Reverse adjustment algorithm for stabilizing FeO content of sintered ore

Abstract

The invention relates to a reverse adjustment algorithm for stabilizing the FeO content of sintered ore. The reverse adjustment algorithm comprises the specific steps that S1, a sintered ore quality-FeO content prediction model is established; s2, determining a sample template and a FeO value stable range; s3, calculating a difference value between the maximum value and the minimum value of each parameter feature in the sample template; s4, FeO content prediction is conducted on sintering data collected at set intervals, and whether adjustment is needed or not is judged; s5, formulating a reverse adjustment algorithm; and S6, carrying out effect verification on the reverse adjustment algorithm. According to the method disclosed by the invention, the sinter FeO content prediction model based on the XGBoost algorithm is established, so that the quality of the sinter is accurately controlled; the FeO content of the sintered ore is stabilized by establishing a sample template, it is ensured that chemical components of the sintered ore are uniform and consistent, and the strict requirements of blast furnace smelting are met; in addition, production fluctuation is reduced through real-time monitoring and a dynamic adjusting mechanism, and the stability and efficiency of sinter production are improved.

Description

Technical Field

[0001] The invention relates to the technical field of sintering production, and in particular to a reverse regulation algorithm for stabilizing the FeO content of sintered ore. Background Art

[0002] Sintered ore is an important raw material for blast furnace ironmaking. Ferrous oxide (FeO), as its key component, has a significant impact on the quality of sintered ore and the efficiency of blast furnace smelting. The fluctuation of FeO content in sintered ore will directly affect the coke ratio and output of the blast furnace, so the stability of FeO content is crucial. However, the formation mechanism of FeO in sintered ore is very complex and affected by many factors, resulting in large fluctuations in its content and uneven distribution, which is difficult to accurately calculate and control using traditional methods.

[0003] At present, the on-site control of FeO content in sintered ore mainly relies on the operator's experience and judgment, that is, estimating the FeO content by observing the brightness and thickness of the red fire layer at the tail of the sintering machine, and then adjusting the fuel ratio and operating parameters. Although this method is effective to a certain extent, it is limited by the operator's technical level and sense of responsibility, as well as the fluctuation of raw fuel quality, and its accuracy and stability are limited.

[0004] The Chinese patent application with publication number CN105276988A discloses "a method for controlling the FeO content of the tail section of a sintering machine", which uses a dual CCD image monitoring system to collect images of the tail section of a sintering machine. The dual CCD image monitoring system includes a color CCD camera and an infrared CCD camera. The color CCD camera collects visible light images, and the infrared CCD camera collects infrared images. The collected images are processed by a processing computer, and finally the FeO content of the sintered ore is measured by a data measurement and analysis system. The data measurement and analysis system includes a fuzzy clustering system and a neural network system. It solves the problems of the existing method for measuring the FeO content of sintered ore, which requires pyrotechnics to rely on experience for judgment, has high labor intensity, and has large deviations in the quality of sintered ore. However, this method has high equipment cost, is easily affected by the environment, and has insufficient continuous operation capacity.

[0005] The Chinese patent application with publication number CN104977316A discloses "a method for determining the trend of FeO content in sintered ore", including: determining the sintering end temperature of solid fuel in the material layer by R and FeO control benchmark data in the sintered ore in the PLC; performing sintering process control at the sintering end position; controlling the material layer state by the end position and the end temperature; determining the physical heat content of sintered ore and the temperature of the ring cooling exhaust gas by the percentage of fixed carbon contained in the solid fuel and the material layer control state; performing corresponding temperature difference analysis by tracking the large flue temperature and the ring cooling exhaust gas temperature to predict the content of FeO in sintered ore; and accurately obtaining the trend of FeO content in sintered ore within the time period when the sintering end position meets the process requirements by comparing the trend between the sintering flue temperature and the ring cooling exhaust gas temperature. It can determine the trend of FeO content in sintered ore online, so as to adjust the FeO content of sintered ore in advance, reduce the consumption of solid fuel in production, and increase the output of sintered ore. However, it only considers the influence of the flue temperature and the temperature of the ring cooling exhaust gas, fails to fully consider the sintering raw materials and other process parameters, and lacks a rapid adjustment plan for FeO fluctuations.

