A method and system for predicting the concentration of co in sintering flue gas coupled with injection

Abstract

The application discloses a kind of sintering flue gas CO concentration prediction method and system coupled with injection, and the real-time parameter data affecting sintering flue gas CO concentration is obtained by the real-time data acquisition module set for;Sintering flue gas CO concentration label generation module is used to input the real-time parameter data in real-time data acquisition module to the neural network model established in advance, and the sintering flue gas CO concentration label is output by neural network model;Sintering flue gas CO concentration prediction value determination module is used to determine the deviation between sintering flue gas CO concentration label and sintering flue gas CO detection value, and whether deviation is within the preset range is judged, by comparing sintering flue gas CO concentration prediction value and preset concentration threshold value finally output in real time, can be triggered according to the high and low of prediction value grading alarm instruction, solve the problem that the reasonable optimization control of injection parameter and sintering process parameter cannot be realized in prior art, so as to realize the problem that sintering flue gas CO concentration is extremely low.

Description

Technical Field

[0001] This invention relates to the field of sintering production technology, specifically to a method and system for predicting CO concentration in sintering flue gas using coupled injection. Background Technology

[0002] The sintering process in steel production is one of the core links in the generation and emission of pollutants. During the sintering process, incomplete combustion of solid fuels, solid-phase reactions of the mixture, and reduction reactions continuously generate harmful gases such as carbon monoxide (CO). CO in sintering flue gas is not only a key pollutant under air pollution control, but also directly reflects the combustion conditions of the sintering bed, fuel utilization efficiency, and production energy consumption level. With increasingly stringent national environmental emission standards and the comprehensive promotion of ultra-low emission transformation in steel enterprises, achieving the ultimate control of CO concentration in sintering flue gas has become a core necessity for major steel sintering production enterprises to improve quality, reduce carbon emissions, and achieve compliant production.

[0003] Currently, most existing CO concentration control technologies for sintering flue gas adopt a post-processing online detection mode. This involves installing gas analyzers in the flue to collect real-time CO concentration data, which can only provide reactive alarms and adjustments after CO concentration exceeds the limit, and cannot predict CO concentration trends in advance. Furthermore, existing technologies do not deeply link the coupled injection process parameters with the core sintering production process parameters in the sintering production control process. They neglect the coupled influence of key parameters such as the particle size distribution of the sintering mixture, the particle size distribution of solid fuel, the moisture content of the mixture, the thickness of the material layer, and the coupled injection flow rates of steam and fuel gas on CO generation.

[0004] For example, patent number CN 121474887 A discloses an intelligent sintering CO emission reduction and energy-saving system and method for sintering machines, including an internal circulation subsystem, an external circulation subsystem, a large flue subsystem, an annular cooling co-processing subsystem, and an intelligent monitoring and control subsystem. The sintering machine's air box is equipped with at least one air box branch pipe. Each subsystem achieves flue gas diversion and circulation through the flue, circulation pipe, and fan. The intelligent monitoring and control subsystem includes a main controller and temperature, pressure, and flow sensors, CO / NO concentration monitors, and regulating valves on each pipe and fume hood. This invention achieves classified flue gas circulation through the collaboration of multiple subsystems. High-carbon flue gas removes CO and NO through the waste heat and catalytic effect of the annular cooler, medium-carbon flue gas is internally circulated and reused, and low-carbon flue gas is purified before emission. The intelligent monitoring and control subsystem collects parameters in real time and performs linkage control, avoiding the lag of traditional control. While improving the flue gas recycling rate and reducing CO and NOx emissions, it ensures stable sintering production. Patent number CN 120393724 A discloses a coupled sintering CO removal device and its operation control method. The coupled sintering CO removal unit includes a desulfurization unit, a dust collector, a flue gas heater, a CO reaction section, and an SCR reactor. By coordinating the control of the coupled sintering CO removal unit during start-up, operation, and shutdown, CO catalyst poisoning and deactivation are avoided, thereby effectively extending the service life of the CO catalyst.

