Method for predicting sulfur dioxide concentration of gas-fired boiler after blending combustion of refined desulfurization regenerated gas

Abstract

The invention discloses a method for predicting the sulfur dioxide concentration of a gas-fired boiler after blending combustion of refined desulfurization regenerated gas. The method comprises the following steps: collecting historical operation data of the gas-fired boiler; preprocessing the historical operation data to obtain an original data set; screening at least ten characteristic variables related to the sulfur dioxide concentration as a model data set; performing dimensionless processing on the model data set, and controlling the order of magnitude of the model data set to be between [0, 1] to obtain an input data set; and constructing a sulfur dioxide concentration prediction model based on a lightweight gradient elevator algorithm, training the lightweight gradient elevator algorithm by using the input data set to obtain a hyper-parameter combination, and inputting the hyper-parameter combination into the lightweight gradient elevator algorithm to obtain a prediction model. The method can predict the sulfur dioxide concentration at the inlet of the desulfurization device of the gas-fired boiler, so that the problem of detection time delay of a traditional measurement method is solved.

Description

Technical Field

[0001] The invention relates to the technical field of flue gas sulfur dioxide concentration prediction, in particular to a method for predicting the sulfur dioxide concentration of a gas-fired boiler after desulfurization regeneration gas is mixed with refined fuel. Background Art

[0002] Coke oven gas, a key byproduct of the coke reduction process, is widely used in industrial and city gas, steelmaking, and the production of gases such as hydrogen and methanol. However, coke oven gas contains various impurities, such as hydrogen sulfide and carbon-based sulfur. Direct use of coke oven gas can cause pipeline corrosion and reduce the quality of steel billets during the steelmaking process. Furthermore, during combustion, hydrogen sulfide and carbon-based sulfur are converted into sulfur dioxide, which can cause air pollution and adversely affect human health. Therefore, to effectively utilize coke oven gas, it is necessary to perform fine desulfurization pretreatment to remove harmful substances such as hydrogen sulfide and carbon-based sulfur.

[0003] Currently, most fine desulfurization equipment is a dry desulfurization composite tower, which is filled with desulfurizers such as iron hydroxide, coke, and activated carbon, and can effectively purify coke oven gas. However, the desulfurizer needs to be regenerated and replaced after a period of use. The conventional method is to manually remove it and then turn it over and regenerate it. This not only increases labor costs and reduces production efficiency, but the escaped gas also causes environmental pollution. To solve this problem, the coke oven gas after hydrogen precipitation can be used as regeneration gas and passed into the desulfurization composite tower. By periodically purging the desulfurizer, harmful impurities are removed, thereby achieving the purpose of continuous production. However, in this process, impurities such as hydrogen sulfide and carbon-based sulfur will be transferred from the desulfurizer to the regeneration gas. Therefore, how to absorb this part of the fine desulfurization regeneration gas becomes a key issue.

[0004] In practical applications, the refined desulfurized regenerated gas can be mixed with blast furnace gas and then transported to a gas-fired boiler to generate electricity. This can effectively utilize the refined desulfurized regenerated gas and save gas consumption. However, the blending process of refined desulfurized regenerated gas containing multiple impurities will lead to environmental problems such as excessive sulfur dioxide emission concentrations. At present, most of the existing technologies directly detect the sulfur dioxide content in the exhaust gas after combustion, and adjust the relevant conditions during the combustion process according to the sulfur dioxide content in the exhaust gas. This scheme has a certain hysteresis when measuring the sulfur dioxide concentration, which leads to the situation that the sulfur dioxide concentration at the outlet of the desulfurization device exceeds the standard cannot be completely avoided. Therefore, there is an urgent need for a prediction method to achieve accurate and rapid prediction of the sulfur dioxide concentration at the inlet of the gas boiler desulfurization system, so as to provide a reference basis for the control measures of the gas boiler after blending with refined desulfurized regenerated gas. Summary of the Invention

[0005] In order to overcome the defects in the prior art, the present invention provides a method for predicting the sulfur dioxide concentration of a gas-fired boiler after desulfurization and regeneration gas is mixed with refined gas. This method can predict the sulfur dioxide concentration at the inlet of the gas-fired boiler, thereby solving the problem of detection time lag in traditional measurement methods and effectively avoiding excessive sulfur dioxide concentration in the exhaust gas.