[0006] In summary, accurate control and rapid adjustment of the FeO content in sintered ore are the key to improving the efficiency of blast furnace ironmaking. However, the existing methods have certain limitations, and further research and development of more accurate and practical control methods are needed. Summary of the invention

[0007] The present invention provides a reverse regulation algorithm for stabilizing the FeO content of sintered ore. By establishing a sintered ore FeO content prediction model based on the XGBoost algorithm, accurate control of the sintered ore quality is achieved. By establishing a sample template to stabilize the FeO content of the sintered ore, the chemical composition of the sintered ore is ensured to be uniform and consistent, meeting the strict requirements of blast furnace smelting. In addition, real-time monitoring and dynamic regulation mechanisms are used to reduce production fluctuations and improve the stability and efficiency of sintered ore production.

[0008] In order to achieve the above object, the present invention adopts the following technical solutions:

[0009] A reverse regulation algorithm for stabilizing the FeO content of sintered ore comprises the following steps:

[0010] Step S1: Establishing a sinter quality-FeO content prediction model;

[0011] Step S2: Determine the sample template and the stable range of FeO value;

[0012] Step S3: Calculate the difference between the maximum and minimum values ​​of each parameter feature in the sample template;

[0013] Step S4: predicting the FeO content of the sintering data collected at regular intervals and determining whether adjustment is required;

[0014] Step S5: formulating a reverse adjustment algorithm;

[0015] Step S6: verify the effect of the reverse regulation algorithm.

[0016] Furthermore, the step S1 is specifically as follows:

[0017] Step S11: Determine the prediction model; use the XGBoost model to predict the FeO content, improve the prediction accuracy by gradually optimizing the model, and accelerate the training process through parallel computing; use the CART decision tree as the base learner, and the XGBoost model is an additive model composed of K base learners, that is:

[0018]

[0019] in represents the predicted value of the i-th sample, i = 1, 2, ..., n, f k represents the base learner, represents the feature vector of the i-th sample,

[0020] The loss function of the XGBoost model is to add a function Ω(f to measure the complexity of the base learner) to the traditional loss function. k ), the loss function formula of the XGBoost model is as follows:

[0021]

[0022] Item 1 is a traditional loss function, which is used to measure the degree of fit of the XGBoost model to the data; the second The function controls the complexity of the model to prevent overfitting of the XGBoost model.

[0023] Furthermore, the optimization of the XGBoost model is to solve each base learner f by minimizing the loss function k ; The XGBoost model uses a step-by-step solution, first solving the first base learner f1, then the second base learner f2, ... and finally the last base learner f t ; By performing Taylor second-order expansion on the traditional loss function, we can solve the basis learner f t When , the loss function of the XGBoost model is simplified as:

[0024]

[0025] Where g i and h i The loss functions are The gradient and Hessian matrix at :

[0026]

[0027]

[0028] Minimize the above loss function to get a new round of f t ;

[0029] Step S12: determine the training and test data sets; the training and test data sets include batching data, sintering process operation data and sintered ore quality data, wherein the ignition temperature, trolley speed and main pipe negative pressure in the sintering process operation data are all adjustable parameters.

[0030] Furthermore, the step S2 is specifically as follows:

[0031] The test data set is trained based on the XGBoost model, and the parameters of the XGBoost model are adjusted to obtain the best training model. The test data set is then predicted to obtain the predicted value of the FeO content, and the normal distribution is processed to determine the mean μ and standard deviation σ; samples with FeO content values ​​in (μ-3σ, μ+3σ) are selected, and samples with a difference of r between the predicted FeO value and the true value are selected from the samples, where r is less than 0.5; the maximum value max and the minimum value min of the FeO predicted values ​​of these samples are calculated, and the stable range of the FeO content is [min, max], and these sample sets are used as sample templates.

[0032] Furthermore, the step S3 is specifically as follows:

[0033] The difference d1, d2, ..., dn between the maximum and minimum values ​​of each parameter feature value in the sample template determined in step S2 is calculated, where n is the number of features in the sample template.

[0034] Furthermore, the step S4 is specifically as follows:

[0035] During the sintering ore production process, sintering data is collected once every period of time, and the collected data is input into the sintering ore FeO content prediction model established in step S1 to obtain the predicted value Y of the FeO content; if Y is within the stable range [min, max] of the FeO content determined in step S2, the sintering production process is determined to be normal and no adjustment is required; if Y is not within the stable range [min, max] of the FeO content determined in step S2, the sintering production process is determined to be abnormal and needs to be adjusted.