[0005] The existing conventional prediction models mentioned above mostly use a single fitting algorithm, and do not perform data time-series matching and model adaptive optimization for the time lag characteristics and dynamic fluctuation characteristics of multiple processes in sintering production. They cannot accurately represent the nonlinear mapping relationship between multi-dimensional production parameters and flue gas CO concentration, resulting in the inability to complete reasonable pre-control of injection parameters and sintering process parameters according to real-time production conditions. It is difficult to achieve stable control of sintering flue gas CO concentration to the extreme, and there are many technical defects such as low prediction accuracy, poor adaptability to operating conditions, lack of model self-correction capability, and strong control lag. Summary of the Invention

[0006] The purpose of this invention is to provide a method and system for predicting CO concentration in sintering flue gas using coupled injection, in order to solve the technical problems in existing sintering flue gas CO concentration control and prediction technologies, such as the lack of deep linkage modeling between coupled injection parameters and core sintering process parameters, the absence of a real-time self-correction mechanism for the model, low accuracy of timing matching of production parameters, inability to achieve reasonable pre-adjustment and control of injection parameters and sintering process parameters, and thus difficulty in achieving stable control of sintering flue gas CO concentration to an extremely low level.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] A method for predicting CO concentration in sintering flue gas via coupled injection includes the following steps:

[0009] S1: Real-time parameter data acquisition: Collect real-time parameter data of all dimensions that affect the generation and emission of CO concentration in sintering flue gas under coupled injection sintering production conditions, including real-time matrix of particle size distribution of sintering mixture, real-time value of moisture content of sintering mixture, real-time value of solid fuel ratio, real-time matrix of particle size distribution of solid fuel, real-time value of average thickness of sintering material layer, real-time value of temperature of sintering mixture, real-time value of sintering batching ratio, real-time value of steam injection flow rate and real-time value of gas injection flow rate during coupled injection;

[0010] S2: Real-time prediction output of neural network model: Input all real-time parameter data collected in S1 into the pre-trained neural network model, receive the real-time CO concentration label of sintering flue gas output by the neural network model, and use the real-time CO concentration label of sintering flue gas as the predicted value of CO concentration of sintering flue gas for output processing; wherein, the neural network model is pre-trained and has a non-linear and precise mapping relationship between various real-time parameter data and CO concentration label of sintering flue gas.

[0011] S3: Predicted Concentration Deviation Verification and Judgment: After the real-time sintering flue gas CO concentration label is output in S2, the concentration deviation between the real-time sintering flue gas CO concentration label and the real-time in-situ detection value of sintering flue gas CO concentration is determined; wherein, the real-time detection value of sintering flue gas CO concentration is synchronously detected in real time by in-situ laser gas analyzers symmetrically installed on the two exhaust flues of sintering, and the two sets of detection data are processed by mean denoising; the calculated concentration deviation value is compared with the preset allowable deviation range for verification and judgment.

[0012] S4: Switching between Predicted Value Output and Model Adaptive Retraining: If the concentration deviation value is within the preset allowable deviation range, the real-time sintering flue gas CO concentration label is directly output as the final predicted sintering flue gas CO concentration value for real-time optimization of sintering production conditions and environmental control; if the concentration deviation value is not within the preset allowable deviation range, it is determined that the current prediction accuracy of the neural network model is not up to standard, and the model adaptive iterative optimization mechanism is immediately triggered to retrain the neural network model based on the latest real-time production data until the model prediction deviation meets the preset requirements;

[0013] S5: Concentration Prediction Value Graded Alarm Control: Real-time comparison of the final output sintering flue gas CO concentration prediction value with the preset concentration threshold, and trigger graded alarm commands according to the prediction value.

[0014] Furthermore, all real-time parameter data collected in S1 are precisely correlated according to the timing logic of the sintering production process using data tracking technology. The timing logic of the sintering production process is as follows: sintering batching and solid fuel detection are in the upstream batching process. After about 7-9 minutes, mixing and granulation, moisture adjustment and detection, and particle size separation of the mixed material are completed. After 10 minutes, the sintering material distribution height and mixed material temperature are detected. After ignition in the furnace, sintering is completed after 30-50 minutes, which corresponds to the detection of CO concentration in the flue gas.

[0015] Furthermore, the neural network model in S2 is configured according to the actual data structure, and the specific construction steps are as follows:

[0016] S201: Obtain N sets of historical sample data that are sufficient for training and testing, and divide the N sets of historical sample data into a training set and a test set in a ratio of 9:1, with the number of historical sample data in the training set being greater than the number of historical sample data in the test set.

