A method for predicting the calorific value of biomass charcoal based on a bivariate feature extraction strategy

Abstract

The present application provides a kind of based on the biomass charcoal calorific value prediction method of bivariate feature extraction strategy, it is related to biomass energy utilization technical field, and prediction method includes the following steps: according to the relationship between calorific value and the spectral data of element in biomass charcoal, determine the X characteristic element of linear correlation between spectral data and calorific value;According to the concentration of element in biomass charcoal and calorific value, determine Y characteristic element by pearson correlation analysis;X variable characteristic is obtained;Y variable characteristic is obtained;According to the relationship between X variable characteristic and Y variable characteristic and calorific value, construct quantitative analysis model;X variable characteristic and Y variable characteristic in the biomass charcoal to be predicted are input into quantitative analysis model, and the predicted calorific value is obtained.The present application can be used for accurate quantitative prediction analysis of biomass charcoal CV by XY bivariate feature extraction method combined with quantitative analysis model, and the accuracy of biomass charcoal calorific value prediction result is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of biomass energy utilization, and particularly relates to a biomass charcoal calorific value prediction method based on a bivariate feature extraction strategy. BACKGROUND

[0002] With the continuous rise of fossil energy prices, the value of biomass energy utilization has attracted more and more attention from energy departments of various countries. Biomass charcoal has become one of the clean energies in China due to its advantages such as cleanliness, and recyclable ash. Calorific value (CV) is an important index for heat efficiency, combustion equipment heat balance, and fuel quality calculation. At present, the calorific value is mainly calculated by measuring the enthalpy difference between reactants and products through an oxygen bomb calorimeter. However, this method is expensive and cumbersome to operate, and cannot meet the real-time monitoring requirements of industrial production.

[0003] Since the calorific value is related to the content of elements in the biomass charcoal, the calorific value can be predicted by the content of elements in the biomass charcoal. At present, the calorific value of biomass charcoal is usually predicted by X-variable feature extraction methods such as competitive adaptive reweighted sampling (CARS), genetic algorithm (GA), and regression coefficient method. These methods mainly extract characteristic spectra according to the root mean square error of cross-validation (RMSECV) of the model constructed by different sampling spectral band combinations. Since the regression model corresponding to these methods is mainly a linear partial least squares regression (PLSR), and the biomass charcoal mainly contains metal element atomic / ion spectral lines and non-metal element atomic / molecular spectral lines, the X-variable feature extraction method can only screen out most of the metal element LIBS characteristic spectra that meet the Lambert-Beer linear law, resulting in low prediction accuracy of the calorific value of biomass charcoal. SUMMARY

[0004] The technical problem to be solved by the present application is to solve the problem of low prediction accuracy of the calorific value of biomass charcoal in the prior art. The present application provides a biomass charcoal calorific value prediction method based on a bivariate feature extraction strategy. The prediction method provides a basis for predicting the calorific value of biomass charcoal by using a bivariate feature extraction method, improves the accuracy of the prediction result, and solves the problem of low prediction accuracy of the calorific value of biomass charcoal in the prior art.

[0005] The technical solution adopted by the present application to solve the technical problem is as follows:

[0006] A biomass charcoal calorific value prediction method based on a bivariate feature extraction strategy, comprising the following steps:

[0007] S1: obtaining the calorific value of biomass charcoal;

[0008] S2: determining X characteristic elements linearly related to the calorific value according to the relationship between the calorific value and the spectral data of elements in the biomass charcoal.

[0009] S3: Based on the calorific value and the concentration of elements in the biochar, the characteristic element Y is determined by Pearson correlation analysis;

[0010] S4: Obtain the elemental feature spectral lines of the X feature elements as X variable features;

[0011] S5: Obtain the elemental feature spectral lines of the Y feature elements as Y variable features;

[0012] S6: Construct a quantitative analysis model based on the characteristics of the X variable and the relationship between the characteristics of the Y variable and the calorific value;

[0013] S7: Input the X variable characteristics and Y variable characteristics of the biochar to be predicted into the quantitative analysis model to obtain the predicted calorific value;

[0014] The quantitative analysis model is in the form of CV=K(X+Y)+B, where CV is the calorific value of biochar, X and Y are the characteristics of variable X and variable Y, respectively, and K and B are both numerical matrices.

[0015] The spectral data of elements in biochar were determined by laser-induced breakdown spectroscopy.

[0016] Optionally, step S1 includes: determining the calorific value of the biochar using an oxygen bomb calorimeter.

