Silicon-manganese alloy dynamic regulation and control method and system in converter direct feeding process and storage medium

Abstract

The invention relates to the technical field of converter smelting, and discloses a silicon-manganese alloy dynamic regulation and control method and system in a converter direct feeding process and a storage medium. The method comprises the following steps: acquiring molten iron components, temperature and furnace gas data, carrying out timestamp alignment and Kalman filtering to obtain synchronized smelting state data, substituting the silicon-manganese concentration, the temperature and the iron oxide content in slag into a rate equation, and carrying out numerical integration to obtain a dynamic oxidation loss curve; adjusting oxygen supply and slagging in the early stage of blowing according to the curve, monitoring the carbon-oxygen product in the middle stage, calculating the remaining decarburization time, determining an alloy adding time window, collecting parameters at the moment, inputting the parameters into the long-short-term memory neural network, predicting the yield, and inversely calculating the total alloy demand; and measuring the actual silicon-manganese content during tapping, calculating the comprehensive deviation ratio, and adding alloy in batches to obtain a target end-point component. The problems of inaccurate oxidation loss prediction, extensive yield estimation and lack of a dynamic feedback adjustment mechanism in the prior art are solved.

Description

Technical Field

[0001] This application relates to the field of converter smelting technology, and in particular to a method, system and storage medium for dynamic control of silicon-manganese alloy in a direct converter process. Background Technology

[0002] The direct converter process refers to a steelmaking method in which steel is directly sent to continuous casting after tapping from the converter via an argon station without refining. In this process, the tapping endpoint fluctuates greatly, making it difficult to accurately predict the oxidation loss of silicon and manganese elements. Existing technologies usually use the empirical coefficient method to estimate the oxidation loss and add silicon and manganese alloy at the tapping stage to adjust the composition. The timing and amount of addition are judged by human experience.

[0003] The shortcomings of the existing technology are that the empirical coefficient method fixes the silicon-manganese oxidation loss rate to a certain range, and fails to accurately calculate the oxidation rate at each moment according to the dynamic changes of temperature, slag condition and oxygen potential in the actual smelting process. This results in a large deviation in the prediction of oxidation loss. In addition, the method of adding the alloy at one time cannot be dynamically adjusted according to the actual addition effect. When the composition deviation is large after the first addition, there is a lack of effective correction means, resulting in low hit rate of the final composition and high alloy consumption.

[0004] Further analysis revealed that even if oxidation loss could be accurately predicted and the total alloy demand determined, the traditional method of calculating the amount added using a fixed yield coefficient was difficult to adapt to changes in smelting conditions due to the nonlinear coupling effect of multiple factors such as steel temperature, slag basicity, and stirring intensity on the yield. This resulted in a deviation between the actual amount added and the actual demand. Moreover, if a one-time addition method was used during the tapping stage, it was impossible to make corrections based on the actual effect. Even if batch addition was used, there was a lack of a closed-loop feedback adjustment mechanism based on real-time composition detection. The addition amount of each batch was still executed according to the initial calculation value, and subsequent batches could not be dynamically optimized based on the actual effect of the previous batch, thus affecting the accuracy of the final composition control. Summary of the Invention

[0005] This application provides a method, system, and storage medium for dynamic control of silicon-manganese alloy in a converter direct-cook process. It is used to establish a dynamic prediction model of silicon-manganese oxidation loss based on metallurgical reaction mechanism and combine it with a long short-term memory neural network to intelligently predict the yield. During the tapping stage, it implements batch dynamic addition and adjusts the subsequent addition amount in a closed loop according to the actual composition detection results of each batch. This solves the problems of inaccurate oxidation loss prediction, coarse yield estimation, and lack of dynamic feedback adjustment mechanism in the prior art, and improves the accuracy of silicon-manganese alloy addition calculation and the hit rate of endpoint composition.

[0006] In a first aspect, this application provides a method for dynamic control of silicon-manganese alloy in a converter direct-cooking process, the method comprising: Step S1: Collect the silicon content, manganese content, carbon content, temperature, and furnace gas composition of molten iron, perform timestamp alignment and Kalman filtering to obtain synchronized multi-source smelting status data; Step S2: Substitute the silicon-manganese concentration, temperature and iron oxide content in the slag from the smelting state data into the oxidation reaction rate equation, calculate the element oxidation rate change at different blowing stages through numerical integration, and obtain the silicon-manganese dynamic oxidation loss curve. Step S3: Adjust the oxygen supply flow rate and lime addition amount in the early stage of blowing according to the oxidation loss curve, monitor the rate of decrease of carbon-oxygen product in the furnace gas in the middle stage of blowing, calculate the remaining decarburization time, and obtain the alloy addition time window under a weak oxidizing atmosphere. Step S4: Collect the steel temperature, slag basicity, and stirring intensity parameters during the addition time window, input them into the pre-trained long short-term memory neural network, output the silicon yield and manganese yield, and calculate the total alloy demand by combining the target composition. Step S5: Measure the actual silicon and manganese content of the molten steel when tapping, calculate the comprehensive deviation rate from the target value, decompose the total alloy demand into multiple batches according to the deviation rate, add them sequentially according to the set time interval and addition rate, and obtain the final silicon and manganese content within the target range.

[0007] Secondly, this application provides a dynamic control system for silicon-manganese alloys in a direct converter process, the dynamic control system for silicon-manganese alloys in a direct converter process includes: The filtering module is used to collect the silicon content, manganese content, carbon content, temperature and furnace gas composition of molten iron, perform timestamp alignment and Kalman filtering to obtain synchronized multi-source smelting status data. The calculation module is used to substitute the silicon-manganese concentration, temperature and iron oxide content in the slag in the smelting state data into the oxidation reaction rate equation, and calculate the changes in element oxidation rate at different blowing stages through numerical integration to obtain the dynamic oxidation loss curve of silicon-manganese. The monitoring module is used to adjust the oxygen supply flow rate and lime addition amount in the early stage of blowing according to the oxidation loss curve, monitor the rate of decrease of carbon-oxygen product in the furnace gas in the middle stage of blowing, calculate the remaining decarburization time, and obtain the alloy addition time window under a weak oxidizing atmosphere. The input module is used to collect the steel temperature, slag basicity, and stirring intensity parameters during the addition time window, input the pre-trained long short-term memory neural network, output the silicon yield and manganese yield, and calculate the total alloy demand by combining the target composition. The decomposition module is used to measure the actual silicon and manganese content of molten steel at the time of tapping, calculate the comprehensive deviation rate from the target value, decompose the total alloy demand into multiple batches according to the magnitude of the deviation rate, and add them sequentially according to the set time interval and addition rate to obtain the final silicon and manganese content within the target range.

[0008] Thirdly, a dynamic control device for silicon-manganese alloy in a converter direct-cooking process is provided, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor invokes the instructions in the memory to cause the dynamic control device for silicon-manganese alloy in the converter direct-cooking process to execute the above-described dynamic control method for silicon-manganese alloy in the converter direct-cooking process.

[0009] Fourthly, a computer-readable storage medium is provided, wherein instructions are stored therein, which, when executed on a computer, cause the computer to perform the aforementioned method for dynamic control of silicon-manganese alloy in the converter direct-flow process.

[0010] The technical solution provided in this application obtains synchronized multi-source smelting state data by collecting silicon, manganese, carbon, temperature, and furnace gas composition data from molten iron, performing timestamp alignment and Kalman filtering. This solves the problem of poor data quality caused by inconsistent sampling frequencies of different sensors and measurement noise interference. Timestamp alignment unifies data from different frequencies to the same time base, achieving time synchronization of multi-source data. Kalman filtering effectively suppresses random measurement noise through two-stage recursive calculation of prediction and update, enabling subsequent oxidation loss prediction and yield calculation to be based on high-quality data input. The silicon-manganese concentration, temperature, and iron oxide content in the slag from the smelting state data are substituted into the oxidation reaction rate equation, and the changes in elemental oxidation rates at different blowing stages are calculated by numerical integration to obtain the dynamic oxidation loss curve of silicon-manganese. Compared with the empirical method of using a fixed oxidation loss rate coefficient in the prior art, this application calculates the temperature-related rate constant based on the Arrhenius equation and combines it with real-time silicon-manganese data. The oxidation rate is dynamically calculated based on the concentration and iron oxide content. The cumulative oxidation loss at each time point is obtained by numerically integrating the rate equation using the fourth-order Runge-Kutta method. This method fully considers the influence of real-time changes in temperature, composition, and slag condition on the oxidation reaction during smelting, transforming oxidation loss prediction from static estimation to dynamic tracking. Based on the oxidation loss curve, the oxygen supply flow rate and lime addition are adjusted in the early stage of blowing, and the furnace gas carbon-oxygen product decline rate is monitored in the middle stage of blowing, and the remaining decarburization time is calculated to obtain the alloy addition time window under a weak oxidizing atmosphere. The oxidation rate of silicon-manganese is controlled by adaptively adjusting the oxygen supply and slag-forming system in the early stage. In the middle stage, the timing of reducing the oxygen supply intensity is quantitatively calculated based on the carbon-oxygen product change rate and carbon content decay equation, so that the furnace is in the optimal metallurgical conditions of low oxygen potential, stable temperature, and low iron oxide content in the slag at the time of alloy addition. Compared with the existing technology that relies on manual experience to judge the timing of addition, this application realizes accurate prediction and active control based on real-time smelting status.

[0011] The steel temperature, slag basicity, and stirring intensity parameters within the input time window are input into a pre-trained long short-term memory neural network (LSTM). The output is the silicon and manganese yields, which are then combined with the target composition to calculate the total alloy requirement. The LSTM neural network, through its input gate, forget gate, and output gate, as well as its cell state mechanism, can capture the nonlinear coupling relationship and long-term dependence of multiple factors such as temperature, slag condition, and stirring on the yield. Compared to existing technologies that use fixed yield coefficients, this application achieves intelligent yield prediction by learning the complex mapping patterns hidden in historical data through neural networks. The alloy addition amount is calculated based on the predicted yield, avoiding systematic bias caused by fixed coefficients. The actual silicon and manganese content of the molten steel at tapping time is measured and compared with the target composition. The overall deviation rate of the target value is used to decompose the total alloy demand into multiple batches based on the deviation rate. The alloys are added sequentially at set time intervals and addition rates to obtain the final silicon-manganese content within the target range. The batch addition strategy adaptively determines the number of batches and the addition ratio of each batch based on the deviation rate. After each batch is added, the actual effect is obtained through rapid component detection and a new deviation value is calculated. The addition amount of subsequent batches is dynamically adjusted based on the new deviation value to form a closed-loop feedback control. Compared with the open-loop method of adding at once or in fixed batches in the prior art, this application achieves gradual approximation of the target composition through real-time detection and dynamic adjustment. When a large deviation occurs after the first batch is added, it can be effectively corrected by subsequent batches, which significantly improves the accuracy of the final composition control and the alloy utilization efficiency. Attached Figure Description

[0012] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments 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.

