A method for online evaluation and intelligent optimization control of oxygen blowing efficiency in converter oxygen lances

Abstract

This invention relates to the field of steelmaking technology in iron and steel metallurgy, specifically to a method for online evaluation and intelligent optimization control of oxygen lance blowing efficiency in a converter. The method includes: inputting pre-processed process data into an evaluation model to obtain an initial oxygen blowing efficiency; analyzing historical process data using a density clustering algorithm to divide it into blowing processes, mapping the pre-processed process data to these processes, and identifying the current blowing process; inputting the pre-processed process data and the current blowing process into an intelligent optimization model, which includes a convolutional neural network and a reinforcement learning network cascaded together. The convolutional neural network generates a condition feature vector based on the pre-processed process data, and the reinforcement learning network has a multi-objective reward function, using the condition feature vector as the state input to calculate the optimal combination of oxygen lance operating parameters. This invention can effectively improve the intelligence level and production efficiency of converter steelmaking.

Description

Technical Field

[0001] This invention relates to the field of steelmaking technology in iron and steel metallurgy, specifically to a method for online evaluation and intelligent optimization control of oxygen blowing efficiency in a converter oxygen lance. Background Technology

[0002] As a core piece of equipment in the steelmaking process, the oxygen lance in a converter directly affects oxygen utilization, smelting cycle, steel quality, and energy consumption. The oxygen lance injects oxygen into the molten pool through a high-speed jet, promoting decarburization and impurity oxidation in the molten iron. The jet characteristics (such as velocity, turbulence, and penetration depth) and the molten pool reaction state (such as slag-metal interface dynamics and bubble distribution) jointly determine the oxygen blowing efficiency. Traditional theory suggests that supersonic jets can enhance molten pool agitation, but excessive jet attenuation or insufficient molten pool reaction can lead to a decrease in oxygen utilization. Therefore, real-time monitoring of the jet's dynamic characteristics and the molten pool response is necessary to achieve precise control of the oxygen blowing efficiency.

[0003] In existing technologies, the assessment of oxygen blowing efficiency in converter oxygen lances largely relies on offline data analysis after smelting, such as back-calculating efficiency using static indicators like oxygen consumption and smelting time. This fails to reflect the dynamic changes in jet characteristics and molten pool reactions in real time, leading to delayed assessments and difficulty in guiding process optimization. Oxygen lance parameter control typically employs methods like adjusting the lance position with a fixed oxygen supply flow rate or adjusting the flow rate with a fixed lance position, relying on manual experience and easily causing the jet to deviate from design conditions, leading to molten pool over-oxidation or splashing. Furthermore, existing jet detection technologies are mostly based on single sensors (such as pressure, temperature, or CO / CO2 concentration), lacking multi-source data fusion analysis of jet velocity fields, temperature fields, and molten pool reactions. This makes it difficult to comprehensively characterize the key factors affecting oxygen blowing efficiency, and the lack of a closed-loop feedback with real-time operating conditions results in insufficient optimization accuracy. It also prevents fault prediction based on real-time changes in oxygen blowing efficiency, increasing the risk of equipment damage. Although existing patented technologies have explored aspects such as jet detection and oxygen utilization improvement, such as non-contact jet measurement systems or dynamic control methods based on CO / CO2, they still lack integrated solutions for real-time evaluation of oxygen blowing efficiency and multi-parameter collaborative optimization, making it difficult to solve the fundamental problems of evaluation lag and passive regulation. Summary of the Invention

[0004] To address the aforementioned technical problems in converter oxygen lance blowing technology, such as lagging oxygen blowing efficiency assessment, passive parameter control, low oxygen utilization, and insufficient intelligent optimization, this invention provides an online assessment and intelligent optimization control method for converter oxygen lance blowing efficiency. This invention primarily utilizes multi-source detection data to achieve online calculation and dynamic tracking of oxygen blowing efficiency, solving the problem of lagging assessment; designs a multi-parameter collaborative intelligent optimization algorithm to achieve adaptive control of oxygen lance position, oxygen supply flow rate, and injection pressure, improving oxygen utilization; establishes a correlation rule between oxygen blowing efficiency and abnormal operating conditions, predicting faults such as nozzle wear and pipeline leakage in advance through efficiency changes, achieving proactive early warning of abnormal operating conditions; realizes a closed-loop linkage between oxygen blowing efficiency assessment and parameter optimization, ensuring that oxygen lance parameters are always matched to the dynamic reaction state of the molten pool in real time, shortening the smelting cycle and reducing production energy consumption; and adapts to steelmaking processes of different converter specifications, improving the method's versatility and practicality, providing technical support for the intelligent upgrading of converter steelmaking.

