Intelligent adjustment control method and system for ferroalloy production operation

Abstract

The application discloses a kind of intelligent adjustment control method and system for ferroalloy preparation operation, belong to metallurgical automation intelligent control field.The method is through infrared imaging and microwave sensing depth fusion, constructs the three-dimensional composite feature map of charge surface in submerged arc furnace, temperature distribution and three-dimensional geometric topography surface are represented simultaneously;Based on the atlas, polar coordinate grid division is carried out, and the temperature standard deviation and center thermal field offset of each region are calculated as the quantitative index of furnace condition stability evaluation;Combined with real-time tapping temperature and alloy composition, the reduction agent correction coefficient is calculated using the heat balance compensation model based on the principle of Gibbs free energy minimization;Local fire point is identified and spatial distribution instruction containing distribution position and time length is generated;Synchronous execution of reduction agent delivery quantity adjustment and distributor action realizes the collaborative closed-loop control of overall heat balance and local fire point suppression.The application realizes the accurate quantification and dynamic compensation of reaction intensity in the furnace, significantly reduces power consumption and production cost.

Description

Technical Field

[0001] This application belongs to the field of metallurgical automation and intelligent control, specifically relating to an intelligent regulation and control method and system for ferroalloy preparation and operation. Background Technology

[0002] With the continuous improvement of automation in the metallurgical industry, ferroalloy preparation technology is playing an increasingly prominent role in the pillar industries of the national economy. As a complex high-temperature thermochemical reaction device, the submerged arc furnace involves a deep coupling of material conversion and energy flow in its production process. Maintaining the stability of thermal and chemical balance within the furnace is a key prerequisite for ensuring alloy quality and reducing power consumption.

[0003] Among them, the charging operation regulation and material distribution control system of the electric arc furnace is the core technology direction for achieving high-efficiency production. This technology aims to improve the operating efficiency and equipment life of the electric arc furnace by dynamically monitoring the state of the material surface inside the furnace and adjusting the raw material ratio and spatial material distribution strategy in real time, so as to ensure that the reduction reaction takes place within the optimal temperature range.

[0004] However, traditional ferroalloy preparation processes rely heavily on manual observation of the furnace mouth flame color to determine the amount of charge to be added, making it difficult to quantify the degree of reduction reaction below the charge surface in real time and with precision. Due to the extreme reaction environment within the furnace and its significant hysteresis and nonlinear characteristics, simply relying on a fixed charge ratio cannot effectively address the dynamic fluctuations in raw material composition, leading to uneven local thermal fields, sparking, and severe heat loss. Furthermore, existing control methods lack the ability to integrate and analyze the thermal field distribution on the charge surface and the feedback data from iron tapping, making it difficult to achieve deep integration of spatial charge distribution commands and thermal balance compensation models. This results in the system's inability to make timely and precise closed-loop adjustments to nonlinear changes in furnace conditions.

[0005] Therefore, a smart regulation and control method and system for the preparation and operation of ferroalloys is desired. Summary of the Invention

[0006] The purpose of this invention is to provide an intelligent adjustment and control method and system for ferroalloy preparation, which can effectively solve the problem that traditional ferroalloy preparation processes cannot make timely and accurate closed-loop adjustments to the nonlinear changes in furnace conditions.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0008] A method for intelligent regulation and control of ferroalloy preparation includes the following specific steps:

[0009] A three-dimensional composite feature map is constructed. The three-dimensional composite feature map is obtained by deep fusion of infrared imaging and microwave sensing to obtain a panoramic thermal field distribution map and three-dimensional geometric shape surface of the material surface in the electric arc furnace. After registration and fusion in the spatial coordinate system, it is generated to simultaneously characterize the absolute temperature of each point on the material surface and its actual altitude in the furnace.

[0010] Based on the three-dimensional composite feature map, the material surface is divided into polar coordinate grids to form multiple independent feature regions. The temperature standard deviation, the offset of the central thermal field relative to the geometric center of the furnace, and the radial temperature gradient of each feature region are calculated as quantitative indicators for evaluating furnace stability.

[0011] Based on the real-time collection of tapping temperature and alloy composition during the tapping process, combined with the quantitative indicators for furnace stability evaluation, the reducing agent correction coefficient for adjusting the reducing agent dosage is calculated through a heat balance compensation model. The heat balance compensation model is constructed based on the principle of minimizing Gibbs free energy.

[0012] Based on the three-dimensional composite feature map, the piercing point with abnormal local temperature rise is identified, and based on the spatial location and morphological features of the piercing point, a spatial fabric instruction containing the fabric position and fabric duration is generated.

[0013] The reducing agent dosage adjustment corresponding to the reducing agent correction coefficient and the material feeder action corresponding to the space material feeding command are executed synchronously to achieve coordinated closed-loop control of overall thermal balance and local spit-fire point suppression in the furnace.

[0014] Furthermore, the specific steps for constructing a three-dimensional composite feature map include:

[0015] Multiple infrared detection units deployed in the space above the inside of the electric arc furnace collect the original radiation signals of the material surface. After non-uniformity correction and blackbody calibration, the signals are converted into temperature matrices. The multi-channel temperature matrices are then transformed by perspective and image stitched together to synthesize a panoramic thermal field distribution map covering the material surface inside the electric arc furnace.

[0016] Multiple frequency-modulated continuous wave microwave radars are used to scan the material surface in a step-by-step manner to obtain the echo signals of the sampling points. The precise distance of each sampling point is obtained by using the difference frequency ranging principle and converted into initial point cloud data. The initial point cloud data is then processed by bilinear interpolation to reconstruct the three-dimensional geometric topography of the material surface inside the electric arc furnace.

[0017] Using the geometric center of the furnace as a reference, the image coordinates of the panoramic thermal field distribution map are mapped onto the three-dimensional grid points of the three-dimensional geometric surface through coordinate transformation. This completes the registration and fusion of temperature data and spatial geometric data, and constructs a three-dimensional composite feature map, which is used to simultaneously characterize the absolute temperature of each point on the material surface and its actual altitude in the furnace.

[0018] Furthermore, the specific steps for dividing the material surface into polar coordinate grids and calculating quantitative indicators for evaluating furnace stability include:

[0019] The panoramic thermal field distribution map is transformed from the image coordinate system to the actual plane rectangular coordinate system of the furnace with the geometric center of the furnace as the origin, and then transformed into the polar coordinate system.

[0020] With the geometric center of the furnace as the origin, the material surface is evenly divided into multiple fan-shaped areas in the circumferential direction, and according to the furnace design, it is divided into multiple concentric rings along the radial radius, including the central area, the middle ring area and the edge area. The characteristic area is formed by the intersection of the fan-shaped areas and the concentric rings.

[0021] Calculate the temperature standard deviation of the temperature values ​​of all pixels in each feature region as a quantitative indicator to measure the reaction stability of that feature region;

[0022] The top 5% of pixels with the highest temperature in the panoramic thermal field distribution map are extracted as the high-temperature point set. The geometric centroid coordinates of the high-temperature point set are calculated using the weighted centroid algorithm, which is used as the location of the central thermal field.