[0013] Figure 1 This is a schematic diagram of an embodiment of the dynamic control method for silicon-manganese alloy in the converter direct-cook process of this application; Figure 2 This is a schematic diagram of the dynamic oxidation loss curve of silicon-manganese in the embodiments of this application; Figure 3 This is a schematic diagram of an embodiment of the dynamic control system for silicon-manganese alloy in the converter direct-flow process of this application. Figure 4 This is a schematic block diagram of the dynamic control equipment for silicon-manganese alloy in the converter direct-flow process of this invention. Detailed Implementation

[0014] This application provides a method, system, and storage medium for dynamic control of silicon-manganese alloys in a converter direct-cash process. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings of this application 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 described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" or "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.

[0015] For ease of understanding, the specific process of the embodiments of this application is described below. Please refer to [link / reference]. Figure 1 One embodiment of the dynamic control method for silicon-manganese alloy in the converter direct-cook process of this application includes: Step S1: Collect the silicon content, manganese content, carbon content, temperature, and furnace gas composition of molten iron, perform timestamp alignment and Kalman filtering to obtain synchronized multi-source smelting status data; Step S2: Substitute the silicon-manganese concentration, temperature and iron oxide content in the slag from the smelting state data into the oxidation reaction rate equation, and calculate the change in element oxidation rate at different blowing stages through numerical integration to obtain the dynamic oxidation loss curve of silicon-manganese. Step S3: Adjust the oxygen supply flow rate and lime addition amount in the early stage of blowing according to the oxidation loss curve, monitor the rate of decrease of carbon-oxygen product in the furnace gas in the middle stage of blowing, calculate the remaining decarburization time, and obtain the alloy addition time window under a weak oxidizing atmosphere. Step S4: Collect the steel temperature, slag basicity, and stirring intensity parameters during the addition time window, input them into the pre-trained long short-term memory neural network, output the silicon yield and manganese yield, and calculate the total alloy demand by combining the target composition. Step S5: Measure the actual silicon and manganese content of the molten steel when tapping, calculate the comprehensive deviation rate from the target value, decompose the total alloy demand into multiple batches according to the magnitude of the deviation rate, add them sequentially according to the set time interval and addition rate, and obtain the final silicon and manganese content within the target range.

[0016] It is understood that the executing entity of this application can be a dynamic control system for silicon-manganese alloys in a converter direct-flow process, or it can be a terminal or a server; the specific implementation is not limited here. This application's embodiment uses a server as an example for illustration.

[0017] Specifically, the composition of molten iron and the parameters of the blowing process are collected using a spectrometer, an infrared thermometer, and a furnace gas analysis system. The spectrometer detects the silicon, manganese, carbon, phosphorus, and sulfur content in the molten iron every 2 seconds, forming a data sequence of molten iron composition. The infrared thermometer collects the temperature of the molten iron at the same frequency, forming a temperature data sequence. The furnace gas analysis system collects the volume fractions of carbon monoxide and carbon dioxide in the flue gas, as well as the oxygen supply pressure and oxygen flow rate, once per second, forming a sequence of blowing process parameters. Because the sampling frequencies of different sensors differ, each data sequence needs to be timestamped and unified to the same time resolution reference. All data were standardized to a 1-second time resolution. Then, the diagonal elements of the process noise covariance matrix and the observation noise covariance matrix of the Kalman filter algorithm were set to 0.001. Recursive filtering calculations were performed on the time-aligned original dataset. The Kalman filter iterates continuously through two stages: prediction and update. In the prediction stage, the state at the current moment is predicted based on the state estimate of the previous moment and the system dynamic model. In the update stage, the predicted state is corrected using the measurement value at the current moment. Through this recursive calculation, measurement noise interference is eliminated, and finally, 21-dimensional synchronized multi-source smelting state data including silicon, manganese, carbon, phosphorus and sulfur concentrations, temperature, furnace gas composition and oxygen supply parameters are obtained.

[0018] Silicon concentration, manganese concentration, molten steel temperature, and ferric oxide content in slag were extracted from synchronized multi-source smelting state data to construct a basic parameter set for the oxidation reaction. The silicon concentration and ferric oxide content were substituted into the silicon oxidation reaction rate equation, which describes the relationship between the silicon oxidation rate and the silicon concentration and ferric oxide content. Simultaneously, the Arrhenius equation was used to calculate the silicon oxidation rate constant. The Arrhenius equation shows that the reaction rate constant has an exponential relationship with temperature. By substituting the molten steel temperature into this equation, the silicon oxidation rate constant under the current temperature conditions was obtained, and a time function of the silicon oxidation rate was established. This function describes the change of the silicon oxidation rate over time. Similarly, the manganese concentration and ferric oxide content were substituted into the manganese redox reaction rate equation. The redox of manganese includes two opposite processes: the oxidation of manganese by ferric oxide and the reduction of manganese oxide by ferric oxide. The manganese oxidation rate constant and the time function were calculated respectively. The manganese reduction rate constant was used to obtain the time function of manganese oxidation rate. The fourth-order Runge-Kutta method was used to numerically integrate the time functions of silicon and manganese oxidation rates over the blowing time interval. The fourth-order Runge-Kutta method is a high-precision numerical integration algorithm. Its basic idea is to calculate the weighted average of the four function values ​​in each time step as the integral value of that step. In specific execution, the time step is set to 0.5 seconds, and the calculation is gradually advanced from the start of blowing. At each step, the oxidation loss increment in that time period is calculated based on the oxidation rate at the current time. The oxidation loss increments of all time periods are accumulated to obtain the cumulative oxidation loss of silicon and manganese. Based on the evolution of these two cumulative oxidation losses over time, the element loss change trajectory of each stage of blowing in the early, middle and late stages is plotted to form the silicon-manganese dynamic oxidation loss curve.

[0019] To determine whether the cumulative oxidation loss of silicon in the ferrosilicon dynamic oxidation loss curve exceeds a preset threshold, if the cumulative oxidation loss exceeds the threshold, it indicates that the initial silicon content of the molten iron is high and requires enhanced oxidation. In this case, the oxygen flow rate is increased from the standard value of 500 cubic meters per minute to 580 cubic meters per minute (a 16% increase), and the oxygen lance position is lowered from the initial height of 1.8 meters to 1.5 meters (a 0.3-meter decrease). By increasing the oxygen supply and lowering the lance position, the penetration of the oxygen jet into the molten pool is improved, promoting rapid silicon oxidation and achieving the desired smelting effect. In the initial stage, oxidation control parameters are strengthened. Then, it is determined whether the cumulative oxidation loss of manganese in the silicon-manganese dynamic oxidation loss curve is lower than the preset threshold. When the cumulative oxidation loss of manganese is lower than the threshold, it indicates that the initial manganese content of the molten iron is low and protection is needed to avoid excessive oxidation. At this time, the oxygen supply flow rate is reduced from the standard value of 500 cubic meters per minute to 450 cubic meters per minute, a reduction of 10%. At the same time, the amount of lime added is increased from the standard ratio of 45 kg per ton of steel to 52 kg per ton of steel, and the amount of dolomite added is increased from 15 kg per ton of steel to 18 kg per ton of steel. By increasing the slag basicity, the iron oxide mass fraction in the slag is controlled from 18%-22% initially to 12%-15%, reducing the oxidation driving force of the anchor. This yields the protection and control parameters for the early stage of blowing. During the middle stage of blowing, the product of the volume fractions of carbon monoxide and carbon dioxide in the furnace gas (the carbon-oxygen product) is continuously monitored. The rate of change of this product over time is calculated. When the rate of change begins to decrease from its peak and reaches a threshold of -50% (volume fraction squared) per minute, it indicates that the decarburization reaction has entered the later stage. At this point, the carbon content is calculated using the carbon content decay equation and the current oxygen flow rate. The remaining decarburization time required to reach the target value is determined by the carbon content decay equation, which describes the exponential decay of carbon content over time. The remaining decarburization time is calculated by substituting the current carbon content, target carbon content, iron oxide content, and oxygen flow rate into the equation. Based on the remaining decarburization time, the oxygen supply intensity is reduced to 300 cubic meters per minute 3 minutes in advance to lower the oxygen potential in the furnace and stabilize the molten steel temperature in the range of 1580℃ to 1620℃. At the same time, the iron oxide content in the slag is controlled to be reduced to a weak oxidation range of 5%-8%, thus obtaining the alloy addition time window under a weak oxidation atmosphere.

[0020] Fifteen parameters were collected at specific times during the alloy addition window, including molten steel temperature, slag basicity, stirring intensity, amount of silicon-manganese alloy added, silicon content in the alloy, manganese content in the alloy, addition temperature, addition rate, slag fluidity index, and oxygen potential index. These 15 parameters formed the yield prediction input parameter set, which was then input into a pre-trained Long Short-Term Memory (LSTM) neural network. The LSM is a special type of recurrent neural network structure, consisting of an input layer, a first hidden layer, a second hidden layer, and an output layer. The first hidden layer contains 64 LSM units, and the second hidden layer contains 32 LSM units. Each LSM unit contains three gate structures: an input gate, a forget gate, and an output gate. The input gate controls the degree to which new information enters the cell state, the forget gate controls the degree to which old information is forgotten, and the output gate controls the degree to which the cell state is output to the hidden state. The process of calculating the activation values ​​of neurons in each layer through forward propagation involves first passing the input parameters to the input layer, and then propagating them layer by layer forward. In each hidden layer, the LSM... The unit calculates the activation values ​​of the three gates based on the current input and the hidden state of the previous time step, updates the cell state, and outputs the hidden state of the current time step. Finally, the output layer generates the output values ​​of two nodes, namely the silicon yield prediction value and the manganese yield prediction value. Based on the target steel grade's required endpoint silicon content and endpoint manganese content, combined with the cumulative oxidation loss of silicon and manganese, as well as the existing silicon and manganese content in the current molten steel, the required additional silicon and manganese content is calculated. Specifically, the endpoint silicon content is subtracted from the existing silicon content in the current molten steel, and the cumulative oxidation loss of silicon is added to obtain the required additional silicon content. This is then multiplied by the total mass of the molten steel to obtain the required additional silicon content. The same method is used to calculate the manganese content. The silicon content is divided by the product of the silicon content in the silicon-manganese alloy and the silicon yield prediction value to obtain the amount of alloy added based on silicon. The manganese content is divided by the product of the manganese content in the silicon-manganese alloy and the manganese yield prediction value to obtain the amount of alloy added based on manganese. The average of the two values ​​is used to obtain the total alloy requirement.