[0005] The technical means employed in this invention are as follows:

[0006] A method for online evaluation and intelligent optimization control of oxygen blowing efficiency in a converter oxygen lance includes the following steps: Collect process data during the converter smelting process. The process data includes oxygen lance jet characteristic parameters, molten pool state parameters, oxygen concentration, carbon monoxide concentration, carbon dioxide concentration and partial pressure in the furnace gas. The process data is preprocessed; The pretreated process data is input into the evaluation model to obtain the initial oxygen blowing efficiency; Historical process data is analyzed using density clustering algorithms and divided into blowing processes. The preprocessed process data is then mapped to the blowing processes to identify the current blowing process. The preprocessed process data and the current blowing process are input into the intelligent optimization model, which includes a convolutional neural network and a reinforcement learning network connected in sequence. The reinforcement learning network has a multi-objective reward function and uses the working condition feature vector as the state input to calculate the optimal combination of oxygen lance operating parameters. The initial oxygen blowing efficiency is used as the core variable of the multi-objective reward function to calculate the reward value after the action is executed.

[0007] Furthermore, the formula for calculating the initial oxygen blowing efficiency is as follows: η = α˙η1 + β˙η2 + γ˙η3 Where α+β+γ=1, η is the initial oxygen blowing efficiency, α is the weighting coefficient of jet utilization efficiency, β is the weighting coefficient of molten pool reaction efficiency, γ is the weighting coefficient of oxygen utilization rate, η1 is the jet utilization efficiency, η2 is the molten pool reaction efficiency, and η3 is the oxygen utilization rate.

[0008] Furthermore, the multi-objective reward function is defined as maximizing oxygen blowing efficiency, minimizing oxygen consumption, and shortening the smelting cycle. The calculation formula for the multi-objective reward function is as follows: R=λ1 R η +λ2 R O2 +λ3 R t Where R is the multi-objective reward function, λ1 is the weighting coefficient of the oxygen blowing efficiency reward, and R η The oxygen blowing efficiency bonus is represented by λ2, which is the weighting coefficient for the oxygen consumption bonus. R O2 The reward is for oxygen consumption, λ3 is the weighting coefficient for the smelting cycle reward, and R is the oxygen consumption reward. t The reward for the smelting cycle is λ1+λ2+λ3=1.

[0009] Furthermore, the reinforcement learning network uses the oxygen lance position, oxygen supply flow rate, blowing pressure, and gas ratio as the decision action space, and makes real-time decisions through the Q-Learning algorithm to output the optimal parameter combination.

[0010] Furthermore, the blowing process is divided into four stages according to the carbon content of the molten pool: slag formation stage, decarburization stage, slag adjustment stage, and endpoint control stage. When the carbon content in the molten pool exceeds the first threshold, it is determined to be in the slag formation stage; When the carbon content in the molten pool is greater than the second threshold and less than or equal to the first threshold, it is determined to be in the decarburization stage; When the carbon content in the molten pool is greater than the third threshold and less than or equal to the second threshold, it is determined to be in the slag adjustment stage; When the carbon content in the molten pool is less than or equal to the third threshold, it is determined to be in the endpoint control stage.

[0011] Furthermore, the method also includes: calculating the oxygen blowing efficiency in real time; when the oxygen blowing efficiency is greater than or equal to the target value, maintaining the current optimal oxygen lance operating parameter combination; when the oxygen blowing efficiency is less than the target value, the intelligent optimization model re-derives the optimal oxygen lance operating parameter combination based on the adjusted data, until the oxygen blowing efficiency is greater than or equal to the target value.

[0012] Furthermore, the method also includes: A rule base for correlation between oxygen blowing efficiency trends and equipment failures is established. By analyzing oxygen blowing efficiency, abnormal operating conditions can be predicted in advance. The rule base includes the following prediction rules: When the oxygen blowing efficiency drops below 75%, the decarburization rate decreases by more than 20%, and the molten pool heating rate decreases by more than 15% simultaneously, it is predicted to be due to wear of the oxygen lance nozzle. When the oxygen blowing efficiency decreases by ≥18% compared to the previous furnace cycle, and the oxygen concentration in the furnace gas increases by more than 5%, it is predicted that there is a leak in the oxygen supply pipeline. When oxygen blowing efficiency fluctuates drastically or the intensity of molten pool stirring is abnormal, it is considered a risk of molten pool splashing. The significant fluctuation in oxygen blowing efficiency is determined based on the standard deviation of the oxygen blowing efficiency, and the formula for calculating the standard deviation is as follows:

[0013] in, The standard deviation of oxygen blowing efficiency. n The number of time steps within the window. η i Let be the oxygen blowing efficiency value at the i-th time step. The standard deviation is the arithmetic mean of the oxygen blowing efficiency within the window. When the standard deviation is greater than or equal to 0.4 and lasts for more than 25 seconds, it is considered that the oxygen blowing efficiency fluctuates drastically. The abnormal molten pool stirring intensity is calculated based on the molten pool stirring intensity, and the calculation formula for the molten pool stirring intensity is:

[0014] in, The intensity of the molten pool stirring. Oxygen flow rate, This refers to the impact depth of the oxygen lance jet. This is the weight of the molten steel. For the blowing time, when the stirring intensity of the molten pool is less than 70% of the target value, it is judged as too weak; when the stirring intensity of the molten pool is greater than 130% of the target value, it is judged as too strong. When any of the following is detected: oxygen lance nozzle wear, oxygen supply pipeline leakage, risk of molten pool splashing, or excessive stirring intensity, the system immediately issues an early warning signal to achieve proactive fault prevention and control.

