Pyrolysis efficiency prediction method based on XGBoost model and simulated annealing algorithm

Abstract

The invention discloses a pyrolysis efficiency prediction method based on an XGBoost model and a simulated annealing algorithm, and the method comprises the steps: collecting sample data, the data comprises data which does not contain the yield of hexane insoluble substances under the same mixing proportion and data which contains the yield of hexane insoluble substances under the same mixing proportion, the difference value of the two is the difference value of the yields of the non-hexane-containing insoluble substance and the n-hexane-containing insoluble substance. According to the method, a polynomial linear regression model is established, a plurality of independent variables are quantified, the influence of the independent variables on dependent variables is studied, a regression equation is made, a result is predicted, and through the fitted regression equation and curve, which pyrolysis products have the most obvious interaction effect between the sample weight and the mixing ratio, and which pyrolysis products have the most obvious mixing ratio are specifically analyzed. And based on the fitted regression formula and the characteristics and composition of the co-pyrolysis product, establishing an optimization model by using a simulated annealing algorithm, and solving the co-pyrolysis mixing proportion so as to improve the product utilization rate and the energy conversion efficiency. And searching a global optimal solution in a large-scale search space through an optimization algorithm.

Description

Technical Field

[0001] The present invention relates to the technical field of pyrolysis process optimization, and in particular to a pyrolysis efficiency prediction method based on an XGBoost model and a simulated annealing algorithm. Background Art

[0002] With the advent of the big data era, machine learning has become particularly important in solving complex problems and mining data value. In particular, the gradient boosting decision tree, as a classic ensemble learning method, enhances the accuracy and generalization ability of prediction by combining multiple decision tree models. However, traditional GBDT often encounters performance bottlenecks when processing large-scale high-dimensional data, resulting in slow calculation speed and high resource consumption.

[0003] Globally, with the continuous growth of energy demand and the pursuit of renewable energy by various countries, pyrolysis technology has attracted widespread attention as a clean and efficient energy conversion method. The pyrolysis process heats biomass or coal under oxygen-deficient conditions to decompose it into gas, liquid and solid products. Due to the high complexity of the pyrolysis process and the diversity of raw materials, pyrolysis reactions with different raw materials and under different conditions require a lot of experiments to optimize their product distribution and efficiency.

[0004] In this context, it becomes crucial to study the yield and quality mechanism of co-pyrolysis of biomass and coal. As one of the core technologies for clean conversion and utilization of coal, the optimization of pyrolysis technology can significantly improve energy utilization efficiency, promote comprehensive resource utilization, and ensure energy security. At the same time, scientific researchers and engineers around the world are committed to developing more intelligent control systems to reduce resource waste and environmental impact during the pyrolysis process. Summary of the invention

[0005] The purpose of the present invention is to propose a pyrolysis efficiency prediction method based on the XGBoost model and simulated annealing algorithm. By introducing the XGBoost model and the simulated annealing algorithm, the distribution and efficiency of the products in the pyrolysis process are analyzed and predicted, and the simulated annealing algorithm is used to adjust and optimize various parameters in the pyrolysis process to improve energy utilization efficiency and product quality. Due to the high dimensionality of the experimental data, XGBoost parallel computing and block processing are used to improve the model calculation speed, maximize the energy conversion efficiency, and reduce the waste of experimental resources.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions:

[0007] A pyrolysis efficiency prediction method based on an XGBoost model and a simulated annealing algorithm, comprising:

[0008] S1, collecting sample data, the data including data of the yield of hexane-free insoluble matter at the same mixing ratio and data of the yield of hexane-containing insoluble matter at the same mixing ratio, wherein the difference between the two is the difference between the yield of n-hexane-free insoluble matter and the yield of n-hexane-containing insoluble matter;

[0009] S2, taking the difference between the yield of n-hexane-free insoluble matter and that of n-hexane-containing insoluble matter as the independent variable and the yield of pyrolysis products as the dependent variable, a polynomial linear regression model was established to calculate the coefficient of each product and determine which pyrolysis products have the most obvious effect on the yield of pyrolysis products;

[0010] S3, determining the change rates of tar, coke residue and water yields at different mixing ratios based on a polynomial linear regression model;

[0011] S4, establishing a simulated annealing algorithm, based on the polynomial fitting formula obtained from the polynomial linear regression model, bringing the data calculated by the polynomial fitting formula into the simulated annealing model to determine the optimal solution of each co-heating combination at different mixing ratios;

[0012] S5, establish a pyrolysis product yield prediction model based on the pyrolysis reaction data collected experimentally, capture key feature information from complex data sets, optimize the prediction performance, and after completing model training, use an independent test set to evaluate the model.