[0021] After tapping begins, the actual temperature, silicon content, and manganese content of the molten steel are detected using an immersion thermometer and a spectrometer. The immersion thermometer measures the temperature directly by inserting a probe into the molten steel, while the spectrometer determines the elemental content by analyzing the spectrum emitted by the molten steel. This yields compositional data at tapping time. The difference between the actual silicon content and the target silicon content is calculated to obtain the silicon deviation value, and the difference between the actual manganese content and the target manganese content is calculated to obtain the manganese deviation value. These two deviation values ​​are then divided by their respective target values ​​to obtain the normalized deviation. Finally, the squares of the two normalized deviations are summed, the square root is taken, and multiplied by 100% to obtain the overall deviation rate. The batch addition strategy is determined based on the numerical range of the overall deviation rate. When the overall deviation rate is greater than or equal to the first threshold of 15%, the deviation is considered large, and the total alloy demand is divided into three batches. When the overall deviation rate is between the second threshold of 5% and the first threshold of 15%, the deviation is considered moderate, and the alloy is added in two batches. When the overall deviation rate is less than the second threshold of 5%, the deviation is considered small, and the alloy is added in a single batch. This yields the batch addition scheme. The vibration frequency of the vibrating feeder is controlled according to the batch addition scheme to adjust the addition speed of each batch. The vibrating feeder drives the material trough to vibrate through electromagnetic vibration, causing the material to flow out at a set speed. The vibration frequency is... There is a calibration relationship between the addition rates. The required vibration frequency value is calculated based on the target addition rate for each batch, and the feeder parameters are set. After adding the first batch of silicon-manganese alloy, the argon blowing flow rate of the argon blowing agitator is set to 250 L / min, and the stirring time is set to 30 seconds. Argon blowing agitation is started to promote the uniform dispersion of alloying elements in the molten steel. Argon blowing agitation accelerates the homogenization of the composition by generating a circulating flow in the molten steel, resulting in the first batch of homogenized molten steel. The silicon and manganese content of the first batch of homogenized molten steel is measured using a rapid composition detection device, and the new deviation values ​​from the target silicon and manganese content are calculated. The amount of silicon-manganese alloy required for the second batch is recalculated based on the new deviation value. Specifically, the new silicon deviation and manganese deviation are multiplied by the steel mass and divided by the corresponding element content and yield prediction value in the alloy, respectively, to obtain the adjusted addition amount for the second batch. The second batch of alloy is added by controlling the vibrating feeder according to the adjusted addition amount for the second batch, and the stirring and measurement steps are repeated. When a third batch exists, the addition amount for the third batch is adjusted according to the measurement deviation and the addition is executed. Through multiple batches of gradual addition and composition feedback adjustment after each batch, the silicon-manganese content in the steel is gradually adjusted to the target range to obtain the final silicon-manganese content within the target range.

[0022] In one specific embodiment, step S1 includes: The silicon, manganese, carbon, phosphorus, and sulfur contents of molten iron were collected using a spectrometer to obtain a sequence of molten iron composition data. The temperature of molten iron was collected by an infrared thermometer to obtain a sequence of molten iron temperature data. The volume fractions of carbon monoxide and carbon dioxide in the flue gas are collected by the furnace gas analysis system, and the oxygen supply pressure and oxygen flow rate are collected simultaneously to obtain the parameter sequence of the blowing process. The hot metal composition data sequence, hot metal temperature data sequence, and blowing process parameter sequence are timestamped and unified to the same time resolution benchmark to obtain the original dataset after time alignment. The parameters of the process noise covariance matrix and the observation noise covariance matrix of the Kalman filter algorithm are set, and the time-aligned original dataset is recursively filtered to eliminate measurement noise interference, so as to obtain synchronized multi-source smelting state data containing silicon, manganese, carbon, phosphorus and sulfur concentrations, temperature, furnace gas composition and oxygen supply parameters.

[0023] Specifically, the spectrometer emits an excitation light source into the molten iron sample, causing the atoms in the sample to transition from the ground state to an excited state. When the excited-state atoms return to the ground state, they emit a spectrum of a specific wavelength. Different elements emit different wavelengths of light. By detecting the intensity of each wavelength in the spectrum, the content of each element in the sample can be determined. The spectrometer continuously detects the content of silicon, manganese, carbon, phosphorus, and sulfur in the molten iron at a sampling frequency of once every 2 seconds. Each detection generates a set of data records containing the content values ​​of the five elements. Over time, these data records are arranged in chronological order to form a sequence of molten iron composition data. The infrared thermometer is based on Planck's law of blackbody radiation: the higher the temperature of an object, the stronger the infrared radiation energy emitted. The infrared thermometer measures the intensity of infrared radiation emitted by the surface of the molten iron in a non-contact manner. Based on the correspondence between radiation intensity and temperature, the temperature of the molten iron is calculated. The infrared thermometer also collects temperature data at a frequency of once every 2 seconds, obtaining a temperature value each time, which is arranged in chronological order to form a sequence of molten iron temperature data. The furnace gas analysis system analyzes the composition of flue gas extracted from the top of the converter. It uses gas chromatography or infrared absorption to determine the volume fraction of carbon monoxide and carbon dioxide in the flue gas. At the same time, it measures the oxygen supply pressure in the oxygen supply pipeline through a pressure sensor and the oxygen flow rate through a flow meter. The sampling cycle of the furnace gas analysis system is once every 1 second. Each sampling generates a set of data records containing four parameters: carbon monoxide volume fraction, carbon dioxide volume fraction, oxygen supply pressure, and oxygen flow rate. These data are arranged in chronological order to form a sequence of blowing process parameters.

[0024] Because the sampling frequency of the spectrometer and infrared thermometer is once every 2 seconds while that of the furnace gas analysis system is once every 1 second, the time points of the three data sequences are not aligned. Therefore, each data sequence needs to be timestamped. A timestamp records the precise time information of the data acquisition moment. Each data record in the molten iron composition data sequence is marked with its acquisition time timestamp. Similarly, each data record in the molten iron temperature data sequence and the blowing process parameter sequence is timestamped. Then, they are unified to the same time resolution benchmark. Specifically, 1 second is chosen as the unified time resolution. For the molten iron composition and temperature data with a sampling frequency of once every 2 seconds, linear interpolation is used between two adjacent sampling points to generate interpolated data at intermediate moments. Linear interpolation assumes that the change between two known data points is linear, and the values ​​at intermediate moments are linearly weighted according to the time distance. Through interpolation, all data sequences have a unified time resolution of one data point per second. The molten iron composition data, temperature data, and blowing process parameter data at the same moment are merged into a complete data record. The data records at all moments constitute the original dataset after time alignment.

[0025] The Kalman filter algorithm is a recursive filtering algorithm used to eliminate random noise in measurement data. The algorithm consists of two phases: prediction and update. In the prediction phase, the prior state estimate for the current moment is predicted based on the system's state transition model and the state estimate from the previous moment. In the update phase, the prior state estimate is corrected using the actual measured value at the current moment to obtain the posterior state estimate. The process noise covariance matrix describes the uncertainty of the system's dynamic process, while the observation noise covariance matrix describes the uncertainty of the measurement process. Setting a smaller value for the diagonal elements of the process noise covariance matrix indicates relatively stable dynamic changes in the system, while setting another smaller value for the diagonal elements of the observation noise covariance matrix indicates relatively high measurement accuracy. The specific process of performing the recursive filtering calculation on the time-aligned original dataset is to proceed step-by-step from the first moment forward. The process proceeds as follows: at each time step, a prediction phase is first executed to calculate the prior state estimate and the prior error covariance estimate. Then, an update phase is executed to calculate the Kalman gain, which reflects the weight of the measured value in the state update. Based on the Kalman gain and the difference between the actual measured value and the prior estimate, the prior estimate is corrected to obtain the posterior state estimate. At the same time, the posterior error covariance estimate is updated. The posterior state estimate at the current time step is used as the initial condition for the prediction at the next time step to continue the recursive calculation. After recursive filtering, the measurement noise in the original dataset is effectively suppressed, resulting in synchronized multi-source smelting state data including silicon, manganese, carbon, phosphorus, and sulfur concentrations, temperature, furnace gas composition, and oxygen supply parameters. The data records at each time step in this dataset have been filtered to remove random fluctuations and retain the true trend of change, and all parameters are perfectly aligned in time.

[0026] In the practical application of the direct converter process, the composition of molten iron fluctuates significantly. A spectrometer detected that the silicon content of a certain batch of molten iron fluctuated between 0.55% and 0.65%, and the manganese content fluctuated between 0.35% and 0.42%. An infrared thermometer measured the molten iron temperature fluctuating between 1310℃ and 1340℃. After the blowing process began, the furnace gas analysis system detected that the carbon monoxide volume fraction gradually increased from 20% initially to a peak of 58% in the middle stage, and the carbon dioxide volume fraction changed from 12% initially to 18% in the middle stage. The oxygen supply flow rate was adjusted from 480 cubic meters per minute to 520 cubic meters per minute. These raw data contain sensor measurement noise, causing unreasonable jumps in values ​​between adjacent moments. By using timestamps, the composition data collected every 2 seconds by the spectrometer and the temperature data collected every 2 seconds by the infrared thermometer are compared with the furnace gas data collected every 1 second by the furnace gas analysis system on the time axis. To generate a data point every second, linear interpolation was performed on the data sequence with a low sampling frequency. Then, the Kalman filter algorithm was applied to set the parameters of the process noise covariance matrix and the observation noise covariance matrix. The recursive filtering calculation started from the first second of blowing. At each moment, the current state was first predicted according to the state transition model, and then the predicted value was corrected by Kalman gain using the actual measured value. After the recursive processing, the fluctuation curve of silicon content became smooth, the random jump of manganese content was eliminated, the measurement noise of temperature data was suppressed, and the changing trend of furnace gas composition was clearly distinguishable. Finally, synchronized multi-source smelting state data containing 21-dimensional parameters was obtained. Each moment in this dataset contains filtered parameters such as silicon, manganese, carbon, phosphorus and sulfur concentrations, molten iron temperature, carbon monoxide volume fraction, carbon dioxide volume fraction, oxygen supply pressure and oxygen flow rate. All parameters are completely synchronized in time and the data quality is significantly better than the original measurement data.

[0027] In one specific embodiment, step S2 includes: The silicon concentration, manganese concentration, molten steel temperature and iron oxide content in slag were extracted from synchronized multi-source smelting state data to obtain a set of basic parameters for oxidation reaction. Substitute the silicon concentration and iron oxide content from the basic parameter set of the oxidation reaction into the silicon oxidation reaction rate equation, and combine the Arrhenius equation to calculate the silicon oxidation rate constant, thus obtaining the silicon oxidation rate time function. Substitute the manganese concentration and iron oxide content from the basic parameter set of the oxidation reaction into the manganese redox reaction rate equation, calculate the manganese oxidation rate constant and manganese reduction rate constant, and obtain the time function of manganese oxidation rate. The fourth-order Runge-Kutta method was used to numerically integrate the time functions of silicon oxidation rate and manganese oxidation rate over the blowing time interval to obtain the cumulative oxidation loss of silicon and manganese. Based on the evolution of the cumulative oxidation loss of silicon and manganese over time, the trajectory of element loss changes in the early, middle and late stages of blowing was plotted to obtain the dynamic oxidation loss curves of silicon and manganese.

[0028] Specifically, when extracting parameters from synchronized multi-source smelting state data, the values ​​of four key fields—silicon concentration, manganese concentration, molten steel temperature, and iron oxide content in the slag—are directly read from each record in the dataset. These four parameters constitute the basic parameter set for the oxidation reaction. The silicon oxidation reaction rate equation describes the relationship between the silicon oxidation rate and the silicon concentration and iron oxide content. This equation shows that the rate of change of silicon concentration over time is equal to the negative silicon oxidation rate constant multiplied by the current silicon concentration multiplied by the iron oxide content to the power of 0.5, where the silicon oxidation rate constant is significantly affected by temperature.