[0006] To achieve the above object, the technical solution adopted by the present invention is: a method for predicting the sulfur dioxide concentration of a gas-fired boiler after desulfurization regeneration gas is mixed with refined gas, which is used to predict the sulfur dioxide concentration at the inlet of the gas-fired boiler, comprising the following steps:

[0007] Collect data, collect historical operating data of gas boilers;

[0008] Obtaining an original data set, preprocessing the historical operation data, eliminating the influence of outliers in the characteristic variables, and obtaining an original data set;

[0009] Obtaining a model data set, calculating the strength of the correlation between characteristic variables in the original data set and sulfur dioxide concentration, and selecting at least ten characteristic variables related to sulfur dioxide concentration as the model data set;

[0010] Acquire an input data set, perform dimensionless processing on the model data set, control the magnitude of the model data set between [0, 1], and eliminate the influence of the magnitude and dimension differences of characteristic variables on the measurement accuracy of the model, thereby obtaining an input data set;

[0011] A training model is provided, wherein a sulfur dioxide concentration prediction model is constructed based on a lightweight gradient boosting machine algorithm, and the lightweight gradient boosting machine algorithm is trained using the input data set to obtain a hyperparameter combination, and the hyperparameter combination is input into the lightweight gradient boosting machine algorithm to obtain a prediction model.

[0012] Among them, the historical operating data include: the temperature and pressure of the fine desulfurization regeneration gas, the temperature and pressure of the low-temperature coal-added inlet gas, the temperature and pressure of the high-temperature coal-added outlet gas, the blast furnace gas flow, temperature and pressure, furnace pressure, oxygen content, total air volume, main steam flow, active power and suction fan inlet air pressure.

[0013] The above technical solution trains a model for predicting sulfur dioxide concentration. Real-time operating data from the gas boiler inlet is then integrated into the model. The model then predicts the sulfur dioxide content of the exhaust gas at the boiler outlet. If the predicted value is below a threshold, normal combustion and emissions are maintained. If the predicted value is above the threshold, an alarm is triggered, prompting operators to adjust the operating status of the gas boiler until the predicted value falls below or equals the threshold. This testing method allows for predicting sulfur dioxide levels before emissions occur, thus preventing the emission of gases with excessive sulfur dioxide content.

[0014] Furthermore, preprocessing the historical operation data includes:

[0015] The density-based clustering algorithm determines the specific values ​​of the neighborhood radius ε and the density threshold Minpts according to the clustering silhouette coefficient s, finds the core object and its dense neighborhood objects, connects the core object and the neighborhood to form a dense area as a cluster, and screens outliers in the historical operation data and removes them. The clustering silhouette coefficient s is calculated as follows:

[0016]

[0017] Among them, s i For a single sample d i The silhouette coefficient, a i Representative sample d i The average distance to other samples in the same cluster, b i For a single sample d i The average distance between the cluster and all samples in its nearest cluster, and n is the number of samples in the historical operation data. The above method can eliminate zero values, error values, and missing values ​​caused by the instability of the boiler data acquisition system, thereby preventing outliers from affecting the training results.

[0018] Furthermore, the mutual information method is used to analyze the correlation between the characteristic variables in the original data set and the sulfur dioxide concentration, and the method is as follows:

[0019] MI(X;Y)=H(X)+H(Y)-H(X,Y);

[0020]

[0021] Where MI(X;Y) is the mutual information between feature variables X and Y; H(X) and H(Y) represent the information entropy of variables X and Y, respectively; H(X,Y) is the joint information entropy of variables X and Y; p(x) and p(y) represent the probability distributions of variables X and Y, respectively; and p(x,y) is the joint probability distribution of variables X and Y.

[0022] Ten characteristic variables are selected from high to low based on the correlation between each characteristic variable and the sulfur dioxide concentration to form the model data set for training the model. In this application, the ten characteristic variables obtained are total air volume, main steam flow, fine desulfurization regeneration gas flow, fine desulfurization regeneration gas temperature, oxygen content, suction fan inlet air pressure, furnace pressure, active power, blast furnace gas flow, and low-temperature coal inlet gas pressure.