[0036] Furthermore, the step S5 is specifically as follows:

[0037] Step S51: If it is determined in step S4 that the sintering production process is abnormal and needs to be adjusted, calculate the differences s1, s2, ……, sn between each eigenvalue and each sample eigenvalue in the sample template, and divide them by the differences d1, d2, ……, dn between the maximum and minimum values of each eigenvalue in the sample template determined in step S2 to normalize each sample data feature, obtaining s1 / d1, s2 / d2, ……, sn / dn, and calculate the Euclidean distance D:

[0038]

[0039] Step S52: Select the top m samples with the smallest Euclidean distance from the sample template according to step S51, calculate the mean value of each eigenvalue in these m samples, set the mean value of the ignition temperature as T0, the mean value of the pallet car speed as S0, and the mean value of the main pipe negative pressure as P0;

[0040] Step S53: Compare the ignition temperature value T, the pallet car speed value S, and the main pipe negative pressure value P collected at regular intervals with the ignition temperature value T0, the pallet car speed value S0, and the main pipe negative pressure value P0 obtained in step S52 to determine the adjustment strategy;

[0041] (1) Adjustment strategy for ignition temperature;

[0042] If T > T0, the adjustment strategy is to decrease the ignition temperature value, and the decreasing ratio of the ignition temperature value is:

[0043] Unit: %;

[0044] If T < T0, the adjustment strategy is to increase the ignition temperature value, and the increasing ratio of the ignition temperature value is:

[0045] Unit: %;

[0046] If T = T0, no adjustment is made;

[0047] (2) Adjustment strategy for pallet car speed:

[0048] If S > S0, the adjustment strategy is to decrease the pallet car speed value, and the decreasing ratio of the pallet car speed value is:

[0049] Unit: %;

[0050] If S < S0, the adjustment strategy is to increase the pallet car speed value, and the increasing ratio of the pallet car speed value is:

[0051] Unit: %;

[0052] If S = S0, no adjustment is made;

[0053] (3)主管负压的调节策略;

[0054] 如果P > P0, the adjustment strategy is to reduce the main duct negative pressure value, and the reduction ratio of the main duct negative pressure value is:

[0055] in %;

[0056] 如果P < P0, the adjustment strategy is to increase the main duct negative pressure value, and the increase ratio of the main duct negative pressure value is:

[0057] in %;

[0058] 如果P = P0, no adjustment is made.

[0059] 进一步的,所述步骤S6具体如下:

[0060] 经过步骤S5后,确定了调节后烧结数据即点火温度、台车速度以及主管负压值,经过一段时间的烧结过程后,再次获得一组烧结数据;基于步骤S1确定的FeO含量预测模型,再次预测FeO的含量,直到FeO含量值处于设定的稳定范围内。

[0061] 与现有技术相比,本发明的有益效果是:

[0062] 通过建立基于XGBoost算法的烧结矿FeO含量预测模型,实现对烧结矿质量的精准控制;通过建立样本模板稳定烧结矿FeO含量,确保了烧结矿的化学成分均匀一致,满足高炉冶炼的严格要求;另外通过实时监测和动态调节机制减少了生产波动,提高了烧结矿生产的稳定性和效率。 附图说明

[0063] Figure 1 是本发明所述一种稳定烧结矿FeO含量的反向调节算法的流程图。

[0064] Figure 2 是本发明所述制定反向调节算法步骤的流程图。 具体实施方式

[0065] 下面结合附图对本发明的具体实施方式作进一步说明:

[0066] 如 Figure 1 所示,本发明所述一种稳定烧结矿FeO含量的反向调节算法,包括如下步骤:

[0067] 步骤S1:建立烧结矿质量-FeO含量预测模型;

[0068] Step S2: Determine the sample template and the stable range of FeO value;

[0069] Step S3: Calculate the difference between the maximum and minimum values ​​of each parameter feature in the sample template;

[0070] Step S4: predicting the FeO content of the sintering data collected at regular intervals and determining whether adjustment is required;

[0071] Step S5: formulating a reverse adjustment algorithm;

[0072] Step S6: verify the effect of the reverse regulation algorithm.