[0013] Preferably, the difference between the yield of n-hexane insoluble matter and the yield of n-hexane insoluble matter in step S1 is expressed as:

[0014] x=T 不含 -T 含

[0015] Among them, T 不含 is the yield without n-hexane insoluble matter at the same mixing ratio, T 含 is the yield with n-hexane insoluble matter at the same mixing ratio, and x is the difference between the yield without n-hexane insoluble matter and the yield with n-hexane insoluble matter.

[0016] Preferably, in step S2, the general formula of the polynomial linear regression model is:

[0017] y=β n x n +…β 2 x 2 +β 1 x+b

[0018] Among them, y is the yield of pyrolysis products, b is a constant term, and x n is the term of the polynomial, β 1 are the polynomial coefficients.

[0019] Preferably, in step S4, the simulated annealing algorithm is:

[0020] Let s = s 1 ,s 2 ,…S n For all possible solutions, find S * Make C(S * )=minC(S i )

[0021]

[0022] Among them, P is the probability that the algorithm selects a worse solution, and T is the simulation parameter of temperature;

[0023] Substitute into the simulated annealing formula:

[0024] ΔC=j(s k +1)-j(s k )

[0025] Among them, when T is very large, P tends to 1, and there is a greater probability of selecting a non-current optimal solution;

[0026] P decreases as T decreases;

[0027] When T approaches 0, P approaches 0 + , the algorithm almost exclusively selects the optimal solution.

[0028] Preferably, the pyrolysis product yield prediction model established in step S5 is an XGBoot regression model, and the formula is:

[0029]

[0030] Where F = {f(x) = w q(x)}(q:R m →{1,2,…T},w∈R T ) is a set of CART decision tree structures, q is the tree structure of samples mapped to leaf nodes, T is the number of leaf nodes, and w is the real number score of leaf nodes.

[0031] Compared with the prior art, the present invention has the following beneficial effects:

[0032] The present invention establishes a polynomial linear regression model, quantifies multiple independent variables, studies their influence on the dependent variable, makes a regression equation, predicts the results, and specifically analyzes which pyrolysis products have the most obvious interaction effect between sample weight and mixing ratio through the fitted regression equation and curve.

[0033] The invention uses a simulated annealing algorithm to establish an optimization model based on the fitted regression formula and the characteristics and composition of the co-pyrolysis products, and solves the co-pyrolysis mixing ratio to improve the product utilization rate and energy conversion efficiency. The global optimal solution is found in the large-scale search space of the optimization algorithm, so as to find the co-pyrolysis mixing ratio when the tar yield is the maximum.

[0034] In order to predict the yield of pyrolysis products based on experimental data, the present invention establishes an XGBoost regression model, adopts MSE as an evaluation criterion, thereby testing the prediction effect, and predicts the products to obtain prediction results. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 It is a polynomial fitting curve diagram of the present invention;

[0036] Figure 2 It is a schematic diagram of the simulated annealing algorithm flow of the present invention;

[0037] Figure 3 It is a diagram of the CS / HN simulated annealing solution process of the present invention;

[0038] Figure 4 It is a fitting curve diagram for predicting tar yield of the decision tree regression model of the present invention;

[0039] Figure 5 It is a fitting curve diagram for predicting tar yield of the random forest regression model of the present invention;

[0040] Figure 6 This is the XGBoot regression model tar yield prediction fitting curve diagram of the present invention

[0041] Figure 7 This is a diagram of the prediction results of the XGBoot model of the present invention. DETAILED DESCRIPTION

[0042] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0043] A pyrolysis efficiency prediction method based on an XGBoost model and a simulated annealing algorithm, comprising:

[0044] S1, collecting sample data, the data including data of the yield of hexane-free insoluble matter at the same mixing ratio and data of the yield of hexane-containing insoluble matter at the same mixing ratio, wherein the difference between the two is the difference between the yield of n-hexane-free insoluble matter and the yield of n-hexane-containing insoluble matter;

[0045] S2, taking the difference between the yield of n-hexane-free insoluble matter and that of n-hexane-containing insoluble matter as the independent variable and the yield of pyrolysis products as the dependent variable, a polynomial linear regression model was established to calculate the coefficient of each product and determine which pyrolysis products have the most obvious effect on the yield of pyrolysis products;

[0046] S3, determining the change rates of tar, coke residue and water yields at different mixing ratios based on a polynomial linear regression model;

[0047] S4, establishing a simulated annealing algorithm, based on the polynomial fitting formula obtained from the polynomial linear regression model, bringing the data calculated by the polynomial fitting formula into the simulated annealing model to determine the optimal solution of each co-heating combination at different mixing ratios;

[0048] S5, establish a pyrolysis product yield prediction model based on the pyrolysis reaction data collected experimentally, capture key feature information from complex data sets, optimize the prediction performance, and after completing model training, use an independent test set to evaluate the model.