[0017] Optionally, step S2 includes:

[0018] S21: Obtain LIBS spectral data of all elements in the biochar;

[0019] S22: Construct a linear PLSR model based on the calorific value and the LIBS spectral data of all elements;

[0020] S23: The X feature elements are derived based on the interactive verification results of the linear PLSR model.

[0021] Optionally, the characteristic element X is Ca, Cr, Mg, and K.

[0022] Optionally, when performing the Pearson correlation analysis in step S3, the element with a highly significant correlation coefficient is the Y feature element.

[0023] Optionally, the Y characteristic element is C, O, H, and Na.

[0024] Optionally, step S6 involves constructing the quantitative analysis model using an artificial neural network algorithm.

[0025] Optionally, the calculation formula for constructing the quantitative analysis model using an artificial neural network algorithm is as follows:

[0026] Formula (1);

[0027] Formula (2);

[0028] Formula (3);

[0029] Where i is the weak predictor, j is the sample index, and D is the sample weight. i,j D represents the training weights for the previous round of samples. i+1,j For training weights in the next round of samples, alpha i Assign weights to each weak predictor, Error i The sum of the weights of all samples for each weak predictor.

[0030] Optionally, the effectiveness of the quantitative analysis model is evaluated by the root mean square error, the average relative error, and the relative standard error.

[0031] Optionally, the root mean square error is calculated using the following formula:

[0032] Formula (4);

[0033] The formula for calculating the average relative error is as follows:

[0034] Formula (5);

[0035] The formula for calculating the relative standard error is as follows:

[0036] Formula (6);

[0037] Where RMSEP is the root mean square error of prediction, RMSE is the root mean square error, Ym is the actual value, Yp is the predicted value, n is the sample size, AREP is the mean relative error, RSDP is the relative standard error, and mean(Yp) is the predicted sample mean.

[0038] The beneficial effects of this invention are:

[0039] The present invention provides a method for predicting the calorific value of biochar based on a bivariate feature extraction strategy. By analyzing the correlation between the CV of biochar and the content of various elements, Y-type variable features related to CV are selected based on the correlation significance analysis results. X-type variable features related to CV are obtained by screening the regression coefficient threshold of a multivariate linear model. Furthermore, the extracted variable features are used to construct a nonlinear neural network model for biochar CV. Thus, the combination of the XY bivariate feature extraction method and the quantitative analysis model can be used for accurate quantitative prediction and analysis of biochar CV, improving the accuracy of biochar calorific value prediction results. Attached Figure Description

[0040] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0041] Figure 1 This is a flowchart illustrating the biomass calorific value prediction method based on a bivariate feature extraction strategy in this invention.

[0042] Figure 2 This is a schematic diagram of the XY bivariate feature extraction method used in this invention;

[0043] Figure 3 These are the LIBS spectra of the straw charcoal in this invention;

[0044] Figure 4 This is the threshold selection spectrum of the regression coefficient method in this invention. Figure 1 ;

[0045] Figure 5 This is the threshold selection spectrum of the regression coefficient method in this invention. Figure 2 ;

[0046] Figure 6 This is the result of bivariate feature extraction in this invention. Figure 1 ;

[0047] Figure 7 This is the result of bivariate feature extraction in this invention. Figure 2 ;

[0048] Figure 8 This is a schematic diagram of the model construction process of the present invention. Figure 1 ;

[0049] Figure 9 This is a schematic diagram of the model construction process of the present invention. Figure 2 ;

[0050] Figure 10 This is a schematic diagram of the model prediction results in this invention. Detailed Implementation

[0051] The present invention will now be described in further detail. The embodiments described below are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0052] To address the issue of low accuracy in predicting the calorific value of biomass char in existing technologies, this invention provides a biomass char calorific value prediction method based on a bivariate feature extraction strategy. (See [link to relevant documentation]). Figure 1 As shown, the prediction method includes the following steps:

[0053] S1: Obtain the calorific value of biochar;

[0054] S2: Based on the relationship between calorific value and spectral data of elements in biochar, determine the X characteristic element that is linearly correlated with calorific value;

[0055] S3: Based on the calorific value and the concentration of elements in biochar, the characteristic elements of Y are determined by Pearson correlation analysis;

[0056] S4: Obtain the elemental characteristic spectral lines of the X feature elements as features of the X variable;

[0057] S5: Obtain the elemental characteristic spectral lines of the Y feature elements as features of the Y variable;

[0058] S6: Construct a quantitative analysis model based on the relationship between the characteristics of variable X and variable Y and calorific value;

[0059] S7: Input the X and Y variable characteristics of the biochar to be predicted into the quantitative analysis model to obtain the predicted calorific value.

