Method for predicting NOx emission concentration of circulating fluidized bed boiler based on LightGBM-CatBoost

Abstract

The invention relates to a method for predicting the NOx emission concentration of a circulating fluidized bed boiler based on LightGBM-CatBoost, and the method comprises the steps: carrying out the data collection and feature extraction, and dividing a training set and a test set; constructing a mean value model L-C; taking the exhaust pressure, the generator power, the instantaneous coal feeding amount and the primary air volume as model input variables, and taking the NOx emission concentration as a target variable to obtain a prediction result of the NOx emission concentration. The method is good in prediction precision and stability, the deviation between the predicted NOx emission concentration value and the actual value is small, the deviation is smaller than 5% when the NOx emission concentration is larger than 120 mg.m <-3 >, it can be guaranteed that the NOx emission concentration values of CFB boilers of different coal-fired units are accurately predicted under the wide load condition, and the method is suitable for popularization and application. And a reliable method and data support are provided for emission and control of NOx emission concentration of the CFB boiler during deep peak regulation.

Description

Technical Field

[0001] The present invention relates to the technical field of circulating fluidized bed boilers, in particular to a circulating fluidized bed boiler NO x Methods for predicting emission concentrations. Background Art

[0002] Circulating Fluidized Bed (CFB) boilers have been widely used in coal-fired power generation systems due to their high combustion efficiency, wide fuel adaptability, excellent peak-shaving capability and low pollutant emissions. Most CFB boilers use Selective Non-catalytic Reduction (SNCR) technology to reduce NO x This technology reduces NO emissions by controlling the amount of ammonia reducing agent injected into the combustion zone in the CFB boiler furnace or the inlet of the cyclone separator. x Reduced to N2, thereby significantly reducing NO x To achieve the consumption of renewable energy, CFB boilers in coal-fired power plants need to undertake peak load regulation. When coal-fired units are performing peak load regulation, the temperature and oxygen content in the furnace of the CFB boiler change, causing NO x The amount of generated gas is higher than that under normal working conditions; at the same time, it makes it difficult for the flue gas temperature to reach the SNCR operation temperature range, resulting in the reduction agent and NO x The reaction rate slows down, making the denitrification system NO x The removal efficiency is reduced, NO x Emission concentration increases. How to control NO in CFB boilers during peak load regulation of coal-fired units? x Predicting and controlling emission concentrations is extremely important and has become a research hotspot.

[0003] In recent years, researchers have proposed CFB boiler NO x Machine learning prediction method for emission concentration. With the rapid development of artificial intelligence technology and the collection and storage of a large amount of industrial data, x The machine learning algorithm prediction of emission concentration provides data support. x Emission concentration prediction mainly focuses on traditional machine learning algorithms and optimization, and its main goal is to improve its prediction accuracy. Ensemble learning algorithms are only for NO x Simple prediction of emissions does not consider the complementarity and adaptability of algorithms for prediction. x There are few reports on the construction of emission concentration prediction models and their generalization capabilities. Summary of the Invention

[0004] Aiming at the deficiencies of the prior art, the present invention provides a circulating fluidized bed boiler NO x Methods for predicting emission concentrations.

[0005] In order to solve the above technical problems, the present invention models a CFB boiler in a 300MW coal-fired power plant and calculates the NO x Characteristic parameters of emissions and NO x The correlation coefficient between emission concentrations is calculated, and the characteristic parameters with strong correlation are selected as the data collected in the model. At the same time, an integrated learning algorithm is used to analyze the NO x The factors affecting emission concentration were analyzed, and the complementarity and adaptability between algorithms were fully utilized to combine and analyze four ensemble learning algorithms: CatBoost, NGBoost, LightGBM, and XGBoost. Six mean ensemble models were constructed using the Boosting method, and the stability, accuracy, and other aspects of the six mean models were compared and analyzed. The present invention provides the following technical solutions:

[0006] The present invention provides a circulating fluidized bed boiler NO based on LightGBM-CatBoost x The method for predicting emission concentration includes the following steps:

[0007] (1) Data collection and feature extraction;

[0008] (2) Preprocess the collected data and divide it into training set and test set;

[0009] (3) Model fusion: Based on the LightGBM model and the CatBoost model, the mean model LightGBM-CatBoost is constructed after integration;

[0010] (4) Select the model evaluation index, and use the mean model LightGBM-CatBoost to evaluate the NO of circulating fluidized bed boiler through training set and test set. x The emission concentration is predicted; exhaust pressure, generator power, instantaneous coal feed rate, and primary air volume are used as model input variables, and NO x The emission concentration is used as the target variable of the model to obtain the NO x Prediction results of emission concentrations.