[0023] Furthermore, the steps to construct a sulfur dioxide concentration prediction model based on the lightweight gradient boosting algorithm include:

[0024] Set the hyperparameters in the lightweight gradient boosting machine algorithm to the range of values: num_leaves to [32, 2048], max_depth to [10, 30], min_child_samples to [1, 15], and learning_rate to [0.01, 0.2];

[0025] Based on the value ranges of num_leaves, max_depth, min_child_samples, and learning_rate, the minimum value of the corresponding hyperparameter is input into the lightweight gradient boosting machine algorithm, and the lightweight gradient boosting machine algorithm is trained using the input data set to calculate the mean square error and R 2 value;

[0026] Perform parameter iteration within the range of num_leaves, max_depth, min_child_samples, and learning_rate, and train the updated lightweight gradient boosting machine algorithm using the input dataset, while calculating the corresponding mean square error and R 2 value;

[0027] The hyperparameters within the value range are searched and multiple hyperparameter combinations and corresponding mean square errors and R are obtained. 2 value, the mean square error tends to be minimum and R 2 The hyperparameter combination with the largest value is taken as the optimal hyperparameter combination. Among them, the lightweight gradient boosting machine algorithm uses the histogram method and the leaf-wise method to process data, which can not only improve the prediction accuracy, but also improve the prediction speed and reduce memory utilization. The lightweight gradient boosting machine algorithm scans multiple line graphs to find the best leaf nodes for segmentation. Compared with the traditional gradient boosting tree model's strategy of segmenting all leaf nodes layer by layer, the lightweight gradient boosting machine algorithm can save the number of nodes and avoid the unnecessary consumption of memory and computation by the layered strategy.

[0028] Furthermore, the step training model also includes:

[0029] The trained prediction model is connected to the gas boiler. The prediction model predicts the sulfur dioxide concentration based on real-time data to obtain a predicted value. Combined with the actual value of the sulfur dioxide concentration measured, the mean square error between the actual value and the predicted value and R 2 Values ​​used to evaluate the performance of the prediction model.

[0030] Furthermore, the hyperparameter combination of the prediction model is changed, and the real-time data of the gas boiler inlet is connected to the changed prediction model. The changed prediction model predicts the sulfur dioxide concentration based on the real-time data to obtain a new prediction value. Combined with the actual value obtained by measuring the sulfur dioxide concentration, the mean square error and R between the real value and the new prediction value are calculated. 2 Compare the performance of prediction models formed by different hyperparameter combinations to select prediction models with better performance. The smaller the mean square error between the true value and the predicted value, the better the R 2 The larger the value, the better the performance of the prediction model.

[0031] Furthermore, the prediction model includes an alarm module for warning when sulfur dioxide concentration exceeds a preset threshold. If the predicted value exceeds the threshold, the alarm module issues an alarm prompting the operator to adjust the operating status of the gas boiler. The alarm module stops sounding the alarm until the predicted value falls below or equals the threshold.

[0032] By means of the above technical solution, the beneficial effects of the present invention are as follows:

[0033] 1. In this application, the correlation between each characteristic variable and the sulfur dioxide concentration is analyzed by the mutual information method, and a sulfur dioxide concentration prediction model is trained based on the characteristic variables with high correlation with the sulfur dioxide concentration. The prediction model can predict the sulfur dioxide content of the exhaust gas based on the real-time operating data at the inlet of the gas boiler, avoiding the problem of time lag in detection of traditional measurement methods, and can realize timely and effective regulation of the operating status of the gas boiler.

[0034] In order to make the above and other objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0036] Figure 1 This is a flow chart of a method for predicting sulfur dioxide concentration in a gas-fired boiler after blending desulfurized regenerated gas with refined fuel in an embodiment of the present invention;

[0037] Figure 2 This is a historical operating data view of the temperature and pressure of the refined desulfurization regeneration gas in an embodiment of the present invention;

[0038] Figure 3 It is a schematic diagram of the lightweight gradient boosting machine algorithm. DETAILED DESCRIPTION

[0039] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0040] It should be noted that, in the description of the present invention, the terms "first," "second," etc., are used solely for descriptive purposes and to distinguish similar objects. There is no order of precedence between the two, nor should they be construed as indicating or implying relative importance. Furthermore, in the description of the present invention, unless otherwise specified, "plurality" means two or more.

[0041] Example: Combination Figure 1 As shown, this embodiment discloses a method for predicting the sulfur dioxide concentration of a gas-fired boiler after mixing desulfurized regenerated gas with refined gas, which is used to predict the sulfur dioxide concentration at the inlet of the gas-fired boiler, comprising the following steps:

[0042] Collect data, collect historical operating data of gas boilers, including: temperature and pressure of refined desulfurization regeneration gas, temperature and pressure of low-temperature coal-added inlet gas, temperature and pressure of high-temperature coal-added outlet gas, blast furnace gas flow, temperature and pressure, furnace pressure, oxygen content, total air volume, main steam flow, active power and suction fan inlet air pressure. Figure 2 The historical operating data of the temperature and pressure of the refined desulfurization regeneration gas are displayed.