[0073] The step S1 is specifically as follows:

[0074] Step S11: Determine the prediction model; the present invention adopts the XGBoost model to predict the FeO content. XGBoost is an efficient gradient boosting decision tree algorithm that improves the prediction accuracy by gradually optimizing the model and supports parallel computing to accelerate the training process; the XGBoost model is an additive model composed of K basis learners, that is:

[0075]

[0076] in represents the predicted value of the i-th sample, i = 1, 2, ..., f k represents the base learner, n, represents the feature vector of the i-th sample, Base learner f k The present invention adopts CART decision tree as the base learner, and has a very good effect on regression problems.

[0077] The loss function of the XGBoost model is to add a function Ω(f to measure the complexity of the base learner) to the traditional loss function. k ), the loss function formula of the XGBoost model is as follows:

[0078]

[0079] Item 1 is a traditional loss function, such as mean square error, mean absolute error, L1 norm loss, etc. It is used to measure the degree of fit of the XGBoost model to the data; the second item The function controls the complexity of the model to prevent overfitting of the XGBoost model.

[0080] The optimization algorithm of the XGBoost model is to solve each base learner f by minimizing the loss function k; The XGBoost model uses a step-by-step solution, first solving the first base learner f1, then the second base learner f2, ... and finally the last base learner f t ; Assume that the current iteration has been t-1 times, and t-1 base learners are obtained. Next, at the tth time, solve the optimal base learner f t Let the predicted value of the model in round t-1 be The calculation formula is as follows:

[0081]

[0082] Then the loss function can be expressed as:

[0083]

[0084] Performing Taylor second-order expansion on the loss function, the loss function can be expressed as:

[0085]

[0086] where g i and h i The loss functions are The gradient and Hessian matrix at :

[0087]

[0088]

[0089] Since in round t, is known, so is a constant, so the loss function in the tth round can be simplified as:

[0090]

[0091] Minimize the above loss function to get a new round of f t .

[0092] Step S12: determine the training and test data sets; the training and test data sets include batching data, sintering process operation data and sintered ore quality data, wherein the sintering process operation data include ignition temperature, trolley speed and main pipe negative pressure, which are all adjustable parameters.

[0093] The step S2 is specifically as follows:

[0094] The test data set is trained based on the XGBoost model, and the parameters of the XGBoost model are adjusted to obtain the best training model. The test data set is then predicted to obtain the predicted value of the FeO content, and the normal distribution is processed to determine the mean μ and standard deviation σ; samples with FeO content values ​​in (μ-3σ, μ+3σ) are selected, and samples with a difference of r between the predicted FeO value and the true value are selected from the samples, where r is less than 0.5; the maximum value max and minimum value min of the FeO predicted values ​​of these samples are calculated, and the stable range of FeO content is [min, max], and these sample sets are used as sample templates.

[0095] The step S3 is specifically as follows:

[0096] The difference d1, d2, ..., dn between the maximum and minimum values ​​of each parameter feature value in the sample template determined in step S2 is calculated, where n is the number of features in the sample template.

[0097] The step S4 is specifically as follows:

[0098] During the sintering ore production process, sintering data is collected once every period of time, and the collected data is input into the sintering ore FeO content prediction model established in step S1 to obtain the predicted value Y of the FeO content; if Y is within the stable range [min, max] of the FeO content determined in step S2, the sintering production process is determined to be normal and no adjustment is required; if Y is not within the stable range [min, max] of the FeO content determined in step S2, the sintering production process is determined to be abnormal and needs to be adjusted.

[0099] The step S5 is specifically as follows:

[0100] Step S51: If step S4 determines that the sintering production process is abnormal and needs to be adjusted, the difference s1, s2, ..., sn between each eigenvalue and each sample eigenvalue in the sample template is calculated, and divided by the difference d1, d2, ..., dn between the maximum and minimum values ​​of each eigenvalue in the sample template determined in step S2, each sample data feature is normalized to obtain s1 / d1, s2 / d2, ..., sn / dn, and the Euclidean distance D is calculated:

[0101]

[0102] Step S52: According to step S51, the first m samples with the smallest Euclidean distance are selected from the sample template, and the mean of each characteristic value in these m samples is calculated. The mean of the ignition temperature is set to T0, the mean of the trolley speed is set to S0, and the mean of the main pipe negative pressure is set to P0.

