Machine learning-based dynamic regulation and control method and system for utilization ratio of sintered return mine cold-pressed block blast furnace

Abstract

The invention discloses a machine learning-based dynamic regulation and control method and system for the utilization ratio of a sintering return mine cold-pressing block blast furnace, and relates to the technical field of blast furnace ironmaking, and the method comprises the following steps: collecting and preprocessing blast furnace multi-source process parameters in real time, calculating the deviation degree between the selected parameters and a forward motion reference value, and generating parameter state distribution feature vectors; fusing the feature vector and original process data, inputting the fused feature vector and original process data into a pre-trained machine learning model, outputting an optimal cold pressing block charging proportion, and converting a proportion instruction into a control signal of a feeding system through a controller to realize proportion dynamic adjustment; the system comprises a data acquisition and preprocessing module, a parameter state feature generation module, an intelligent decision module and an instruction execution regulation and control module. According to the method, the parameters reflecting the operation state of the blast furnace are calculated and classified, the input quality and decision accuracy of the model are remarkably improved, real-time and self-adaptive accurate regulation and control of the cold pressing block proportion are achieved, and reasonable utilization of resources is promoted while smooth operation of the blast furnace is guaranteed.

Description

Technical Field

[0001] This invention relates to the field of blast furnace ironmaking technology, and in particular to a method and system for dynamic control of the utilization ratio of sintered return ore cold-pressed briquettes in blast furnaces based on machine learning. Background Technology

[0002] A blast furnace uses steel plates as its outer shell, lined with refractory bricks. Blast furnace ironmaking boasts excellent technical and economic indicators, a simple process, high production capacity, high labor productivity, and low energy consumption. Therefore, iron produced using this method accounts for the vast majority of the world's total iron production. Blast furnace smelting is the core ironmaking process in the steel industry. It refers to the process of reducing iron ore to pig iron (liquid iron-carbon alloy) through physicochemical reactions under high-temperature conditions in a vertical cylindrical metallurgical unit called a blast furnace, using iron ore as the main raw material, while simultaneously producing slag and blast furnace gas as byproducts.

[0003] Blast furnaces will remain the primary means of ironmaking for a considerable period, and their stable operation is crucial for ensuring high production efficiency, energy conservation, and low costs. Cold-pressed sintered ore is already being used in several blast furnaces in China. As an increasingly important blast furnace charge, the rational utilization of cold-pressed ore contributes to the rational recycling of resources and reduces production costs. However, due to differences in physicochemical properties (such as morphology, compressive strength, and low-temperature reduction pulverization index) compared to traditional sintered ore and pellets, the proportion of cold-pressed ore charged into the furnace significantly affects key parameters such as furnace temperature, bed permeability, and gas distribution.

[0004] Blast furnace operation is characterized by strong coupling of multiple parameters, large hysteresis, and nonlinearity. Adjustments to the utilization ratio of cold-pressed briquettes often lag behind changes in furnace conditions. An excessively high ratio worsens the permeability of the burden, while an excessively low ratio fails to fully realize economic benefits, making it difficult to achieve optimal matching and reduce blast furnace operating costs. Currently, the proportion of cold-pressed briquettes fed into the furnace relies heavily on operator experience, placing high demands on operators and resulting in heavy workloads. Therefore, a precise, real-time control method and system are urgently needed.

[0005] Machine learning technology has provided new ideas for the control and optimization of blast furnaces. For example, invention patent CN120230889A discloses a method for optimizing the structure of blast furnace burden based on intelligent predictive mixing control. This mainly involves a method for optimizing the structure of high-grade ultra-low titanium vanadium-titanium magnetite blast furnace burden, and obtaining optimal burden droplet performance. This method does not involve cold briquetting and only involves control based on the variation of droplet intervals in the high-temperature zone, resulting in insufficient adaptability. Another example is invention patent CN120700221A, which discloses a dynamic control method for blast furnace operation, adjusting blast furnace raw materials based on monitoring the particle size of ore and coke and the flow rate of cold air. This method does not involve cold briquetting, has only one monitoring parameter, and dynamic control will fail if the monitoring of these three parameters fails. Finally, invention patent CN114819587B discloses a method and system for evaluating and predicting the activity of the blast furnace hearth based on big data. This method does not involve cold briquetting or burden control.