[0023] The Euclidean distance between the central thermal field and the preset coordinates of the furnace geometric center is calculated to obtain the central thermal field offset, which serves as the basis for determining whether the airflow inside the furnace is deflected.

[0024] Furthermore, the specific steps for calculating the reducing agent correction coefficient using the heat balance compensation model include:

[0025] Real-time acquisition of actual iron tapping temperature, actual alloy composition, raw material moisture content, ambient temperature and humidity, and electrode input power during the iron tapping process;

[0026] The actual tapping temperature, actual alloy composition, raw material moisture content, ambient temperature and humidity, and electrode input power are input into a thermal balance compensation model with a pre-set thermodynamic database. Based on the principle of minimizing Gibbs free energy, the theoretical target tapping temperature and theoretical target alloy composition at thermodynamic equilibrium under the current operating conditions are solved through iterative calculation.

[0027] The deviation between the theoretical target tapping temperature and the actual tapping temperature, as well as the deviation between the theoretical target alloy composition and the actual alloy composition, are calculated as the reduction potential gap.

[0028] Based on the reduction potential energy gap and combined with the material surface radiative heat loss estimated by the panoramic thermal field distribution map, the reducing agent correction coefficient is calculated by a multiple regression model that includes temperature adjustment weight coefficient, composition adjustment weight coefficient and heat loss compensation coefficient.

[0029] Furthermore, the surface radiative heat loss estimated from the panoramic thermal field distribution map further includes:

[0030] The average temperature of the material surface inside the electric arc furnace was calculated based on the panoramic thermal field distribution map.

[0031] According to the Stefan-Boltzmann law, the radiative heat loss power is calculated using the average temperature of the material surface, the ambient temperature, the total area of ​​the material surface, and the pre-stored material surface emissivity.

[0032] The radiative heat loss power is normalized to obtain the normalized surface radiative heat loss, which is then used as input to calculate the reducing agent correction coefficient.

[0033] Furthermore, the steps for identifying the piercing points and generating spatial cloth instructions specifically include:

[0034] The three-dimensional composite feature map is scanned in real time, and the connected component analysis algorithm is used to merge adjacent pixels with temperatures exceeding the preset high temperature threshold into high temperature connected regions.

[0035] When the highest temperature of the high-temperature connected area in multiple consecutive thermal images exceeds the preset ignition determination threshold, and its projected area exceeds the proportion threshold of the total area of ​​the material surface, it is determined to be an ignition point.

[0036] Extract the average coordinates of the geometric center of the ignition point in the actual plane coordinates of the furnace, and convert them into polar coordinates with the geometric center of the furnace as the origin to obtain the radial distance and azimuth angle, which are used as the target position of the material distributor.

[0037] Based on the difference between the highest temperature at the ignition point and the ignition judgment threshold, the thickness of the material to be covered is dynamically determined, and combined with the projected area of ​​the ignition point, the total volume of the material to be covered and the covering time are calculated.

[0038] Furthermore, the steps for generating space cloth instructions specifically include:

[0039] Extract the material level height at the coordinates of the ignition point from the three-dimensional composite feature map;

[0040] Based on the fixed vertical distance from the rotation center of the feeder chute to the bottom of the furnace, the radial distance of the ignition point, and the material level height, the inclination angle of the chute is calculated through geometric relationships to ensure that the material falls accurately to the target point.

[0041] The azimuth angle, radial distance, chute inclination angle, material placement duration, and preset unloading flow rate of the target location are combined to generate a complete spatial material placement command.

[0042] Furthermore, the method also includes trend prediction and advance adjustment steps based on long short-term memory neural networks:

[0043] Extract continuous thermal field evolution data from historical databases to construct a training dataset. Each sample contains a sequence of feature vectors corresponding to the thermal field distribution maps collected at multiple past time steps, as well as the actual thermal field change trend in the future period as a label.

[0044] Construct and train a long short-term memory neural network to learn the temporal patterns of thermal field evolution and output a prediction of future thermal field trends;

[0045] Real-time data on the evolution of the thermal field during the current period is collected and input into a trained long short-term memory neural network to obtain a prediction of the material surface temperature trend over a future period.

[0046] If the prediction results show that the temperature in the center area of ​​the material surface has a continuous downward trend, then while the current furnace condition is still within the normal range, the reducing agent correction coefficient should be increased in advance by the predetermined amount.

[0047] Furthermore, the method also includes an electrode adjustment linkage step:

[0048] Based on the characteristic regions divided by the polar coordinate grid, extract the average temperature or standard deviation of the characteristic regions corresponding to the electrode positions of each phase;

[0049] When the average temperature or the standard deviation of temperature exceeds the preset reaction intensity threshold, it is determined that the reaction intensity below the electrode of that phase is too high.

[0050] While simultaneously adjusting the amount of reducing agent and operating the feeder, a command is sent to the electrode lifting mechanism to raise the phase electrode upward by a preset amount in order to reduce the intensity of the local reaction.

[0051] A smart adjustment and control system for ferroalloy preparation includes:

[0052] The multi-dimensional situational awareness unit is used to obtain a panoramic thermal field distribution map and three-dimensional geometric shape surface of the material surface inside the electric arc furnace through the deep fusion of infrared imaging and microwave sensing, and to construct a three-dimensional composite feature map containing temperature dimension and spatial geometric dimension.

[0053] The regional thermal field quantization analysis unit is used to divide the material surface in the three-dimensional composite feature map into polar coordinate grids, forming multiple feature regions, and to calculate the temperature standard deviation of each feature region and the offset of the central thermal field relative to the geometric center of the furnace.

[0054] The thermal balance compensation model unit has a built-in thermodynamic database based on the principle of minimizing Gibbs free energy. It is used to calculate the reducing agent correction coefficient for adjusting the amount of reducing agent added, based on the real-time collected iron tapping temperature and alloy composition, combined with the furnace condition index output by the regional thermal field quantitative analysis unit.

[0055] The local anomaly point identification and cloth placement instruction generation unit is used to identify the piercing point based on the three-dimensional composite feature map and automatically calculate the spatial cloth placement instruction, which includes cloth placement position, cloth placement duration, and chute tilt angle.

[0056] The collaborative closed-loop control unit is used to synchronously execute the adjustment of the reducing agent dosage corresponding to the reducing agent correction coefficient and the action of the fabric distributor corresponding to the space fabric distribution command.

[0057] In summary, this application includes at least one of the following beneficial technical effects:

[0058] 1. This invention, through the deep fusion of infrared imaging and microwave sensing, transforms the traditional reliance on the vague experience of manually observing the fire color into high-resolution temperature matrix and spatial topography data. By finely dividing the material surface into multiple characteristic regions and calculating the central thermal field offset, real-time quantitative monitoring of the reduction reaction intensity in the furnace is achieved, effectively eliminating subjective errors in manual judgment and significantly reducing the range of furnace temperature fluctuations.