[0015] Furthermore, the method also includes: A self-iterative update system for the model is established, which stores the process data, oxygen blowing efficiency, optimal oxygen lance operation parameter combination and smelting effect of each smelting process. The newly stored data is input into the intelligent evaluation model for model retraining, and the model parameters of the convolutional neural network and the weight parameters and multi-objective reward function of the reinforcement learning network are optimized.

[0016] Compared with the prior art, the present invention has the following advantages: This invention achieves real-time quantitative evaluation of oxygen blowing efficiency, breaking through the traditional offline evaluation mode and enabling real-time tracking of dynamic changes in efficiency during the blowing process. Through multi-parameter collaborative intelligent optimization, it significantly improves oxygen utilization and oxygen blowing efficiency, shortens the smelting cycle, reduces production energy consumption, and enhances molten pool reaction efficiency. An abnormal operating condition prediction mechanism provides accurate early warnings for faults such as nozzle wear, pipeline leakage, and molten pool splashing, improving production stability and reducing the probability of abnormal operating conditions. It replaces manual experience-based control, reducing the labor intensity of operators. The detection module seamlessly integrates with existing converter production systems, requiring no large-scale equipment modifications, exhibiting strong compatibility, easy industrialization, and low cost.

[0017] Based on the above reasons, this invention can be widely promoted in the fields of iron and steel metallurgy and steelmaking. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating the online evaluation and intelligent optimization control method for oxygen blowing efficiency of a converter oxygen lance according to the present invention.

[0020] Figure 2 This is a schematic diagram of the unit used for data acquisition in this invention.

[0021] Figure 3 This is a schematic diagram of the evaluation and optimization closed-loop control system in this invention.

[0022] Figure 4 This is a diagram of the convolutional neural network architecture in this invention. Detailed Implementation

[0023] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. 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 should fall within the scope of protection of the present invention.

[0024] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0025] This invention employs a multi-source data acquisition, oxygen blowing efficiency quantitative evaluation, dynamic operating condition identification, multi-parameter collaborative optimization, closed-loop control execution, and abnormal operating condition early warning process. By integrating data from non-contact jet detection, furnace gas composition detection, and molten pool status detection, an oxygen blowing efficiency evaluation model is established. Combined with an improved convolutional neural network and reinforcement learning (CNN-RL) hybrid algorithm, intelligent optimization of oxygen lance parameters is achieved. At the same time, a fault prediction mechanism is established to realize intelligent control of the entire blowing process.

[0026] like Figure 1 As shown, this invention provides a method for online evaluation and intelligent optimization control of oxygen blowing efficiency in a converter oxygen lance, comprising the following steps: S1. Collect process data during the converter smelting process. The process data includes oxygen lance jet characteristic parameters, molten pool state parameters, oxygen concentration, carbon monoxide concentration, carbon dioxide concentration and partial pressure in the furnace gas.

[0027] like Figure 2 As shown, a multi-dimensional detection system was built to synchronously collect core data during the blowing process, with a collection frequency of ≥10kHz, to achieve data transmission and preprocessing without delay.

[0028] Jet characteristic detection unit: A non-contact detection unit composed of a laser Doppler velocimeter, a particle image velocimeter, and an infrared thermal imager is used to collect oxygen lance jet characteristics such as jet velocity field, temperature field, included angle, and turbulence intensity.

[0029] Furnace gas composition detection unit: Employs a high-temperature extraction mass spectrometer and an infrared gas analyzer to detect the concentrations and partial pressures of O2, CO, CO2, and SO2 in the furnace gas in real time.

[0030] Molten pool condition detection unit: It detects the molten pool condition in converter steelmaking by collecting data on the stirring intensity and reaction intensity of the molten pool through furnace vibration sensors and sound sensors, combined with the molten pool temperature collected by thermocouples.

[0031] The parameters of the molten pool state include: molten pool carbon content, molten pool temperature, molten pool heating rate, decarburization rate, molten pool stirring intensity, and molten steel weight.

[0032] Oxygen lance operation detection unit: Collects real-time oxygen lance position, oxygen supply flow rate, blowing pressure, cooling water flow rate and other oxygen lance operation and equipment operating parameters.

[0033] The oxygen lance jet characteristics include: oxygen flow rate, oxygen supply intensity, oxygen lance jet impact depth, oxygen lance position, oxygen supply flow rate, blowing pressure, and gas ratio.

[0034] Oxygen lance operation parameter acquisition unit: responsible for acquiring the above-mentioned oxygen lance related parameters.

[0035] The feedback arrow pointing from the data preprocessing module to the molten pool condition detection unit is used to feed back the preprocessed molten pool condition characteristics and operating condition identification results to the molten pool condition detection unit, thereby realizing closed-loop correction of molten pool temperature, liquid level, pressure and other state parameters, eliminating errors caused by sensor noise and detection delay, and improving the accuracy of condition detection and the robustness of operating condition identification.

[0036] S2. Preprocess the process data.

[0037] Data preprocessing module: Preprocesses the collected data by interpolating missing values, removing outliers, and standardizing data to eliminate the impact of abnormal data and provide a high-quality data foundation for oxygen efficiency assessment.

[0038] S3. Input the pretreated process data into the evaluation model to obtain the initial oxygen blowing efficiency.