[0006] In summary, directly applying publicly available machine learning models to the control of blast furnace cold briquetting ratios often faces the following problems:

[0007] 1) Blast furnace process parameters have high dimensionality and noise, resulting in poor quality of model input features and weak model adaptability;

[0008] 2) The control model has weak interpretability and is difficult to integrate with expert knowledge in the field of cold-pressed blocks;

[0009] 3) For the specific control target of the proportion of cold-pressed blocks, there is a lack of targeted feature engineering methods, and the control model is unable to capture key state information, resulting in insufficient prediction accuracy and robustness.

[0010] Therefore, this invention proposes a machine learning-based method and system for dynamic control of the utilization ratio of sintered return ore cold-pressed briquetting blast furnace to solve the problems existing in the prior art. Summary of the Invention

[0011] To address the aforementioned problems, the present invention aims to propose a machine learning-based method and system for controlling the utilization ratio of sintered return ore cold-pressed briquettes in blast furnaces. By calculating and classifying key process parameters of the blast furnace and generating parameter state distribution feature vectors, these vectors are fused with the original data to provide high-quality input for the machine learning model, ultimately achieving precise, real-time, and adaptive control of the cold-pressed briquette feed ratio.

[0012] To achieve the objectives of this invention, the invention is implemented through the following technical solution: a method for dynamic control of the utilization ratio of blast furnaces for sintered return ore cold-pressed briquettes based on machine learning, comprising the following steps:

[0013] Step 1: Data Acquisition and Preprocessing

[0014] In the blast furnace control system, multi-source process parameter data related to the smooth operation of the blast furnace are collected in real time, and the collected data is cleaned and normalized.

[0015] Step 2: Parameter State Quantization and Feature Generation

[0016] For each process parameter in the multi-source process parameter data, calculate the deviation between its current value and the preset blast furnace operating reference value for that parameter; classify each parameter into a first state, a second state, or a third state according to the preset deviation threshold range; count the number of parameters classified into each state and generate a parameter state distribution feature vector that characterizes the overall operating state of the blast furnace.

[0017] Step 3: Intelligent Proportional Decision Making

[0018] The pre-processed multi-source process parameter data is fused with the parameter state distribution feature vector to form a comprehensive feature vector. The comprehensive feature vector is then input into a pre-trained cold briquette ratio control machine learning model, which outputs the optimal cold briquette feed ratio corresponding to the current blast furnace state.

[0019] Step 4: Command Execution and Control

[0020] The output of the optimal cold-pressed block feeding ratio is converted into control commands for the blast furnace charging system by the controller, thereby dynamically adjusting the actual feeding ratio of the cold-pressed blocks.

[0021] A further improvement is made in the following: In step one, the multi-source process parameter data includes: blast furnace operating parameters within a historically set time period, physicochemical parameters of the cold-pressed briquettes already in the furnace, and physicochemical parameters of the cold-pressed briquettes to be in the furnace. The blast furnace operating parameters are selected from at least six of the following: air volume, air temperature, air pressure, oxygen enrichment, blast humidity, furnace top pressure, furnace top temperature, furnace core temperature, cooling wall water temperature difference, burden pressure difference, burden permeability index, utilization coefficient, fuel ratio, coal ratio, coke ratio, gas utilization rate, smelting intensity, and the weight, basicity, drum index, low-temperature reduction pulverization index (RDI+3.15), and reduction index of sinter, pellets, and lump ore in the ore batch. The physicochemical parameters of the cold-pressed briquettes already in the furnace and the cold-pressed briquettes to be in the furnace are selected from at least three of the following: composition, basicity, compressive strength, low-temperature reduction pulverization index (RDI+3.15), reduction index, and particle size.