[0103] Step S53: Compare the ignition temperature value T, trolley speed value S, and main duct negative pressure value P collected at regular intervals with the ignition temperature value T0, trolley speed value S0, and main duct negative pressure value P0 obtained in Step S52 to determine the adjustment strategy;

[0104] (1) Adjustment strategy for ignition temperature;

[0105] If T > T0, the adjustment strategy is to decrease the ignition temperature value, and the decrease ratio of the ignition temperature value is:

[0106] Unit: %;

[0107] If T < T0, the adjustment strategy is to increase the ignition temperature value, and the increase ratio of the ignition temperature value is:

[0108] Unit: %;

[0109] If T = T0, no adjustment is made;

[0110] (2) Adjustment strategy for trolley speed:

[0111] If S > S0, the adjustment strategy is to decrease the trolley speed value, and the decrease ratio of the trolley speed value is:

[0112] Unit: %;

[0113] If S < S0, the adjustment strategy is to increase the trolley speed value, and the increase ratio of the trolley speed value is:

[0114] Unit: %;

[0115] If S = S0, no adjustment is made;

[0116] (3) Adjustment strategy for main duct negative pressure;

[0117] If P > P0, the adjustment strategy is to decrease the main duct negative pressure value, and the decrease ratio of the main duct negative pressure value is:

[0118] Unit: %;

[0119] If P < P0, the adjustment strategy is to increase the main duct negative pressure value, and the increase ratio of the main duct negative pressure value is:

[0120] Unit: %;

[0121] If P = P0, no adjustment is made.

[0122] The specific steps of Step S6 are as follows:

[0123] After step S5, the adjusted sintering data, namely the ignition temperature, trolley speed and main pipe negative pressure value, are determined. After a period of sintering process, a set of sintering data is obtained again. Based on the FeO content prediction model determined in step S1, the FeO content is predicted again until the FeO content value is within the set stable range.

[0124] The following embodiments are implemented on the premise of the technical solution of the present invention, and provide detailed implementation methods and specific operation processes, but the protection scope of the present invention is not limited to the following embodiments.

[0125] [Example]

[0126] The minimum value of FeO content in the sample template of the embodiment is 80.1, the maximum value is 9.24, and the stable range is [80.1%, 9.24%]. The difference between the maximum and minimum values ​​of each parameter feature in the sample template is d1, d2, ..., dn, and n is 28 in this embodiment.

[0127] During the sintering ore production process, sintering data is collected every once in a while. In this embodiment, the collected sintering data are 1024.52, 2, -12.09, 105.11, -11.19, 532.59, 322.85, 178.43, 175.3, 141.42, 250.68, 163.33, 231.35, 310.84, 215.75, 281.13, 109.88, 46, 63.25, 19.43, 13.4, 8.71, 11.73, 1.34 , 13.4, 13.4, 5.75, 5.75, where the first three data are the ignition temperature T value (1024.52℃), the trolley speed S value (2m / min) and the main pipe negative pressure P value (-12.09KPa). The collected data are input into the sintered ore FeO content prediction model established in step S1, and the predicted value Y of the FeO content is 9.40%. Obviously, 9.40% is not within the stable range of the FeO content [80.1%, 9.24%] determined in step S2. It is determined that the sintering production process is abnormal and needs to be adjusted.

[0128] In this embodiment, the process of the reverse adjustment algorithm is as follows: Figure 2 shown.

[0129] In this embodiment, the collected sintering data has 28 eigenvalues. The difference s1, s2, ..., s28 between each eigenvalue and each sample eigenvalue in the sample template is calculated, and the difference d1, d2, ..., d28 between the maximum and minimum values ​​of each eigenvalue in the sample template is divided. Each sample data is normalized to s1 / d1, s2 / d2, ..., s28 / d28, and the Euclidean distance D is calculated as follows:

[0130]

[0131] Since the ignition temperature, main pipe negative pressure, and trolley speed in the sample data eigenvalues are all adjustable parameters, the top 10 samples with the smallest Euclidean distance are selected from the sample template, and the mean value of each eigenvalue in these 10 samples is calculated. Among them, the mean value T0 of the ignition temperature is 999.74 °C, the mean value S0 of the trolley speed is 2 m / min, and the mean value P0 of the main pipe negative pressure is -11.16 KPa.