[0059] 2. This invention constructs a heat balance compensation model based on tapping feedback, which dynamically adjusts the reducing agent correction coefficient according to the actual physicochemical state of the molten pool. This precise closed-loop adjustment mechanism greatly optimizes the reducing agent ratio and significantly reduces heat loss caused by localized ignition points. Actual operation data shows that after adopting the method described in this invention, the unit alloy power consumption is significantly reduced, while the effective utilization rate of the reducing agent is significantly improved, greatly reducing production costs.

[0060] 3. This invention, through the synchronous linkage of spatial material distribution commands and the feeding system, can quickly eliminate localized scorching points, achieving a smooth material surface distribution and uniform thermal evolution. This stable thermal regime effectively avoids drastic fluctuations in the furnace airflow and localized overheating, reducing thermal shock and chemical erosion of the furnace lining. It is estimated that the service life of the submerged arc furnace lining is significantly extended, and equipment maintenance costs are significantly reduced. Attached Figure Description

[0061] Figure 1 This is a schematic diagram of the overall intelligent adjustment and control method for the preparation and operation of ferroalloys;

[0062] Figure 2 This is a schematic diagram illustrating the principle of fusing thermal balance compensation model with multi-dimensional situational awareness;

[0063] Figure 3 This is a flowchart of the construction of a three-dimensional composite feature map and the quantitative analysis of the regional thermal field based on the fusion of infrared imaging and microwave sensing;

[0064] Figure 4 It is a closed-loop control logic diagram based on the spatial fabric command generated by local ignition point identification and the synchronous adjustment of reducing agent dosage. Detailed Implementation

[0065] To make the objectives, technical solutions, and advantages of this invention clearer, the following description is provided in conjunction with the appendix. Figure 1 To be continued Figure 4 The present invention will be further described in detail below with reference to specific embodiments.

[0066] like Figure 1 As shown, the intelligent regulation and control method for ferroalloy preparation is implemented according to the following steps:

[0067] First, in step S1, a three-dimensional composite feature map is constructed. In the intelligent regulation and control method for ferroalloy preparation, the first step is to establish a data foundation that reflects the state of the material surface inside the submerged arc furnace. To this end, the system acquires a panoramic thermal field distribution map and a three-dimensional geometric surface of the material surface inside the submerged arc furnace through deep fusion of infrared imaging and microwave sensing, and constructs a three-dimensional composite feature map. This map will serve as the core data source for subsequent regional thermal field analysis and closed-loop regulation. The specific implementation steps are as follows:

[0068] Step S101: Deploy the infrared detection unit and microwave radar equipment:

[0069] Four sets of infrared detection units are symmetrically deployed in the upper space inside the electric arc furnace, with each set distributed at a 90-degree angle at the observation hole position below the furnace top cover. Each infrared detection unit contains a long-wave infrared thermal imager, with its operating wavelength set between 8 micrometers and 14 micrometers to avoid the absorption peaks of carbon dioxide and water vapor in the flue gas inside the furnace.

[0070] The long-wave infrared thermal imager has a preset resolution of 640×480 pixels and a temperature measurement range covering 600 degrees Celsius to 2000 degrees Celsius. To cope with high-temperature and high-dust environments, each long-wave infrared thermal imager is enclosed in a protective cover with a water-cooling jacket, and the cooling water flow rate is maintained at 15 to 20 liters per minute to ensure that the core temperature of the long-wave infrared thermal imager is below 45 degrees Celsius.

[0071] Meanwhile, the front of the protective cover is equipped with a nitrogen purging port. By continuously injecting industrial nitrogen at a pressure of 0.3 MPa, a positive pressure air curtain is formed in front of the lens to prevent dust from adhering.

[0072] For synchronous acquisition of material level and shape, the system employs four frequency-modulated continuous wave microwave radars, with their center frequencies set between 76 GHz and 81 GHz. Each frequency-modulated continuous wave microwave radar is mounted on a mechanical scanning mechanism, and performs step-by-step scanning within a horizontal 360-degree and vertical 60-degree range.

[0073] Step S102, Infrared thermal imaging acquisition and preprocessing: The raw radiation signal acquired by the long-wave infrared thermal imager is converted into a real-time temperature matrix by the internal processor after non-uniformity correction and blackbody calibration.

[0074] Non-uniformity correction is used to eliminate image noise caused by inconsistent detector pixel responses, while blackbody calibration converts the radiation signal into absolute temperature values ​​using a built-in reference blackbody, ensuring temperature measurement accuracy. This temperature matrix is ​​transmitted to the central control unit via industrial Ethernet at a frequency of 25 frames per second.

[0075] Step S103: Panoramic thermal field distribution map synthesis. The central control unit performs perspective transformation and image stitching on the received four images. The perspective transformation is based on the pre-calibrated internal and external parameters of the camera to correct each image to the same reference viewpoint; the image stitching uses feature point matching and fusion algorithms to eliminate geometric distortion caused by the installation angle and ghosting in overlapping areas, and finally synthesizes a panoramic thermal field distribution map covering the material surface inside the electric arc furnace.

[0076] The panoramic thermal distribution map is stored in the form of a two-dimensional temperature matrix, with each pixel corresponding to the real-time temperature of a specific location on the material surface.

[0077] Step S104: Microwave radar scanning and echo processing. After the microwave pulse is emitted, it is reflected by the material surface and captured by the receiving antenna. The central control unit uses the difference frequency ranging principle of frequency-modulated continuous wave radar to calculate the precise distance of each sampling point on the material surface by calculating the frequency difference between the emitted signal and the echo signal.

[0078] The number of sampling points was set to 2000 per cycle to ensure the resolution of the topography reconstruction. The three-dimensional coordinates (angles and distances converted to spatial coordinates) of all sampling points constituted the initial point cloud data.

[0079] Step S105: 3D material surface topography reconstruction. The initial point cloud data is processed by bilinear interpolation algorithm and meshed on the horizontal projection surface to reconstruct the 3D geometric topography of the material surface inside the electric arc furnace.

[0080] The three-dimensional geometric surface is represented by a mesh model, reflecting the elevation and undulation of each point on the material surface, providing a spatial reference for subsequent analysis of material layer thickness and airflow distribution.

[0081] Step S106 involves fusing multi-source information to construct a three-dimensional composite feature map. The panoramic thermal field distribution map generated in step S103 is registered and fused with the three-dimensional geometric surface reconstructed in step S105 in a spatial coordinate system. The registration and fusion process uses the geometric center of the furnace as a reference, and maps the pixel coordinates of the temperature matrix to three-dimensional grid points through coordinate transformation, thus constructing a three-dimensional composite feature map containing temperature data and spatial geometric data.

[0082] The three-dimensional composite feature map not only records the absolute temperature of each point on the material surface, but also correlates the actual altitude of that point in the furnace, providing a three-dimensional spatial benchmark for subsequent reaction intensity assessment.

[0083] In summary, after completing the three-dimensional composite feature map constructed in step S1, the system has a panoramic thermal field distribution map and three-dimensional geometric surface of the material surface inside the current electric arc furnace. However, the three-dimensional composite feature map alone is not enough to directly drive the actuator to make adjustments. It is also necessary to conduct in-depth quantitative analysis on these raw data to extract key indicators that can characterize the stability of the furnace condition.