[0049] In order to explore whether there is an interactive effect between n-hexane insoluble matter (INS) and the mixing ratio, which has an important influence on the yield of pyrolysis products, the yield of each product was obtained by product / sample, and then it was found that there were outliers in the data, so the outliers were corrected;

[0050] The difference expression between the yield of n-hexane insoluble matter and the yield of n-hexane insoluble matter in step S1 is:

[0051] x=T 不含 -T 含

[0052] Among them, T 不含 is the yield without n-hexane insoluble matter at the same mixing ratio, T 含 is the yield with n-hexane insoluble matter at the same mixing ratio, and x is the difference between the yield without n-hexane insoluble matter and the yield with n-hexane insoluble matter;

[0053] After the processed data is obtained in step S2, the polynomial linear regression method is used for analysis to further explore the interaction effect between n-hexane insoluble matter (INS) and the mixing ratio, and the polynomial interpolation is used to evaluate which pyrolysis product yield is most significantly affected by this interaction effect. A polynomial linear regression model is established for analysis, with the difference in yield between containing and not containing n-hexane insoluble matter as the independent variable, and the yield of the pyrolysis product as the dependent variable, through the formula:

[0054] y=β n x n +…β 2 x 2 +β 1 x+b

[0055] Among them, y is the yield of pyrolysis products, b is a constant term, and x n is the term of the polynomial, β 1 are the polynomial coefficients;

[0056] The processed data is brought into the polynomial linear regression model, and the fitting formula is solved as follows:

[0057] CN / HN:y=-19.71x 4 +22.88x 3 -9.149x 2 +1.484x-0.08393

[0058] CS / HS:y=7.314x 4 -7.569x 3 +2.548x 2 -0.3177x+0.0109

[0059] CS / SM:y=-37.38x 4 +40.47x 3 -14.5x 2 +1.962x-0.07746

[0060] GA / HN:y=-13.06x 4 +14.65x 3 -5.634x 2 +0.8845x-0.04879

[0061] GA / HN:y=25.02x 4 -25.71x 3 +8.578x 2 -1.056x+0.042

[0062] GA / SM:y=-172.8x 4 +177.4x 3 -58.01x 2 +6.752x-0.2554

[0063] RH / HN:y=-10.79x 4 +9.88x 3 -2.891x 2 +0.3511x-0.01928

[0064] RH / SM:y=-46.27x 4 +47.96x 3 -16.23x 2 +2.058x-0.08101

[0065] SD / HS:y=-16.04x 4 +18.49x 3 -7.418x 2 +1.218x-0.06951

[0066] SD / SM:y=37.77x 4 -40.56x 3 +14.42x 2 -1.936x+0.07489

[0067] The transformation range of the fitting curve is calculated, and the results are shown in Table 1:

[0068] Table 1 Transformation range of fitting curve

[0069]

[0070] In the pyrolysis experiment, there is an interaction effect between n-hexane insoluble matter (INS) and the mixing ratio, which has an important influence on the yield of pyrolysis products. If there is an interaction effect, the change rate of tar, coke residue and water yield under different ratios is calculated, and the results are shown in Table 2, Table 3 and Table 4:

[0071] Table 2 Variation of tar under different ratios

[0072]

[0073]

[0074] Table 3 Variation of coke slag under different proportions

[0075]

[0076] Table 4 Variation of water content at different ratios

[0077]

[0078] Through the detailed analysis of Table 3, we can clearly see the significant impact of different ratios on yield. These tables not only reveal the complex relationship between ratio and yield, but also point out the direction of optimization.

[0079] First, it was observed that the yield of tar had a more significant interaction effect than the yield of slag and water. This means that the yield of tar fluctuated greatly when the ratio changed. Specifically, when the RH / HN ratio reached 0.4, the yield of tar showed a significant peak, indicating that the production efficiency of tar was the highest under this ratio.

[0080] Similarly, when the GA / SM ratio is set to 0.4, the slag yield also shows a significant interaction effect. This shows that by adjusting this ratio, the amount of slag generated can be effectively controlled, thereby optimizing the quality and performance of the product. Under other ratios, the slag yield changes relatively smoothly, indicating that these ratios have little effect on the slag yield.

[0081] However, unlike tar and coke residue, water yield is almost insensitive to changes in the mix ratio, which means that no matter how the mix ratio is adjusted, the water yield will remain at a relatively stable level.