[0022] Further improvements are made in that: the dynamic control range of the furnace feed ratio of the cold-pressed block is 0% to 50%, the alkalinity range of the cold-pressed block is 1.0 to 3.0, the particle size range is 10 to 60 mm, the compressive strength range is 1800 to 5000 N, the low-temperature reduction pulverization index range is 45% to 100%, and the reduction index range is 40% to 100%.

[0023] A further improvement is made in the following: In step two, the specific rules for classifying parameters according to the degree of deviation are as follows: parameters with a deviation between 0% and 5% are classified as the first state, parameters with a deviation between 5% and 10% are classified as the second state, and parameters with a deviation greater than or equal to 10% are classified as the third state.

[0024] A further improvement is that, in step two, the process of generating the parameter state distribution feature vector further includes: assigning different initial weight values ​​to parameters of different categories, performing weighted statistics on the number of parameters in each category based on the initial weight values, and dynamically adjusting the weight values ​​according to the correlation between the historical classification results of each parameter and the subsequent smooth operation of the blast furnace.

[0025] A further improvement is made in step three, where the training process of the machine learning model for controlling the proportion of cold-pressed blocks includes:

[0026] S1. Obtain a training dataset from the blast furnace history database. The training dataset includes historical multi-source process parameter data, parameter state distribution feature vector generated in step two, and the proportion of cold-pressed blocks fed into the furnace corresponding to the optimal forward operation state of the blast furnace during that historical period, which serves as the label.

[0027] S2. Preprocess and feature-select the training dataset obtained in step S1;

[0028] S3. Using the filtered historical multi-source process parameter data and its corresponding parameter state distribution feature vector as input features, and the cold-pressed block feed ratio as the prediction target, a machine learning algorithm is used for supervised training to obtain a machine learning model for cold-pressed block ratio control.

[0029] A further improvement is that the machine learning algorithm is selected from at least one of gradient boosting decision tree, random forest, support vector machine, and long short-term memory neural network.

[0030] A further improvement is that, in step four, the controller is a proportional-integral-derivative controller, a fuzzy logic controller, a neural network controller, or an expert system, and the control command is used to adjust the feeding speed of the conveyor belt scale or the opening of the batching valve.

[0031] A machine learning-based dynamic control system for the utilization ratio of blast furnaces for sintered return ore cold briquettes includes:

[0032] The data acquisition and preprocessing module is used to perform the steps in step one;

[0033] The parameter state feature generation module is used to execute step two.

[0034] The intelligent decision-making module has a built-in machine learning model for adjusting the proportion of cold-pressed blocks, which is used to execute step three.

[0035] The instruction execution control module is used to execute step four.

[0036] Further improvements include a human-machine interface that displays multi-source process parameter data, parameter state distribution feature vectors, comprehensive feature vectors, optimal cold-pressed block feed ratio, and blast furnace running state prediction information in real time, and receives manual intervention commands.

[0037] The beneficial effects of this invention are as follows: This invention deeply integrates and effectively transforms the expert experience in the field of blast furnace ironmaking with machine learning technology. By quantifying the qualitative judgment of the operator on the furnace condition into a parameter state distribution feature vector based on the calculation of process parameter deviation and state classification, it provides the model with high-quality input features that have both physical meaning and statistical representation, which significantly improves the effectiveness and interpretability of the model input information. It fundamentally improves the common problems of poor feature quality and ambiguous decision basis of traditional data-driven models under complex blast furnace working conditions, making the intelligent decision-making process more in line with production reality.

[0038] Furthermore, this invention constructs an accurate and continuously optimized adaptive intelligent decision-making mechanism. It takes a comprehensive vector that integrates multi-source raw data and parameter state features as input, and uses a machine learning model to continuously learn the complex mapping relationship between the optimal proportion of cold-pressed blocks and real-time operating conditions in the multivariable, strongly coupled, and nonlinear dynamic process of the blast furnace. In particular, by introducing a dynamic weight adjustment mechanism, it can adaptively optimize the contribution of each parameter in the decision-making based on the correlation between the historical state of the parameters and the subsequent evolution of the furnace conditions. Thus, the model not only has high-precision real-time decision-making capabilities, but also has long-term adaptability that can continuously improve and optimize itself with the production process.