[0017] S202: Preprocess the historical sample input data, initialize the weights and biases of the preliminary neural network model, and use the preprocessed historical sample input data in the training set as the input to the preliminary neural network model to obtain the corresponding CO concentration label network value of sintering flue gas; wherein, the preprocessing includes normalization processing and smoothing processing.

[0018] S203: Determine the calculated value of the loss function based on the CO concentration label network value of the sintering flue gas and the historical CO concentration label value of the sintering flue gas in the preprocessed training set, and determine whether the calculated value of the loss function is less than the preset value. If the calculated value of the loss function is greater than the preset value, proceed to S204; if the calculated value of the loss function is less than the preset value, proceed to S205.

[0019] S204: Calculate the loss function value and backpropagate to update the network weight values ​​and biases, then repeat steps S202 to S203;

[0020] S205: Input the test set into the preliminary neural network model. When the test set feedback results meet the preset conditions, use the preliminary neural network as the final neural network model.

[0021] Furthermore, the training set in S201 includes historical sample input data and historical sample output data; among which, the historical sample input data includes historical values ​​of the real-time matrix of sintering mixture particle size distribution, historical values ​​of sintering mixture moisture content, historical values ​​of solid fuel ratio, historical values ​​of the real-time matrix of solid fuel particle size distribution, historical values ​​of the average thickness of sintering material layer, historical values ​​of sintering batching ratio, historical values ​​of sintering mixture temperature, historical values ​​of sintering batching ratio, historical values ​​of steam injection flow rate and gas injection flow rate during coupled injection.

[0022] Furthermore, the normalization process in S202 preprocessing uses a 0-1 normalization model to normalize the historical sample input data. Specifically, the 0-1 normalization model is as follows:

[0023]

[0024] In the formula, x represents the historical sample input data, max represents the maximum value of the historical sample input data, and min represents the minimum value of the historical sample input data.

[0025] Furthermore, the smoothing process in the S202 preprocessing uses a mean smoothing model, which is specifically as follows:

[0026]

[0027] In the formula, n is the nth data point of the historical sample input data, and y is the moving smoother setting value.

[0028] Furthermore, the neural network model used in S2 is an ANN neural network model, specifically:

[0029]

[0030] In the formula, It is the target output. This is the current input amount. Let m be the weighted neuron connecting the i-th input and the j-th hidden input, and m be the number of input neurons. Let J be the connection weight between the j-th hidden neuron and the output neuron. For the activation function of hidden neurons, For the output neuron activation function, For the j-th bias hidden neuron, H is the bias of the output neuron, and HN is the number of hidden neurons in the output neuron.

[0031] Furthermore, the graded alarm instructions in S5 are as follows: when the predicted CO concentration of sintering flue gas is greater than the first preset exceedance threshold, a first exceedance warning alarm instruction is sent to the sintering production control terminal and the operation and maintenance management personnel terminal; when the predicted CO concentration of sintering flue gas is less than the second preset low load threshold, a second abnormal operating condition alarm instruction is sent to the target control object; wherein, the value of the first preset exceedance threshold is greater than the value of the second preset low load threshold, realizing full-dimensional alarm control of two-way abnormal operating conditions of high and low CO concentration.

[0032] This invention provides another technical solution: a coupled injection sintering flue gas CO concentration prediction system, including a real-time data acquisition module, a sintering flue gas CO concentration label generation module, and a sintering flue gas CO prediction value determination module; the real-time data acquisition module is used to acquire real-time parameter data affecting the sintering flue gas CO concentration; the sintering flue gas CO concentration label generation module is used to input the real-time parameter data from the real-time data acquisition module into a pre-established neural network model, and output a sintering flue gas CO concentration label through the neural network model; the sintering flue gas CO concentration prediction value determination module is used to determine the deviation value between the sintering flue gas CO concentration label and the sintering flue gas CO detection value, and determine whether the deviation value is within a preset range.

[0033] Compared with the prior art, the beneficial effects of the present invention are:

[0034] 1. The method and system for predicting CO concentration in sintering flue gas by coupled injection of the present invention is the first to deeply couple and model the dual matrix parameters of particle size of sintering mixture and particle size of solid fuel with the parameters of steam and gas flow rate of coupled injection and the core process parameters of sintering. It accurately characterizes the nonlinear mapping relationship between multiple parameters and CO concentration in sintering flue gas, breaks through the technical bottleneck of single parameter control and incomplete modeling dimensions in the existing technology, and predicts in line with the actual production conditions on site.