[0084] Therefore, the system will proceed to step S2, where it will further refine the material surface into smaller regions, calculate the temperature standard deviation of each region and the central thermal field offset, thereby transforming the fuzzy thermal image into quantifiable furnace condition evaluation parameters, providing a basis for subsequent compensation and adjustment decisions. For example... Figure 3 As shown, the specific implementation steps are as follows:

[0085] Step S201: Polar coordinate projection and region division of the thermal field distribution map. The central control unit first transforms the panoramic thermal field distribution map generated in step S103 from the image coordinate system to the actual plane rectangular coordinate system of the furnace with the geometric center of the furnace as the origin. This transformation is based on pre-calibrated camera intrinsic and extrinsic parameters and is achieved through inverse perspective transformation and distortion correction, ensuring that the image coordinates of each pixel are accurately mapped to the actual plane coordinates within the furnace. .

[0086] These planar coordinates were then converted to polar coordinates. ,in The distance to the geometric center of the furnace. This is the azimuth angle. After conversion, with the geometric center of the furnace as the origin, the material surface is evenly divided into 8 sector areas in the circumferential direction, and the angle range of each sector area is fixed at 45 degrees.

[0087] At the same time, based on the furnace design radius This radius value is pre-stored in the system parameters, further subdividing the material surface radially into three concentric rings: the central area corresponds to a radius... to Scope (excluding) ), the corresponding radius of the Middle Ring Road to Scope (excluding) ), the radius corresponding to the edge region to The range.

[0088] This polar coordinate grid division creates 24 independent feature regions, each uniquely defined by a specific angular and radial interval.

[0089] Step S202: Calculation and stability determination of temperature standard deviation for each feature region. For each feature region defined in step S201, the central control unit extracts the temperature values ​​of all pixels within that feature region in the actual plane coordinates of the furnace. The temperature dataset that constitutes this feature region, where arrive , This represents the total number of pixels within this feature region. Calculate the arithmetic mean of these temperature values:

[0090]

[0091] And temperature standard deviation:

[0092]

[0093] Temperature standard deviation As a core quantitative indicator for measuring the stability of the reaction in this characteristic region, its value reflects the degree of temperature dispersion within the characteristic region.

[0094] The system has preset fluctuation thresholds corresponding to the current smelting product. The fluctuation threshold is pre-calibrated and stored in the system parameter table based on the smelting characteristics and historical operating data of different ferroalloys. For example, for high-carbon ferromanganese smelting, It can be set to 45 degrees Celsius.

[0095] When the temperature standard deviation of a certain characteristic region Exceed At that time, the central control unit determines that there is a risk of material segregation or uneven airflow distribution in the characteristic area.

[0096] Step S203: Extraction of the central thermal field and calculation of the centroid coordinates. The central control unit sorts all the pixels in the panoramic thermal field distribution map in descending order of temperature value and selects the top 5% of the pixels with the highest temperature as the set of high-temperature points. These points represent the core reaction zone where the reduction reaction in the furnace is most intense.

[0097] For each pixel in the set of high-temperature points, obtain its coordinates in the actual plane coordinates of the furnace. and its temperature value The weighted centroid algorithm is used to calculate the geometric centroid coordinates of the set of high-temperature points. The calculation formula is as follows:

[0098]

[0099]

[0100] The summation symbol Iterate through all high-temperature points. The pixel temperature value. This is a planar coordinate system. The centroid coordinates reflect the central location of the high-temperature reaction core region, i.e., the central thermal field.

[0101] Step S204, Calculation of central thermal field offset and determination of flow deviation, using the geometric centroid coordinates of the high-temperature point set calculated in step S203. Preset coordinates with the geometric center of the furnace In comparison, the coordinates of the furnace's geometric center are determined by the furnace design parameters, and are typically taken as... .

[0102] Calculate the central thermal field offset using the following formula. :

[0103]

[0104] Central thermal field offset The magnitude of this value directly reflects the degree to which the high-temperature reaction core zone inside the furnace deviates from the geometric center of the furnace. A preset radial deviation threshold is set within the system. This threshold is set according to the furnace size; for example, for a 10-meter diameter submerged arc furnace, It can be set to 0.5 meters, or adjusted according to the radius of the electrode distribution circle. When Exceed When the temperature is high, it indicates that the central thermal field inside the furnace is significantly deviated from the geometric center of the furnace, and the airflow distribution inside the furnace shows a deviation phenomenon. At this time, the system immediately triggers a thermal field imbalance warning signal and uses the central thermal field offset as the basis for subsequent material distribution adjustment.

[0105] Step S205, Radial Temperature Gradient Calculation and Thermal Field Uniformity Evaluation: The system further calculates the radial temperature gradient between each feature region, specifically the rate of temperature difference change between the central region and the edge region. First, the average temperature of all pixels in the central region is calculated. and the average temperature of all pixels in the edge region. .

[0106] Then determine the representative radii of the central and peripheral regions, usually taking the median radius of the central region. Median radius of the edge region , Design radius of the furnace. Radial temperature gradient. Calculate using the following formula:

[0107]

[0108] The radial temperature gradient is measured in degrees Celsius per meter, representing the temperature change per unit distance along the radial direction. The system's preset uniformity threshold is 150 degrees Celsius per meter, an empirical value derived from metallurgical thermal principles and extensive field operational data.

[0109] If the calculation yields If the temperature exceeds 150 degrees Celsius per meter, the system determines that the current heat field distribution is severely uneven. This situation indicates that the reduction reaction inside the furnace is stratified or the airflow channel is abnormal, requiring forced intervention through subsequent steps.

[0110] In summary, step S2 transforms the original thermal field image data into a series of quantitative indicators for evaluating furnace stability, such as temperature standard deviation, central thermal field offset, and radial temperature gradient. These quantitative indicators accurately characterize the intensity distribution and stability of the current reduction reaction in the furnace, providing a clear decision-making basis for the dynamic adjustment of the thermal balance compensation model in step S3 and the precise generation of spatial material distribution commands in step S4.

[0111] Regarding step S3, the construction of the heat balance compensation model and the calculation of the reducing agent correction coefficient, after completing the quantitative analysis of the regional thermal field in step S2, the system grasps the stability characteristics of the current furnace condition (i.e., the quantitative indicators for furnace condition stability evaluation). However, it is still necessary to combine the actual physicochemical state of the iron tapping process to dynamically evaluate and compensate the overall driving force of the reduction reaction in the furnace. Therefore, the system proceeds to step S3, as follows... Figure 2 As shown, by real-time monitoring of the iron temperature and alloy composition, a heat balance compensation model based on the principle of minimizing Gibbs free energy is constructed to calculate the reducing agent correction coefficient K, providing a quantitative basis for subsequent feeding adjustments. The specific implementation steps are as follows:

[0112] Step S301: Real-time data acquisition during the tapping process. A dual-band infrared thermometer is installed above the tapping trough. The two working bands of the thermometer are set to 8μm~14μm and 3.8μm~4.0μm, respectively. By detecting the radiation intensity of these two bands and calculating their ratio, the influence of smoke, dust, water vapor, etc., on the measurement accuracy is eliminated using the colorimetric thermometry principle. The thermometer continuously acquires the temperature data at the molten pool outlet at a sampling frequency of 1Hz, and records it as the measured tapping temperature. .