[0011] Preferably, the NO x Includes: NO, NO2 and N2O.

[0012] Preferably, in step (1), according to the circulating fluidized bed boiler NO x Characteristic parameters of emissions and NO xThe correlation coefficient between emission concentrations is used, and the characteristic parameters with strong correlation are selected as the data collected in the model.

[0013] Preferably, the characteristic parameters with strong correlation include: exhaust pressure, generator power, instantaneous coal feed rate and primary air volume.

[0014] Preferably, in step (2), the ratio of the training set to the test set is 7:3;

[0015] Preferably, in step (2), after dividing into a training set and a test set, each group of data in the training set and the test set is normalized.

[0016] Preferably, in step (3), based on the stacked ensemble machine learning method, the Boosting method is used to construct the mean model LightGBM-CatBoost.

[0017] Preferably, in step (4), the model evaluation index includes: goodness of fit R 2 , root mean square error RMSE and mean absolute error MAE.

[0018] Preferably, the RMSE value of the mean model LightGBM-CatBoost is 0.16, the MAE value is 0.12, and R 2 The value is 0.97.

[0019] Preferably, in step (4), the circulating fluidized bed boiler NO x The deviation between the predicted and true emission concentrations.

[0020] The technical solution of the present invention achieves the following beneficial technical effects:

[0021] To achieve NO reduction in circulating fluidized bed boilers of coal-fired units under variable operating conditions x In order to accurately predict the emission concentration, this paper takes the circulating fluidized bed (CFB) boiler of a 300MW coal-fired unit as the research object and constructs a CFB boiler NOx prediction algorithm using six algorithms of data-driven ensemble learning. x Emission concentration prediction model and CFB boiler NO x The effects of CFB boiler load, coal feed rate and primary air volume on the NO emission concentration of CFB boiler were studied. x The prediction model was validated using goodness of fit, root mean square error, and mean absolute error to determine the impact of emission concentration. The results showed that compared with the other five models, the LightGBM-CatBoost (LC) model had a mean absolute error and root mean square error of 0.12 and 0.16, respectively, and a goodness of fit of 0.97, indicating the best prediction accuracy and stability.

[0022] Furthermore, as the load of the CFB boiler changes, NO x The maximum relative error between the predicted and actual emission concentrations is less than 13.9%, and the relative error of most predicted values ​​is less than 3%. When the coal feed rate changes, the NO2 emissions predicted by the XC (XGBoost-CatBoost) and CNLX (CatBoost-NGBoost-LightGBM-XGBoost) models are significantly different. x The maximum relative error between the emission concentration value and the actual value is 9%, which is 13.9% less than that of the LC model. x The maximum relative error between the emission concentration value and the true value is less than 13%. x Concentration higher than 120 mg·m -3 When the LC model predicts NO x The maximum deviation between the emission concentration value and the true value is less than 5%, which is less than the maximum deviation of 10% of the XC and CNLX models. In summary, the method of the present invention can ensure the NOx emission of CFB boilers of different coal-fired units under wide load conditions. x The emission concentration value is accurately predicted. This achievement provides new ideas for SNCR low-NOx combustion and control, and provides important theoretical basis and data support for the implementation of flexible transformation of CFB boilers in future engineering practice.

[0023] The present invention uses exhaust pressure, generator power, instantaneous coal feed rate and primary air volume as model input variables, and uses the Boosting method to construct the mean model LightGBM-CatBoost based on the integration of LightGBM model and CatBoost model to calculate NO x The emission concentration is predicted, so that NO x The emission concentration value has the smallest deviation from the actual value. x Emission concentration is greater than 120 mg·m -3 The deviation is less than 5%. The method of the present invention can ensure the NO x The method based on the mean integrated model of the present invention has good applicability and effectiveness in predicting CFB boiler pollutant emissions, and provides a reference for the NOx prediction of CFB boilers during the deep peak regulation period. x It provides reliable methods and data support for emission concentration and control. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 is the CFB boiler NO in Example 1 of the present invention x Heat map of the relationship between emissions and various parameters;

[0025] Figure 2 is NO in Example 2 of the present invention x Emission concentration forecast roadmap;

[0026] Figure 3 The load change of NO under different models in Example 3 of the present invention is x Effect diagram of emission concentration; among them, (a) CN combination model; (b) LC combination model; (c) LX combination model; (d) XN combination model; (e) XC combination model; (f) CNLX combination model; Figure 4-6 same;

[0027] Figure 4 The effect of coal feeding rate on NO under different models in Example 3 of the present invention is shown in Figure 3. x Effect diagram of emission concentration;

[0028] Figure 5 The effect of air volume change on NO in different models in Example 3 of the present invention is shown in Figure 3. x Effect diagram of emission concentration;

[0029] Figure 6 For different models NO in Example 3 of the present invention x Prediction value deviation analysis chart.