[0060] For ease of understanding, this invention preferably uses straw charcoal as the biochar; and preferably uses an oxygen bomb calorimeter to determine the calorific value of the biochar; the spectral data of elements in the biochar can be determined by laser-induced breakdown spectroscopy (LIBS); LIBS is an emerging technology that uses the atomic or ion spectra emitted when a laser directly ablates the sample surface to generate plasma for detection; currently, in the process of quantitative analysis of biomass biomass characteristics based on LIBS, the conventional variable feature extraction methods are mainly X independent variable feature extraction type methods, such as competitive adaptive reweighted sampling method, genetic algorithm, and regression coefficient method, etc. This type of method mainly extracts characteristic spectra based on the root mean square error (RMSECV) of cross-validation of models constructed from different combinations of sampling spectral bands. Since the regression model corresponding to this type of method is mainly linear partial least squares regression (PLSR), and biochar, such as straw charcoal, mainly contains atomic / ionic spectral lines of metal elements and atomic / molecular spectral lines of non-metal elements, the X independent variable feature extraction method can only screen out most of the LIBS characteristic spectra of metal elements that conform to the Lambert-Beer linear law. The X characteristic elements screened in this invention are the elements whose spectral data are linearly related to their calorific value, and the X variable features are the elemental characteristic spectral lines of the screened X characteristic elements.

[0061] Due to matrix effects, existing conventional variable feature extraction methods, namely single X-variable feature extraction methods, have significant errors, resulting in low accuracy in predicting the calorific value of biomass char. Therefore, this invention adds a Y-variable feature extraction method to the traditional X-variable feature extraction method, thereby improving the accuracy of biomass char calorific value prediction by combining the Y-variable feature extraction method with the traditional X-variable feature extraction method.

[0062] Specifically, in order to extract the X variable characteristics, this invention first measures the calorific value of biochar using an oxygen bomb calorimeter, obtains the spectral data of all elements in the biochar using LIBS, analyzes the calorific value and the spectral data of all elements, and screens out elements whose spectral data are linearly correlated with the calorific value. These elements are the X characteristic elements; the spectral data of these X characteristic elements, that is, the elemental characteristic spectral lines, are the X variable characteristics.

[0063] This invention further obtains the concentration of all elements in biochar, wherein the concentration of each element can be obtained using existing elemental analyzers, ICP-MS, etc.; the calorific value of biochar and the corresponding concentration of all elements in biochar are analyzed, and the correlation between the calorific value of biochar and the concentration of each element in biochar is obtained by Pearson correlation analysis. The optimal combination of spectral lines for most non-metallic elements is obtained through significant correlation analysis. This optimal combination of spectral lines can be linear or non-linear. Based on the correlation analysis results, elements with a high correlation between concentration and calorific value are identified as Y characteristic elements; further, the elemental characteristic spectral lines of Y characteristic elements in biochar are obtained by LIBS, which are the Y variable characteristics.

[0064] Finally, a model is built based on the extracted X and Y variable features and calorific value. The modeling method can adopt the existing neural network modeling method to obtain a quantitative analysis model. When it is necessary to predict the calorific value of biochar, the X and Y variable features of biochar are first obtained through LIBS, and the obtained X and Y variable features are input into the above quantitative analysis model. After the calculation of the quantitative analysis model, the predicted calorific value of the biochar can be obtained.

[0065] The present invention provides a method for predicting the calorific value of biochar based on a bivariate feature extraction strategy. By analyzing the correlation between the CV of biochar and the content of various elements, Y-type variable features related to CV are selected based on the correlation significance analysis results. X-type variable features related to CV are obtained by screening the regression coefficient threshold of a multivariate linear model. Furthermore, the extracted variable features are used to construct a nonlinear neural network model for biochar CV. Thus, the combination of the XY bivariate feature extraction method and the quantitative analysis model can be used for accurate quantitative prediction and analysis of biochar CV, improving the accuracy of biochar calorific value prediction results.

[0066] Specifically, the preferred step S2 of the present invention includes:

[0067] S21: Obtain LIBS spectral data of all elements in biochar;

[0068] S22: Construct a linear PLSR model based on calorific value and LIBS spectral data of all elements;

[0069] S23: Determine the X feature elements based on the interactive verification results of the linear PLSR model.