[0113] Simultaneously, an online fluorescence spectrometer was installed next to the molten iron trough, automatically extracting molten iron samples every 10 minutes for compositional analysis. The analyzer uses X-ray fluorescence spectroscopy to detect the mass fraction of major elements such as silicon, manganese, and chromium, as well as the content of impurity elements such as phosphorus and sulfur in the molten iron. The detection results are recorded as the actual alloy composition. subscript To represent different elements.

[0114] In addition, the system obtains the moisture content of coke and ore in real time through an online moisture meter installed on the raw material conveyor belt, and records it as the raw material moisture content. Ambient temperature is obtained by an ambient temperature and humidity sensor installed near the furnace body. and ambient humidity The input power of each phase electrode is collected by the electrode electrical parameter measurement module, and the average value is recorded as the electrode input power. All data is transmitted to the central control unit via industrial Ethernet.

[0115] Step S302: Construction of the thermal balance compensation model and calculation of the theoretical target value. The central control unit has a pre-installed thermodynamic database of the iron alloy reduction reaction. The thermodynamic database contains the following thermodynamic data:

[0116] Standard molar Gibbs free energy of formation for each reactant and product Polynomial coefficients varying with temperature; molar isobaric heat capacity of each substance. Polynomial coefficients as a function of temperature; phase transition temperature and enthalpy of each substance. The relationship between the equilibrium constant and temperature for each major reduction reaction (such as MnO + C = Mn + CO, SiO2 + 2C = Si + 2CO, etc.).

[0117] These data are derived from the "Handbook of Metallurgical Thermodynamics" and well-known experimental results in the field, and are stored in a thermodynamic database in tabular form or as fitting formulas.

[0118] The thermal balance compensation model is based on the principle of minimizing Gibbs free energy. Its core idea is that for a multiphase, multi-component closed system composed of given elements, the total Gibbs free energy of the system is minimized when equilibrium is reached under isothermal and isobaric conditions. The model considers the molten pool of the submerged arc furnace as a system containing a metallic phase, a slag phase, and a gas phase, each of which is composed of several components.

[0119] Model calculation of theoretical target temperature and theoretical target components The specific steps are as follows:

[0120] Input current operating parameters: measured tapping temperature Actual alloy composition Moisture content of raw materials Ambient temperature Ambient humidity and electrode input power And the raw material ratio data for the current batch.

[0121] Based on the alloy composition, raw material ratio, and raw material moisture content, calculate the total molar amount of each element entering the molten pool, and thus determine the total material composition vector of the system. .

[0122] Set the initial iteration temperature The thermodynamic database is invoked to calculate the Gibbs free energy of all possible components at the initial iteration temperature.

[0123] Establish the total Gibbs free energy function of the system. It is a function of temperature and phase composition. At a given temperature, It is a linear combination of the molar numbers of each component, and the coefficients are the chemical potentials of each component.

[0124] Using the Lagrange multiplier method, under the conditions of element mass conservation and non-negativity constraints, the solution is obtained. The minimum phase composition is used to obtain the equilibrium phase composition and alloy composition at that temperature.

[0125] Estimate the total enthalpy change of the system based on the electrode input power, adjust the temperature, and repeat the iterative process of "setting the iterative temperature → calculating the Gibbs free energy → solving for the equilibrium phase composition" until the calculated equilibrium temperature matches the input temperature. The temperature obtained at this point is the theoretical target temperature. The corresponding alloy composition is the theoretical target composition. .

[0126] The calculation process is implemented by writing an iterative solution program, and convergence can be accelerated using the Newton-Raphson method.

[0127] Step S303: Calculate the potential energy gap; the central control unit will calculate the actual tapping temperature. Compared with the theoretical target temperature Compare and calculate temperature deviation :

[0128]

[0129] Actual alloy composition With theoretical target components Compare and calculate the deviation of each element's composition. :

[0130]

[0131] Temperature deviation This reflects the difference between the heat released by the reduction reaction in the furnace and the theoretical requirement: if This indicates that the actual heat release is insufficient, and a reducing agent needs to be added to enhance the reaction; if This indicates excessive heat release, and the reducing agent can be appropriately reduced. (Composition deviation) This reflects whether the reduction of each element is insufficient or excessive: if , representing element The actual content is lower than the theoretical value, indicating insufficient reduction, requiring an increase in reducing agent; conversely, the actual content is higher than the theoretical value, requiring a decrease in reducing agent.

[0132] Step S304, reducing agent correction factor Based on the aforementioned reduction potential energy gap and combined with the estimated surface radiative heat loss from the material surface thermal field distribution, the calculation model of the reducing agent correction coefficient is calculated using the following formula. :

[0133]

[0134] The meanings of each symbol are as follows:

[0135] The temperature regulation weighting coefficient is dimensionless and reflects the degree of influence of temperature deviation on the reducing agent requirement. The temperature regulation weighting coefficient is determined by collecting historical operating data and using a multiple linear regression method: Using the amount of reducing agent adjusted in actual production (after normalization) as the dependent variable, the regression yielded... Value. Usually. The value range is from 0.3 to 0.5.

[0136] The theoretical target temperature calculated in step S302 is expressed in degrees Celsius. The actual tapping temperature in step S301 is expressed in degrees Celsius. This is the component adjustment weighting coefficient, dimensionless, reflecting the degree of influence of component deviation on the reducing agent requirement. It is also determined through multiple linear regression, and its value typically ranges from 0.4 to 0.6. The number of elements, including at least silicon, manganese, and chromium. Let be the correlation factor for element i, representing the amount of reducing agent adjustment required for each 1 percentage point deviation in the mass fraction of element i. It is dimensionless, expressed as the mass ratio of reducing agent to ore. This correlation factor is calculated based on the stoichiometric relationship of the reduction reaction. For example, for manganese, the reduction reaction is MnO + C = Mn + CO, and the amount of carbon required to reduce 1 kg of manganese is... kg, therefore Other elements follow the same principle, with their specific values ​​pre-calculated and stored in the system.

[0137] and The meaning is the same as in step S303, which are elements respectively. The theoretical target composition and the actual alloy composition. This is a dimensionless heat loss compensation coefficient that reflects the impact of surface radiative heat loss on the reducing agent demand. The heat loss compensation coefficient is determined through on-site calibration, and its value typically ranges from 0.1 to 0.2. The normalized surface radiative heat loss is dimensionless, and its calculation process is as follows:

[0138] First, based on the panoramic thermal field distribution map obtained in step S1, the average temperature of the material surface inside the electric arc furnace is calculated. (Unit: Kelvin). According to the Stefan-Boltzmann law, the power loss due to surface radiative heat is:

[0139]

[0140] in The surface emissivity is 0.8 to 0.9. It is the Stefan-Boltzmann constant. ; This refers to the total area of ​​the material surface, in square meters. The ambient temperature is expressed in Kelvin.