[0029] The Arrhenius equation shows that the reaction rate constant has an exponential relationship with temperature. This constant is equal to the exponential factor multiplied by the negative activation energy of the natural constant, divided by the gas constant multiplied by the exponent of the absolute temperature. The exponential factor is 2.5 × 10^6 seconds, the activation energy is 180 kJ / mol, and the gas constant is 8.314 J / mol / Kelvin. The absolute temperature needs to be converted by adding 273.15 degrees Celsius. Substituting the molten steel temperature from the basic parameters of the oxidation reaction into the Arrhenius equation, we first calculate the exponential term (negative activation energy divided by the gas constant and then by the absolute temperature), then calculate the natural constant raised to the power of the exponent, and finally multiply by the exponential factor to obtain the silicon oxidation rate constant under the current temperature conditions. Substitute the calculated silicon oxidation rate constant value, along with the silicon concentration and iron oxide content extracted from the basic parameter set of the oxidation reaction, into the silicon oxidation reaction rate equation. First, calculate the iron oxide content to the power of 0.5, then multiply this value by the silicon concentration, and finally multiply by the silicon oxidation rate constant and take the negative value to establish a functional expression of the silicon oxidation rate with respect to time. This functional expression calculates the silicon oxidation rate at each time step based on the current silicon concentration and iron oxide content.

[0030] The redox process of manganese involves two opposite reactions: oxidation, where manganese is oxidized by iron oxide to form manganese oxide; and reduction, where manganese oxide is reduced by iron to form manganese. The rate equation shows that the rate of change of manganese concentration over time is equal to the negative manganese oxidation rate constant multiplied by the manganese concentration multiplied by the iron oxide content, plus the manganese reduction rate constant multiplied by the manganese oxide content multiplied by the iron activity. The manganese oxidation rate constant is calculated using the Arrhenius equation, with a pre-exponential factor of 1.2 × 10⁵ and an exponential term calculated as -165000 divided by the gas constant multiplied by the absolute temperature. The manganese reduction rate constant has a pre-exponential factor of 3.5 × 10⁴ and an exponential term calculated as -155000 divided by the gas constant multiplied by the absolute temperature. Substituting the molten steel temperature into the formula for the manganese oxidation rate constant, the exponential term is calculated first, then the natural constant is raised to the power of the exponent, and finally multiplied by the pre-exponential factor of 1.2 × 10⁵ to obtain the manganese oxidation rate constant. Similarly, the manganese reduction rate constant is obtained by substituting the molten steel temperature into the formula. Substituting the manganese oxidation rate constant, manganese reduction rate constant, and manganese concentration, iron oxide content, manganese oxide content, and iron activity extracted from the basic parameters of the oxidation reaction into the manganese redox reaction rate equation, we first calculate the manganese oxidation term, which is the negative manganese oxidation rate constant multiplied by the manganese concentration and iron oxide content. Then we calculate the manganese reduction term, which is the manganese reduction rate constant multiplied by the manganese oxide content and iron activity. The two terms are added together to obtain the manganese oxidation rate. A functional expression of the manganese oxidation rate with respect to time is established. This function calculates the net oxidation rate of manganese at each time point based on the current manganese concentration, iron oxide content, manganese oxide content, and iron activity.

[0031] The fourth-order Runge-Kutta method is a numerical integration algorithm used to solve differential equations numerically over a time interval. The time interval, from the start to the end of the blowing process, is divided into multiple smaller intervals, each with a time step of 0.5 seconds. Given the initial silicon concentration from the initial moment, four slope values ​​are calculated at each time step to estimate the concentration change within that time step. The first slope is equal to the silicon oxidation rate function calculated at the current moment and the current silicon concentration. The second slope is calculated by constructing an intermediate moment and an intermediate concentration value. The intermediate moment is the current moment plus 0.5 times the time step, and the intermediate concentration value is the current silicon concentration plus 0.5 times the time step multiplied by the first slope. Then, the temperature, iron oxide content, and intermediate concentration value at the intermediate moment are substituted into the silicon oxidation rate function to obtain the second slope. The third slope is also calculated by constructing an intermediate moment and an intermediate concentration value. The intermediate moment is still the current moment plus 0.5 times the time step, but the intermediate concentration value becomes the current silicon concentration plus 0.5 times the time step multiplied by the second slope. These values ​​are then substituted into the silicon oxidation rate function to obtain the third slope. The fourth slope is calculated by adding the full time step to the current time. The concentration value is the current silicon concentration plus the full time step multiplied by the third slope. These values ​​are then substituted into the silicon oxidation rate function to calculate the fourth slope.

[0032] After obtaining the four slopes, the silicon concentration at the next time step is equal to the current silicon concentration plus the time step size divided by 6, then multiplied by the sum of the first slope plus twice the second slope plus twice the third slope plus the fourth slope. This weighted averaging method achieves fourth-order accuracy. This calculation process is repeated from the start time to the end time, updating the silicon concentration at each time step according to the above calculation method of the four slopes and the weighted average formula, ultimately obtaining a sequence of silicon concentration values ​​at each time step during the blowing process. The initial silicon concentration minus the silicon concentration at each time step is the silicon oxidation loss at that time step, and the initial silicon concentration at the end of the blowing process minus the silicon concentration at the end time step is the cumulative silicon oxidation loss. The same fourth-order Runge-Kutta numerical integration process is performed on the manganese oxidation rate time function. Starting from the initial manganese concentration, four slopes are calculated at each time step, and the manganese concentration is updated according to the weighted average formula, advancing step by step until the end of the blowing process, obtaining a sequence of manganese concentration values ​​at each time step. The initial manganese concentration minus the manganese concentration at the end time step is the cumulative manganese oxidation loss.

[0033] The blowing process was divided into three stages: the early stage (0-5 minutes), the middle stage (6-12 minutes), and the late stage (13 minutes to the end). The cumulative oxidation losses of silicon and manganese were extracted for each moment within each stage. A two-dimensional coordinate system was established with time on the horizontal axis and cumulative oxidation loss on the vertical axis. The data points for the cumulative oxidation loss of silicon at each moment were marked on the coordinate system, and these data points were connected sequentially in time to form the oxidation loss curve for silicon. Similarly, the data points for the cumulative oxidation loss of manganese at each moment were marked and connected to form the oxidation loss curve for manganese. These two curves together constitute the dynamic oxidation loss curve of silicon and manganese. This curve visually shows that in the early stage of blowing, due to the high iron oxide content in the slag and the suitable temperature, the oxidation reaction is active, and the oxidation rate of silicon and manganese is fast, resulting in a rapid increase in cumulative loss and a steep upward trend, with a large curve slope. In the middle stage, the oxidation reaction reaches its peak, and silicon and manganese continue to oxidize rapidly, with the cumulative loss continuing to increase, and the curve maintaining a large slope. In the later stages, as the decarburization reaction proceeds, the iron oxide content in the slag gradually decreases, the driving force of the oxidation reaction weakens, the oxidation rate of silicon and manganese slows down, the cumulative loss increases gradually and the slope of the curve gradually decreases and eventually becomes horizontal.

[0034] Figure 2 This is a schematic diagram of the dynamic oxidation loss curve of silicon-manganese in an embodiment of this application. Figure 2 As shown in the figure, the dynamic oxidation loss curves of silicon and manganese demonstrate the changes in the cumulative oxidation loss of silicon and manganese over time during converter blowing. The horizontal axis represents blowing time (in minutes), and the vertical axis represents the cumulative oxidation loss (in percentage). The solid line represents the oxidation loss of silicon, and the dashed line represents the oxidation loss of manganese. The figure shows that in the early blowing stage (0-5 minutes), due to the high iron oxide content in the slag and suitable temperature, the oxidation rate of silicon and manganese is rapid, and the curve shows a steep upward trend. In the middle blowing stage (6-12 minutes), the oxidation reaction reaches its peak, and silicon and manganese continue to oxidize rapidly, maintaining a large slope in the curve. In the later blowing stage (13-18 minutes), as the iron oxide content in the slag gradually decreases, the driving force of the oxidation reaction weakens, the oxidation rate of silicon and manganese slows down, and the slope of the curve gradually decreases and tends to flatten.

[0035] In one specific embodiment, step S3 includes: Determine whether the cumulative oxidation loss of silicon in the dynamic oxidation loss curve of silicon-manganese exceeds the preset threshold. If it does, increase the oxygen supply flow rate from the standard value to the preset ratio, and at the same time lower the oxygen lance position to the preset height to obtain the enhanced oxidation control parameters in the early stage of blowing. Determine whether the cumulative oxidation loss of manganese in the dynamic oxidation loss curve of silicon-manganese is lower than the preset threshold. If it is lower, reduce the oxygen supply flow rate from the standard value by the preset ratio, and at the same time increase the amount of lime and dolomite added to control the iron oxide content in the slag to decrease to the target range, and obtain the protection and control parameters in the early stage of blowing. During the mid-stage of blowing, the product of the volume fraction of carbon monoxide and the volume fraction of carbon dioxide in the furnace gas is continuously monitored, and the rate of change of the product over time is calculated. When the rate of change starts to decrease from the peak and the rate of decrease reaches the judgment value, the remaining decarbonization time required for the carbon content to reach the target value is calculated by combining the carbon content decay equation and the current oxygen flow rate. Based on the remaining decarburization time, the oxygen supply intensity is reduced in advance for a preset duration to lower the oxygen potential in the furnace and stabilize the temperature of the molten steel. At the same time, the iron oxide content in the slag is controlled to be reduced to a weak oxidation range, thus obtaining the alloy addition time window under a weak oxidation atmosphere.

[0036] Specifically, to determine whether the cumulative oxidation loss of silicon in the ferrosilicon dynamic oxidation loss curve exceeds a preset threshold, the cumulative oxidation loss value of silicon at the end of the early blowing stage (5 minutes) is read from the curve. This value is then compared with the preset threshold, which is set based on the initial silicon content of the molten iron. When the initial silicon content is higher than 0.6%, the threshold is set to 25% of the initial silicon content. The comparison operation is performed using the difference method. The cumulative oxidation loss of silicon is calculated by subtracting the preset threshold. If the difference is positive, it indicates that the cumulative oxidation loss exceeds the threshold and requires enhanced oxidation. If the difference is negative or zero, it indicates that the threshold has not been exceeded and no adjustment is needed. When the judgment result exceeds the threshold, the standard oxygen supply flow rate of 500 cubic meters per minute is retrieved. This standard value is multiplied by the enhancement ratio coefficient of 1.16 to calculate the adjusted oxygen supply flow rate of 580 cubic meters per minute. At the same time, the initial oxygen lance position height of 1.8 meters is retrieved. This height is subtracted from the preset reduction height of 0.3 meters to calculate the adjusted oxygen lance position of 1.5 meters. These two adjusted values ​​are the control parameters for enhanced oxidation in the early stage of blowing. By increasing the oxygen supply flow rate and reducing the lance position height, the penetration of the oxygen jet into the molten pool is enhanced, thereby promoting the rapid oxidation of silicon.