[0030] Figure 7 This is a conceptual diagram of the Boosting algorithm involved in Example 2 of the present invention. DETAILED DESCRIPTION

[0031] To make the purpose, technical solutions, and technical effects of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0032] The following is a specific example of the circulating fluidized bed boiler NO based on LightGBM-CatBoost of the present invention. x The method for predicting emission concentrations is explained in detail.

[0033] Example 1 Data Collection and Feature Extraction

[0034] 1.1 Input conditions

[0035] The present invention takes the CFB boiler of a 300MW coal-fired unit as the research object, wherein the composition of the CFB boiler includes a single furnace, a full-hanging structure, a full steel frame Π-shaped layout, equipped with two 50% capacity primary fans, secondary fans and induced draft fans, and two electric bag dust collectors. The superheated steam of the CFB boiler is generated by the superheater, and the reheated steam is generated by the reheater. The cross-sectional dimensions of the CFB furnace are 28.3m×39.9m, with natural circulation and a membrane water-cooled wall. The main design parameters of the CFB boiler are shown in Table 1. In order to ensure that the efficiency of the CFB boiler is 91.3%, the parameters such as the CFB boiler superheater outlet steam pressure and the reheater inlet and outlet steam pressure are all selected from the rated output working condition values ​​in Table 1. Coal quality characteristics will affect NO during the combustion process x However, in actual engineering, it's difficult for power plants to capture changes in coal quality properties in real time. Furthermore, within the timeframe of the modeling data used in this invention, coal quality properties remain constant. Because nitrogen content and volatile matter content are correlated during coal combustion, this invention uses design coal quality property values ​​as model inputs, as shown in Table 2.

[0036] Table 1300MW coal-fired unit CFB design parameters

[0037] parameter unit Maximum continuous output condition Rated output conditions Boiler evaporation capacity <![CDATA[t·h -1 ]]> 1165 1104 Superheater outlet steam pressure MPa 17.4 17.4 Superheater outlet steam temperature ℃ 541 541 Reheater outlet steam pressure MPa 3.56 3.48 Reheater inlet steam pressure MPa 3.73 3.65 Reheater inlet steam temperature ℃ 541 541 Reheater outlet steam temperature ℃ 329 323

[0038] Table 2 Coal quality characteristics

[0039]

[0040] 1.2 Data Collection and Feature Extraction

[0041] CFB boiler NO x The emission concentration dataset used is derived from actual values ​​from the online monitoring system of a CFB boiler at a 300 MW coal-fired unit. The relevant variables are shown in Table 3. The dataset includes 19 hours of minute-by-minute sensor values ​​distributed over three days, involving 25 feature variables and one target variable. N4–N9 represent the primary air volume of the five duct igniters when the CFB boiler load changes; N10–N12 represent the exhaust pressure of the three exhaust valves when the CFB boiler load changes; and N15–N25 represent the openings of the secondary air box air volume valves at different locations in the furnace when the CFB boiler load changes.

[0042] Table 3 CFB boiler NO x Key emission characteristic parameters

[0043] Label describe unit N1 Generator power MW N2 <![CDATA[O2 concentration in flue gas]]> ％ N3 Instantaneous coal feed <![CDATA[t·h -1 ]]> N4~N9 Primary air volume <![CDATA[t·h -1 ]]> N10~N12 Exhaust pressure Pa N13 Furnace flue gas temperature ℃ N14 Tower clean flue gas temperature ℃ N15~N25 Secondary air door opening ％ Y <![CDATA[NO x Emission concentration]]> <![CDATA[mg·m -3 ]]>

[0044] NO produced by coal combustion x It mainly includes nitric oxide (NO), nitrogen dioxide (NO2) and a small amount of nitrous oxide (N2O). The fuel-type NO produced during combustion xThe concentration is positively linearly correlated with the excess air coefficient and instantaneous coal feed rate. Figure 1 Describes the relationship between instantaneous coal feed rate, primary air volume and NO x The correlation strength between emission concentrations. Most of the characteristic variables in this dataset are highly correlated, but there are still some redundant features in the dataset. These features are closely related to NO x The emission concentration has little correlation, which can be simplified in the model learning process.