[0141] Then calculate the After normalization, we get :

[0142]

[0143] in and The minimum and maximum radiative heat loss values ​​recorded during historical operation are updated regularly to ensure... Between 0 and 1.

[0144] Reducing agent correction factor The value range is set between 0.85 and 1.15. When the calculated value exceeds this range, the boundary value is used. When this occurs, it indicates that the amount of reducing agent needs to be increased; when When this occurs, it indicates that the amount of reducing agent needed should be reduced.

[0145] Step S305, reducing agent correction factor The output and limiting processing, the central control unit will calculate the... The value is limited to ensure it remains within the safe range of 0.85 to 1.15. (Processed value) The value is sent to the batching scale controller via industrial Ethernet, serving as the basis for adjusting the reducing agent dosage in subsequent step S5. Simultaneously, the system transmits the calculated input parameters ( ), Output reducing agent correction factor The timestamps are also stored in the historical database for periodic re-regression of the weighting coefficients. This enables online adaptive optimization of the model.

[0146] Through step S3 above, the system establishes a correction coefficient from the iron tapping data to the reducing agent. This complete mapping enables dynamic compensation for the furnace's thermal balance. The reducing agent correction coefficient... This will serve as a key input, working in conjunction with the spatial fabric command generated in step S4, to synchronize the adjustment of the reducing agent dosage and fabric position in step S5, thereby forming a closed-loop control.

[0147] For step S4, local spiking point identification and spatial material placement command generation, after completing the thermal balance compensation model construction and obtaining the reducing agent correction coefficient in step S3. Afterwards, the system has a basis for adjusting the total amount of reducing agent.

[0148] However, for abnormal phenomena such as localized flash points, precise spatial material placement is required for targeted intervention. To address this, the system proceeds to step S4, where real-time analysis of the three-dimensional composite feature map identifies flash points with abnormally high local temperatures and generates corresponding spatial material placement commands to rapidly suppress localized flash points on the material surface. The specific implementation steps are as follows:

[0149] Step S401, Identification and determination of splatter points: The central control unit performs real-time scanning of the three-dimensional composite feature map constructed in step S106 and uses a connected component analysis algorithm to identify local high-temperature areas.

[0150] First, a high temperature pre-screening threshold is set, which is 1100 degrees Celsius. Pixels with temperatures exceeding this threshold are marked as candidate high temperature points.

[0151] Then, through eight-neighbor connectivity analysis, adjacent candidate high-temperature points are grouped into independent high-temperature connected regions, with each region representing a potential candidate ignition point.

[0152] For each identified high-temperature connected region, the system calculates the proportion of its projected area to the total area of ​​the material surface inside the electric arc furnace. The total area of ​​the material surface is based on the furnace design radius. The calculation shows that, The area of ​​the region is obtained by multiplying the number of pixels contained in the feature region by the actual area corresponding to each pixel. The actual area corresponding to each pixel is determined by the coordinate transformation relationship in step S201.

[0153] The system sets the threshold for judging fire. Temperature (Celsius). A high-temperature connected region is considered to have experienced a sparking failure when it simultaneously meets both of the following conditions; this high-temperature connected region is the characteristic region of the sparking point:

[0154] 1. The highest temperature of this feature region in 10 consecutive thermal images all exceeded [a certain value]. ;

[0155] 2. The projected area of ​​this feature region exceeds 2% of the total area of ​​the material surface.

[0156] Monitoring for 10 consecutive frames is to eliminate transient interference and ensure the continuity of the sparking phenomenon. Once a characteristic region is identified as a sparking point, the system immediately records the boundary and features of that region.

[0157] Step S402, Extraction of Flame Point Coordinates and Calculation of Fabric Position: For feature areas identified as flame points, the central control unit extracts the coordinates of their geometric center in polar coordinates. First, the average coordinates of all pixels within this feature area in the actual plane coordinates of the furnace are calculated. Then convert the average coordinates to polar coordinates:

[0158]

[0159] in The distance from the ignition point to the geometric center of the furnace. This is the azimuth angle. (Coordinates) That is, the target position of the fabric feeder.

[0160] Based on the mechanical structure of the fabric feeder, the rotation angle command of the fabric feeder directly corresponds to the azimuth angle. This means the fabric feeder needs to rotate to that azimuth angle; the telescopic displacement command corresponds to... This refers to the radial distance the fabric feeder's chute needs to extend. The system uses these two parameters as the basic command for the fabric position.

[0161] Step S403: Calculation of chute inclination angle and coverage thickness. To ensure that the material accurately covers the area directly above the ignition point, the chute inclination angle needs to be adjusted according to the material level at that location. The system obtains the ignition point coordinates from the 3D geometric surface reconstructed in step S105. Material level height This height is based on the bottom of the furnace.

[0162] The feeder chute is mounted on a rotatable and variable-amplitude mechanism, and the vertical distance from its center of rotation to the bottom of the furnace is... Fixed installation parameters are pre-stored in the system. The chute's inclination angle... Defined as the angle between the chute and the horizontal plane, it is calculated using the following geometric relationship:

[0163]

[0164] in The vertical distance from the rotation center of the chute to the bottom of the furnace is denoted as . This refers to the material level at the ignition point. Radial distance from the ignition point to the geometric center of the furnace. Inclination angle of the chute. Ensure that the material falls accurately to the target point as it slides down the chute.

[0165] The material coverage thickness is automatically calculated based on the degree to which the temperature exceeds the limit in the flame-scalding zone. The degree to which the temperature exceeds the limit is defined. ,in This represents the highest temperature of this feature region in the most recent frame. Coverage thickness. Determined according to the following rules:

[0166] like Degrees Celsius, then set the coverage thickness. millimeters;

[0167] like Degrees Celsius, then set the coverage thickness. Millimeters.

[0168] This rule ensures that the more severe the scorching, the thicker the covering material.

[0169] Step S404: Fabric instruction set generation. Based on the above calculation results, the central control unit generates a complete space fabric instruction set, which includes the following parameters:

[0170] Fabric position: Rotation angle of the rotary fabric feeder and the radial distance from the ignition point to the geometric center of the furnace ;

[0171] Fabric covering time: Calculated based on the volume of material to be covered and the unloading flow rate. Volume of material to be covered. ,in The projected area of ​​the characteristic region of the ignition point. For coverage thickness. Discharge flow rate. The system presets the feeder gate opening and material characteristics, for example, it can be set to a constant value of 50 kg / s. Feeding time. The unit is seconds.

[0172] Discharge flow rate: In this embodiment, the discharge flow rate adopts a preset constant value, which is set by the operator according to the material type and stored in the system.

[0173] The generated instruction set is sent to the fabric feeder servo drive system via industrial Ethernet, waiting for the synchronous execution of step S5.

[0174] In step S4, the system transforms the qualitative identification of the fire-sting phenomenon into specific spatial fabric instructions, providing complete execution parameters for subsequent precise fabric adjustments.