[0070] Specifically, the present invention preferably uses the regression coefficient method (RC) as the feature extraction method for the X variable. By setting different regression coefficient thresholds, variable features are screened, and linear partial least squares regression (PLSR) models with different combinations of variable features are constructed. The optimal combination of element spectral lines is selected as the extraction result based on the RMSECV value of the model.

[0071] The preferred characteristic element X of this invention is Ca, Cr, Mg and K. Similarly, the characteristic variable X is the spectral data of Ca, Cr, Mg and K.

[0072] In the preferred step S3 of this invention, when performing Pearson correlation analysis, the element with a highly significant correlation coefficient (p<0.01) is the Y characteristic element, and the preferred Y characteristic elements are C, O, H and Na. Similarly, the Y variable characteristics are the spectral data of C, O, H and Na.

[0073] In the preferred step S6 of this invention, a quantitative analysis model is constructed using an artificial neural network algorithm.

[0074] The Genetic Algorithm Optimization and Adaptive Augmentation Artificial Neural Network Algorithm (GA-BP-Adaboost) is an improved nonlinear artificial neural network model. On the one hand, it uses a genetic algorithm (GA) to optimize the boundary function parameters w and b. On the other hand, it uses multiple weak predictors to train each sample in the dataset and continuously adjusts the sample weights D. At the same time, it assigns a weight alpha to each weak predictor to calculate the final strong predictor result.

[0075] Specifically, the calculation formula for constructing a quantitative analysis model using artificial neural network algorithms is as follows:

[0076] Formula (1);

[0077] Formula (2);

[0078] Formula (3);

[0079] Where i is the weak predictor, j is the sample index, and D is the sample weight. i,j D represents the training weights for the previous round of samples. i+1,j For training weights in the next round of samples, alpha i Assign weights to each weak predictor, Error i The sum of the weights of all samples for each weak predictor.

[0080] If a sample is accurately predicted (less than the relative error rate threshold), the weight of that sample decreases, the alpha sign in formula (2) becomes negative, and the corresponding Error value remains unchanged (+0); if a sample is not accurately predicted (greater than the relative error rate threshold), the weight of that sample increases, the alpha sign in formula (2) becomes positive, and the corresponding Error value increases (+D). i,j ).

[0081] The effectiveness of the preferred quantitative analysis model of this invention is evaluated by root mean square error, average relative error, and relative standard error.

[0082] Specifically, the formula for calculating the root mean square error is as follows:

[0083] Formula (4);

[0084] The formula for calculating the average relative error is as follows:

[0085] Formula (5);

[0086] The formula for calculating the relative standard error is as follows:

[0087] Formula (6);

[0088] Where RMSEP is the root mean square error of prediction, RMSE is the root mean square error, Ym is the actual value, Yp is the predicted value, n is the sample size, AREP is the mean relative error, RSDP is the relative standard error, and mean(Yp) is the predicted sample mean.

[0089] The smaller the RMSE, the better the modeling effect; the smaller the RSDP and AREP, the higher the model prediction accuracy.

[0090] In summary, this invention addresses the shortcomings of traditional X-variable feature extraction methods in the quantitative analysis of biochar CV using LIBS. It proposes an XY bivariate feature extraction method. First, the correlation between biochar CV and the content of various elements is analyzed. Based on the significance analysis results, Y-type feature variables related to CV are selected, mainly including the broadened spectral bands of C, O, H, and Na elemental analysis lines. Simultaneously, X-type feature variables related to CV are obtained by screening the regression coefficient thresholds of a multivariate linear model, mainly including the spectral lines of Ca, Cr, Mg, and K elemental analysis lines. Furthermore, a nonlinear neural network model for biochar CV is constructed using the extracted feature variables, and its AREP and RSDP values ​​are significantly lower than those reported in related literature. The results show that the XY bivariate feature extraction method combined with the GA-BP-Adaboost model can be used for accurate quantitative prediction and analysis of biochar CV.

[0091] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0092] In the preferred measurement process of this invention, the LIBS system ablation source is a Q-switched Nd:YAG solid-state laser with an excitation wavelength of 1064 nm, a maximum frequency of 2 Hz, a pulse delay of 10 ns, and a maximum excitation energy of 100 mJ. After laser focusing, the normal to the sample surface is incident and induces plasma. A collimating lens at a 45° angle relative to the horizontal direction receives the plasma signal and transmits it to the fiber optic channel. The fiber optic output is connected to a 7-channel charge-coupled spectrometer with a resolution of 0.05 nm and a detection wavelength range of 187 nm-980 nm. In this invention, to address the influence of laser pulse energy fluctuations on spectral line intensity, the laser energy is set to 30 mJ, the number of repeated laser ablation cycles at a single point is set to 3, and the single-point spot size is set to 200 μm. To avoid bremsstrahlung emission, the detector delay time relative to the laser pulse is 0.7 μs.