[0035] 2. The CO concentration prediction method and system for sintering flue gas by coupled injection of the present invention adds a deviation verification mechanism between the predicted concentration and the actual detected concentration and an adaptive retraining mechanism for the model. It can correct the neural network model in real time according to the fluctuation of production conditions, solve the problems of poor adaptability of traditional prediction models and the decay of prediction accuracy over long-term operation, and ensure that the prediction results are accurate and stable in the long term.

[0036] 3. The CO concentration prediction method and system for sintering flue gas by coupled injection of the present invention strictly matches the time lag characteristics of multi-process production in sintering, realizes accurate correlation of multi-process parameters through data tracking technology, eliminates prediction errors caused by time misalignment, and combines image recognition to accurately collect granular matrix data, greatly improving the accuracy of basic data collection.

[0037] 4. The sintering flue gas CO concentration prediction method and system of the present invention with coupled injection is equipped with a graded CO concentration alarm mechanism, which can realize early warning of excessive CO concentration and synchronous reminder of abnormal production conditions. It supports operation and maintenance personnel to complete the pre-adjustment and control of injection parameters and sintering process parameters in advance, reduce CO generation from the source, and truly achieve extremely low and stable control of sintering flue gas CO concentration, which is suitable for the long-term control needs of ultra-low emission sintering in the steel industry. Attached Figure Description

[0038] Figure 1 This is a flowchart of the prediction method according to an embodiment of the present invention;

[0039] Figure 2 This is a schematic diagram of the neural network model structure in an embodiment of the present invention. Detailed Implementation

[0040] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0041] To address the limitation of existing technologies that cannot achieve optimal control of injection parameters and sintering process parameters to minimize CO concentration in sintering flue gas, this invention provides a coupled injection method for predicting CO concentration in sintering flue gas. Figure 1 As shown, the process steps are as follows:

[0042] Step 1: Obtain real-time parameter data affecting the CO concentration in sintering flue gas; among which, the real-time parameter data includes the real-time matrix of particle size distribution of sintering mixture, the real-time value of moisture content of sintering mixture, the real-time value of solid fuel ratio, the real-time matrix of particle size distribution of solid fuel, the real-time value of average thickness of sintering material layer, the real-time value of temperature of sintering mixture, the real-time value of sintering batching ratio, and the real-time values ​​of steam injection flow rate and gas injection flow rate during coupled injection.

[0043] For example, the real-time particle size distribution matrix of the sintered mixture in this step is obtained by the following method:

[0044] (1) Acquire images of the sintered mixture located after the sintering secondary mixer at preset time intervals. Control the camera to capture on-site images by calling a specially designed algorithm. Optimize the acquired images using image preprocessing technology, including noise reduction and contrast enhancement, to ensure image clarity and detail.

[0045] (2) Image segmentation technology is used to subdivide the original image into multiple sub-regions and magnify each sub-region so that the fine structure of the sintered mixture can be clearly presented.

[0046] (3) Based on the segmented image, the particle size of the sintering mixture is automatically detected and classified using an image recognition algorithm (convolutional neural network), and the particle size range of the mixture is divided.

[0047] (4) The segmented data is passed to the correction model, which corrects the data based on predefined rules and algorithms. The corrected data is output to provide reliable support for on-site production decisions.

[0048] Specifically, the proportions of each particle size of the sintered mixture, d1<0.5mm, 0.5mm≤d2<1mm, 1mm≤d3<3mm, 3mm≤d4<5mm, 5mm≤d5<8mm, and d6≥8mm, are obtained through an algorithm and then aggregated to obtain a real-time particle size distribution matrix of the sintered mixture.

[0049] In specific embodiments, considering that under stable production conditions, various data are relatively stable, those skilled in the art can also determine the real-time particle size distribution matrix of the sintered mixture by analyzing the data in each image and then averaging all the data based on multiple images of the sintered mixture surface taken within a certain period of time. For example, the real-time particle size distribution matrix of the sintered mixture obtained from all image data within 5 seconds can be used as the real-time matrix within these 5 seconds. The data obtained in this way has higher accuracy.