[0045] Figure 1 The horizontal and vertical coordinates are both feature parameter labels, and the value in each square is the correlation coefficient of the two corresponding feature parameters. Figure 1 The correlation coefficients are mainly concentrated in the range of 0.5 to 1.0, indicating that the correlations between the features are strong. x The correlation coefficient of NO generation reached 0.69, indicating that the primary air volume has an important influence on the NO generation during combustion. x The concentration has a great influence. Exhaust pressure, instantaneous coal feeding rate and generator power all have high correlation coefficients, so it is necessary to use the above characteristics as model input. In addition, the amount of O2 in flue gas is related to NO x The correlation coefficient is 0.15, indicating that the O2 concentration in the flue gas has a significant effect on the boiler NO x The generation impact is small, so it is simplified.

[0046] Example 2 Model construction and calculation

[0047] 2.1 Calculation method

[0048] Based on the stacked ensemble machine learning method, six mean ensemble models were constructed by combining four ensemble learning models: CN (CatBoost-NGBoost) combination model, LC (LightGBM-CatBoost) combination model, LX (LightGBM-XGBoost) combination model, XN (XGBoost-NGBoost) combination model, XC (XGBoost-CatBoost) combination model,

[0049] CNLX (CatBoost-NGBoost-LightGBM-XGBoost) combined model was used to analyze the NO x The emission concentration was modeled and calculated, and the stability and accuracy of the six combination models were compared and analyzed to determine the optimal combination model.

[0050] The model uses exhaust pressure, generator power, instantaneous coal feed rate, and primary air volume as model input variables. xThe emission concentration is the target variable. x The emission concentration prediction model framework is as follows Figure 2 As shown. After preprocessing, the original data set obtained 4956 groups of samples as NO x Emission prediction data, of which 3476 groups of samples are used as training sets and 1489 groups of samples are used as test sets. The data of each group of training and test sets must be normalized to eliminate the impact of the difference between different variable dimensions on modeling. The formula is as follows:

[0051]

[0052] Where: x is the measured data; x n is the normalized result; x max and x min are the maximum and minimum values ​​in the data respectively.

[0053] The normalized results are fed into four models. Boosting allows for simultaneous training of multiple models, using a validation set for early stopping to prevent overfitting. The four models are then combined to construct an average model, and the average prediction result is taken as the final prediction. Model prediction performance can be evaluated by adding an evaluation function.

[0054] Prediction model construction method: Since the modeling of the present invention has the characteristics of high-dimensional big data, an integrated learning model that is good at handling complex and changeable data is selected. This type of model can effectively improve the accuracy of the model, overfitting or underfitting, and other problems. Boosting method (such as Figure 7 )By combining multiple super weak classifiers to produce a strong learner, after multiple iterations, the Boosting algorithm can combine the weak rules of each iteration into a strong decision rule until the predetermined number of learners or the predetermined prediction accuracy is reached.

[0055] 2.2 Model Evaluation Method

[0056] By adding the goodness of fit (R 2 ), root mean square error (RMSE), and mean absolute error (MAE) are used as evaluation indicators of the model, among which the goodness of fit (R 2 ) coefficient value is between 0 and 1. The closer the value is to 1, the better the fit between the model and the data. In order to ensure the prediction accuracy and generalization ability of the model, this paper mainly uses RMSE and R 2 As the main evaluation parameter. 2The calculation formula is shown in formula (2). The root mean square error (RMSE) reflects the average difference between the model prediction value and the actual value. The RMSE calculation formula is shown in formula (3). The mean absolute error (MAE) represents the average absolute error between the prediction value and the actual value. The calculation formula is shown in formula (4).

[0057]

[0058] Where: n is the sample size; y i is the measured value of the sample; y′ i is the predicted value; is the average value of all test samples.