[0093] The biochar used in this invention is straw charcoal; 79 straw charcoal samples were purchased from Nanjing Zhironglian Technology Co., Ltd., including rice straw charcoal and corn straw charcoal. After drying in a 45℃ oven, the samples were pulverized using a blade grinder and passed through a 75μm sieve before being placed in resealable bags for later use. The elemental contents of N, C, S, H, and O in the samples were determined using an elemental analyzer; the elemental contents of K, Na, Ca, Mg, Pb, Cr, Cu, Zn, and P were determined using ICP-MS; and the CV was determined using an oxygen bomb calorimeter. The results are shown in Table 1.

[0094] Table 1

[0095] Element composition and properties Number of samples Range Mean ± standard deviation N (g / kg) 79 0.63-1.41 1.01±0.22 C (g / kg) 79 39.94-60.2 51.43±4.22 S (g / kg) 79 0.2-0.71 0.28±0.07 H (g / kg) 79 0.49-2.4 1.45±0.56 O (g / kg) 79 36.23-57.15 45.83±4.56 CV (MJ / kg) 79 15.68-21.17 18.73±1.3 K (g / kg) 79 4.62-13.42 9.09±2.81 Na (g / kg) 79 0.34-5.64 2.57±1.38 Ca (g / kg) 79 2.02-10.78 5.75±2.04 Mg (g / kg) 79 0.8-7.41 3.96±1.9 Pb (mg / kg) 79 1.11-11.53 5.69±2.89 Cr (mg / kg) 79 2.79-28.51 12.51±6.51 Cu (mg / kg) 79 6.67-19.06 11.9±3.48 Zn (mg / kg) 79 27.01-624.2 84.25±101.28 P (mg / kg) 79 1.47-2.91 2.17±0.38

[0096] The general form of the quantitative analysis model is CV = K(X + Y) + B, where CV is the calorific value of straw charcoal, and X and Y are the X-variable feature and Y-variable feature, respectively. For this invention, the X-variable feature and Y-variable feature each represent a set of data, and the corresponding feature extraction methods can be classified as X-variable feature extraction method and Y-variable feature extraction method; K and B are both numerical matrices; the XY bivariate feature extraction method used in this invention follows the procedure described below. Figure 2 As shown.

[0097] First, the Y-variable feature extraction method was used to obtain the spectral peak band combinations of non-metallic elements such as C and O related to the CV of straw charcoal. Through correlation analysis between CV and straw charcoal element concentration, the broadened bands of elemental analysis lines with extremely significant correlation (P<0.01) were selected as Y-variable features. Subsequently, the regression coefficient method (RC) was used as the X-variable feature extraction method. By setting different regression coefficient thresholds, feature variables were screened, and linear partial least squares regression (PLSR) models with different combinations of feature variables were constructed. Based on the RMSECV value of the model, the optimal combination of metallic element spectral lines was selected as the extraction result. Since the extracted features include linear and nonlinear variables, a nonlinear precise quantitative analysis model for straw charcoal CV was finally constructed.

[0098] The relevant calculation formulas in the model building process are as follows:

[0099] Formula (1);

[0100] Formula (2);

[0101] Formula (3);

[0102] Where i is the weak predictor, j is the sample index, and D is the sample weight. i,j D represents the training weights for the previous round of samples. i+1,j For training weights in the next round of samples, alpha i Assign weights to each weak predictor, Error i The sum of the weights of all samples for each weak predictor.

[0103] If a sample is accurately predicted (less than the relative error rate threshold), the weight of that sample decreases, the alpha sign in formula (2) becomes negative, and the corresponding Error value remains unchanged (+0); if a sample is not accurately predicted (greater than the relative error rate threshold), the weight of that sample increases, the alpha sign in formula (2) becomes positive, and the corresponding Error value increases (+D). i,j ).

[0104] Where i and j are the weak predictor and sample number, respectively, and Error is the sum of the weights of all samples for each weak predictor. If a sample is accurately predicted (less than the relative error rate threshold), the weight of that sample decreases, the alpha sign in formula (2) becomes negative, and the corresponding Error value remains unchanged (+0); if a sample is not accurately predicted (greater than the relative error rate threshold), the weight of that sample increases, the alpha sign in formula (2) becomes positive, and the corresponding Error value increases (+Di,j).