[0039] Breaking away from the traditional offline evaluation model, the formula for calculating the initial oxygen blowing efficiency is as follows: η = α·η1 + β·η2 + γ·η3 Where η is the initial oxygen blowing efficiency, α is the weighting coefficient of jet utilization efficiency, β is the weighting coefficient of molten pool reaction efficiency, γ is the weighting coefficient of oxygen utilization rate, η1 is the jet utilization efficiency, η2 is the molten pool reaction efficiency, and η3 is the oxygen utilization rate. Where α+β+γ=1, the weighting coefficients are determined by the analytic hierarchy process combined with the actual smelting process. The jet utilization efficiency is the percentage of the total energy of the supersonic jet ejected from the oxygen lance nozzle that is actually transferred to the molten pool for stirring, decarburization, and slag formation. Its calculation formula is:

[0040] in, K For comprehensive correction factors, v c The axial velocity of the jet when it reaches the surface of the molten pool. v 0 represents the oxygen lance nozzle exit velocity.β The coefficient representing the influence of the jet angle. θ This is the actual jet angle. θ 0 represents the nozzle's designed jet angle. γ The turbulence intensity influence coefficient is... I The turbulence intensity at the jet exit. I 0 represents the reference turbulence intensity for an ideal smooth nozzle.

[0041] Molten pool reaction efficiency η 2 represents the percentage of oxygen that actually participates in beneficial metallurgical reactions such as carbon, silicon, manganese, and phosphorus in the oxygen entering the molten pool:

[0042] in, r c This represents the actual carbon-oxygen reaction rate. r c,max This represents the maximum oxygen reaction rate. dT / dt φ(CO) / φ(CO2) is the rate of change of the molten pool temperature, and φ(CO) / φ(CO2) is the volume ratio of carbon monoxide to carbon dioxide in the furnace gas.

[0043] The formula for calculating oxygen utilization rate is:

[0044] in, The effective oxygen consumption for participating in metallurgical reactions such as decarburization, desiliconization, demanganese removal, and dephosphorization. This represents the total amount of oxygen entering the converter. Oxygen utilization rate characterizes the overall efficiency of oxygen utilization; the higher the value, the less oxygen is wasted, and the better the economics of the blowing process.

[0045] The oxygen blowing efficiency is classified according to the following standards: Excellent: η≥90%, Good: 80%≤η<90%, Medium: 60%≤η<80%, Poor: η<60%, so as to realize the real-time quantification and grade determination of oxygen blowing efficiency.

[0046] S4. Analyze historical process data using density clustering algorithm, divide it into blowing processes, map the preprocessed process data to the blowing processes, and identify the current blowing process.

[0047] A digital twin model of the converter blowing process was established, and real-time collected data was mapped to the digital twin to restore the dynamic process of oxygen jet from the oxygen lance and reaction in the molten pool. The blowing conditions were divided by density clustering algorithm, and the blowing process was divided into slag formation stage, decarburization stage, slag adjustment stage and endpoint control stage. A corresponding oxygen blowing efficiency target value was set for each condition, which solved the problem of traditional condition identification relying on manual labor and having low accuracy.

[0048] When the carbon content in the molten pool exceeds the first threshold, it is determined to be in the slag formation stage. The first threshold is approximately 3.0%.

[0049] When the carbon content in the molten pool is greater than the second threshold and less than or equal to the first threshold, it is determined to be in the decarburization stage. The second threshold is 0.6% to 1.0%.

[0050] When the carbon content in the molten pool is greater than the third threshold and less than or equal to the second threshold, it is determined to be in the slag adjustment stage. The third threshold is the sum of the target carbon content and the allowable deviation of the carbon content. The target carbon content is 0.05% to 0.15%.

[0051] When the carbon content in the molten pool is less than or equal to the third threshold, it is determined to be the endpoint control stage, and the temperature is close to the target temperature range.

[0052] S5. Input the preprocessed process data and the current blowing process into the intelligent optimization model. The intelligent optimization model includes a convolutional neural network and a reinforcement learning network connected in sequence. The convolutional neural network generates a working condition feature vector based on the preprocessed process data. The reinforcement learning network has a multi-objective reward function and uses the working condition feature vector as the state input to calculate the optimal combination of oxygen lance operating parameters. The initial oxygen blowing efficiency is used as the core variable of the multi-objective reward function to calculate the reward value after the action is executed.

[0053] The intelligent optimization module enables collaborative adaptive optimization of multiple parameters of the oxygen lance, overcoming the limitations of low optimization accuracy and poor real-time performance of existing single-model optimization.

[0054] Convolutional neural networks automatically extract the core influencing factors of oxygen blowing efficiency (such as jet angle, CO / CO2 ratio, and molten pool stirring intensity) and generate operating condition feature vectors. For example... Figure 4 As shown, the convolutional neural network (CNN) used in this invention adopts a classic feature extraction architecture of 3 convolutional layers + 2 fully connected layers, which is adapted to feature extraction of multi-dimensional data in converter processes. Based on the preprocessed process data, a working condition feature vector is generated. The specific calculation steps for the convolutional neural network (CNN) to generate the working condition feature vector are as follows: The first step is to organize the preprocessed process data into a 1×N one-dimensional input vector and input it into the CNN input layer. The preprocessed process data has been standardized to [0,1], with no outliers or missing values.