[0037] To determine whether the cumulative oxidation loss of manganese in the ferrosilicon dynamic oxidation loss curve is lower than a preset threshold, the cumulative oxidation loss value of manganese is read from the curve at the end of the early blowing stage (5 minutes). This value is then compared with the preset threshold, which is set based on the initial manganese content of the molten iron. When the initial manganese content is lower than 0.3%, the threshold is set to 15% of the initial manganese content. The comparison is performed using a difference method, calculating the preset threshold minus the cumulative oxidation loss of manganese. If the difference is positive, it indicates that the cumulative oxidation loss is lower than the threshold and protection is needed; if the difference is negative or zero, it indicates that it is not lower than the threshold and no adjustment is needed. When the judgment result is below the threshold, the standard oxygen supply flow rate of 500 cubic meters per minute is retrieved. Multiplying this standard value by a reduction coefficient of 0.9 yields an adjusted oxygen supply flow rate of 450 cubic meters per minute. Simultaneously, the standard lime addition amount of 45 kg per ton of steel is retrieved. Multiplying this standard value by an increase coefficient of 1.156 yields an adjusted lime addition amount of 52 kg per ton of steel. The standard dolomite addition amount of 15 kg per ton of steel is retrieved. Multiplying this standard value by an increase coefficient of 1.2 yields an adjusted dolomite addition amount of 18 kg per ton of steel. Increasing the addition of lime and dolomite increases the slag basicity. Increased basicity leads to a decrease in the activity of iron oxide in the slag, thereby reducing the driving force for manganese oxidation. The iron oxide mass fraction in the slag is controlled from the initial range of 18% to 22% to a range of 12% to 15%. The adjusted oxygen supply flow rate, lime addition amount, dolomite addition amount, and target iron oxide mass fraction range are the protection and control parameters for the early stage of blowing.

[0038] During the mid-stage of the blowing process, when continuously monitoring the volume fractions of carbon monoxide and carbon dioxide in the furnace gas, the current volume fraction values ​​of carbon monoxide and carbon dioxide are read from the data collected every second by the furnace gas analysis system. These two values ​​are multiplied to obtain the carbon-oxygen product (C-O-product). The C-O-product reflects the intensity of the decarbonization reaction; the more vigorous the decarbonization reaction, the larger the C-O-product value. Continuous monitoring means that the C-O-product is calculated every second, forming a data sequence of C-O-product changes over time. To calculate the rate of change of the product with respect to time, the current C-O-product value is subtracted from the C-O-product value of the previous second, and the difference is divided by the time interval of 1 second. This yields the first derivative of the C-O-product with respect to time, i.e., the rate of change. A positive rate of change indicates that the C-O-product is increasing, while a negative rate of change indicates that the C-O-product is decreasing. The absolute value of the rate of change indicates the speed of change in the C-O-product. To determine when the rate of change begins to decline from its peak, the calculated rate of change value per second needs to be recorded. When the rate of change value at a certain moment is less than the rate of change value of the previous second, it indicates that the rate of change has begun to decline. Continue monitoring of the rate of change at subsequent moments to see if it continues to decline. If the rate of change declines for three consecutive seconds, it is confirmed that the rate of change has begun to decline from its peak. When the rate of decline reaches a threshold, the rate of change, i.e., the second derivative, is calculated. Subtracting the rate of change of the previous second from the current rate of change and then dividing by the time interval of one second yields the rate of decline. The threshold is set at -50 volume fraction squared per minute, which needs to be converted to -0.833 volume fraction squared per second. When the calculated rate of decline is less than -0.833, it indicates that the rate of decline has reached the threshold, at which point the decarbonization reaction enters its later stage.

[0039] When calculating the remaining decarburization time required to reach the target carbon content by combining the carbon content decay equation and the current oxygen flow rate, the carbon content decay equation describes the exponential decay of carbon content over time. Specifically, the current carbon content equals the initial carbon content multiplied by the negative decarburization rate constant (natural constant) multiplied by the exponent of the cumulative oxygen flow time. The decarburization rate constant is 3.2 × 10⁻⁴ m³ / min. The current carbon content value is read from synchronized multi-source smelting status data, and the target carbon content value is read from the target steel grade requirements. For ordinary carbon steel, the target carbon content range is 0.03% to 0.05%, with a midpoint of 0.04%. The current oxygen flow rate is read from the furnace gas analysis system. The current carbon content is divided by the target carbon content value to obtain the ratio. Taking the natural logarithm of this ratio yields the logarithm, which is equal to the decarburization rate constant multiplied by the oxygen flow rate multiplied by the remaining decarburization time. Therefore, the remaining decarburization time equals the logarithm divided by the decarburization rate constant and then divided by the oxygen flow rate. The calculation process involves first calculating the ratio, then taking the logarithm, and finally dividing by the product of the decarbonization rate constant and the oxygen flow rate to obtain the remaining decarbonization time in minutes.

[0040] When reducing the oxygen supply intensity based on a preset duration of 3 minutes for the remaining decarburization time, the calculated remaining decarburization time is subtracted by 3 minutes to obtain the time elapsed since the start of the adjustment. A timer is set to trigger an oxygen flow rate adjustment command after this duration, adjusting the oxygen flow rate from the current value to 300 cubic meters per minute. Reducing the oxygen supply intensity lowers the partial pressure of oxygen in the furnace, leading to a decrease in oxygen potential. Oxygen potential is defined as the logarithm of the dissolved oxygen activity in molten steel; a decrease in oxygen potential means a decrease in the dissolved oxygen concentration in the molten steel, creating a weakly oxidizing atmosphere. Simultaneously, the molten steel temperature is monitored. When the temperature deviates from the range of 1580°C to 1620°C, the temperature is stabilized by fine-tuning the oxygen supply flow rate and the oxygen lance position. If the temperature is too high, the lance position is appropriately raised to reduce heat input; if the temperature is too low, the lance position is appropriately lowered to increase heat input. Monitoring the iron oxide content in the slag is crucial. When the iron oxide mass fraction exceeds 8%, the amount of reducing substances such as carbon powder added to the slag-forming materials is increased to consume the iron oxide. When the iron oxide mass fraction is below 5%, the amount of reducing substances added is reduced. Controlling the iron oxide mass fraction in the slag within the range of 5% to 8% is considered a weak oxidation range. The alloy addition time window is defined as the period from the moment the oxygen supply flow adjustment command is triggered to the end of the calculated remaining decarburization time. During this period, the furnace is in a weak oxidation atmosphere, the molten steel temperature is stable within the target range, and the iron oxide content in the slag is within the weak oxidation range. Adding silicon-manganese alloy under these conditions yields the highest recovery rate.

[0041] In one specific embodiment, step S4 includes: At each moment of the alloy addition time window, the following parameters are collected: molten steel temperature, slag basicity, stirring intensity, amount of silicon-manganese alloy added, silicon content in the alloy, manganese content in the alloy, addition temperature, addition rate, slag fluidity index, and oxygen potential index, to obtain the set of input parameters for yield prediction. The yield prediction input parameter set is input into a pre-trained long short-term memory neural network. The neural network contains an input layer, a first hidden layer, a second hidden layer, and an output layer. The activation values ​​of neurons in each layer are calculated through forward propagation to obtain the predicted silicon yield and the predicted manganese yield. Based on the target steel grade's required final silicon and manganese content, combined with the cumulative oxidation loss of silicon and manganese, and the existing silicon and manganese content in the molten steel, calculate the required additional silicon and manganese content. Divide the mass of silicon by the product of the silicon content and the predicted silicon yield in the silicon-manganese alloy to obtain the amount of alloy added based on silicon. Divide the mass of manganese by the product of the manganese content and the predicted manganese yield in the silicon-manganese alloy to obtain the amount of alloy added based on manganese. Calculate the average of the two to obtain the total alloy requirement.

[0042] Specifically, when collecting parameters at the alloy addition time window, the temperature value is obtained by inserting an immersion rapid thermometer into the molten steel to measure the temperature value. The slag basicity value is obtained by sampling and analyzing the slag composition, which is the mass fraction of calcium oxide divided by the mass fraction of silicon dioxide. The currently set argon blowing flow rate value is read from the control system of the argon blowing agitator and the agitation intensity value is obtained by converting it according to the correspondence between flow rate and intensity. The amount of silicomanganese alloy to be added is obtained by accumulating the planned batches of silicomanganese alloy and the expected amount added in each batch. The chemical analysis values ​​of silicon and manganese content in the batch of silicomanganese alloy are read from the quality inspection report of silicomanganese alloy. The addition temperature value is obtained by measuring the surface temperature of the alloy to be added with an infrared thermometer. The addition speed value is obtained by converting the vibration frequency setting value of the vibrating feeder. The slag fluidity index value is obtained by measuring the time it takes for the slag to flow through the standard orifice at a specific temperature and looking up the table. The oxygen potential index value is obtained by taking the logarithm of the dissolved oxygen concentration in the molten steel. These 10 parameter values ​​are arranged in order to form a 15-dimensional vector. The first 10 dimensions are steel temperature, slag basicity, stirring intensity, amount of silicon-manganese alloy added, silicon content in the alloy, manganese content in the alloy, addition temperature, addition rate, slag fluidity index, and oxygen potential index. The other 5 dimensions are filled with auxiliary parameters such as carbon content, sulfur content, phosphorus content, oxygen supply flow rate, and oxygen lance position of the current molten steel, forming a complete set of input parameters for yield prediction.

[0043] When the yield prediction input parameter set is input into the pre-trained Long Short-Term Memory (LSTM) neural network, the input layer contains 15 neurons, each receiving a numerical value of the corresponding dimension from the input parameter set. The input layer passes the received 15 numerical vectors to the first hidden layer, which contains 64 LSM units. Each LSM unit contains three gating structures: an input gate, a forget gate, and an output gate, as well as a cell state. The input gate controls how much new input information enters the cell state, the forget gate controls how much old information in the cell state is forgotten, and the output gate controls how much information in the cell state is output to the hidden state. Each LSM unit in the first hidden layer first calculates the activation value of the input gate by multiplying the input vector by the input gate weight matrix, adding the result of multiplying the hidden state by the input gate's cyclic weight matrix at the previous time step, and then compressing the result to between 0 and 1 using the sigmoid activation function. The activation value of the forget gate is calculated similarly by multiplying the input vector by the forget gate weight matrix, adding the result of multiplying the hidden state by the forget gate's cyclic weight matrix at the previous time step, and then using the sigmoid activation function to obtain the forget gate activation value. When calculating candidate cell states, the input vector is multiplied by the cell state weight matrix, and the result of multiplying the previous hidden state by the cell state cyclic weight matrix is ​​added. The result is then compressed to between -1 and +1 using the tanh activation function to obtain the candidate cell state. When updating cell states, the previous cell state is multiplied by the forget gate activation value for selective forgetting, and the result of multiplying the candidate cell state by the input gate activation value for selective memorization is added to obtain the new cell state at the current time step. The output gate activation value is calculated by multiplying the input vector by the output gate weight matrix, adding the result of multiplying the previous hidden state by the output gate cyclic weight matrix, and then applying the sigmoid activation function. The current hidden state equals the new cell state, which is then multiplied by the output gate activation value using the tanh activation function. After each of the 64 long short-term memory units in the first hidden layer completes the above calculations, it outputs 64 hidden state values, forming a 64-dimensional vector that is then passed to the second hidden layer.