[0039] Finally, this invention achieves a complete real-time dynamic closed-loop control from perception, analysis, decision-making to execution, generating clear economic benefits. Under the premise of stable blast furnace operation, this method and system can significantly increase the utilization rate of sintered return ore cold-pressed briquettes and reduce their feed fluctuations, thereby promoting the resource recycling of solid waste. Practical applications show that this invention can effectively stabilize key blast furnace process parameters (such as permeability), reduce fuel consumption, improve utilization coefficient, and promote a rational and green recycling of sintered return ore. Attached Figure Description

[0040] Figure 1 This is a schematic diagram of the process of the blast furnace utilization ratio dynamic control method based on machine learning for sintering return ore cold pressing briquetting according to the present invention. Detailed Implementation

[0041] 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.

[0042] It should be noted that the technical means not described in detail in the following embodiments are all conventional means in the art, are not the key points of the invention, and will not be elaborated upon.

[0043] Example 1

[0044] See Figure 1 This embodiment provides a machine learning-based method for dynamic control of the utilization ratio of sintered return ore cold-pressed briquetting blast furnace. This method is applied to a 3800m³ blast furnace and deployed on a server in the blast furnace main control room. It enables data interaction between the system and the blast furnace's existing distributed control system (DCS), process monitoring system, and raw material analysis database, achieving real-time acquisition of process parameters and issuance of control commands. The data acquisition and command issuance cycle is set to 1 minute, and the specific steps include:

[0045] Step 1: Data Acquisition and Preprocessing

[0046] Real-time acquisition of multi-source process parameter data, totaling 20 core parameters, divided into two main categories:

[0047] Blast furnace operating parameters (14 types): air volume, air temperature, air pressure, oxygen enrichment, blast humidity, furnace top pressure, furnace top temperature (average of four points), charge column pressure difference, charge column permeability index, gas utilization rate (CO2 / (CO+CO2)), fuel ratio, coke ratio, coal ratio, and comprehensive smelting intensity.

[0048] Cold-pressed block parameters (6 types): basicity (CaO / SiO2), compressive strength, particle size (20-50mm proportion), low-temperature reduction pulverization index (RDI+3.15), reduction index (RI) of the cold-pressed blocks to be put into the furnace, and the real-time proportion of the cold-pressed blocks already put into the furnace (as known input parameters).

[0049] Data cleaning: For parameter values ​​that are missing for more than 5 consecutive periods, linear interpolation of the previous and next valid values ​​is used to fill them in; for outliers that jump instantaneously (defined as the difference between the current value and the previous value exceeding 3 times the standard deviation of the parameter’s historical data), the valid value from the previous moment is used to replace them.

[0050] Normalization: The Min-Max normalization method is used to determine the maximum and minimum values ​​of each parameter based on historical data from the past year, and all real-time collected data are linearly mapped to the [0,1] interval (for example, the historical range of air volume is [5800,7200] m). 3 / min, then the air volume at a certain moment is 6500m³. 3 The normalized value of / min is (6500-5800) / (7200-5800)=0.5).

[0051] Step 2: Parameter State Quantization and Feature Generation

[0052] First, assign a "blast furnace smooth operation benchmark value Bi" to each of the 20 parameters in step one. This value is taken from the median of all parameter values ​​from all data within the past six months that were defined as a "stable smooth operation period" of the blast furnace (standard: fuel ratio consistently below 510 kg / t, and permeability index fluctuation rate < 5%). Examples of benchmark values ​​for some key parameters are as follows:

[0053] Air volume (B1): 6500 m³ / min

[0054] Wind temperature (B2): 1200℃

[0055] Differential pressure of the material column (B4): 165 kPa

[0056] Breathability index (B5): 45

[0057] Gas utilization rate (B6): 48.5%

[0058] Fuel ratio (B7): 505 kg / t

[0059] Cold-pressed block alkalinity (B8): 1.8

[0060] Cold-pressed block compressive strength (B9): 2500N

[0061] For any parameter i at time t, its normalized value Vi norm (t) is denormalized to obtain the actual value Vi(t), and its deviation from the reference value Bi is calculated as Di(t):

[0062] Di(t) = |Vi(t) - Bi| / Bi × 100%

[0063] (Note: For indicators such as fuel ratio that are expected to decrease, the absolute value is still taken when calculating the deviation, and the optimization direction is reflected through learning during model training.)