[0050] In addition, the real-time moisture content of the sintering mixture in this step can be collected by a moisture content detector (infrared or microwave moisture detector) installed above the conveyor belt after the secondary mixer. The real-time solid fuel ratio is obtained using a data tracking method. The particle size distribution matrix of the sintering solid fuel can be obtained by an image recognition and processing system installed on the solid fuel feeding belt corresponding to the sintering batch. A specially designed algorithm is used to control a camera to capture on-site images, and image preprocessing techniques are used to optimize the acquired images, including noise reduction and contrast enhancement, to ensure image clarity and detail. Image segmentation technology is used to subdivide the original image into multiple sub-regions, and each sub-region is magnified to clearly present the fine structure of the sintering solid fuel. Based on the segmented images, an image recognition algorithm (convolutional neural network) is used to automatically detect and classify the particle size of the sintering solid fuel and divide the solid fuel into particle size intervals. The segmented data is fed into a correction model, which corrects the data based on predefined rules and algorithms. The proportions of each particle size of the sintered solid fuel (d1<0.5mm, 0.5mm≤d2<1mm, 1mm≤d3<2mm, 2mm≤d4<3mm, 3mm≤d5<5mm, d6≥5mm) are obtained through an algorithm and aggregated to obtain the real-time particle size distribution matrix of the sintered solid fuel. The material ratio includes iron-containing mixed ore p1, limestone p2, dolomite p3, quicklime p4, cold return ore p5, and the aforementioned solid fuel ratio p6, expressed as a percentage, also obtained using data tracking methods. The real-time average thickness of the sintering layer can be measured using a layer thickness gauge. The real-time temperature of the sintering mixture can be obtained using a temperature sensor. The real-time steam injection flow rate and gas injection flow rate during coupled blowing are detected by the flow orifice plate on the sintering coupled blowing device. The real-time CO concentration of the sintering flue gas is detected in real time by an in-situ laser gas analyzer installed on the two sintering exhaust ducts.

[0051] It should be noted that the above-mentioned detection data have a temporal correspondence. Specifically, sintering batching and solid fuel detection are in the upstream batching process. After about 7-9 minutes, mixing and granulation, moisture adjustment and detection, and particle size separation of the mixture are completed. After about 10 minutes, sintering material distribution (material distribution height detection and mixture temperature detection) takes place. Subsequently, after ignition in the furnace, sintering is completed in 30-50 minutes (corresponding to flue gas CO concentration detection). The real-time parameter data in this step can be correlated with each other using data tracking technology.

[0052] Step 2: Input the real-time parameter data from Step 1 into the pre-established neural network model, and receive the real-time sintering flue gas CO concentration label output by the neural network model. Output the real-time sintering flue gas CO concentration label as the predicted value of sintering flue gas CO concentration. The neural network model contains the mapping relationship between the real-time parameter data and the sintering flue gas CO concentration label.

[0053] This step, before labeling the sintering flue gas CO concentration as the "sintering flue gas CO concentration", also includes:

[0054] (1) Determine the deviation between the CO concentration label of sintering flue gas and the real-time CO concentration detection value of sintering flue gas. The CO concentration of sintering flue gas is detected in real time by an in-situ laser gas analyzer installed on the two exhaust flues of sintering, and the average value is processed to determine whether the above deviation value is within the preset range.

[0055] (2) When the deviation value is within the preset range, proceed to the step of “outputting the real-time sintering flue gas CO concentration label as the predicted value of sintering flue gas CO concentration”; when the deviation value is not within the preset range, retrain the neural network model.

[0056] The type of neural network model in this embodiment can be set as needed, such as using ANN, CNN, or other common algorithm models. The specific model can be set according to the actual data structure and obtained through the following steps:

[0057] Step 201: Obtain N sets of historical sample data suitable for training and testing; divide the N sets of historical sample data into a training set and a test set according to a preset ratio, with the number of historical sample data sets in the training set being greater than the number of historical sample data sets in the test set. The sample size for obtaining samples suitable for training and testing can be [10000, 100000]. The ratio of the training set to the test set can be set to 9:1.