[0175] For step S5, the dosage of reducing agent and the fabric are controlled synchronously in a closed loop; after generating the spatial fabric instruction in step S4, the system has obtained the reducing agent correction coefficient. And specific fabric parameters. For example... Figure 4 As shown, to translate these adjustment commands into actual actions and achieve deep stabilization of the furnace condition, the system proceeds to step S5, where a complete closed-loop control loop is constructed by synchronously adjusting the reducing agent dosage and the charge feeder's actions. Simultaneously, a trend prediction algorithm based on a Long Short-Term Memory (LSTM) neural network and an electrode adjustment linkage mechanism are introduced to achieve early intervention and multi-dimensional collaborative control. The specific implementation steps are as follows:

[0176] Step S501: Synchronous adjustment of the feeding ratio; the central control unit transmits the reducing agent correction coefficient calculated in step S304 via industrial Ethernet. Send to the batching scale controller. The batching scale controller receives the data. After the value is determined, the amount of coke or semi-coke added should be adjusted immediately.

[0177] The adjustment process is achieved by changing the output frequency of the batching scale's frequency converter: when When the value is greater than 1, increase the output frequency of the batching scale inverter and increase the amount of reducing agent added; when When the value is less than 1, reduce the output frequency of the batching scale inverter and decrease the amount of reducing agent added.

[0178] The batching scale controller has a pre-set curve showing the relationship between frequency and reducing agent dosage. This curve is obtained through on-site calibration to ensure adjustment accuracy. The system is required to switch the reducing agent ratio within 30 seconds to ensure timely furnace condition response.

[0179] In step S502, the fabric distributor performs an action. While adjusting the proportion, the central control unit sends the spatial fabric distribution command generated in step S404 to the fabric distributor servo drive system.

[0180] Driven by a servo motor, the rotary fabric feeder rapidly rotates to θ, achieving an angular positioning accuracy of 0.1 degrees to ensure accurate fabric placement. The actuator drives the chute to adjust its amplitude according to the extension distance ρ, while simultaneously adjusting the tilt angle calculated in step S403. Adjust the chute inclination angle.

[0181] The material feeder gate opens according to the discharge flow command, and discharges material according to the preset discharge flow rate Q and material feeding time t. Through the synchronous linkage between the feeding ratio and the material feeding position, the system increases the overall reduction power by increasing the total amount of reducing agent, and maintains the flatness of the material surface by locally pressing the material to extinguish the spiking points.

[0182] Step S503: Trend Prediction and Early Adjustment Based on LSTM. The system introduces a trend prediction algorithm based on a Long Short-Term Memory (LSTM) neural network into the closed-loop control loop to proactively detect changes in furnace conditions and intervene in advance. The training and prediction process of this algorithm is as follows:

[0183] Training Data Construction: Thermal field evolution data for continuous operation periods is extracted from historical databases. Each sample contains a panoramic thermal field distribution map collected every minute over the past 4 hours, totaling 240 time steps. The thermal field map at each time step is downsampled or key area temperature features are extracted, such as the average temperature of the central area and the temperature of each sector, to form an input feature vector sequence. The corresponding label represents the actual thermal field change trend within the next 30 minutes, specifically quantified as the rate of change of the central area temperature over the next 30 minutes or whether a sustained temperature decrease event will occur in the future. The training dataset needs to cover historical data under different operating conditions, including typical furnace conditions such as normal fluctuations, flashover, and flow deviation.

[0184] Network Structure Design: The Long Short-Term Memory (LSTM) neural network employs a multi-layer structure. The input layer dimension corresponds to the feature vector dimension at each time step, the hidden layer contains several LSTM units, and the output layer is a fully connected layer, outputting the temperature prediction value or trend classification result for the next 30 minutes. In this embodiment, the network output is a sequence of predicted central area temperatures for the next 30 minutes or a trend indicator value.

[0185] Loss function design: If regression prediction is used, the loss function is defined as the mean squared error (MSE), which is the squared difference between the predicted and actual temperature values; if classification prediction is used, the cross-entropy loss function is employed. The training objective is to minimize the loss function so that the network can accurately predict future thermal trends.

[0186] Model Training: The constructed training dataset is input into the Long Short-Term Memory (LSTM) neural network. Iterative training is performed using the Adam optimizer, with a batch size of 32 and an initial learning rate of 0.001. The training epochs are stopped based on the validation set error. After training, the LSTM model parameters are fixed and deployed in the central control unit.

[0187] Online Prediction Application: The system collects real-time thermal field evolution data from the past 4 hours. After preprocessing using the same methods as the training data, the data is input into an LSTM model to obtain a prediction of the material surface temperature trend for the next 30 minutes. If the prediction shows a continuous downward trend in the central area temperature, the system will intervene in advance, even if the current tapping temperature is still within the normal range, and adjust the reducing agent correction factor by a predetermined amount. The predetermined range can be dynamically determined based on the predicted rate of descent, for example, adjusted upwards for every 10 degrees Celsius decrease. Value 0.02.

[0188] Step S504, Electrode Adjustment Linkage: The system is equipped with an electrode adjustment linkage module to coordinate with the charge feeding adjustment to achieve deep stability of the furnace condition. When infrared imaging shows that the reaction intensity below a certain phase electrode is too high, the system dynamically fine-tunes the insertion depth of that phase electrode while adjusting the charge feeding.

[0189] The criteria for determining excessive reaction intensity are as follows: based on the characteristic regions divided in step S2, extract the average temperature or standard deviation of the temperature of the corresponding region of the phase electrode. If the average temperature or standard deviation exceeds the preset reaction intensity threshold, it is determined that the reaction intensity is too high.

[0190] The preset threshold is pre-calibrated according to different smelting varieties and electrode positions. For example, for an electrode with a diameter of 1.2 meters, if the average temperature of its corresponding area exceeds 1450 degrees Celsius, it is judged as an excessively high reaction intensity.

[0191] Once the system determines that the reaction intensity is too high, it sends a command to the electrode lifting mechanism to raise the phase electrode by 50 mm to 100 mm. The specific lifting range is determined based on the degree of temperature exceedance: the greater the temperature exceedance, the greater the lifting range. The lifting action is performed synchronously with the fabric adjustment to reduce the local reaction intensity below the electrode, preventing excessive electrode consumption and localized overheating.

[0192] Through the coordinated execution of step S5, the system achieves multi-level closed-loop control, from overall proportion adjustment to local material distribution intervention, and then to trend prediction and early intervention and electrode linkage compensation, ensuring long-term stable and efficient operation of the furnace.

[0193] The intelligent regulation and control method for ferroalloy preparation described in this embodiment was applied to an industrial application verification on a 32000 kVA submerged arc furnace. The statistical results of 90 days of continuous operation are as follows:

[0194] (1) Power consumption index: Before application, the average power consumption per unit alloy was 3850 kWh / t, and after application it was reduced to 3520 kWh / t, a decrease of about 8.6%, saving about 4.8 million kWh of electricity per year.

[0195] (2) Reducing agent consumption: Before application, the average coke consumption was 520 kg / t, which was reduced to 485 kg / t after application, a decrease of about 6.7%, saving about 700 tons of coke per year.

[0196] (3) Furnace lining life: The average service life of the furnace lining before application is 18 months. After application, it is expected to be extended to more than 24 months, an increase of about 33%.