[0064] Classification rules:

[0065] If 0%≤Di(t)<5%, then parameter i is classified as the first state (“excellent”) at time t.

[0066] If 5%≤Di(t)<10%, then parameter i is classified as the second state (“good”) at time t.

[0067] If Di(t)≥10%, then parameter i is classified as the third state ("difference") at time t.

[0068] Weighted statistics and dynamic weight adjustment:

[0069] Initial weight allocation: Based on the importance of the process and expert experience, initial weights Wi(0) are assigned to 20 parameters. The initial weights of key parameters (permeability index, fuel ratio, gas utilization rate, cold block compressive strength) are 1.5; the initial weights of important parameters (air volume, air temperature, air pressure, material column pressure difference) are 1.2; and the initial weights of other general parameters are 1.0.

[0070] Weighted eigenvector generation: At time t, calculate the number of each state parameter after weighting.

[0071] N 优 (t) = ΣWi(t) (sum over all parameters i in the first state)

[0072] N 良 (t) = ΣWi(t) (sum over all parameters i in the second state)

[0073] N 差 (t) = ΣWi(t) (sum over all parameters i in the third state)

[0074] Generate the feature vector of the state distribution of the basic parameters: [N 优 (t),N 良 (t),N 差 (t)];

[0075] Dynamic weight adjustment: The weight is adaptively adjusted once every natural week. The frequency of each parameter i being classified as "poor" (third state) in the past week is calculated, and a Pearson correlation analysis is performed with the rate of change of the average value of the core smooth operation indicator of the blast furnace, "permeability index", relative to the previous week (considered as an indicator of the degree of deterioration). If the correlation coefficient between the "poor class frequency" of a parameter and the "degree of permeability deterioration" is greater than 0.7 (strong positive correlation), it indicates that the deterioration of the parameter has a significant impact on the furnace condition, and its weight Wi is increased by 10%; if the correlation coefficient is less than -0.7, the weight is decreased by 10%. The weight adjustment has an upper limit and a lower limit (e.g., 0.5 to 2.0) to prevent the weight of individual parameters from getting out of control.

[0076] Step 3: Intelligent Proportional Decision Making

[0077] The 20-dimensional normalized original parameter vector obtained in step one is compared with the 3-dimensional weighted parameter state distribution feature vector [N] generated in step two. 优 (t),N 良 (t),N 差 The features are concatenated (t) to form a 23-dimensional comprehensive feature vector X. final (t).

[0078] X final (t) is input into a pre-trained gradient boosting decision tree (GBDT) model for cold-pressed block proportion control, and outputs a value between 0 and 50, which is the optimal cold-pressed block proportion P recommended by the system at the current moment. set (t) (unit: %)

[0079] The model training process is as follows:

[0080] Data preparation: Collect approximately 500,000 valid minute-level historical data points from this blast furnace over the past two years. For each data point, generate its corresponding comprehensive feature vector X using the methods described in steps one and two above. finalhist ;

[0081] Labeling: A team of experts, consisting of three senior operators and process engineers, reviewed historical data to identify periods of continuous and stable blast furnace operation (8 consecutive hours without abnormal furnace alarms). The average proportion of cold-pressed briquettes fed into the furnace during these periods, which achieved the best overall economic benefits (lowest iron cost), was used as the label value P_label for the starting point of those periods. Approximately 30,000 high-quality labeled samples were obtained.