[0058] The training set in this step includes historical sample input data and historical sample output data; among which, the historical sample input data includes historical values ​​of the real-time matrix of sintering mixture particle size distribution, historical values ​​of sintering mixture moisture content, historical values ​​of solid fuel ratio, historical values ​​of the real-time matrix of solid fuel particle size distribution, historical values ​​of the average thickness of sintering material layer, historical values ​​of sintering batching ratio, historical values ​​of sintering mixture temperature, historical values ​​of sintering batching ratio, and historical values ​​of steam injection flow rate and gas injection flow rate during coupled injection;

[0059] Step 202: Preprocess the historical sample input data, initialize the weights and biases of the preliminary neural network model, and use the preprocessed historical sample input data from the training set as input to the preliminary neural network model to obtain the corresponding CO concentration label network value for sintering flue gas. The preprocessing includes normalization and smoothing. Specifically, the normalization process uses a 0-1 normalization model to normalize the historical sample input data. The 0-1 normalization model is as follows:

[0060]

[0061] In the formula, x represents the historical sample input data, max represents the maximum value of the historical sample input data, and min represents the minimum value of the historical sample input data.

[0062] The smoothing process uses a mean smoothing model, which is as follows:

[0063]

[0064] In the formula, n is the nth data point of the historical sample input data, and y is the moving smoother setting value.

[0065] Step 203: Determine the calculated value of the loss function based on the CO concentration label network value of the sintering flue gas and the historical CO concentration label value of the sintering flue gas in the preprocessed training set, and determine whether the calculated value of the loss function is less than the preset value. If the calculated value of the loss function is greater than the preset value, proceed to step 204; if the calculated value of the loss function is less than the preset value, proceed to step 205.

[0066] Step 204: Calculate the loss function value and backpropagate to update the network weights and biases, then repeat steps 202 to 203;

[0067] Step 205: Input the test set into the preliminary neural network model. When the test set feedback results meet the preset conditions, use the preliminary neural network as the final neural network model.

[0068] In this embodiment, the neural network model selected is the ANN neural network model, such as... Figure 2 As shown, specifically:

[0069]

[0070] in, It is the target output. This is the current input amount. Let m be the weighted neuron connecting the i-th input and the j-th hidden input, and m be the number of input neurons. Let J be the connection weight between the j-th hidden neuron and the output neuron. For the activation function of hidden neurons, For the output neuron activation function, For the j-th bias hidden neuron, H is the bias of the output neuron, and HN is the number of hidden neurons in the output neuron.

[0071] Based on the above method, this embodiment of the invention further includes graded alarm control for concentration prediction values: by comparing the final output predicted CO concentration value of sintering flue gas with a preset concentration threshold in real time, graded alarm commands are triggered according to the level of the predicted value; specifically as follows:

[0072] When the predicted CO concentration of sintering flue gas is greater than the first preset threshold, a first alarm command is sent to the target object; when the predicted CO concentration of sintering flue gas is less than the second preset threshold, a second alarm command is sent to the target object; wherein, the first preset threshold is greater than the second preset threshold, and the target object can be the sintering production control terminal and the operation and maintenance management personnel terminal.

[0073] To further illustrate the above embodiments, the present invention also provides a coupled-injection sintering flue gas CO concentration prediction system for performing the above-described coupled-injection sintering flue gas CO concentration prediction method, specifically including: a real-time data acquisition module, a sintering flue gas CO concentration tag generation module, and a sintering flue gas CO prediction value determination module; wherein:

[0074] The real-time data acquisition module is used to acquire real-time parameter data that affects the CO concentration of sintering flue gas. The real-time parameter data includes the real-time matrix of particle size distribution of sintering mixture, the real-time value of moisture content of sintering mixture, the real-time value of solid fuel ratio, the real-time matrix of particle size distribution of solid fuel, the real-time value of average thickness of sintering material layer, the real-time value of temperature of sintering mixture, the real-time value of sintering batching ratio, and the real-time values ​​of steam injection flow rate and gas injection flow rate during coupled injection.

[0075] The sintering flue gas CO concentration label generation module is used to input real-time parameter data into a pre-established neural network model and receive the sintering flue gas CO concentration label output by the neural network model; wherein, the neural network model contains the mapping relationship between the real-time parameter data and the sintering flue gas CO concentration label;

[0076] The sintering flue gas CO concentration prediction module is used to determine the deviation between the sintering flue gas CO concentration label and the sintering flue gas CO detection value, and to determine whether the deviation is within a preset range. When the deviation is within the preset range, the sintering flue gas CO concentration label is used as the predicted value of the sintering flue gas CO concentration. When the deviation is not within the preset range, the neural network model is retrained.