[0059] Example 3 Prediction Results and Analysis

[0060] 3.1 Model Validation

[0061] In order to screen out the model with better performance, six models were verified and compared on the same training set and test set. All six methods were established based on the mean model. Table 4 shows three typical performance indicators of the model. As can be seen from Table 4, the RMSE of the LC model is 0.16, which is the smallest value among the six models, reflecting that the prediction accuracy of this model is higher than that of the other models. At the same time, the MAE of the LC model and the CN model is the same, 0.12, and the MAE is the smallest among the six models. The CN and LC models R 2 The coefficient is 0.97, indicating that both models have a high goodness of fit to the data and capture most of the changes. 2 The RMSE value is 0.94, indicating that the data fit of the model is poor and the model variability explanation ability of the model is weak. At the same time, the RMSE value is the largest at 0.22. In summary, since the LC model has the smallest RMSE and the largest R 2 , and MAE is the smallest, indicating that this model has the best prediction accuracy and stability among the six models.

[0062] Table 4 Comparison of model evaluation indicators

[0063]

[0064] 3.2 Effect of load change on NO x Impact of emission concentration

[0065] Figure 3 For model NO under variable load conditions xTrends in the variation between predicted and actual emission concentration values. When the unit operates between 20% and 26% operating conditions, the deviations between the predicted and actual values ​​of the six models are all within 5%. When the unit varies between 26% and 90% operating conditions, the model prediction deviations gradually increase, with the CN model's maximum deviation from the actual value reaching 30%. When the unit operates between 50% and 85% operating conditions, 13.8% of the sample points have deviations greater than 35%. Within this range, 13.6% of the sample points for the LX and XN models each have deviations greater than 30%. Within this range, 10% of the sample points for the CNLX model have deviations greater than 30%. Within this range, 5% of the sample points for the LC model have deviations greater than 28%. When the unit operates above 90% operating conditions, 16% of the sample points for the CN, XN, XC, and CNLX models each have deviations greater than 6%. Within this range, 18.7% of the sample points for the LX model have deviations greater than 4.5%. Within this range, 12% of the sample points for the LC model have deviations greater than 4%. In summary, the deviation between the predicted value and the actual value of the LC model is the smallest during the unit load change process, showing good stability.

[0066] 3.3 Effect of coal feed rate change on NO x Impact of emission concentration

[0067] Figure 4 The effect of coal feed rate on NO x The relationship between the predicted value and the actual value of emission concentration. -1 The six models showed no significant difference within the range. When the coal feed rate was greater than 120t·h -1 There are different degrees of deviation between the model prediction value and the actual value. The three models LX, XN, and CNLX are in the range of 120-150t·h -1 Within the interval, more than 14.2% of the sample points have a deviation greater than 13.2%; when the coal feed rate is greater than 120t·h -1 When the coal feeding rate is greater than 120t·h, the deviation of more than 8% of the sample points of the XC model is greater than 13.5%; -1 When the coal feed rate is within this range, 6% of the predicted sample points have an error greater than 13%. When the CN model predicts the coal feed rate within this range, 7.6% of the predicted sample points have a deviation greater than 15%. Therefore, when only observing the change in coal feed rate, the XC model has a smaller deviation between the predicted and actual values, showing better accuracy.

[0068] 3.4 Effect of primary air volume change on NO x Impact of emission concentration

[0069] Figure 5 Shows the effect of primary air volume change on NO xThe relationship between the predicted value and the actual value of the emission concentration. -1 The deviation between the predicted values ​​and the actual values ​​of each model is small within the interval. When the primary air volume is between 180 and 260 t·h -1 The difference between the models gradually becomes obvious when the air volume of LX and XN models is 190~260t·h -1 25% of the predicted values ​​within the range deviate from the actual values ​​by more than 17.5%; 22% of the sample points of the XC and CNLX models within this interval have predicted values ​​that deviate from the actual values ​​by more than 10.4%; 17.4% of the predicted values ​​of the CN model have predicted values ​​that deviate from the actual values ​​by more than 13%; and 10% of the sample points of the LC model have predicted values ​​that deviate from the actual values ​​by more than 7.6%. In summary, when the primary air volume of the system changes, the prediction stability of the LC model is stronger than that of the other five models, and it is more suitable for predicting NO under the condition of primary air volume changes. x emission.

[0070] 3.5, NO x Emission concentration prediction

[0071] In order to study the NO x The deviation between the predicted and actual emission concentration values ​​was determined by using information such as load, coal feed rate, and primary air volume under nine different operating conditions (the specific values ​​are shown in Table 5). Figure 6 The NO of CFB boiler under nine working conditions x Emission concentration.