[0082] Model Training and Validation: The GBDT algorithm was implemented using the LightGBM framework. The total samples were randomly divided into training, validation, and test sets in an 8:1:1 ratio. Mean squared error (MSE) was used as the loss function, and hyperparameters were optimized through grid search. The final model parameters were: learning rate 0.05, number of decision trees 500, and maximum tree depth 7. The model's performance metrics on the independent test set were: mean absolute error (MAE) of the predicted proportions of 0.52% and coefficient of determination R0. 2 The accuracy reaches 0.89, meeting the precision requirements for precise on-site control of blast furnaces;

[0083] Step 4: Command Execution and Control

[0084] A 10-minute adjustment cycle is defined. At the beginning of each cycle, the model outputs the recommended ratio P. set Read the actual proportion P fed back by the cold-pressed block batching belt scale in the feeding system. curr ;

[0085] A proportional-integral (PI) controller is used for closed-loop regulation. The input of the controller is the deviation between the setpoint and the actual value, e(t) = P. set -P curr The controller parameters were determined on-site as follows: proportional gain Kp = 0.8, integral time Ti = 15 minutes. The controller calculates and outputs a standard current signal of 4-20mA based on the deviation e(t). This signal directly acts on the variable frequency belt scale under the cold-pressed block batching silo. By linearly adjusting the belt speed, the actual proportion of cold-pressed blocks entering the furnace is smoothly adjusted to the set value P within approximately 30-40 minutes without overshoot. set Nearby, dynamic, closed-loop optimization of the proportion is achieved.

[0086] To objectively evaluate the control method of this embodiment, it was continuously implemented on the 3800m³ blast furnace for three months. By comparing the key production indicators before and after implementation, its effectiveness was verified.

[0087] The average proportion of cold-pressed briquettes fed into the furnace has steadily increased from 3.5% before implementation to 5.2%;

[0088] The hourly fluctuation range of the cold-pressed block ratio was significantly narrowed from ±2.8% before implementation to ±1.1%, and the stability was greatly improved;

[0089] The variance of the blast furnace charge permeability index decreased by 32%, and the furnace condition stability was enhanced.

[0090] The overall fuel ratio of the blast furnace decreased from 508 kg / t to 502 kg / t;

[0091] The blast furnace utilization coefficient increased from 2.45 t / (m³) 3 ·d) Slightly increased to 2.48t / (m 3 ·d);

[0092] The above data shows that the method described in this embodiment successfully achieves precise and adaptive dynamic control of the proportion of cold-pressed briquettes fed into the furnace. While improving the utilization rate of returned ore resources, it also ensures and optimizes the smooth operation of the blast furnace and its economic and technical indicators.

[0093] Example 2

[0094] This embodiment provides a system for implementing the dynamic control method of blast furnace utilization ratio of sintered return ore cold-pressed briquettes provided in Embodiment 1. The system adopts a modular design and is deployed in the industrial server in the blast furnace main control room. It communicates with the blast furnace basic automation system, the Laboratory Information Management System (LIMS), and the charging PLC control system via industrial Ethernet to form a real-time monitoring-decision-control closed loop. The system modules and functions are as follows:

[0095] Data acquisition and preprocessing module

[0096] Function: To perform step one in Example 1;

[0097] Implementation: This module includes a dedicated data interface service that actively reads blast furnace operating parameters such as air volume and air temperature from the DCS real-time database at 1-minute intervals, periodically obtains physicochemical parameters such as chemical composition and strength of cold-pressed blocks from the LIMS database, and reads the proportion of cold-pressed blocks already fed into the furnace from the batching PLC. The module has built-in data cleaning (such as interpolation and outlier removal) and normalization processing subroutines, and publishes the processed standardized data stream to the internal data bus.

[0098] Parameter state feature generation module

[0099] Function: Corresponds to step two in Example 1;

[0100] Implementation: This module receives preprocessed parameter data from the data bus. Its core is a state quantization engine, which internally stores the "forward baseline value" and current weight coefficient of each parameter. The engine calculates the deviation of each parameter in real time and classifies it according to preset thresholds (0-5%, 5-10%, ≥10%). Another statistical subroutine performs weighted counting of parameters in the three states according to weights, generating a three-dimensional feature vector of [excellent, good, poor]. In addition, this module includes a weight management unit that automatically executes a dynamic weight adjustment algorithm at a predetermined period (e.g., weekly) to update the parameter weight library.

[0101] Intelligent decision-making module

[0102] Function: This corresponds to step three in Example 1.