[0043] The historical operation data is processed by a density-based clustering algorithm, and the specific values ​​of the neighborhood radius ε and the density threshold Minpts are determined according to the clustering silhouette coefficient s. Outliers in the historical operation data are screened out and removed. The screening steps are as follows:

[0044] Determine the core object and its dense neighborhood objects, connect the core object and the dense neighborhood objects to determine the neighborhood radius ε; give a density threshold Minpts to determine the size of the dense area; combine the neighborhood radius ε with the density threshold Minpts to calculate the clustering silhouette coefficient s of the gas boiler historical operation data. The calculation method of the clustering silhouette coefficient s is as follows:

[0045]

[0046] Where s i For a single sample d i The silhouette coefficient, a i Representative sample d i The average distance to other samples in the same cluster, b iFor a single sample d i The average distance between all samples in its nearest cluster, n, and the number of samples in the historical operating data, is calculated. This method removes zero values, erroneous values, and missing values ​​caused by instability in the boiler data acquisition system, thus preventing outliers from affecting the training results. After removing outliers, the remaining historical operating data forms the original dataset.

[0047] The Cluster Silhouette Coefficient (CSC) is a method for evaluating clustering effectiveness. It combines cohesion and separation. It can be used to evaluate the impact of different algorithms, or different algorithm operation modes, on clustering results based on the same raw data.

[0048] The neighborhood radius refers to the range of influence in a graph centered on a node. In this application, it is represented by ε, which represents the set of all nodes that are ε steps away from the central node. The neighborhood radius is used to measure the distance and correlation between nodes. The larger the neighborhood radius ε, the larger the influence range of the node and the greater the degree. In addition, the larger the neighborhood radius, the smaller the clustering coefficient and the weaker the correlation between nodes.

[0049] The density threshold is a parameter used to control the compactness of clustering. In density clustering, a higher density threshold results in more small clusters, while a lower density threshold results in fewer but larger clusters. In maximum density point clustering, when the density value of an object is greater than the density threshold, the object is considered to belong to a relatively fixed cluster; otherwise, the object is temporarily considered to be noise.

[0050] The mutual information method is used to analyze the correlation between the characteristic variables in the original data set and the sulfur dioxide concentration. The method is as follows:

[0051] MI(X;Y)=H(X)+H(Y)-H(X,Y);

[0052]

[0053] Where MI(X;Y) is the mutual information between feature variables X and Y; H(X) and H(Y) represent the information entropy of variables X and Y, respectively; H(X,Y) is the joint information entropy of variables X and Y; p(x) and p(y) represent the probability distributions of variables X and Y, respectively; and p(x,y) is the joint probability distribution of variables X and Y.

[0054] According to the correlation between each characteristic variable and the sulfur dioxide concentration, ten characteristic variables are selected from high to low to form a model data set for training the model. The steps include: calculating the probability distribution of each characteristic variable, then calculating the joint probability distribution between each characteristic variable and the sulfur dioxide concentration, and then deriving the mutual information between each characteristic variable and the sulfur dioxide concentration. The mutual information values ​​between each characteristic variable and the sulfur dioxide concentration are arranged from large to small, and according to the mutual information values ​​between each characteristic variable and the sulfur dioxide concentration, ten characteristic variables are selected from high to low to form a model data set for training the model. The ten characteristic variables obtained are total air volume, main steam flow, fine desulfurization regeneration gas flow, fine desulfurization regeneration gas temperature, oxygen content, suction fan inlet air pressure, furnace pressure, active power, blast furnace gas flow, and low-temperature coal plus inlet gas pressure.

[0055] Because there are many feature variables and their magnitudes vary widely, using these data directly as input parameters can affect prediction accuracy. Therefore, we need to use normalization to make the feature variables dimensionless, controlling the magnitude of the dataset to between [0, 1] to eliminate the impact of magnitude and dimension differences on model measurement accuracy. Finally, we organize the feature variables in the processed model dataset into a format that the Python program can understand, forming the input dataset.