[0072] Table 5 Input conditions under different working conditions

[0073]

[0074] Depend on Figure 6 It can be seen that when the unit load exceeds 80%, the maximum error between the predicted and actual values ​​of the CN and XN models reaches 15%, while the deviations of the other models above this load are all less than 5%. When the LX model predicts under the operating conditions of 20-70%, the deviation between the predicted and actual values ​​is 9% at multiple sample points, indicating that the model's prediction stability is poor under this load. The XC model and CNLX model perform well in the operating range of 25-70%, with maximum errors of 6.8% and 8.3%, respectively, indicating good model stability in this operating range. The LC model has a maximum deviation of 7% when operating under the operating condition of 60%, but the maximum deviation does not exceed 3% under other operating conditions. Therefore, this model has higher accuracy and stability than the other models.

[0075] In summary, the present invention adopts an integrated learning algorithm to analyze the NO xThe mean ensemble model of emission concentration was established by screening the 25 variables in the data set and NO x The variables with strong correlation with the concentration change are used as the input parameters of the model and the NO x The results show that compared with the other five mean models, the LC model has the lowest RMSE value of 0.16 and MAE value of 0.12, both of which are the lowest among the six models, and R 2 The value of 0.97 indicates that the model has a good goodness of fit. When the load changes, more than 90% of the sample points in the model prediction value have a deviation of less than 5%; when the primary air volume changes, more than 90% of the predicted sample points have a deviation of less than 7.6%. Finally, the model was verified using nine different load conditions. The results showed that the LC model only had a 7% deviation between the predicted value and the actual value when the operating conditions were 60%, and the deviation was less than 3% under the remaining operating conditions, showing good stability. x Emission control and NO under the conditions of changes in boiler load, coal volume, air volume and other parameters during deep peak regulation x The concentration prediction provides theoretical and technical guidance, and also offers a reference for data-driven modeling of CFB boiler pollutant emissions.

[0076] Obviously, the above embodiments are merely examples for clarity of explanation and are not intended to limit the implementation methods. Those skilled in the art will readily appreciate that other variations or modifications based on the above descriptions are possible. It is not necessary and impossible to enumerate all implementation methods here. Obvious variations or modifications arising therefrom remain within the scope of protection of the claims of this patent application.

Claims

1. A circulating fluidized bed boiler NO based on LightGBM-CatBoost x The method for predicting emission concentration is characterized in that: The following steps are involved: (1) Data collection and feature extraction; (2) Preprocess the collected data and divide it into training set and test set; (3) Model fusion: Based on the LightGBM model and the CatBoost model, the mean model LightGBM-CatBoost is constructed after integration; (4) Select model evaluation indicators and use the mean model LightGBM-CatBoost to evaluate the NOx of circulating fluidized bed boilers through training and test sets. x The emission concentration is predicted; exhaust pressure, generator power, instantaneous coal feed rate, and primary air volume are used as model input variables, and NO x The emission concentration is used as the model target variable to obtain the circulating fluidized bed boiler NO x Prediction results of emission concentration.

2. The prediction method according to claim 1, characterized in that: The NO x Includes: NO, NO2 and N2O.

3. The prediction method according to claim 1, characterized in that: In step (1), according to the circulating fluidized bed boiler NO x Emission characteristic parameters and NO x The correlation coefficient between emission concentrations is used to select characteristic parameters with strong correlation as the data collected in the model.

4. The prediction method according to claim 3, characterized in that: The characteristic parameters with strong correlation include: exhaust pressure, generator power, instantaneous coal feed rate and primary air volume.

5. The prediction method according to claim 1, characterized in that: In step (2), the ratio of the training set to the test set is 7:

3.

6. The prediction method according to claim 1, characterized in that: In step (2), after dividing into training set and test set, each group of data in the training set and the test set is normalized.

7. The prediction method according to claim 1, characterized in that: In step (3), based on the stacked ensemble machine learning method, the Boosting method is used to construct the mean model LightGBM-CatBoost.

8. The prediction method according to claim 1, characterized in that: In step (4), the model evaluation index includes: goodness of fit R 2 , root mean square error RMSE and mean absolute error MAE.

9. The prediction method according to claim 8, characterized in that: The mean model LightGBM-CatBoost has an RMSE value of 0.16, a MAE value of 0.12, and an R 2 The value is 0.

97.

10. The prediction method according to claim 1, characterized in that: In step (4), the circulating fluidized bed boiler NO x The deviation between the predicted and true values ​​of emission concentrations.

Images