[0055] Step 2: Shallow Feature Extraction (First Convolutional Layer + First Pooling Layer): The first convolutional layer performs a convolution operation on the input vector using a 3×1 convolution kernel, calculated as: Conv1(x) = ReLU(ω1) x+b1), where ω1 is the kernel weight of the first convolutional layer and b1 is the bias term of the first convolutional layer, extracting the basic numerical features of the input data.

[0056] The first pooling layer performs max pooling on the feature map output by the first convolutional layer, retaining the maximum value within each pooling window and removing redundant information to obtain a shallow feature vector with compressed dimensions.

[0057] Step 3: Extraction of mid-level correlation features (second convolutional layer + second pooling layer): The second convolutional layer uses 64 3×1 convolutional kernels to perform secondary convolution on the shallow feature vectors. The calculation formula is: Conv2(x) = ReLU(ω2). Conv1_out+b2) is used to explore the preliminary correlation between different process parameters. Here, ω2 is the kernel weight of the second convolutional layer, b2 is the bias term of the second convolutional layer, and Conv1_out is the output of the first pooling layer.

[0058] The second pooling layer performs max pooling again to further compress the feature dimensions, resulting in the mid-layer correlation feature vector.

[0059] Step 4: Deep Core Feature Extraction (Third Convolutional Layer): The third convolutional layer uses 128 3×1 convolutional kernels to perform three convolutions on the feature vectors from the middle layer. The calculation formula is: Conv3(x) = ReLU(ω3) Conv2_out+b3) is used to extract the core collaborative features that affect oxygen blowing efficiency. Here, ω3 is the kernel weight of the third convolutional layer, b3 is the bias term of the third convolutional layer, and Conv2_out is the output of the second pooling layer.

[0060] Step 5: Feature vector flattening and fusion (first fully connected layer): The two-dimensional feature map output from the third convolutional layer is flattened into a one-dimensional feature vector, which is then input into the first fully connected layer. The calculation formula is: FC1(x) = ReLU(ω4) Conv3_out_flatten+b4) achieves deep fusion of all extracted features, where Conv3_out_flatten is the flattened feature, ω4 is the convolution kernel weight of the first fully connected layer, and b4 is the bias term of the first fully connected layer.

[0061] Step 6: Output of operating condition feature vector (second fully connected layer): The second fully connected layer performs a linear transformation on the output of the first fully connected layer 1, calculated as: FC2(x) = ω5 FC1_out+b5 outputs a 1×128 feature vector of the working conditions. This vector contains the core features of all process data and is directly input into the reinforcement learning network as a state input for calculating the optimal combination of oxygen lance operating parameters. Here, ω5 is the convolution kernel weight of the second fully connected layer, b5 is the bias term of the second fully connected layer, and FC1_out is the output of the first fully connected layer.

[0062] The reinforcement learning network uses maximizing oxygen blowing efficiency, minimizing oxygen consumption, and shortening the smelting cycle as multi-objective reward functions. The oxygen lance position, oxygen supply flow rate, blowing pressure, and gas ratio are used as the action space, and the Q-Learning algorithm makes real-time decisions to output the optimal parameter combination.

[0063] The formula for calculating the multi-objective reward function is: R=λ1 R η +λ2 R O2 +λ3 R t Where R is the multi-objective reward function, λ1 is the weighting coefficient of the oxygen blowing efficiency reward, and R η The oxygen blowing efficiency bonus is represented by λ2, which is the weighting coefficient for the oxygen consumption bonus. R O2 The reward is for oxygen consumption, λ3 is the weighting coefficient for the smelting cycle reward, and R is the oxygen consumption reward. t The smelting cycle bonus is λ1 + λ2 + λ3 = 1. A higher oxygen blowing efficiency bonus is better, and its calculation formula is: R η =Δ[C] / V O2 Δ[C] represents the change in carbon content, V O2 This refers to oxygen consumption. The smaller the oxygen consumption reward, the better; it should be constructed as a positive reward. The formula for calculating the oxygen consumption reward is: R O2 =1 V O2 / V O2, max V O2, max This is for maximizing oxygen consumption. The shorter the smelting cycle reward, the better; it should be constructed as a positive reward system. The formula for calculating the smelting cycle reward is: R t =1 t / t max t is the smelting time, t max This is the maximum smelting time.

[0064] As a preferred embodiment of the present invention, a built-in supersonic jet design constraint is incorporated to ensure that the main oxygen flow is always in a supersonic state, thereby preventing the deterioration of jet characteristics.

[0065] like Figure 3 As shown, the present invention further includes the following steps: S6. Calculate oxygen blowing efficiency in real time. When the oxygen blowing efficiency is greater than or equal to the target value, maintain the current optimal combination of oxygen lance operating parameters. When the oxygen blowing efficiency is less than the target value, the intelligent optimization model re-derives the optimal combination of oxygen lance operating parameters based on the adjusted data until the oxygen blowing efficiency is greater than or equal to the target value.

[0066] Specifically, the preferred range for oxygen blowing efficiency is 85% to 95%, with the optimal value being 90%.