[0077] In summary, this invention provides a method and system for predicting CO concentration in sintering flue gas using coupled injection. For the first time, it deeply couples and models the particle size of the sintering mixture and the particle size of the solid fuel with the flow parameters of the injected steam and fuel gas, as well as the core sintering process parameters. This accurately represents the nonlinear mapping relationship between multiple parameters and the CO concentration in sintering flue gas, overcoming the technical bottlenecks of single-parameter control and incomplete modeling dimensions in existing technologies. The prediction closely matches the actual production conditions. Secondly, this invention adds a mechanism for verifying the deviation between the predicted and actual detected concentrations and for adaptive model retraining. This allows for real-time correction of the neural network model based on fluctuations in production conditions, solving the problems of poor adaptability and long-term accuracy decay in traditional prediction models, ensuring long-term accurate and stable prediction results. Furthermore, this invention includes a graded CO concentration alarm mechanism, enabling simultaneous alerts for CO concentration exceeding limits and abnormal production conditions. This supports maintenance personnel in pre-optimizing and controlling injection and sintering process parameters, reducing CO generation at the source, and truly achieving extremely low and stable CO concentration control in sintering flue gas, meeting the long-term ultra-low emission control requirements of the steel industry.

[0078] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for predicting CO concentration in sintering flue gas using coupled injection, characterized in that, Includes the following steps: S1: Real-time parameter data acquisition: Collect real-time parameter data of all dimensions that affect the generation and emission of CO concentration in sintering flue gas under coupled injection sintering production conditions, including real-time matrix of particle size distribution of sintering mixture, real-time value of moisture content of sintering mixture, real-time value of solid fuel ratio, real-time matrix of particle size distribution of solid fuel, real-time value of average thickness of sintering material layer, real-time value of temperature of sintering mixture, real-time value of sintering batching ratio, real-time value of steam injection flow rate and real-time value of gas injection flow rate during coupled injection; S2: Real-time prediction output of neural network model: Input all real-time parameter data collected in S1 into the pre-trained neural network model, receive the real-time CO concentration label of sintering flue gas output by the neural network model, and use the real-time CO concentration label of sintering flue gas as the predicted value of CO concentration of sintering flue gas for output processing; wherein, the neural network model is pre-trained and has a non-linear and precise mapping relationship between various real-time parameter data and CO concentration label of sintering flue gas. S3: Predicted Concentration Deviation Verification and Judgment: After the real-time sintering flue gas CO concentration label is output in S2, the concentration deviation between the real-time sintering flue gas CO concentration label and the real-time in-situ detection value of sintering flue gas CO concentration is determined; wherein, the real-time detection value of sintering flue gas CO concentration is synchronously detected in real time by in-situ laser gas analyzers symmetrically installed on the two exhaust flues of sintering, and the two sets of detection data are processed by mean denoising; the calculated concentration deviation value is compared with the preset allowable deviation range for verification and judgment. S4: Switching between Predicted Value Output and Model Adaptive Retraining: If the concentration deviation value is within the preset allowable deviation range, the real-time sintering flue gas CO concentration label is directly output as the final predicted sintering flue gas CO concentration value for real-time optimization of sintering production conditions and environmental control; if the concentration deviation value is not within the preset allowable deviation range, it is determined that the current prediction accuracy of the neural network model is not up to standard, and the model adaptive iterative optimization mechanism is immediately triggered to retrain the neural network model based on the latest real-time production data until the model prediction deviation meets the preset requirements; S5: Concentration Prediction Value Graded Alarm Control: Real-time comparison of the final output sintering flue gas CO concentration prediction value with the preset concentration threshold, and trigger graded alarm commands according to the prediction value.

2. The method for predicting CO concentration in sintering flue gas using coupled injection as described in claim 1, characterized in that: All real-time parameter data collected in S1 are precisely correlated according to the sintering production process timing logic using data tracking technology; the sintering production process timing logic is as follows: Sintering batching and solid fuel testing are located in the upstream batching process. After about 7-9 minutes, mixing and granulation, moisture adjustment and testing, and particle size classification of the mixture are completed. After 10 minutes, the sintering material distribution height and mixture temperature are tested. After ignition in the furnace, sintering is completed in 30-50 minutes, which corresponds to the detection of CO concentration in the flue gas.