[0044] The second hidden layer contains 32 Long Short-Term Memory (LSTM) units, receiving a 64-dimensional vector from the first hidden layer as input. Each LSM unit also contains an input gate, a forget gate, an output gate, and a cell state. It processes the input data according to the same computational flow as the first hidden layer, except for the dimension of the weight matrix. Each LSM unit in the second hidden layer multiplies the 64-dimensional input vector by its respective weight matrix, updates the cell state using a gating mechanism based on the previous hidden state, and outputs the hidden state. The 32 hidden state values ​​output by the 32 LSM units in the second hidden layer form a 32-dimensional vector, which is then passed to the output layer. The output layer contains two neurons: the first neuron predicts the silicon yield, and the second neuron predicts the manganese yield. Each output neuron multiplies the received 32-dimensional vector by its output weight matrix and adds a bias vector to obtain its output value. The value calculated by the first output neuron is the predicted silicon yield, and the value calculated by the second output neuron is the predicted manganese yield. The entire forward propagation process starts from the input layer, and the data passes through the first hidden layer, the second hidden layer and finally reaches the output layer. The calculation of each layer is based on the weight parameters of the current layer and the data passed from the previous layer. Nonlinear mapping is achieved through a combination of matrix multiplication and activation functions, and finally the 15-dimensional input parameters are mapped to 2-dimensional output prediction values.

[0045] When calculating the required additional silicon and manganese content based on the target steel grade's specified endpoint silicon and manganese content, the endpoint silicon and manganese content requirements are retrieved from the target steel grade's technical standard document. The cumulative oxidation loss of silicon and manganese are retrieved from the endpoint of the silicon-manganese dynamic oxidation loss curve; these values ​​represent the total amount of silicon and manganese lost due to oxidation reactions from the start of smelting to the current moment. The existing silicon and manganese content in the molten steel are retrieved from the current moment record of the synchronized multi-source smelting status data; these values ​​represent the actual concentrations of silicon and manganese present in the molten steel. To calculate the required additional silicon content, the endpoint silicon content is subtracted from the existing silicon content in the molten steel to obtain the silicon content gap. This gap is then added to the cumulative oxidation loss of silicon to obtain the total silicon concentration gap that needs to be supplemented by alloying. Finally, this gap is multiplied by the molten steel mass to obtain the required additional silicon content. The molten steel mass is read from the converter's weighing system in tons. When calculating the required amount of manganese, first calculate the manganese content at the endpoint, then subtract the existing manganese content in the current molten steel to obtain the manganese content gap. Add the cumulative oxidation loss of manganese to obtain the total manganese concentration gap that needs to be supplemented by adding alloys. Finally, multiply by the mass of molten steel to obtain the required amount of manganese to be added.

[0046] When calculating the amount of silicon to be added to the alloy by dividing the mass of silicon by the product of the silicon content and the predicted silicon yield in the ferrosilicon-manganese alloy, the silicon content is read from the ferrosilicon-manganese alloy quality inspection report, and the predicted silicon yield is read from the output of the long short-term memory neural network. Multiplying the two yields the mass fraction of silicon that can be effectively converted into molten steel per unit mass of alloy. Dividing the mass of silicon by this mass fraction gives the amount of alloy added based on the silicon requirement. Similarly, when calculating the amount of manganese to be added to the alloy by dividing the mass of manganese by the product of the manganese content and the predicted manganese yield in the ferrosilicon-manganese alloy, the manganese content is read from the ferrosilicon-manganese alloy quality inspection report, and the predicted manganese yield is read from the output of the long short-term memory neural network. Multiplying the two yields the mass fraction of manganese that can be effectively converted into molten steel per unit mass of alloy. Dividing the mass of manganese by this mass fraction gives the amount of alloy added based on the manganese requirement. Since silicon and manganese coexist in silicon-manganese alloys, the alloy addition amount calculated based on silicon is usually not exactly the same as that calculated based on manganese. The arithmetic mean of the two addition amounts is obtained by adding the two values ​​together and dividing by 2. This yields the total alloy requirement that takes into account both silicon and manganese requirements.

[0047] In one specific embodiment, step S5 includes: After tapping begins, the actual temperature, actual silicon content, and actual manganese content of the molten steel are detected by an immersion thermometer and a spectrometer to obtain the composition detection data at the time of tapping. Calculate the silicon deviation value between the actual silicon content and the target silicon content, and the manganese deviation value between the actual manganese content and the target manganese content. Squat the two deviation values, sum them, and then take the square root to obtain the comprehensive deviation rate. The batch addition strategy is determined based on the numerical range of the comprehensive deviation rate. When the comprehensive deviation rate is greater than or equal to the first threshold, the total alloy demand is divided into three batches. When the comprehensive deviation rate is between the first and second thresholds, it is divided into two batches. When the comprehensive deviation rate is less than the second threshold, it is added in a single batch, thus obtaining the batch addition scheme. The vibration frequency of the vibrating feeder is controlled according to the batch addition scheme to adjust the addition speed of each batch. After each batch is added, argon blowing and stirring are performed and the composition deviation is measured. The addition amount of subsequent batches is adjusted according to the deviation value to obtain the final silicon-manganese content within the target range.

[0048] Specifically, after tapping begins, the actual temperature, silicon content, and manganese content of the molten steel are measured using an immersion thermometer and a spectrometer. Tapping refers to the process of pouring the molten steel, smelted in the converter, into the ladle. The first measurement is performed 10 to 15 seconds after the start of this process. The immersion thermometer is a rapid temperature measuring device. A protective tube with a thermocouple is inserted into the molten steel. The thermocouple senses the temperature of the molten steel and transmits the electrical signal to the display device. The entire temperature measurement process is completed within 5 seconds, yielding the actual temperature value of the molten steel. The spectrometer uses spark direct-reading spectroscopy. An electric arc is excited on the surface of the molten steel, causing it to evaporate and ionize. The ionized atoms emit characteristic spectra. The spectrometer measures the spectral intensity and calculates the elemental content using a standard curve. The entire analysis process is completed within 15 seconds, yielding the actual silicon content and manganese content values. These three actual measurements constitute the composition detection data at the time of tapping.

[0049] When calculating the silicon deviation (actual silicon content vs. target silicon content) and the manganese deviation (actual manganese content vs. target manganese content), the target silicon and manganese content values ​​are read from the target steel grade technical requirements document. The silicon deviation is obtained by subtracting the target silicon content from the actual silicon content value, and the manganese deviation is obtained by subtracting the target manganese content from the actual manganese content value. A positive deviation value indicates that the actual content is higher than the target, requiring a reduction in subsequent additions; a negative deviation value indicates that the actual content is lower than the target, requiring an increase in subsequent additions. The absolute value of the deviation reflects the degree of deviation. The relative silicon deviation is obtained by dividing the silicon deviation by the target silicon content value, and the relative manganese deviation is obtained by dividing the manganese deviation by the target manganese content value. The two relative deviations are squared separately, added together, and the square root of the sum is taken. Finally, multiplying by 100% converts the result to a percentage to obtain the comprehensive deviation rate. The comprehensive deviation rate considers the deviations in both silicon and manganese elements, combining the two deviations into a single scalar index through square root calculation.

[0050] When determining the batch addition strategy based on the numerical range of the overall deviation rate, a first threshold of 15% and a second threshold of 5% are set as judgment boundaries. The calculated overall deviation rate is compared with the first threshold of 15%. If the overall deviation rate is greater than or equal to 15%, the deviation is considered large, and the total alloy demand is divided into three batches for addition. The first batch addition amount is equal to the total alloy demand multiplied by 0.60. The second batch addition amount is equal to the total alloy demand multiplied by a coefficient between 0.25 and 0.30, determined based on the actual deviation after the first batch addition. The third batch addition amount is equal to the total alloy demand multiplied by a coefficient between 0.10 and 0.15, determined based on the actual deviation after the second batch addition. The overall deviation rate is compared with the second threshold of 5%. If the overall deviation rate is less than 5%, the deviation is considered small, and the entire total alloy demand is added at once, with the addition rate controlled within the range of 10 to 15 kg per second. If the overall deviation rate is between 5% and 15% (greater than or equal to 5% but less than 15%), the deviation is considered moderate. The total alloy demand is then divided into two batches. The first batch is equal to the total alloy demand multiplied by 0.65, and the second batch is equal to the total alloy demand multiplied by 0.35. The batch addition scheme uses different batch divisions and addition ratios based on the deviation rate. The greater the deviation, the more batches are added, and the smaller the proportion of each batch is added, in order to leave more room for adjustment.

[0051] When adjusting the feeding speed of each batch by controlling the vibration frequency of the vibrating feeder according to the batch feeding plan, the vibrating feeder is a feeding device that drives the material trough to vibrate through electromagnetic vibration, causing the material to flow out at a set speed. There is a linear correlation between the vibration frequency and the feeding speed. Based on the target feeding speed values ​​for each batch determined by the batch feeding plan, the required vibration frequency is calculated using a calibration formula. The calibration formula is: vibration frequency equals 35 Hz plus 2.5 times the feeding speed, where the unit of feeding speed is kilograms per second and the unit of vibration frequency is Hz. The calculated vibration frequency value is input into the feeder control system, and the feeder vibrates at this frequency. The silicon-manganese alloy flows from the silo through the trough into the ladle, and the feeding process continues until the predetermined amount for that batch has been added. After the first batch is added, the stirring current of the argon blowing agitator is set to 250 L / min and the stirring time to 30 seconds, and the argon blowing agitator is started. The argon blowing agitator drives the flow of molten steel by generating a circulating flow, accelerating the diffusion and homogenization of alloying elements in the molten steel. After stirring, wait 35 to 40 seconds to allow the composition to be fully homogenized. Then, use an immersion thermometer and a spectrometer to measure the temperature and composition of the molten steel again. The measurement time should be controlled within 15 seconds to obtain the silicon and manganese content values ​​of the first batch of homogenized molten steel.

[0052] After measuring the silicon and manganese content of the first batch of homogenized molten steel using a rapid component analysis device, the new deviations from the target silicon and manganese content are calculated. The new silicon deviation is obtained by subtracting the target silicon content from the measured silicon content, and the new manganese deviation is obtained by subtracting the target manganese content from the measured manganese content. Based on these new deviations, the required amount of silicon-manganese alloy to be added for the second batch is recalculated using the same method: multiplying the new silicon deviation by the molten steel mass, dividing by the silicon content in the alloy, and then dividing by the predicted silicon yield to obtain the silicon-based addition amount for the second batch; multiplying the new manganese deviation by the molten steel mass, dividing by the manganese content in the alloy, and then dividing by the predicted manganese yield to obtain the manganese-based addition amount for the second batch. The average of these two values ​​is then used to obtain the adjusted addition amount for the second batch. The vibration frequency is calculated according to the adjusted addition amount for the second batch, and the vibrating feeder is used to add the second batch of alloy. After addition, the stirring and measurement process is repeated, maintaining a stirring flow rate of 250 L / min and a stirring time of 30 seconds. The composition is measured after 30 seconds. When a third batch is present, the addition amount of the third batch is adjusted according to the deviation measured after the second batch is added, and the addition is carried out. Each batch follows a closed-loop process of adding, stirring, waiting, measuring, calculating the deviation, and adjusting the addition amount of the next batch. By approaching the target range batch by batch, the silicon and manganese content in the molten steel is finally brought to the target range, and the final silicon and manganese content within the target range is obtained.