[0067] S7. Establish a rule base for the correlation between oxygen blowing efficiency trends and equipment failures. By analyzing the continuous variation characteristics of oxygen blowing efficiency, abnormal operating conditions can be predicted in advance. The rule base includes the following prediction rules: When the oxygen blowing efficiency continues to decline and the jet angle exceeds the preset threshold, it is predicted that the oxygen lance nozzle is worn.

[0068] Specifically, if the cumulative decrease in oxygen blowing efficiency exceeds the threshold within a continuous period and there is no upward trend, such as when the oxygen blowing efficiency drops from ≥90% to <75%; the decarburization rate decreases by more than 20%; and the molten pool heating rate decreases by more than 15%, it is determined to be oxygen lance nozzle wear.

[0069] When the oxygen blowing efficiency drops sharply and the O2 concentration in the furnace gas rises sharply, it is predicted that there is a leak in the oxygen supply pipeline.

[0070] Specifically, when the oxygen blowing efficiency decreases by ≥18% compared to the previous furnace cycle, and the oxygen concentration in the furnace gas increases by more than 5%, it is determined that there is a leak in the oxygen supply pipeline.

[0071] When oxygen blowing efficiency fluctuates drastically and the stirring intensity of the molten pool is abnormal, it is predicted to be a risk of molten pool splashing.

[0072] The significant fluctuation in oxygen blowing efficiency is determined based on the standard deviation of the oxygen blowing efficiency. The formula for calculating the standard deviation is:

[0073] in, The standard deviation of oxygen blowing efficiency. n The number of time steps within the window. η i Let be the oxygen blowing efficiency value at the i-th time step. The arithmetic mean of the oxygen blowing efficiency within the window is used. When the standard deviation is greater than or equal to 0.4 and lasts for more than 25 seconds, it is judged as a violent fluctuation in oxygen blowing efficiency.

[0074] Abnormal molten pool stirring intensity is based on calculations of molten pool stirring intensity. The formula for calculating molten pool stirring intensity is:

[0075] in, The intensity of the molten pool stirring. Oxygen flow rate, This refers to the impact depth of the oxygen lance jet. This is the weight of the molten steel. For the blowing time, when the stirring intensity of the molten pool is less than 70% of the target value, it is judged as too weak; when the stirring intensity of the molten pool is greater than 130% of the target value, it is judged as too strong. Upon detecting any of the above abnormal trends, the system immediately issues an early warning signal to proactively prevent and control faults. It also outputs an emergency parameter adjustment plan to further facilitate proactive fault prevention and control.

[0076] S8. Establish a self-iterative update system for the model, storing process data, oxygen blowing efficiency, optimal oxygen lance operating parameter combinations, and smelting results for each smelting process. Newly stored data is then input into the intelligent evaluation model for retraining, optimizing the model parameters of the convolutional neural network and the multi-objective reward function of the reinforcement learning network. This ensures the model continuously adapts to changes in the production process, improving long-term optimization accuracy.

[0077] Example 1 This embodiment provides an online evaluation and intelligent optimization control method for the oxygen blowing efficiency of the converter oxygen lance during the slag formation stage (blowing 0-3 min). Using a 180-ton top-and-bottom combined blowing converter steelmaking production system as the application object, all embodiments and comparative examples use the same molten iron raw materials (molten iron containing 4.5% carbon, 0.10% phosphorus, and a temperature of 1300℃), auxiliary material ratios (lime, dolomite), and smelting targets (converter endpoint carbon content of 0.05% and a temperature of 1650℃). The embodiments include the following steps: S1. Collect jet velocity field: velocity is 800m / s, furnace gas CO / CO2 ratio is 0.8, and molten pool temperature is 1350℃.

[0078] S2. The above-collected data is processed by imputing missing values, removing outliers, and standardizing.

[0079] S3. The evaluation model calculates the initial oxygen blowing efficiency. The preprocessed data is input into the evaluation model, and the initial oxygen blowing efficiency is 78%, which is judged as the medium category. The target of this embodiment is ≥85%.

[0080] S4. Analyze historical process data using density clustering algorithms to identify the current slag formation stage.

[0081] S5. Input the preprocessed data and the slag-forming stage data into the intelligent optimization model. The CNN generates a feature vector of the working condition. The reinforcement learning network uses the initial oxygen blowing efficiency as the core variable of the multi-objective reward function to calculate the optimal combination of oxygen lance operating parameters: lance position raised by 50mm, oxygen supply flow rate adjusted to 36000m³. 3 / h, the blowing pressure is adjusted to 0.8MPa.

[0082] S6. Implement closed-loop control. After execution, the oxygen blowing efficiency increases to 88% within 3 seconds, which is judged as good and remains stable; the slag formation time is shortened by 0.5 minutes, there is no back-drying or slag formation, and the oxygen utilization rate is increased by 9%.