3. The method for predicting CO concentration in sintering flue gas using coupled injection as described in claim 1, characterized in that: The neural network model in S2 is set according to the actual data structure, and the specific construction steps are as follows: S201: Obtain N sets of historical sample data that are sufficient for training and testing, and divide the N sets of historical sample data into a training set and a test set in a ratio of 9:1, with the number of historical sample data in the training set being greater than the number of historical sample data in the test set. S202: Preprocess the historical sample input data, initialize the weights and biases of the preliminary neural network model, and use the preprocessed historical sample input data in the training set as the input to the preliminary neural network model to obtain the corresponding CO concentration label network value of sintering flue gas; wherein, the preprocessing includes normalization processing and smoothing processing. S203: Determine the calculated value of the loss function based on the CO concentration label network value of the sintering flue gas and the historical CO concentration label value of the sintering flue gas in the preprocessed training set, and determine whether the calculated value of the loss function is less than the preset value. If the calculated value of the loss function is greater than the preset value, proceed to S204; if the calculated value of the loss function is less than the preset value, proceed to S205. S204: Calculate the loss function value and backpropagate to update the network weight values ​​and biases, then repeat steps S202 to S203; S205: Input the test set into the preliminary neural network model. When the test set feedback results meet the preset conditions, use the preliminary neural network as the final neural network model.

4. The method for predicting CO concentration in sintering flue gas using coupled injection as described in claim 3, characterized in that: The training set in S201 includes historical sample input data and historical sample output data. The historical sample input data includes historical values ​​of the real-time matrix of sintering mixture particle size distribution, historical values ​​of sintering mixture moisture content, historical values ​​of solid fuel ratio, historical values ​​of the real-time matrix of solid fuel particle size distribution, historical values ​​of the average thickness of sintering material layer, historical values ​​of sintering batching ratio, historical values ​​of sintering mixture temperature, historical values ​​of sintering batching ratio, historical values ​​of steam injection flow rate and gas injection flow rate during coupled injection.

5. The method for predicting CO concentration in sintering flue gas using coupled injection as described in claim 3, characterized in that: The normalization process in S202 preprocessing uses a 0-1 normalization model to normalize the historical sample input data. Specifically, the 0-1 normalization model is as follows: In the formula, x represents the historical sample input data, max represents the maximum value of the historical sample input data, and min represents the minimum value of the historical sample input data.

6. The method for predicting CO concentration in sintering flue gas using coupled injection as described in claim 3, characterized in that: The smoothing process in S202 preprocessing uses a mean smoothing model, which is as follows: In the formula, n is the nth data point of the historical sample input data, and y is the moving smoother setting value.

7. The method for predicting CO concentration in sintering flue gas using coupled injection as described in claim 1, characterized in that: The neural network model used in S2 is an ANN neural network model, specifically: In the formula, It is the target output. This is the current input amount. Let m be the weighted neuron connecting the i-th input and the j-th hidden input, and m be the number of input neurons. Let J be the connection weight between the j-th hidden neuron and the output neuron. For the activation function of hidden neurons, For the output neuron activation function, For the j-th bias hidden neuron, H is the bias of the output neuron, and HN is the number of hidden neurons in the output neuron.

8. The method for predicting CO concentration in sintering flue gas using coupled injection as described in claim 1, characterized in that: The hierarchical alarm commands in S5 are as follows: When the predicted CO concentration of sintering flue gas exceeds the first preset threshold, a first over-limit warning alarm command is sent to the sintering production control terminal and the operation and maintenance management personnel terminal; when the predicted CO concentration of sintering flue gas is less than the second preset low load threshold, a second abnormal operating condition alarm command is sent to the target controlled object; wherein, the value of the first preset threshold is greater than the value of the second preset low load threshold, realizing full-dimensional alarm control of two-way abnormal operating conditions of high and low CO concentration.

9. A coupled-injection sintering flue gas CO concentration prediction system, used to implement the coupled-injection sintering flue gas CO concentration prediction method according to claim 1, characterized in that, It includes a real-time data acquisition module, a sintering flue gas CO concentration label generation module, and a sintering flue gas CO prediction value determination module; the real-time data acquisition module is used to acquire real-time parameter data that affects the CO concentration of sintering flue gas; the sintering flue gas CO concentration label generation module is used to input the real-time parameter data in the real-time data acquisition module into a pre-established neural network model, and output the sintering flue gas CO concentration label through the neural network model. The CO concentration prediction module for sintering flue gas is used to determine the deviation between the CO concentration label of sintering flue gas and the CO detection value of sintering flue gas, and to determine whether the deviation is within a preset range.

Images