[0103] Implementation: This module is the "brain" of the system. It has a built-in pre-trained machine learning model for regulating the proportion of cold-pressed blocks (LightGBM GBDT model in this embodiment). The module's feature fusion unit concatenates the normalized original parameter vector (20-dimensional) obtained from the data bus with the parameter state distribution feature vector (3-dimensional) received from the feature generation module to form a comprehensive feature vector. The model inference unit loads the comprehensive feature vector, runs the model, and outputs the optimal recommended value for the proportion of cold-pressed blocks entering the furnace. This module also includes a model monitoring interface, which can be used for online updates or model reloading.

[0104] Instruction Execution Control Module

[0105] Function: This corresponds to step four in Example 1.

[0106] Implementation: This module receives the proportional setpoint output from the intelligent decision module. It integrates a digital PI controller, which compares the setpoint with the actual value fed back from the feeding PLC, calculates the control quantity, and converts the control quantity into the corresponding equipment instruction (such as the frequency setpoint of the frequency converter or the valve opening instruction). The instruction is then sent to the PLC of the blast furnace feeding control system through the standard OPC protocol or Modbus TCP protocol, which ultimately drives the belt scale frequency converter or the batching valve to perform the action and complete the proportional adjustment.

[0107] Human-Computer Interface (HMI)

[0108] Function: Provides operators with a monitoring and intervention interface;

[0109] Implementation: Developed using Web technology, the interface is displayed on the main control room's monitoring screen and on engineers' workstations. The main areas of the interface include:

[0110] Real-time data dashboard: Dynamically scrolling display of key process parameters (original values ​​and status classification labels), bar chart of parameter status distribution feature vectors, real-time scale suggestion values ​​and actual execution values ​​output by the model;

[0111] Trend Analysis View: Displays historical trend curves for key indicators such as cold-pressed briquette ratio, fuel ratio, and air permeability index;

[0112] System status monitoring: Displays the operating status of each module, model confidence, communication link status, etc.

[0113] Manual intervention panel: Provides a "model suggestion follow / manual setting" mode switch, allowing experienced operators to manually set the scale under special working conditions, and has operation permission management and operation log recording functions.

[0114] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A machine learning-based method for dynamically controlling the utilization ratio of sintered return ore cold-pressed briquettes in blast furnaces, characterized in that, Includes the following steps: Step 1: In the blast furnace control system, collect multi-source process parameter data related to the smooth operation of the blast furnace in real time, and clean and normalize the collected data. Step 2: For each process parameter in the multi-source process parameter data, calculate the deviation between its current value and the preset blast furnace operating reference value for that parameter; Based on the preset deviation threshold range, each parameter is classified into a first state, a second state, or a third state; the number of parameters classified into each state is counted, and a parameter state distribution feature vector representing the overall operating state of the blast furnace is generated. Step 3: The pre-processed multi-source process parameter data and parameter state distribution feature vector are fused to form a comprehensive feature vector. The comprehensive feature vector is then input into the pre-trained cold briquette ratio control machine learning model, which outputs the optimal cold briquette feed ratio corresponding to the current blast furnace state. Step 4: The output optimal cold-pressed block feeding ratio is converted into a control command for the blast furnace charging system through the controller, and the actual feeding ratio of cold-pressed blocks is dynamically adjusted.

2. The method for dynamic control of blast furnace utilization ratio of sintered return ore cold-pressed briquettes based on machine learning according to claim 1, characterized in that: In step one, the multi-source process parameter data includes: blast furnace operating parameters within a historical set time period, physicochemical parameters of cold-pressed briquettes already in the furnace, and physicochemical parameters of cold-pressed briquettes to be in the furnace. The blast furnace operating parameters are selected from at least six of the following: air volume, air temperature, air pressure, oxygen enrichment, blast humidity, furnace top pressure, furnace top temperature, furnace core temperature, cooling wall water temperature difference, burden column pressure difference, burden column permeability index, utilization coefficient, fuel ratio, coal ratio, coke ratio, gas utilization rate, smelting intensity, and the weight, basicity, drum index, low-temperature reduction pulverization index, and reduction degree index of sinter, pellets, and lump ore in the ore batch. The physicochemical parameters of the cold-pressed briquettes already in the furnace and the cold-pressed briquettes to be in the furnace are selected from at least three of the following: composition, basicity, compressive strength, low-temperature reduction pulverization index, reduction degree index, and particle size.