[0083] Example 2 This embodiment also provides an online evaluation and intelligent optimization control method for the oxygen blowing efficiency of the converter oxygen lance during the decarburization stage (3-8 min of blowing). Using a 180-ton top-and-bottom combined blowing converter steelmaking production system as the application object, all embodiments and comparative examples use the same molten iron raw materials (molten iron containing 4.5% carbon, 0.10% phosphorus, and a temperature of 1300℃), auxiliary material ratios (lime, dolomite), and smelting targets (converter endpoint carbon content of 0.05% and a temperature of 1650℃). The embodiments include the following steps: S1. Collect relevant data such as furnace gas CO concentration (25%) and molten pool stirring intensity.

[0084] S2. The above-collected data is processed by imputing missing values, removing outliers, and standardizing.

[0085] S3. The evaluation model calculates the initial oxygen blowing efficiency. The pre-processed data is input into the evaluation model, and the initial oxygen blowing efficiency is 86%, which is judged as good. The target of this embodiment is ≥90%.

[0086] S4. Density clustering identifies the current blowing process, identifies that it is currently in the decarburization stage, and determines that the stirring intensity of the molten pool is insufficient and the CO concentration of the furnace gas is low.

[0087] S5. Input the preprocessed data and the decarbonization stage data into the intelligent evaluation model, using the initial oxygen blowing efficiency as the core variable of the reward function, and output the optimal parameters: lower the nozzle position by 30mm and adjust the oxygen supply flow rate to 38000m³. 3 / h, the blowing pressure is adjusted to 0.9MPa.

[0088] S6. After implementation, the oxygen blowing efficiency increased to 92%, which was rated as excellent; the carbon-oxygen reaction rate increased by 12%; the decarburization time was shortened by 1 minute; and the oxygen consumption per furnace was reduced by 4 m³. 3 / t, improving the uniformity of molten pool temperature.

[0089] Example 3 This embodiment also provides an online evaluation and intelligent optimization control method for converter oxygen lance blowing efficiency based on anomaly prediction. Taking a 180-ton top-and-bottom combined blowing converter steelmaking production system as the application object, all embodiments and comparative examples use the same molten iron raw materials (molten iron containing 4.5% carbon, 0.10% phosphorus, and a temperature of 1300℃), auxiliary material ratios (lime, dolomite), and smelting targets (converter endpoint carbon content of 0.05% and a temperature of 1650℃). The embodiments include the following steps: S1. The system detected that the oxygen blowing efficiency decreased continuously for 10 seconds (from 91% to 82%), and the jet angle increased from 10° to 16°.

[0090] S2. Preprocess the data on the downward trend and the change in the jet angle.

[0091] S3. The current oxygen blowing efficiency continues to decline to 82%.

[0092] S4. Density clustering identifies the current blowing process and identifies it as being in the decarburization stage, but anomaly prediction is made by combining it with the association rule base (this can be regarded as a further subdivision of the state of the blowing process).

[0093] S5. The anomaly prediction module determines the issue as oxygen lance nozzle wear based on the associated rule base and outputs emergency optimization parameters: increase the blowing pressure by 0.05 MPa and fine-tune the oxygen supply flow rate to 39000 m³ / h. 3 / h (At this point, the CNN-RL model uses the current reduced oxygen blowing efficiency as the core variable of the reward function and outputs the adjustment action).

[0094] S6. After execution, the oxygen blowing efficiency is restored to 88%, preventing further deterioration of nozzle wear, enabling early prediction of nozzle wear, continuing smelting, and reducing production interruption losses.

[0095] The online evaluation and intelligent optimization control method for oxygen blowing efficiency of converter oxygen lance of the present invention has significant advantages over existing technologies such as traditional manual control, single parameter optimization and big data parameter recommendation in terms of core indicators such as oxygen blowing efficiency, oxygen utilization rate, smelting cycle and oxygen consumption. At the same time, it has the ability to predict abnormal working conditions, which can effectively improve the level of intelligence and production efficiency of converter steelmaking.

[0096] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0097] In the above embodiments of the present invention, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0098] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for online evaluation and intelligent optimization control of oxygen blowing efficiency in a converter oxygen lance, characterized in that, Includes the following steps: Collect process data during the converter smelting process. The process data includes oxygen lance jet characteristic parameters, molten pool state parameters, oxygen concentration, carbon monoxide concentration, carbon dioxide concentration and partial pressure in the furnace gas. The process data is preprocessed; The pretreated process data is input into the evaluation model to obtain the initial oxygen blowing efficiency; Historical process data is analyzed using density clustering algorithms and divided into blowing processes. The preprocessed process data is then mapped to the blowing processes to identify the current blowing process. The preprocessed process data and the current blowing process are input into the intelligent optimization model, which includes a convolutional neural network and a reinforcement learning network connected in sequence. The reinforcement learning network has a multi-objective reward function and uses the working condition feature vector as the state input to calculate the optimal combination of oxygen lance operating parameters. The initial oxygen blowing efficiency is used as the core variable of the multi-objective reward function to calculate the reward value after the action is executed.

2. The method for online evaluation and intelligent optimization control of oxygen blowing efficiency in a converter oxygen lance according to claim 1, characterized in that, The formula for calculating the initial oxygen blowing efficiency is: η = α˙η1 + β˙η2 + γ˙η3 Where α+β+γ=1, η is the initial oxygen blowing efficiency, α is the weighting coefficient of jet utilization efficiency, β is the weighting coefficient of molten pool reaction efficiency, γ is the weighting coefficient of oxygen utilization rate, η1 is the jet utilization efficiency, η2 is the molten pool reaction efficiency, and η3 is the oxygen utilization rate.

