stacking model to obtain the biomass pyrolysis activation energy prediction result, and complete the analysis; The step of constructing the stacking model includes: S2.1, selecting a preset number of models as primary models of the stacking model; In the present application, the preset number of models is 2; the primary models include: random forest model (RF) and decision tree model (DT); S2.2, performing hyperparameter optimization operation on the primary models of the stacking model; The hyperparameter optimization operation method is: using grid search method (Grid SearchCV) to traverse all combinations in the given hyperparameter range, and selecting the optimal parameters through cross-validation; The given hyperparameter range includes: the number of decision trees (n_estimators) in the random forest model (RF) for obtaining the stability and prediction accuracy of the model; the maximum number of features (max_features) of each tree in the random forest model for considering the maximum number of input features; the minimum number of samples (min_samples_split) for node splitting in the decision tree model (DT); and the maximum depth (max_depth) of the tree in the decision tree model (DT) for maximizing the fitting ability of the model;​​ In this embodiment, the optimal parameters are selected by cross-validation: the number of decision trees (n_estimators) in the random forest model (RF) is set to 50; the maximum number of features (max_features) in each tree in the random forest model is set to 1.0, indicating that all features are used; the minimum number of samples for splitting nodes (min_samples_split) in the decision tree model (DT) is set to 2; the maximum depth of the tree (max_depth) in the decision tree model (DT) is set to None; S2.3, the approximate analysis results, element analysis results and raw material type classification in the training set are used as the input features of the primary model of the stacked model optimized by the optimal hyperparameters, and the activation energy data is used as the true label for prediction, to obtain the prediction results of the primary model; S2.4, the prediction results of the primary model are used as the input features of the secondary model, and the activation energy data is used as the true label of the secondary model for prediction, to obtain the final prediction results; The secondary model is a linear regression model.

[0014] S3, the model is verified using the test set; The verification method of the model is: The test set is used to verify the single model and the stacked model respectively, and the verification indicators include the determination coefficient (R²) and the root mean square error (RMSE); In this embodiment, the single model includes: random forest model (RF), adaptive boosting regression model (ABR), gradient boosting regression model (GBR), extreme random tree regression model (ETR), K nearest neighbor regression model (KNR), decision tree model (DT), extreme gradient boosting model (XGB), light gradient boosting machine (LGBM) and category boosting model (CBR); In this embodiment, the stacked model includes: adaptive boosting regression model (ABR), gradient boosting regression model (GBR) and extreme random tree regression model (ETR); category boosting model (CBR), extreme gradient boosting model (XGB) and neural network regressor (NNR); The test set verification effect is shown in Table 1; Table 1: Verification effect of each model From Table 1, it can be seen that the present application performs best on the test set, with R² as high as 0.8952, indicating that the prediction ability of the present application is significantly better than other models; the RMSE is only 20.08, which is the best comprehensive performance among all models, showing its excellent generalization ability and stability, showing that the present application has high prediction accuracy and good generalization ability.

[0015] The above merely describes the embodiments of the present application, and further describes the purposes, technical solutions and beneficial effects of the present application in detail, and it should be understood that the above merely describes the specific embodiments of the present application and is not used to limit the present application, and any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims

1. A method for analyzing biomass pyrolysis activation energy based on a stacking model, characterized in that: The following steps are involved: S1. Collect biomass approximate analysis results, elemental analysis results, raw material type classification, and activation energy data through open source data, and perform preprocessing operations on the collected data to obtain training and test sets with preset proportions; The preprocessing operations include: missing value processing, non-numerical feature encoding and data partitioning operations; S2. Input the training set into the constructed stacking model to obtain the prediction results of biomass pyrolysis activation energy and complete the analysis; The stacking model includes: a random forest model, a decision tree model and a linear regression model; S3. Use the test set to verify the model; The method of verifying the model is: using a test set to verify the single model and the stacked model respectively, and the verification indicators include: determination coefficient and root mean square error.

2. The biomass pyrolysis activation energy analysis method based on a stacking model according to claim 1, characterized in that: The open source data is used to collect the approximate analysis results, elemental analysis results, raw material type classification and activation energy data of the biomass, and preprocessing operations are performed on the collected data to obtain training sets and test sets with preset proportions. The approximate analysis results include: ash content, expressed as a mass percentage; the elemental analysis results include: carbon, hydrogen, nitrogen and oxygen contents, expressed as a mass percentage, and the H / C and O / C mass ratios are calculated based on this; the raw material type classification includes: lignin biomass, herbaceous biomass, animal-derived biomass, agricultural waste and industrial / urban organic waste, and fruit shells; the activation energy data includes: pyrolysis activation energy values ​​calculated by FWO, KAS and Friedman methods.

3. The biomass pyrolysis activation energy analysis method based on a stacking model according to claim 1, characterized in that: The steps of inputting the training set into the constructed stacking model to obtain the prediction result of biomass pyrolysis activation energy and completing the analysis include: S2.

1. Select a preset number of models as primary models for the stacking model; Primary models include: random forest model, decision tree model; S2.

2. performing hyperparameter optimization on the primary model of the stacked model; The hyperparameter optimization operation is to use the grid search method to traverse all combinations within the given hyperparameter range and select the optimal parameters through cross-validation; The given hyperparameter ranges include: the number of decision trees in the random forest model; the maximum number of features per tree in the random forest model; the minimum number of samples for node splitting in the decision tree model; the maximum depth of the tree in the decision tree model; S2.

3. Use the approximate analysis results, elemental analysis results, and raw material type classification in the training set as the primary model input features of the stacking model after optimal hyperparameter optimization, and use the activation energy data as the true label to perform predictions on each model to obtain the prediction results of the primary model; S2.

4. Using the prediction results of the primary model as the input features of the secondary model, and using the activation energy data as the true label of the secondary model for prediction, to obtain the final prediction results; the secondary model is a linear regression model.

Images