[0132] The sintering process operation data is collected every 2 hours. Among them, the ignition temperature T value is 1024.52 °C, the trolley speed S value is 2 m / min, and the main pipe negative pressure P value is -12.09 KPa. Compare with the ignition temperature mean value T0 (999.74 °C), trolley speed S0 value (2 m / min), and main pipe negative pressure P0 value (-11.16 KPa) to determine the adjustment strategy.

[0133] (1) Adjustment strategy for the ignition temperature.

[0134] Since T = 1024.52 °C, T0 = 999.74 °C, and T > T0, the adjustment strategy is to lower the ignition temperature value. The reduction ratio of the ignition temperature value is:

[0135]

[0136] That is, reduce by 12.1% - 24.2%.

[0137] (2) Adjustment strategy for the trolley speed.

[0138] Since S = 2 m / min, S0 = 2 m / min, and S = S0, the adjustment strategy is no adjustment.

[0139] (3) Adjustment strategy for the main pipe negative pressure.

[0140] Since P = -12.09 KPa, P0 = -11.16 KPa, and P < P0, the adjustment strategy is to increase the main pipe negative pressure value. The increase ratio of the main pipe negative pressure value is:

[0141]

[0142] That is, increase by 3.85% - 7.7%.

[0143] Verification of the reverse adjustment algorithm effect:

[0144] In this embodiment, after adjustment, the ignition temperature, trolley speed and main pipe negative pressure value of the sintering process operation data after adjustment are determined. After a period of sintering process, a set of sintering data 999.742, 2, -12.09, 109.906, -12.052, 505.182, 336.436, 183.492, 236.152, 154.992, 282.974, 204. 828, 253.328, 320.136, 255.75, 285.036, 104.35, 52.62, 60.95, 17.508, 16.408, 10.806, 11.604, 0.804, 13.08, 13.08, 6.43, 4.83, Based on the FeO content prediction model, the predicted FeO content is 8.95%, and the FeO content value is in a stable range.

[0145] 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 reverse regulation algorithm for stabilizing the FeO content of sintered ore, characterized in that: The steps include: Step S1: Establishing a sinter quality-FeO content prediction model; Step S2: Determine the sample template and the stable range of FeO value; Step S3: Calculate the difference between the maximum and minimum values ​​of each parameter feature in the sample template; Step S4: predicting the FeO content of the sintering data collected at regular intervals and determining whether adjustment is required; Step S5: formulating a reverse adjustment algorithm; Step S6: verify the effect of the reverse regulation algorithm.

2. The reverse regulation algorithm for stabilizing the FeO content of sintered ore according to claim 1, characterized in that: The step S1 is specifically as follows: Step S11: Determine the prediction model; use the XGBoost model to predict the FeO content, improve the prediction accuracy by gradually optimizing the model, and accelerate the training process through parallel computing; use the CART decision tree as the base learner, and the XGBoost model is an additive model composed of K base learners, that is: in represents the predicted value of the i-th sample, i = 1, 2, ..., n, f k represents the base learner, represents the feature vector of the i-th sample, The loss function of the XGBoost model is to add a function Ω(f to measure the complexity of the base learner) to the traditional loss function. k ), the loss function formula of the XGBoost model is as follows: Item 1 is a traditional loss function, which is used to measure the degree of fit of the XGBoost model to the data; the second The function controls the complexity of the model to prevent overfitting of the XGBoost model.

3. The reverse adjustment algorithm for stabilizing the FeO content of sintered ore according to claim 2, characterized in that: The optimization of the XGBoost model is to solve each base learner f by minimizing the loss function k ; The XGBoost model uses a step-by-step solution, first solving the first base learner f1, then the second base learner f2, and finally the last base learner f t ; By performing Taylor second-order expansion on the traditional loss function, we can solve the basis learner f t When , the loss function of the XGBoost model is simplified as: Where g i and h i The loss functions are The gradient and Hessian matrix at ; Minimize the above loss function to get a new round of f t ; Step S12: determine the training and test data sets; the training and test data sets include batching data, sintering process operation data and sintered ore quality data, wherein the ignition temperature, trolley speed and main pipe negative pressure in the sintering process operation data are all adjustable parameters.