[0082] In order to solve the problem of optimizing the co-thermal mixing ratio to improve product utilization and energy conversion efficiency, it can be seen from the appendix of the question that the yield of tar needs to be increased as much as possible, and the yield of tar can be increased by optimizing the co-thermal mixing ratio. The simulated annealing algorithm is a heuristic optimization algorithm used to find the global optimal solution in a large-scale search space. It is inspired by the solid annealing process, in which the material is heated and then cooled to change its structure and reach a certain ideal state;

[0083] Let s = s 1 ,s 2 ,…S n For all possible solutions, find S * Make C(S * )=minC(S i )

[0084]

[0085] Among them, P is the probability that the algorithm selects a worse solution, and T is the simulation parameter of temperature;

[0086] The formula for simulated annealing is:

[0087] ΔC=j(s k +1)-j(s k )

[0088] When T is large, P tends to 1, and there is a greater probability of selecting a non-optimal solution;

[0089] P decreases as T decreases;

[0090] When T approaches 0, P approaches 0 + , the algorithm almost only selects the optimal solution, which is equivalent to the greedy algorithm.

[0091] After in-depth data analysis and calculation, the preprocessed data was brought into the simulated annealing model based on the obtained polynomial fitting formula. The simulated annealing algorithm, with its efficient optimization performance and flexibility, demonstrated its unique advantages in this scenario. Through the iterative optimization of the simulated annealing algorithm, the optimal solution of each co-heating combination at different mixing ratios was calculated. The results are shown in Table 5. In this process, the algorithm continuously searches in the solution space, and gradually approaches the global optimal solution by simulating the temperature drop and state update in the physical annealing process.

[0092] Table 5 Comparison table of proportions at maximum tar yield

[0093]

[0094] Decision Tree Regression Model:

[0095] Calculate the two parts D of each feature value division 1 and D 2 The sum of squared errors is selected, and the one with the smallest sum of squared errors is selected as the optimal feature and the optimal split point. The formula is as follows:

[0096]

[0097] Among them, c 1 D 1 The sample output mean, c 2 D 2 The sample output mean of .

[0098] Random Forest Regression Model:

[0099] The random forest algorithm, with its core concept of ensemble learning, cleverly integrates multiple decision trees into one. These decision trees are independent and grow independently in the data forest. Each tree has the ability to make detailed judgments and accurate classifications of samples. After processing the data, each decision tree will output a classification result. Finally, the random forest gathers the "wisdom" of these independent decision trees and uses majority voting to determine the most reliable and accurate classification result. The mathematical logic behind it can be expressed as the following formula:

[0100]

[0101] Among them, h i is an individual learner, and T is the depth of the current decision tree subtree.

[0102] XGBoot regression model:

[0103] XGBoost is an algorithm based on gradient boosting tree, so we must first understand the principle of gradient boosting tree. Gradient boosting tree is an ensemble learning method that iteratively trains a series of weak learners (usually decision trees) step by step. Each iteration attempts to correct the error of the previous iteration, and finally combines these weak learners into a strong learner. The formula is:

[0104]

[0105] Where F = {f(x) = w q(x)}(q:R m →{1,2,…T},w∈R T ) is a set of CART decision tree structures, q is the tree structure of samples mapped to leaf nodes, T is the number of leaf nodes, w is the real number score of leaf nodes. When constructing the XGBoost model, it is necessary to find the optimal parameters according to the principle of minimizing the objective function to establish the optimal model. The objective function of the XGBoost model can be divided into the error function term L and the model complexity function term Ω.

[0106] Based on the pyrolysis reaction data collected in the experiment, in order to achieve efficient and accurate prediction of the pyrolysis product yield, three different machine learning regression models were constructed: decision tree regression, random forest regression, and XGBoost regression model. The selection of these models aims to utilize their respective unique algorithmic characteristics and advantages, so that they can capture key feature information from complex data sets to optimize prediction performance. After the model training is completed, the model is evaluated using an independent test set. In order to quantify the predictive performance of the model, the mean square error (MSE) was selected as the evaluation indicator. The MSE formula is as follows:

[0107]

[0108] Where n is the number of samples; y i is the true value; is the predicted value.

[0109] MSE can intuitively reflect the difference between the model prediction value and the actual value. The smaller the value, the better the model prediction effect. Using MSE as the evaluation index, the MSE test set results are calculated as shown in Table 6:

[0110] Table 6 Comparison of MSE of regression models

[0111]

[0112] By comparing the MSE values ​​and fitting curves of the three models, it was found that the XGBoost regression model performed well in predicting the yield of pyrolysis products. Specifically, the MSE value of the XGBoost regression model was 0.000539368, which was significantly lower than the other two models, indicating that the model was better at capturing complex relationships and patterns in the data.