[0053] In one specific embodiment, the vibration frequency of the vibrating feeder is controlled according to the batch addition scheme to adjust the addition speed of each batch. After each batch is added, argon blowing and stirring are performed and the composition deviation is measured. The addition amount of subsequent batches is adjusted according to the deviation value to obtain the endpoint silicon-manganese content within the target range, including: Based on the target addition speed of each batch determined by the batch addition plan, calculate the vibration frequency value required by the vibrating feeder, control the vibrating feeder to add the first batch of silicon-manganese alloy according to the vibration frequency value, and obtain the state of completion of the first batch addition. After the first batch of additions is completed, the stirring flow rate and stirring time of the argon blowing stirring device are set, and the argon blowing stirring is started to promote the uniform dispersion of alloying elements in the molten steel, thus obtaining the first batch of homogenized molten steel. The silicon and manganese contents of the first batch of homogenized molten steel are measured by a rapid component detection device. The new deviation values ​​from the target silicon and manganese contents are calculated. Based on the new deviation values, the required amount of silicon-manganese alloy to be added for the second batch is recalculated to obtain the adjusted addition amount for the second batch. The second batch of alloy was added by adjusting the amount added according to the second batch and controlling the vibratory feeder. The stirring and measurement steps were repeated. When a third batch was present, the amount added for the third batch was adjusted according to the measurement deviation and the addition was performed to obtain the final silicon-manganese content within the target range.

[0054] Specifically, when calculating the required vibration frequency value for the vibratory feeder based on the target addition speed of each batch determined by the batch addition plan, the target addition speed value for the first batch is read from the batch addition plan. This speed is calculated based on the first batch addition amount and the expected addition time. The first batch addition amount equals the total alloy demand multiplied by the corresponding batch ratio coefficient. The expected addition time is set based on production experience as the addition amount divided by a speed value within the range of 8 to 12 kg / s. There is a linear relationship between the vibration frequency and the addition speed of the vibratory feeder, which has been calibrated on-site. The calibration formula is: vibration frequency equals 35 Hz plus 2.5 multiplied by the addition speed, where 35 Hz is the base frequency corresponding to the lowest vibration state of the feeder, 2.5 is the conversion factor between frequency and speed in Hz / s per kilogram, and the addition speed is in kilograms per second. Substituting the first batch target addition speed value into this calibration formula, the frequency increment is first calculated by multiplying 2.5 by the addition speed, and then the frequency increment is added to the base frequency of 35 Hz to obtain the required vibration frequency value. When the vibrating feeder operates according to the calculated vibration frequency value, the frequency value is input into the feeder's electromagnetic vibration controller. The controller adjusts the frequency of the excitation current to make the electromagnet vibrate at that frequency. The vibration of the electromagnet is transmitted to the hopper through a spring system. The vibration of the hopper causes the silicon-manganese alloy particles in it to move forward in a jumping motion with periodic acceleration. The moving speed is proportional to the vibration frequency. The silicon-manganese alloy particles at the end of the hopper continuously fall into the ladle to complete the feeding. During the feeding process, the weighing sensor below the hopper monitors the changes in the hopper's mass in real time. When the cumulative feeding mass reaches the predetermined amount for the first batch, the controller automatically stops the feeder vibration, and the first batch of silicon-manganese alloy is completely fed, indicating that the first batch feeding is complete.

[0055] After the first batch of argon-blowing agitator was completed, the agitation flow rate and time were set. The agitator consists of an argon inlet at the bottom of the ladle and permeable bricks. The airflow interacts with the molten steel to generate a circulating flow that drives the molten steel. The agitation flow rate was set to 250 L / min. This flow rate was determined based on the ladle capacity and the quality of the molten steel. If the flow rate was too low, the agitation intensity would be insufficient and the element diffusion would be slow. If the flow rate was too high, the molten steel flow rate would be too fast, causing inclusions in the slag layer. The agitation time was set to 30 seconds. This time was calculated based on the diffusion rate of the ferrosilicon manganese alloy in the molten steel and the geometry of the ladle. The diffusion rate is related to temperature, concentration gradient, and agitation intensity. The geometry determines the diffusion path length. The diffusion path length divided by the diffusion rate gives the time required for complete homogenization. The agitation flow rate of 250 L / min and the agitation time of 30 seconds were input into the argon-blowing agitator control system. The molten steel began to flow. The newly added ferrosilicon manganese alloy diffused downwards from the top of the ladle and was simultaneously carried by the circulating molten steel to diffuse circumferentially and radially. Silicon and manganese elements dissolved from the surface of the alloy particles into the molten steel and dispersed to various locations with the flow of the molten steel. After stirring for 30 seconds, the control system automatically cuts off the power and stops stirring. At this time, the alloying elements are basically evenly distributed in the molten steel, but there is still a slight concentration gradient. It is necessary to let it stand for an additional 35 to 40 seconds to completely eliminate the concentration gradient through natural diffusion. After the standing time is over, the first batch of homogenized molten steel is obtained.

[0056] When measuring the silicon and manganese content of the first batch of homogenized molten steel using a rapid component analysis device, the device includes a sampling probe, a spectral analysis module, and a data processing module. The sampling probe is inserted to a depth of approximately 200 mm below the surface of the molten steel in the ladle. The sampling chamber at the probe tip draws in the molten steel sample, and an electric arc is generated on the sample surface by an excitation electrode. The high temperature of the arc causes the sample to evaporate and ionize. The ionized silicon and manganese atoms transition and emit spectral lines of characteristic wavelengths. The spectral analysis module separates the light of different wavelengths through grating spectral dispersion, and a photomultiplier tube detects the intensity of each wavelength. The data processing module converts the light intensity into elemental content based on a pre-stored standard working curve. The characteristic spectral line wavelength of silicon is 251.6 nm, and the characteristic spectral line wavelength of manganese is 257.6 nm. The entire detection process completes the output of silicon and manganese content values ​​within 15 seconds. When calculating the new deviations from the target silicon and manganese contents, the new silicon deviation is obtained by subtracting the target silicon content value from the technical requirements of the target steel grade from the measured silicon content value, and the new manganese deviation is obtained by subtracting the target manganese content value from the measured manganese content value. When recalculating the required amount of silicon-manganese alloy for the second batch based on these new deviations, the new silicon deviation represents the difference between the silicon content and the target after the first batch addition, and the new manganese deviation represents the difference between the manganese content and the target, which needs to be compensated for by the second batch addition. When calculating the second batch addition based on silicon element requirements, the new silicon deviation is multiplied by the mass of molten steel to obtain the required mass of silicon element, then divided by the silicon content in the silicon-manganese alloy to obtain the required alloy mass to provide that mass of silicon element, and finally divided by the predicted silicon yield to account for losses during the addition process, ultimately yielding the second batch addition amount based on silicon element requirements. When calculating the second batch addition amount based on manganese, the new manganese deviation value is multiplied by the mass of molten steel to obtain the required manganese mass. This is then divided by the manganese content in the silicon-manganese alloy to obtain the alloy mass required to provide that mass of manganese. Finally, this is divided by the predicted manganese yield to account for losses, resulting in the second batch addition amount based on manganese demand. The second batch addition amount based on silicon and the second batch addition amount based on manganese are added together and divided by 2 to obtain the arithmetic mean, which is the second batch adjustment addition amount.

[0057] When adding the second batch of alloy using the vibratory feeder according to the adjusted addition amount for the second batch, the target addition speed for the second batch is calculated by dividing the adjusted addition amount for the second batch by the expected addition time. The expected addition time for the second batch is usually shorter than that for the first batch because the addition amount is smaller, and the speed range is controlled between 6 and 8 kg / s. The vibration frequency is calculated by substituting the target addition speed for the second batch into the calibration formula, and the frequency is input into the feeder controller to start the addition. The addition stops when the cumulative added mass reaches the adjusted addition amount for the second batch, as monitored by the silo weighing sensor. After the second batch is added, the stirring and measurement process is repeated. Argon blowing is started by setting the stirring flow rate to 250 L / min and the stirring time to 30 seconds. After stirring, the mixture is allowed to stand for 30 seconds, and the silicon and manganese contents of the molten steel are measured using a rapid composition detection device. The deviation from the target value is calculated. If the batch addition plan specifies the existence of a third batch, the addition amount for the third batch is adjusted according to the deviation measured after the addition of the second batch. The calculation method is the same as for the second batch: the deviation value is multiplied by the mass of the molten steel, divided by the content of the corresponding element in the alloy, and then divided by the predicted yield value. The average of the two calculation results for silicon and manganese is used to obtain the addition amount for the third batch. The vibration frequency was calculated based on the amount added in the third batch, and the addition was carried out. The addition speed of the third batch was controlled within the range of 4 to 6 kg per second to achieve fine adjustment. After the addition was completed, the mixture was stirred and measured again. At this time, the silicon and manganese contents in the molten steel were very close to the target values, and the deviation was within the allowable range to obtain the final silicon and manganese content within the target range.

[0058] The above describes the dynamic control method for silicon-manganese alloy in the converter direct-cash process in the embodiments of this application. The following describes the dynamic control system for silicon-manganese alloy in the converter direct-cash process in the embodiments of this application. Please refer to [link to relevant documentation]. Figure 3 One embodiment of the dynamic control system for silicon-manganese alloy in the converter direct-flow process of this application includes: The filtering module is used to collect the silicon content, manganese content, carbon content, temperature and furnace gas composition of molten iron, perform timestamp alignment and Kalman filtering to obtain synchronized multi-source smelting status data. The calculation module is used to substitute the silicon-manganese concentration, temperature and iron oxide content in the slag in the smelting state data into the oxidation reaction rate equation, and calculate the changes in element oxidation rate at different blowing stages through numerical integration to obtain the dynamic oxidation loss curve of silicon-manganese. The monitoring module is used to adjust the oxygen supply flow rate and lime addition amount in the early stage of blowing according to the oxidation loss curve, monitor the rate of decrease of carbon-oxygen product in the furnace gas in the middle stage of blowing, calculate the remaining decarburization time, and obtain the alloy addition time window under a weak oxidizing atmosphere. The input module is used to collect the steel temperature, slag basicity, and stirring intensity parameters during the addition time window, input the pre-trained long short-term memory neural network, output the silicon yield and manganese yield, and calculate the total alloy demand by combining the target composition. The decomposition module is used to measure the actual silicon and manganese content of molten steel at the time of tapping, calculate the comprehensive deviation rate from the target value, decompose the total alloy demand into multiple batches according to the magnitude of the deviation rate, and add them sequentially according to the set time interval and addition rate to obtain the final silicon and manganese content within the target range.

[0059] above Figure 3 The dynamic control system for silicon-manganese alloy in the converter direct-cooking process of this invention will be described in detail from the perspective of modular functional entities. The dynamic control equipment for silicon-manganese alloy in the converter direct-cooking process of this invention will be described in detail from the perspective of hardware processing.

[0060] Reference Figure 4 This invention also provides a dynamic control device for silicon-manganese alloy in a direct converter process. This dynamic control device can be a server, and its internal structure can be as follows: Figure 4 As shown. The dynamic control equipment for silicon-manganese alloy in the direct-cash converter process includes a processor, memory, display screen, input device, network interface, and database connected via a system bus. The processor in this computer design provides computing and control capabilities. The memory of the dynamic control equipment for silicon-manganese alloy in the direct-cash converter process includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database of the dynamic control equipment for silicon-manganese alloy in the direct-cash converter process stores the data corresponding to this embodiment. The network interface of the dynamic control equipment for silicon-manganese alloy in the direct-cash converter process is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements the above-described method.

[0061] The present invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when the instructions are executed on a computer, cause the computer to perform the steps of the dynamic control method for silicon-manganese alloy in the converter direct-cash process.