[0056] A sulfur dioxide concentration prediction model is constructed based on the lightweight gradient boosting algorithm and trained using the input dataset. The steps are as follows:

[0057] Set the hyperparameters in the lightweight gradient boosting machine algorithm to the range of values: num_leaves to [32, 2048], max_depth to [10, 30], min_child_samples to [1, 15], and learning_rate to [0.01, 0.2];

[0058] Based on the value ranges of num_leaves, max_depth, min_child_samples, and learning_rate, the minimum value of the corresponding hyperparameter is input into the lightweight gradient boosting machine algorithm, and the lightweight gradient boosting machine algorithm is trained using the input data set to calculate the mean square error and R 2 value;

[0059] Perform parameter iteration within the range of num_leaves, max_depth, min_child_samples, and learning_rate, and train the updated lightweight gradient boosting machine algorithm using the input dataset, while calculating the corresponding mean square error and R 2 value;

[0060] Repeat the above parameter iteration steps to search for hyperparameters within the value range and obtain multiple hyperparameter combinations and corresponding mean square errors and R 2 value, the mean square error tends to be minimum and R 2 The hyperparameter combination with the largest value is regarded as the best hyperparameter combination.

[0061] The lightweight gradient boosting machine algorithm uses the histogram method and the leaf-wise method to process data, which can not only improve the prediction accuracy, but also increase the prediction speed and reduce memory utilization. Figure 3 As shown in the figure, the lightweight gradient boosting machine algorithm scans multiple line graphs to find the best leaf nodes for segmentation. Compared with the traditional gradient boosting tree model's strategy of segmenting all leaf nodes layer by layer, the lightweight gradient boosting machine algorithm can save the number of nodes and avoid the unnecessary consumption of memory and computation by the hierarchical strategy.

[0062] The trained prediction model is connected to the gas boiler. The prediction model predicts the sulfur dioxide concentration based on real-time data to obtain a predicted value. Combined with the actual value of the sulfur dioxide concentration measured, the mean square error between the actual value and the predicted value and R 2 Values ​​used to evaluate the performance of the prediction model.

[0063] Change the hyperparameter combination of the prediction model, connect the real-time data of the gas boiler inlet to the changed prediction model, and use the changed prediction model to predict the sulfur dioxide concentration based on the real-time data to obtain a new predicted value. Combined with the actual value obtained by measuring the sulfur dioxide concentration, calculate the mean square error and R between the true value and the new predicted value. 2 Compare the performance of prediction models formed by different hyperparameter combinations to select prediction models with better performance.

[0064] The mean square error is a measure of the difference between the estimated value and the estimated value. The smaller the mean square error between the true value and the predicted value, the better the performance of the prediction model. 2 The R value is an indicator used to measure the goodness of fit of the model. It represents the percentage of squared correlation between the model's predicted values ​​and the actual values. 2 The value range is from 0 to 1, and the larger the value, the better the model fit.

[0065] The prediction model also includes an alarm module for warning when sulfur dioxide concentration exceeds a preset threshold. If the predicted value exceeds the threshold, the alarm module issues an alarm, prompting the operator to adjust the operating status of the gas boiler. The alarm module stops sounding the alarm until the predicted value falls below or equals the threshold.

[0066] The above technical solution trains a model for predicting sulfur dioxide concentration. Real-time operating data from the gas boiler inlet is then integrated into the model. The model then predicts the sulfur dioxide content of the exhaust gas at the boiler outlet. If the predicted value is below a threshold, normal combustion and emissions are maintained. If the predicted value is above the threshold, an alarm is triggered, prompting operators to adjust the operating status of the gas boiler until the predicted value falls below or equals the threshold. This testing method allows for predicting sulfur dioxide levels before emissions occur, thus preventing the emission of gases with excessive sulfur dioxide content.

[0067] Specific embodiments are used in the present invention to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core ideas. At the same time, for those skilled in the art, according to the ideas of the present invention, there may be changes in the specific implementation methods and application scopes. In summary, the contents of this specification should not be understood as limiting the present invention.

Claims

1. A method for predicting sulfur dioxide concentration in a gas-fired boiler after mixing desulfurized regeneration gas with refined gas, which is used to predict the concentration of sulfur dioxide at the inlet of the gas-fired boiler, characterized in that: The following steps are involved: Collect data, collect historical operation data of gas boilers; Obtaining an original data set, preprocessing the historical operation data, eliminating the influence of outliers in the characteristic variables, and obtaining an original data set; Acquire a model data set, calculate the strength of the correlation between the characteristic variables in the original data set and the sulfur dioxide concentration, and select at least ten characteristic variables related to the sulfur dioxide concentration as the model data set; Acquire an input data set, perform dimensionless processing on the model data set, control the order of magnitude of the model data set between [0, 1], and obtain an input data set; A training model is provided, wherein a sulfur dioxide concentration prediction model is constructed based on a lightweight gradient boosting machine algorithm, and the lightweight gradient boosting machine algorithm is trained using the input data set to obtain a hyperparameter combination, and the hyperparameter combination is input into the lightweight gradient boosting machine algorithm to obtain a prediction model.