4. The reverse regulation algorithm for stabilizing the FeO content of sintered ore according to claim 1, characterized in that: The step S2 is specifically as follows: The test data set is trained based on the XGBoost model, and the parameters of the XGBoost model are adjusted to obtain the best training model. The test data set is then predicted to obtain the predicted value of the FeO content, and the normal distribution is processed to determine the mean μ and standard deviation σ; samples with FeO content values ​​in (μ-3σ, μ+3σ) are selected, and samples with a difference of r between the predicted FeO value and the true value are selected from the samples, where r is less than 0.5; the maximum value max and minimum value min of the FeO predicted values ​​of these samples are calculated, and the stable range of FeO content is [min, max], and these sample sets are used as sample templates.

5. The reverse adjustment algorithm for stabilizing the FeO content of sintered ore according to claim 1, characterized in that: The step S3 is specifically as follows: The difference d1, d2, ..., dn between the maximum and minimum values ​​of each parameter feature value in the sample template determined in step S2 is calculated, where n is the number of features in the sample template.

6. The reverse adjustment algorithm for stabilizing the FeO content of sintered ore according to claim 1, characterized in that: The step S4 is specifically as follows: During the sintering ore production process, sintering data is collected once every period of time, and the collected data is input into the sintering ore FeO content prediction model established in step S1 to obtain the predicted value Y of 6FeO content; if Y is within the stable range [min, max] of the FeO content determined in step S2, the sintering production process is judged to be normal and no adjustment is required; if Y is not within the stable range [min, max] of the FeO content determined in step S2, the sintering production process is judged to be abnormal and needs adjustment.

7. The reverse regulation algorithm for stabilizing the FeO content of sintered ore according to claim 1, characterized in that: The step S5 is specifically as follows: Step S51: If step S4 determines that the sintering production process is abnormal and needs to be adjusted, the difference s1, s2, ..., sn between each eigenvalue and each sample eigenvalue in the sample template is calculated, and divided by the difference d1, d2, ..., dn between the maximum and minimum values ​​of each eigenvalue in the sample template determined in step S2, each sample data feature is normalized to obtain s1 / d1, s2 / d2, ..., sn / dn, and the Euclidean distance D is calculated: Step S52: Select the top m samples with the smallest Euclidean distance from the sample template according to Step S51, calculate the mean value of each eigenvalue among these m samples, set the mean value of the ignition temperature as T0, the mean value of the trolley speed as S0, and the mean value of the main pipe negative pressure as P0; Step S53: Compare the ignition temperature value T, trolley speed value S, and main pipe negative pressure value P collected at regular intervals with the ignition temperature value T0, trolley speed value S0, and main pipe negative pressure value P0 obtained in Step S52 to determine the adjustment strategy; (1) Adjustment strategy for ignition temperature; If T > T0, the adjustment strategy is to decrease the ignition temperature value, and the decrease ratio of the ignition temperature value is: The unit is %; If T < T0, the adjustment strategy is to increase the ignition temperature value, and the increase ratio of the ignition temperature value is: The unit is %; If T = T0, no adjustment is made; (2) Adjustment strategy for trolley speed: If S > S0, the adjustment strategy is to decrease the trolley speed value, and the decrease ratio of the trolley speed value is: The unit is %; If S < S0, the adjustment strategy is to increase the trolley speed value, and the increase ratio of the trolley speed value is: The unit is %; If S = S0, no adjustment is made; (3) Adjustment strategy for main pipe negative pressure; If P > P0, the adjustment strategy is to decrease the main pipe negative pressure value, and the decrease ratio of the main pipe negative pressure value is: The unit is %; If P < P0, the adjustment strategy is to increase the main pipe negative pressure value, and the increase ratio of the main pipe negative pressure value is: The unit is %; If P = P0, no adjustment is made.

8. The reverse adjustment algorithm for stabilizing the FeO content of sintered ore according to claim 1, characterized in that: The specific content of Step S6 is as follows: After Step S5, the adjusted sintering data, namely the ignition temperature, trolley speed, and main pipe negative pressure value, are determined. After a period of sintering process, a set of sintering data is obtained again; based on the FeO content prediction model determined in Step S1, the content of FeO is predicted again until the FeO content value is within the set stable range.

Images