[0062] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0063] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a silicon-manganese alloy dynamic control device (which can be a personal computer, server, or network device, etc.) in a converter direct-flow process to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0064] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. 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 of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for dynamic control of silicon-manganese alloy in a converter direct-cook process, characterized in that, The method includes: Step S1: Collect the silicon content, manganese content, carbon content, temperature, and furnace gas composition of molten iron, perform timestamp alignment and Kalman filtering to obtain synchronized multi-source smelting status data; Step S2: Substitute the silicon-manganese concentration, temperature and iron oxide content in the slag from the smelting state data into the oxidation reaction rate equation, calculate the element oxidation rate change at different blowing stages through numerical integration, and obtain the silicon-manganese dynamic oxidation loss curve. Step S3: Adjust the oxygen supply flow rate and lime addition amount in the early stage of blowing according to the oxidation loss curve, monitor the rate of decrease of carbon-oxygen product in the furnace gas in the middle stage of blowing, calculate the remaining decarburization time, and obtain the alloy addition time window under a weak oxidizing atmosphere. Step S4: Collect the steel temperature, slag basicity, and stirring intensity parameters during the addition time window, input them into the pre-trained long short-term memory neural network, output the silicon yield and manganese yield, and calculate the total alloy demand by combining the target composition. Step S5: Measure the actual silicon and manganese content of the molten steel when tapping, calculate the comprehensive deviation rate from the target value, decompose the total alloy demand into multiple batches according to the deviation rate, add them sequentially according to the set time interval and addition rate, and obtain the final silicon and manganese content within the target range.

2. The method for dynamic control of silicon-manganese alloy in the converter direct-flow process according to claim 1, characterized in that, Step S1 includes: The silicon, manganese, carbon, phosphorus, and sulfur contents of molten iron were collected using a spectrometer to obtain a sequence of molten iron composition data. The temperature of molten iron was collected by an infrared thermometer, and a sequence of molten iron temperature data was obtained. The volume fractions of carbon monoxide and carbon dioxide in the flue gas are collected by the furnace gas analysis system, and the oxygen supply pressure and oxygen flow rate are collected simultaneously to obtain the parameter sequence of the blowing process. The molten iron composition data sequence, molten iron temperature data sequence, and blowing process parameter sequence are timestamped and unified to the same time resolution reference to obtain the original dataset after time alignment. The process noise covariance matrix and observation noise covariance matrix parameters of the Kalman filter algorithm are set, and the time-aligned original dataset is subjected to recursive filtering calculation to eliminate measurement noise interference, thereby obtaining synchronized multi-source smelting state data containing silicon, manganese, carbon, phosphorus and sulfur concentrations, temperature, furnace gas composition and oxygen supply parameters.

3. The method for dynamic control of silicon-manganese alloy in the converter direct-flow process according to claim 1, characterized in that, Step S2 includes: The silicon concentration, manganese concentration, molten steel temperature and iron oxide content in slag were extracted from the synchronized multi-source smelting state data to obtain the basic parameter set for the oxidation reaction. Substitute the silicon concentration and iron oxide content in the basic parameter set of the oxidation reaction into the silicon oxidation reaction rate equation, and calculate the silicon oxidation rate constant by combining the Arrhenius equation to obtain the silicon oxidation rate time function. Substitute the manganese concentration and iron oxide content in the basic parameter set of the oxidation reaction into the manganese redox reaction rate equation, calculate the manganese oxidation rate constant and manganese reduction rate constant, and obtain the manganese oxidation rate time function. The fourth-order Runge-Kutta method was used to numerically integrate the time functions of silicon oxidation rate and manganese oxidation rate over the blowing time interval to obtain the cumulative oxidation loss of silicon and manganese. Based on the evolution of the cumulative oxidation loss of silicon and manganese over time, the trajectory of element loss changes in the early, middle and late stages of blowing is plotted to obtain the dynamic oxidation loss curve of silicon and manganese.

4. The method for dynamic control of silicon-manganese alloy in the converter direct-flow process according to claim 1, characterized in that, Step S3 includes: Determine whether the cumulative oxidation loss of silicon in the silicon-manganese dynamic oxidation loss curve exceeds a preset threshold. If it does, increase the oxygen supply flow rate from the standard value by a preset ratio, and at the same time lower the oxygen lance position by a preset height to obtain the enhanced oxidation control parameters in the early stage of blowing. Determine whether the cumulative oxidation loss of manganese in the silicon-manganese dynamic oxidation loss curve is lower than a preset threshold. If it is lower, reduce the oxygen supply flow rate from the standard value by a preset ratio, and at the same time increase the amount of lime and dolomite added to control the iron oxide content in the slag to decrease to the target range, thereby obtaining the protection and control parameters for the early stage of blowing. During the mid-stage of blowing, the product of the volume fraction of carbon monoxide and the volume fraction of carbon dioxide in the furnace gas is continuously monitored, and the rate of change of the product over time is calculated. When the rate of change starts to decrease from the peak and the rate of decrease reaches the judgment value, the remaining decarbonization time required for the carbon content to reach the target value is calculated by combining the carbon content decay equation and the current oxygen flow rate. Based on the remaining decarburization time, the oxygen supply intensity is reduced in advance for a preset duration to lower the oxygen potential in the furnace and stabilize the temperature of the molten steel. At the same time, the iron oxide content in the slag is controlled to be reduced to a weak oxidation range, thus obtaining the alloy addition time window under a weak oxidation atmosphere.

5. The method for dynamic control of silicon-manganese alloy in the converter direct-flow process according to claim 4, characterized in that, Step S4 includes: At each time window of the alloy addition, the following parameters are collected: molten steel temperature, slag basicity, stirring intensity, amount of silicon-manganese alloy added, silicon content in the alloy, manganese content in the alloy, addition temperature, addition rate, slag fluidity index, and oxygen potential index, to obtain the set of input parameters for yield prediction. The set of predicted yield parameters is input into a pre-trained long short-term memory neural network, which includes an input layer, a first hidden layer, a second hidden layer, and an output layer. The activation values ​​of neurons in each layer are calculated through forward propagation to obtain the predicted yield values ​​of silicon and manganese. Based on the target steel grade's required final silicon and manganese content, combined with the cumulative oxidation loss of silicon and manganese, and the existing silicon and manganese content in the molten steel, calculate the required additional silicon and manganese content. The amount of silicon added is obtained by dividing the mass of silicon by the product of the silicon content in the silicon-manganese alloy and the predicted silicon yield. The amount of manganese added is obtained by dividing the mass of manganese by the product of the manganese content in the silicon-manganese alloy and the predicted manganese yield. The total alloy requirement is obtained by averaging the two.

6. The method for dynamic control of silicon-manganese alloy in the converter direct-flow process according to claim 1, characterized in that, Step S5 includes: After tapping begins, the actual temperature, actual silicon content, and actual manganese content of the molten steel are detected by an immersion thermometer and a spectrometer to obtain the composition detection data at the time of tapping. Calculate the silicon deviation value between the actual silicon content and the target silicon content, and the manganese deviation value between the actual manganese content and the target manganese content. Squat the two deviation values, sum them, and then take the square root to obtain the comprehensive deviation rate. The batch addition strategy is determined based on the numerical range of the comprehensive deviation rate. When the comprehensive deviation rate is greater than or equal to the first threshold, the total alloy demand is divided into three batches. When the comprehensive deviation rate is between the first threshold and the second threshold, it is divided into two batches. When the comprehensive deviation rate is less than the second threshold, it is added in a single batch, thus obtaining the batch addition scheme. According to the batch addition scheme, the vibration frequency of the vibrating feeder is controlled to adjust the addition speed of each batch. After each batch is added, argon blowing and stirring are performed and the composition deviation is measured. The addition amount of subsequent batches is adjusted according to the deviation value to obtain the final silicon-manganese content within the target range.

7. The method for dynamic control of silicon-manganese alloy in the converter direct-flow process according to claim 6, characterized in that, The process involves controlling the vibration frequency of the vibrating feeder according to the batch addition scheme to adjust the addition speed of each batch. After each batch is added, argon blowing and stirring are performed, and the composition deviation is measured. The addition amount of subsequent batches is adjusted based on the deviation value to obtain the final silicon-manganese content within the target range, including: Based on the target addition speed of each batch determined by the batch addition scheme, calculate the vibration frequency value required by the vibrating feeder, control the vibrating feeder to add the first batch of silicon-manganese alloy according to the vibration frequency value, and obtain the first batch addition completion status. After the first batch of additions is completed, the stirring intensity and stirring time of the argon blowing stirring device are set, and the argon blowing stirring is started to promote the uniform dispersion of alloying elements in the molten steel, so as to obtain the first batch of mixed and homogenized molten steel. The silicon and manganese contents of the first batch of homogenized molten steel are measured by a rapid component detection device. The new deviation values ​​from the target silicon and manganese contents are calculated. Based on the new deviation values, the required amount of silicon-manganese alloy to be added for the second batch is recalculated to obtain the adjusted addition amount for the second batch. The second batch of alloy is added by adjusting the amount added according to the second batch and controlling the vibratory feeder. The stirring and measurement steps are repeated. When a third batch is present, the amount added for the third batch is adjusted according to the measurement deviation and the addition is performed to obtain the final silicon-manganese content within the target range.

8. A dynamic control system for silicon-manganese alloy in a converter direct-cook process, characterized in that, A method for dynamically controlling silicon-manganese alloy in a converter direct-cooking process as described in any one of claims 1-7, wherein the silicon-manganese alloy dynamic control system in the converter direct-cooking process comprises: The filtering module is used to collect the silicon content, manganese content, carbon content, temperature and furnace gas composition of molten iron, perform timestamp alignment and Kalman filtering to obtain synchronized multi-source smelting status data. The calculation module is used to substitute the silicon-manganese concentration, temperature and iron oxide content in the slag in the smelting state data into the oxidation reaction rate equation, and calculate the changes in element oxidation rate at different blowing stages through numerical integration to obtain the dynamic oxidation loss curve of silicon-manganese. The monitoring module is used to adjust the oxygen supply flow rate and lime addition amount in the early stage of blowing according to the oxidation loss curve, monitor the rate of decrease of carbon-oxygen product in the furnace gas in the middle stage of blowing, calculate the remaining decarburization time, and obtain the alloy addition time window under a weak oxidizing atmosphere. The input module is used to collect the steel temperature, slag basicity, and stirring intensity parameters during the addition time window, input the pre-trained long short-term memory neural network, output the silicon yield and manganese yield, and calculate the total alloy demand by combining the target composition. The decomposition module is used to measure the actual silicon and manganese content of molten steel at the time of tapping, calculate the comprehensive deviation rate from the target value, decompose the total alloy demand into multiple batches according to the magnitude of the deviation rate, and add them sequentially according to the set time interval and addition rate to obtain the final silicon and manganese content within the target range.

9. A dynamic control device for silicon-manganese alloy in a converter direct-flow process, characterized in that, The method includes a memory and a processor, wherein the memory stores a computer program that can run on the processor, and the processor executes the computer program to implement the dynamic control method for silicon-manganese alloy in the converter direct-flow process according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is run by the processor, it causes the processor to execute the dynamic control method for silicon-manganese alloy in the converter direct-cook process as described in any one of claims 1 to 7.

Images