2. The method for predicting sulfur dioxide concentration in a gas-fired boiler after mixing desulfurized regenerated gas with refined gas as claimed in claim 1, characterized in that: Preprocessing the historical operation data includes: The density-based clustering algorithm determines the specific values ​​of the neighborhood radius ε and the density threshold Minpts according to the clustering silhouette coefficient s, finds the core object and its dense neighborhood objects, connects the core object and the neighborhood, forms a dense area as a cluster, and screens outliers in the historical operation data and removes them. The calculation method of the clustering silhouette coefficient s is as follows: Among them, s i For a single sample d i The silhouette coefficient, a i Representative sample d i The average distance to other samples in the same cluster, b i For a single sample d i The average distance to all samples in its nearest cluster, n is the number of samples in the historical running data.

3. The method for predicting sulfur dioxide concentration in a gas-fired boiler after mixing desulfurized regenerated gas with refined gas as claimed in claim 2, characterized in that: In the step of calculating the strength of the correlation between the characteristic variables in the original data set and the sulfur dioxide concentration, the mutual information method is used, and the calculation formula is as follows: MI(X;Y)=H(X)+H(Y)-H(X,Y); Among them, MI(X;Y) is the mutual information between feature variables X and Y; H(X) and H(Y) represent the information entropy of variables X and Y respectively; H(X,Y) is the joint information entropy of variables X and Y; p(x) and p(y) represent the probability distribution of variables X and Y respectively; p(x,y) is the joint probability distribution of variables X and Y.

4. The method for predicting sulfur dioxide concentration in a gas-fired boiler after mixing desulfurized regenerated gas with refined gas as claimed in claim 1, characterized in that: The steps to build a sulfur dioxide concentration prediction model based on the lightweight gradient boosting machine algorithm include: Set the value range of the hyperparameters in the lightweight gradient boosting machine algorithm: num_leaves is [32,2048], max_depth is [10,30], min_child_samples is [1,15], learning_rate is [0.01,0.2]; Based on the value ranges of num_leaves, max_depth, min_child_samples, and learning_rate, the minimum values ​​of the corresponding hyperparameters are input into the lightweight gradient boosting machine algorithm, and the lightweight gradient boosting machine algorithm is trained using the input data set to calculate the mean square error and R 2 value; Perform parameter iteration within the range of num_leaves, max_depth, min_child_samples, and learning_rate, train the updated lightweight gradient boosting machine algorithm using the input data set, and calculate the corresponding mean square error and R 2 value; The hyperparameters within the value range are searched and multiple hyperparameter combinations and corresponding mean square errors and R are obtained. 2 value, the mean square error tends to be minimum and R 2 The hyperparameter combination with the largest value is regarded as the best hyperparameter combination.

5. The method for predicting sulfur dioxide concentration in a gas-fired boiler after mixing desulfurized regenerated gas with refined gas as claimed in claim 4, characterized in that: Steps to train the model also include: The real-time data of the gas boiler inlet is connected to the trained prediction model. The prediction model predicts the sulfur dioxide concentration based on the real-time data to obtain the predicted value. The real value obtained by actually measuring the sulfur dioxide concentration is combined to calculate the mean square error between the real value and the predicted value and R 2 value.

6. The method for predicting sulfur dioxide concentration in a gas-fired boiler after mixing desulfurized regenerated gas with refined gas as claimed in claim 5, characterized in that: Steps to train the model also include: Change the hyperparameter combination of the prediction model, connect the real-time data of the gas boiler inlet to the changed prediction model, and use the changed prediction model to predict the sulfur dioxide concentration based on the real-time data to obtain a new prediction value. Combined with the actual value obtained by measuring the sulfur dioxide concentration, calculate the mean square error and R between the real value and the new prediction value. 2 value.

7. The method for predicting sulfur dioxide concentration in a gas-fired boiler after mixing desulfurized regenerated gas with refined gas as claimed in claim 1, characterized in that: The prediction model also includes an alarm module for warning that the sulfur dioxide concentration exceeds a preset threshold.

Images