[0105] The model's performance is primarily evaluated by the root mean square error (RMSE), average relative error (ARE), and relative standard error (RSD), calculated as shown in the formulas below. A smaller RMSE indicates better modeling performance; smaller RSDP and AREP indicate higher model prediction accuracy.

[0106] Formula (4);

[0107] Formula (5);

[0108] Formula (6).

[0109] The LIBS spectral analysis is as follows:

[0110] The main elements of straw carbon are in the range of 187 nm to 980 nm, and the average spectrum is as follows: Figure 3 As shown. A search of the NIST atomic spectroscopy database confirmed that straw charcoal contains non-metallic elements C, O, H, N, S, and P; nutritional metallic elements K, Na, Ca, and Mg; and heavy metal elements Cu, Zn, Cr, and Pb. During crop growth, K, Na, Ca, Mg, Cu, Zn, and Pb mainly exist as metal cations, while Cr mainly exists as an anion (CrO4). 2- Cr2O7 2- Heavy metals (Pb, Cu, and Zn) enter various organs and tissues of crops via ion penetration or the action of transport proteins. However, due to the plant's own antitoxicity mechanisms, most heavy metals are selectively retained by suberin and fructose in the cell wall, while K, Na, Ca, Mg, and Cr are more easily transported into the cell by transport proteins to serve as plant growth factors. Therefore, K, Na, Ca, Mg, and Cr may affect the formation of cellulose, hemicellulose, and lignin, thereby influencing the carbon content (CV) of straw charcoal.

[0111] Compared to nonmetallic and heavy metal elements, nutrient metals K, Ca, Na, and Mg exhibit higher emission line intensities. This may be because nutrient metals have lower ionization energies, making it easier for them to undergo energy level transitions to excited states, resulting in a greater number of atoms in excited states per unit volume and thus stronger spectral line intensities. Furthermore, since K, Na, Ca, Mg, and Cr participate in the physiological growth processes of crops, their LIBS analytical lines may show higher sensitivity to the CV of straw charcoal.

[0112] The bivariate feature extraction process is as follows:

[0113] First, the Y-variable feature extraction method was used to obtain the spectral peak bands of characteristic elements that were significantly correlated with CV. The results are shown in Table 2. CV showed a strong correlation with the concentrations of C, O, H, and Na, with highly significant correlation coefficients (p < 0.01); a relatively low correlation with the concentrations of Ca, K, and Cr, with significant correlation coefficients (p < 0.05); and no correlation with the concentrations of S, Mg, Zn, Pb, N, Cu, and P. Since C, O, and H in straw char mainly exist in the form of elemental carbon, aromatic rings, carboxyl groups, ether bonds, and silicon-oxygen bonds, they can significantly improve the CV of straw char. Therefore, the LIBS spectral peak bands of C, O, H, and Na were selected as the Y-variable feature extraction results.

[0114] Table 2

[0115] N C S H O K Na Ca Mg Pb Cr Cu Zn P CV 0.1 0.85b -0.22 0.46b -0.84b 0.25a -0.33b -0.29a -0.17 -0.16 -0.22a -0.07 -0.17 0.03

[0116] Where a represents p < 0.05; b represents p < 0.01.

[0117] Subsequently, the X-variable feature extraction method was used to obtain the larger values ​​of the regression coefficients of the full-band PLSR model as CV feature variables, and the results are as follows: Figure 4 As shown. When the regression coefficient threshold is set to 15×10... -5 10×10 -5 5×10 -5 4×10 -5 3×10 -5 2×10 -5 and 1×10 -5 At that time, the RMSECV result of the PLSR model it constructed was as follows: Figure 5 As shown in the figure. The results indicate that the RMSECV value first decreases and then increases as the threshold gradually decreases. This may be because when RMSECV decreases, the analytical line spectra of linearly correlated elements Ca, Cr, Mg, and K in the CV model are gradually selected; while when RMSECV increases, the spectra of nonlinearly correlated elements and noise information in the CV model are gradually selected. When the threshold is 4 × 10⁻⁶... -5 When the RMSECV drops to its minimum value of 0.61, the number of 49 selected feature variables is the feature extraction result of variable X.