[0113] Based on the above analysis, the XGBoost regression model was finally selected as the preferred tool for predicting the yield of pyrolysis products. This model not only has a low prediction error, but also can accurately reflect the inherent laws of experimental data, providing strong support for the optimization of the pyrolysis reaction process and the improvement of product yield, and predicting the pyrolysis yield.

[0114] XGBoost greatly improves the training speed and efficiency of the model by using more efficient regularization mechanisms, block-based processing of large data sets, and smarter parallel computing strategies. This improvement has made XGBoost quickly become a mainstream tool in the field of machine learning, and is widely used in many fields such as financial forecasting, recommendation systems, and genetic data analysis. Its efficiency and scalability make it particularly good at dealing with complex problems and large-scale data;

[0115] As a random optimization algorithm based on the physical annealing process, the simulated annealing algorithm searches for the global optimal solution in the solution space through probabilistic search. This algorithm imitates the process of a substance gradually cooling down and solidifying into a crystal after melting at high temperature. It gradually reduces randomness by controlling the "temperature" to ensure that the searched solution is close to the global optimal solution.

[0116] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A pyrolysis efficiency prediction method based on XGBoost model and simulated annealing algorithm, characterized in that: include: S1, collecting sample data, the data including data of the yield of hexane-free insoluble matter at the same mixing ratio and data of the yield of hexane-containing insoluble matter at the same mixing ratio, wherein the difference between the two is the difference between the yield of n-hexane-free insoluble matter and the yield of n-hexane-containing insoluble matter; S2, taking the difference between the yield of n-hexane-free insoluble matter and that of n-hexane-containing insoluble matter as the independent variable and the yield of pyrolysis products as the dependent variable, a polynomial linear regression model was established to calculate the coefficient of each product and determine which pyrolysis products have the most obvious effect on the yield of pyrolysis products; S3, determining the change rates of tar, coke residue and water yields at different mixing ratios based on a polynomial linear regression model; S4, establishing a simulated annealing algorithm, based on the polynomial fitting formula obtained from the polynomial linear regression model, bringing the data calculated by the polynomial fitting formula into the simulated annealing model to determine the optimal solution of each co-heating combination at different mixing ratios; S5, establish a pyrolysis product yield prediction model based on the pyrolysis reaction data collected experimentally, capture key feature information from complex data sets, optimize the prediction performance, and after completing model training, use an independent test set to evaluate the model.

2. The method for predicting pyrolysis efficiency based on the XGBoost model and the simulated annealing algorithm according to claim 1, characterized in that: The difference expression between the yield of n-hexane insoluble matter and the yield of n-hexane insoluble matter in step S1 is: x=T 不含 -T 含 Among them, T 不含 is the yield without n-hexane insoluble matter at the same mixing ratio, T 含 is the yield with n-hexane insoluble matter at the same mixing ratio, and x is the difference between the yield without n-hexane insoluble matter and the yield with n-hexane insoluble matter.

3. The method for predicting pyrolysis efficiency based on XGBoost model and simulated annealing algorithm according to claim 1, characterized in that: The general formula of the polynomial linear regression model in step S2 is: y=β n x n +…β2x 2 +β1x+b Among them, y is the yield of pyrolysis products, b is a constant term, and x n is the term of the polynomial, and β1 is the polynomial coefficient.

4. The method for predicting pyrolysis efficiency based on XGBoost model and simulated annealing algorithm according to claim 1, characterized in that: In step S4, the simulated annealing algorithm is: Let s=s1,s2,…S n For all possible solutions, find S * Make C(S * )=minC(S i ) Among them, P is the probability that the algorithm selects a worse solution, and T is the simulation parameter of temperature; Substitute into the simulated annealing formula: ΔC=j(s k +1)-j(s k ) Among them, when T is very large, P tends to 1, and there is a greater probability of selecting a non-current optimal solution; P decreases as T decreases; When T approaches 0, P approaches 0 + , the algorithm almost exclusively selects the optimal solution.

5. The method for predicting pyrolysis efficiency based on XGBoost model and simulated annealing algorithm according to claim 1, characterized in that: In step S5, the pyrolysis product yield prediction model is established as the XGBoot regression model, and the formula is: Where F = {f(x) = w q(x) }(q:R m →{1,2,…T},w∈R T ) is a set of CART decision tree structures, q is the tree structure of samples mapped to leaf nodes, T is the number of leaf nodes, and w is the real number score of leaf nodes.

Images