3. The method for dynamic control of blast furnace utilization ratio of sintered return ore cold-pressed briquettes based on machine learning according to claim 2, characterized in that: The dynamic control range of the furnace feed ratio of the cold-pressed block is 0% to 50%, the alkalinity range of the cold-pressed block is 1.0 to 3.0, the particle size range is 10 to 60 mm, the compressive strength range is 1800 to 5000 N, the low-temperature reduction pulverization index range is 45% to 100%, and the reduction index range is 40% to 100%.

4. The method for dynamic control of blast furnace utilization ratio of sintered return ore cold-pressed briquettes based on machine learning according to claim 1, characterized in that: In step two, the specific rules for classifying parameters according to the degree of deviation are as follows: parameters with a deviation between 0% and 5% are classified as the first state, parameters with a deviation between 5% and 10% are classified as the second state, and parameters with a deviation greater than or equal to 10% are classified as the third state.

5. The method for dynamic control of blast furnace utilization ratio of sintered return ore cold-pressed briquettes based on machine learning according to claim 1, characterized in that: In step two, the process of generating the parameter state distribution feature vector further includes: assigning different initial weight values ​​to parameters of different categories, performing weighted statistics on the number of parameters in each category based on the initial weight values, and dynamically adjusting the weight values ​​according to the correlation between the historical classification results of each parameter and the subsequent smooth operation of the blast furnace.

6. The method for dynamic control of blast furnace utilization ratio of sintered return ore cold-pressed briquettes based on machine learning according to claim 1, characterized in that: In step three, the training process of the machine learning model for controlling the proportion of cold-pressed blocks includes: S1. Obtain a training dataset from the blast furnace history database. The training dataset includes historical multi-source process parameter data, parameter state distribution feature vector generated in step two, and the proportion of cold-pressed blocks fed into the furnace corresponding to the optimal forward operation state of the blast furnace during that historical period, which serves as the label. S2. Preprocess and feature-select the training dataset obtained in step S1; S3. Using the filtered historical multi-source process parameter data and its corresponding parameter state distribution feature vector as input features, and the cold-pressed block feed ratio as the prediction target, a machine learning algorithm is used for supervised training to obtain a machine learning model for cold-pressed block ratio control.

7. The method for dynamic control of blast furnace utilization ratio of sintered return ore cold-pressed briquettes based on machine learning according to claim 6, characterized in that: The machine learning algorithm is selected from at least one of gradient boosting decision tree, random forest, support vector machine, and long short-term memory neural network.

8. The method for dynamic control of blast furnace utilization ratio of sintered return ore cold-pressed briquettes based on machine learning according to claim 1, characterized in that: In step four, the controller is a proportional-integral-derivative controller, a fuzzy logic controller, a neural network controller, or an expert system, and the control command is used to adjust the feeding speed of the conveyor belt scale or the opening of the batching valve.

9. A system for implementing the machine learning-based dynamic control method for the utilization ratio of sintered return ore cold-pressed briquettes in blast furnaces according to any one of claims 1-8, characterized in that, include: The data acquisition and preprocessing module is used to perform the steps in step one; The parameter state feature generation module is used to execute step two. The intelligent decision-making module has a built-in machine learning model for adjusting the proportion of cold-pressed blocks, which is used to execute step three. The instruction execution control module is used to execute step four.

10. A machine learning-based dynamic control system for the utilization ratio of blast furnaces for sintered return ore cold pressing into briquettes, as described in claim 9, is characterized in that: It also includes a human-machine interface, which displays multi-source process parameter data, parameter state distribution feature vector, comprehensive feature vector, optimal cold-pressed block feeding ratio, and blast furnace running state prediction information in real time, and receives manual intervention commands.

Images