[0118] In summary, the CV feature variable results obtained by the XY bivariate feature extraction method are as follows: Figure 6 As shown. The Y variable feature extraction method mainly includes the broadened wavelength bands of C, O, H, and Na elemental analysis lines, while the X variable feature extraction method mainly includes the spectra of Ca, Cr, Mg, and K elemental analysis lines, and the two methods have no overlap. PLSR models for X univariate, Y univariate, and XY bivariate features are constructed respectively, and... Figure 7The results show that the bivariate feature model has the smallest RMSECV value, indicating that the XY bivariate feature extraction method can successfully obtain CV feature variables. However, since this method obtains a large number of nonlinear feature variables, it is necessary to construct a nonlinear multiple regression model to improve the model's predictive performance.

[0119] The process of constructing and predicting the nonlinear feature model is as follows:

[0120] Before constructing the GA-BP-Adaboost feature model, it is necessary to optimize the parameters of the BP-ANN model using GA and Adaboost to build a more robust nonlinear feature model. The relevant parameter settings are as follows: GA parameters: number of iterations = 20, population size = 20; Adaboost parameters: number of weak predictors = 20, number of double hidden layers = 1, input layer, hidden layer, and output layer transfer functions = tansig, tansig, and purelin respectively, training function = trainbr, learning rate and learning objective = 0.01.

[0121] For the GA-BP part, when the crossover probability is set to 0.8-0.95 and the step size is 0.05, and the mutation probability is set to 0.05-0.5 and the step size is 0.05, the corresponding average fitness values ​​are as follows: Figure 8 As shown, the average fitness value decreases as the mutation probability gradually decreases and the crossover probability gradually increases. When the mutation probability and crossover probability are 0.1 and 0.95 respectively, the average fitness drops to a minimum of 80.98. Since the fitness function is the sum of absolute prediction errors, a smaller fitness indicates more accurate model predictions. For the BP-Adaboost part, since the relative error rate (RE) of the 20 weak predictor models for each training sample ranges from 0 to 7.15%, the RE threshold is set to 0.05 to 0.30, with a step size of 0.05. The results are as follows. Figure 9 As shown, it can be observed that as the RE value increases, the AREP and RSDP values ​​generally show a trend of first decreasing and then increasing. Therefore, the mutation probability, crossover probability, and RE value are set to 0.1, 0.95, and 0.01, respectively, for further construction of the GA-BP-Adaboost model.

[0122] Furthermore, the extracted feature variable data were subjected to principal component analysis (with a threshold parameter set to 99.93%), and a GA-BP-Adaboost model was constructed based on the optimized parameters. The model results were compared with those of the linear PLSR model, and the results are as follows: Figure 10As shown, compared with the PLSR feature model, the predicted values ​​of most samples (1-3, 5, 7-17, 19, 20) in the GA-BP-Adaboost feature model are closer to the measured values, with AREP and RSDP decreasing by 0.82% and 0.91%, respectively. The results indicate that the nonlinear GA-BP-Adaboost model performs better. This may be because the XY bivariate feature extraction method obtains a large number of nonlinear feature variables, while the neural network model, through deep training of multiple neurons, can better fit the nonlinear feature variables. Therefore, combining the XY bivariate feature extraction method with the GA-BP-Adaboost nonlinear multiple regression model can significantly improve the performance of the traditional X-variate feature model and can be used for accurate prediction of straw charcoal CV.

[0123] In summary, the biomass char calorific value prediction method based on a bivariate feature extraction strategy provided by this invention first analyzes the correlation between the CV of straw char and the content of various elements, selecting Y-type feature variables that are highly significantly correlated with CV (P<0.01). This method primarily obtains the broadened analytical line bands of C, O, H, and Na elements existing in the forms of elemental carbon, aromatic rings, and carboxyl groups. Simultaneously, X-type feature variables related to CV are obtained by screening the regression coefficient threshold of the partial least squares regression (PLSR) model. When the threshold is 4×10... -5 The root mean square error (RMSECV) of the cross-validation model was reduced to its lowest value, and the corresponding variables were mainly the analytical line spectra of Ca, Cr, Mg, and K elements involved in crop physiological growth. Based on the extracted XY dual-feature variables, a genetic algorithm-optimized and adaptively enhanced artificial neural network (GA-BP-Adaboost) model was constructed. When the mutation probability, crossover probability, and relative error rate (RE) were set to 0.1, 0.95, and 0.01, respectively, the optimal model's mean relative error (AREP) and relative standard error (RSDP) were 2.39% and 2.97%, respectively, which were 0.82% and 0.91% lower than the XY-PLSR model. The results indicate that the XY bivariate feature extraction method combined with the GA-BP-Adaboost model can provide a methodological basis for accurate quantitative prediction and analysis of biochar CV during industrial use.