3. The method for online evaluation and intelligent optimization control of oxygen blowing efficiency in a converter oxygen lance according to claim 1, characterized in that, The multi-objective reward function is defined as maximizing oxygen blowing efficiency, minimizing oxygen consumption, and shortening the smelting cycle. The calculation formula for the multi-objective reward function is as follows: R=λ1 R η +λ2 R O2 +λ3 R t Wherein, R is a multi-objective reward function, λ1 is a weight coefficient of oxygen blowing efficiency reward, R η is the oxygen blowing efficiency reward, λ2 is a weight coefficient of oxygen consumption reward, R O2 is the oxygen consumption reward, λ3 is a weight coefficient of smelting cycle reward, R t is the smelting cycle reward, λ1+λ2+λ3=1.

4. The method for online evaluation and intelligent optimization control of oxygen blowing efficiency of converter oxygen lance according to claim 1, characterized in that, The reinforcement learning network uses the oxygen lance position, oxygen supply flow rate, blowing pressure, and gas ratio as the decision action space, and makes real-time decisions through the Q-Learning algorithm to output the optimal parameter combination.

5. The method for online evaluation and intelligent optimization control of oxygen blowing efficiency in a converter oxygen lance according to claim 1, characterized in that, The blowing process is divided into four stages according to the carbon content of the molten pool: slag formation stage, decarburization stage, slag adjustment stage, and endpoint control stage. When the carbon content in the molten pool exceeds the first threshold, it is determined to be in the slag formation stage; When the carbon content in the molten pool is greater than the second threshold and less than or equal to the first threshold, it is determined to be in the decarburization stage; When the carbon content in the molten pool is greater than the third threshold and less than or equal to the second threshold, it is determined to be in the slag adjustment stage; When the carbon content in the molten pool is less than or equal to the third threshold, it is determined to be in the endpoint control stage.

6. The method for online evaluation and intelligent optimization control of oxygen blowing efficiency in a converter oxygen lance according to claim 1, characterized in that, The method further includes: calculating the oxygen blowing efficiency in real time; when the oxygen blowing efficiency is greater than or equal to the target value, maintaining the current optimal oxygen lance operating parameter combination; when the oxygen blowing efficiency is less than the target value, the intelligent optimization model re-derives the optimal oxygen lance operating parameter combination based on the adjusted data, until the oxygen blowing efficiency is greater than or equal to the target value.

7. The method for online evaluation and intelligent optimization control of oxygen blowing efficiency in a converter oxygen lance according to claim 1, characterized in that, The method further includes: A rule base for correlation between oxygen blowing efficiency trends and equipment failures is established. By analyzing oxygen blowing efficiency, abnormal operating conditions can be predicted in advance. The rule base includes the following prediction rules: When the oxygen blowing efficiency drops below 75%, the decarburization rate decreases by more than 20%, and the molten pool heating rate decreases by more than 15% simultaneously, it is predicted to be due to wear of the oxygen lance nozzle. When the oxygen blowing efficiency decreases by ≥18% compared to the previous furnace cycle, and the oxygen concentration in the furnace gas increases by more than 5%, it is predicted that there is a leak in the oxygen supply pipeline. When oxygen blowing efficiency fluctuates drastically or the intensity of molten pool stirring is abnormal, it is considered a risk of molten pool splashing. The significant fluctuation in oxygen blowing efficiency is determined based on the standard deviation of the oxygen blowing efficiency, and the formula for calculating the standard deviation is as follows: in, The standard deviation of oxygen blowing efficiency. n The number of time steps within the window. η i Let be the oxygen blowing efficiency value at the i-th time step. The standard deviation is the arithmetic mean of the oxygen blowing efficiency within the window. When the standard deviation is greater than or equal to 0.4 and lasts for more than 25 seconds, it is considered that the oxygen blowing efficiency fluctuates drastically. The abnormal molten pool stirring intensity is calculated based on the molten pool stirring intensity, and the calculation formula for the molten pool stirring intensity is: in, The intensity of the molten pool stirring. Oxygen flow rate, This refers to the impact depth of the oxygen lance jet. This is the weight of the molten steel. For the blowing time, when the stirring intensity of the molten pool is less than 70% of the target value, it is judged as too weak; when the stirring intensity of the molten pool is greater than 130% of the target value, it is judged as too strong. When any of the following is detected: oxygen lance nozzle wear, oxygen supply pipeline leakage, risk of molten pool splashing, or excessive stirring intensity, the system immediately issues an early warning signal to achieve proactive fault prevention and control.

8. The method for online evaluation and intelligent optimization control of oxygen blowing efficiency in a converter oxygen lance according to claim 1, characterized in that, The method further includes: A self-iterative update system for the model is established, which stores the process data, oxygen blowing efficiency, optimal oxygen lance operation parameter combination and smelting effect of each smelting process. The newly stored data is input into the intelligent evaluation model for model retraining, and the model parameters of the convolutional neural network and the weight parameters and multi-objective reward function of the reinforcement learning network are optimized.

Images