[0124] This invention accurately obtains the CV-sensitive feature element variables of straw char based on the XY bivariate feature extraction strategy, and constructs a nonlinear LIBS quantitative analysis model of straw char CV using a genetic algorithm optimization and adaptive enhancement artificial neural network algorithm (GA-BP-Adaboost). This study provides a reliable analysis strategy for straw char fuel quality evaluation and industrial field process analysis.

[0125] Based on the above-described preferred embodiments of the present invention, and through the foregoing description, those skilled in the art can make various changes and modifications without departing from the inventive concept. The technical scope of this invention is not limited to the contents of the specification, but must be determined according to the scope of the claims.

Claims

1. A method for predicting the calorific value of biochar based on a two-variable feature extraction strategy, characterized in that, The method comprises the following steps: S1: obtaining the calorific value of the biomass charcoal; S2: determining an X characteristic element linearly related to the calorific value according to the relationship between the calorific value and the spectral data of the elements in the biomass charcoal; S3: determining a Y characteristic element according to the concentration of the elements in the biomass charcoal through pearson correlation analysis; S4: obtaining the element characteristic spectral line of the X characteristic element as an X variable characteristic; S5: obtaining the element characteristic spectral line of the Y characteristic element as a Y variable characteristic; S6: constructing a quantitative analysis model according to the relationship between the X variable characteristic and the Y variable characteristic and the calorific value; S7: inputting the X variable characteristic and the Y variable characteristic in the biomass charcoal to be predicted into the quantitative analysis model to obtain a predicted calorific value; wherein the quantitative analysis model is in the form of CV=K(X+Y)+B, wherein CV is the calorific value of the biomass charcoal, X and Y are the X variable characteristic and the Y variable characteristic respectively, and K and B are digital matrices; The spectral data of the elements in the biomass charcoal are determined by laser-induced breakdown spectroscopy.

2. The method of predicting the heat value of biochar based on a bivariate feature extraction strategy as claimed in claim 1, wherein, Step S1 comprises: determining the calorific value of the biomass charcoal by an oxygen bomb calorimeter.

3. The method of predicting the heat value of biochar based on a bivariate feature extraction strategy as claimed in claim 1, wherein, Step S2 comprises: S21: obtaining the LIBS spectral data of all elements in the biomass charcoal; S22: constructing a linear PLSR model according to the calorific value and the LIBS spectral data of all elements; S23: obtaining the X characteristic element according to the cross-validation result of the linear PLSR model.

4. The method of predicting the heat value of biochar based on a bivariate feature extraction strategy as claimed in claim 3, wherein, The X characteristic element is Ca, Cr, Mg and K.

5. The method of predicting the heat value of biochar based on a bivariate feature extraction strategy as claimed in claim 1, wherein, When the pearson correlation analysis is performed in step S3, the element with a correlation coefficient showing extreme significance is the Y characteristic element.

6. The method of predicting the heat value of biochar based on a bivariate feature extraction strategy as claimed in claim 5, wherein, The Y characteristic element is C, O, H and Na.

7. The method of predicting the calorific value of biochar based on a bivariate feature extraction strategy according to any one of claims 1-6, characterized in that, Step S6 constructs the quantitative analysis model through an artificial neural network algorithm.

8. The method of predicting the heat value of biochar based on a bivariate feature extraction strategy as claimed in claim 7, wherein, The calculation formula of the quantitative analysis model constructed through the artificial neural network algorithm is as follows: Formula (1); Formula (2); Formula (3); where i is a weak predictor, j is a sample index, D is a sample weight, D i,j is the weight of the previous round of sample training, D i+1,j is the weight of the next round of sample training, alpha i is the weight assigned to each weak predictor, Error i is the sum of all sample weights for each weak predictor.

9. The method of predicting the heat value of biochar based on a bivariate feature extraction strategy as claimed in claim 8, wherein, The effect of the quantitative analysis model is evaluated by the root mean square error, the average relative error and the relative standard deviation.

10. The method of predicting the heat value of biochar based on a bivariate feature extraction strategy as claimed in claim 9, wherein, The calculation formula of the root mean square error is as follows: Equation (4); The calculation formula of the average relative error is as follows: Formula (5); The calculation formula of the relative standard deviation is as follows: Equation (6); wherein RMSEP is the prediction root mean square error, RMSE is the root mean square error, Ym is the actual value, Yp is the predicted value, n is the sample size, AREP is the average relative error, RSDP is the relative standard deviation, and mean(Yp) is the mean value of the predicted samples.

Images