[0197] (4) Control of the burning point: Before application, the burning phenomenon occurred about 12 times per day on average, and the treatment time was about 15 minutes each time; after application, the average number of burning events per day decreased to less than 3 times, the treatment time per event was shortened to less than 3 minutes, and the burning point suppression efficiency was increased by about 80%.

[0198] (5) Temperature fluctuation range: Before application, the temperature fluctuation range in the central area of ​​the furnace was ±85℃. After application, the temperature fluctuation range was reduced to ±35℃, and the thermal stability was significantly improved.

[0199] The application verification data above show that the technical solution proposed in this invention can effectively reduce power consumption and production costs, extend the service life of the furnace lining, and improve the stability and economic benefits of the operation of the submerged arc furnace.

[0200] In summary, the intelligent regulation and control method for ferroalloy preparation and operation proposed in this application constructs a three-dimensional feature map of the material surface through the fusion of infrared imaging and microwave sensing, achieving high-precision perception of the thermal field and morphology. Based on regional quantitative analysis, quantitative indicators for furnace stability evaluation are extracted, and a thermal balance compensation model is established by combining real-time feedback of tapping temperature and alloy composition, and the reducing agent correction coefficient is calculated. For local anomalies such as flash points, charging instructions are generated, and closed-loop control is achieved through the synchronous linkage of charging ratio and charging execution. This method also introduces trend prediction algorithms and electrode adjustment mechanisms to achieve early intervention and multi-dimensional collaboration.

[0201] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention. Therefore, the embodiments should be regarded as exemplary and non-limiting in all respects.

[0202] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A method for intelligent regulation and control of ferroalloy preparation and operation, characterized in that, Includes the following steps: A three-dimensional composite feature map is constructed. The three-dimensional composite feature map is obtained by deep fusion of infrared imaging and microwave sensing to obtain a panoramic thermal field distribution map and three-dimensional geometric shape surface of the material surface in the electric arc furnace. After registration and fusion in the spatial coordinate system, it is generated to simultaneously characterize the absolute temperature of each point on the material surface and its actual altitude in the furnace. Based on the three-dimensional composite feature map, the material surface is divided into polar coordinate grids to form multiple independent feature regions. The temperature standard deviation, the offset of the central thermal field relative to the geometric center of the furnace, and the radial temperature gradient of each feature region are calculated as quantitative indicators for evaluating furnace stability. Based on the real-time collection of tapping temperature and alloy composition during the tapping process, combined with the quantitative indicators for furnace stability evaluation, the reducing agent correction coefficient for adjusting the reducing agent dosage is calculated through a heat balance compensation model. The heat balance compensation model is constructed based on the principle of minimizing Gibbs free energy. Based on the three-dimensional composite feature map, the piercing point with abnormal local temperature rise is identified, and based on the spatial location and morphological features of the piercing point, a spatial fabric instruction containing the fabric position and fabric duration is generated. Synchronously execute the adjustment of reducing agent dosage corresponding to the reducing agent correction coefficient and the action of the material distributor corresponding to the space material distribution command to achieve coordinated closed-loop control of overall thermal balance and local spit-out point suppression in the furnace; The specific steps for constructing a three-dimensional composite feature map include: Multiple infrared detection units deployed in the space above the inside of the electric arc furnace collect the original radiation signals of the material surface. After non-uniformity correction and blackbody calibration, the signals are converted into temperature matrices. The multi-channel temperature matrices are then transformed by perspective and image stitched together to synthesize a panoramic thermal field distribution map covering the material surface inside the electric arc furnace. Multiple frequency-modulated continuous wave microwave radars are used to scan the material surface in a step-by-step manner to obtain the echo signals of the sampling points. The precise distance of each sampling point is obtained by using the difference frequency ranging principle and converted into initial point cloud data. The initial point cloud data is then processed by bilinear interpolation to reconstruct the three-dimensional geometric topography of the material surface inside the electric arc furnace. Using the geometric center of the furnace as a reference, the image coordinates of the panoramic thermal field distribution map are mapped to the three-dimensional grid points of the three-dimensional geometric surface through coordinate transformation, thereby completing the registration and fusion of temperature data and spatial geometric data and constructing a three-dimensional composite feature map, which is used to simultaneously characterize the absolute temperature of each point on the material surface and its actual altitude in the furnace. The specific steps for calculating the reducing agent correction factor using the thermal balance compensation model include: Real-time acquisition of actual iron tapping temperature, actual alloy composition, raw material moisture content, ambient temperature and humidity, and electrode input power during the iron tapping process; The actual tapping temperature, actual alloy composition, raw material moisture content, ambient temperature and humidity, and electrode input power are input into a thermal balance compensation model with a pre-set thermodynamic database. Based on the principle of minimizing Gibbs free energy, the theoretical target tapping temperature and theoretical target alloy composition at thermodynamic equilibrium under the current operating conditions are solved through iterative calculation. The deviation between the theoretical target tapping temperature and the actual tapping temperature, as well as the deviation between the theoretical target alloy composition and the actual alloy composition, are calculated as the reduction potential gap. Based on the reduction potential energy gap and combined with the material surface radiative heat loss estimated by the panoramic thermal field distribution map, the reducing agent correction coefficient is calculated by a multiple regression model that includes temperature adjustment weight coefficient, composition adjustment weight coefficient and heat loss compensation coefficient.

2. The intelligent regulation and control method for ferroalloy preparation and operation according to claim 1, characterized in that, The specific steps for dividing the material surface into polar coordinate grids and calculating quantitative indicators for evaluating furnace stability include: The panoramic thermal field distribution map is transformed from the image coordinate system to the actual plane rectangular coordinate system of the furnace with the geometric center of the furnace as the origin, and then transformed into the polar coordinate system. With the geometric center of the furnace as the origin, the material surface is evenly divided into multiple fan-shaped areas in the circumferential direction, and according to the furnace design, it is divided into multiple concentric rings along the radial radius, including the central area, the middle ring area and the edge area. The characteristic area is formed by the intersection of the fan-shaped areas and the concentric rings. Calculate the temperature standard deviation of the temperature values ​​of all pixels in each feature region as a quantitative indicator to measure the reaction stability of that feature region; The top 5% of pixels with the highest temperature in the panoramic thermal field distribution map are extracted as the high-temperature point set. The geometric centroid coordinates of the high-temperature point set are calculated using the weighted centroid algorithm, which is used as the location of the central thermal field. The Euclidean distance between the central thermal field and the preset coordinates of the furnace geometric center is calculated to obtain the central thermal field offset, which serves as the basis for determining whether the airflow inside the furnace is deflected.

3. The intelligent regulation and control method for ferroalloy preparation and operation according to claim 1, characterized in that, The surface radiative heat loss estimated by the panoramic thermal field distribution map specifically includes: The average temperature of the material surface inside the electric arc furnace was calculated based on the panoramic thermal field distribution map. According to the Stefan-Boltzmann law, the radiative heat loss power is calculated using the average temperature of the material surface, the ambient temperature, the total area of ​​the material surface, and the pre-stored material surface emissivity. The radiative heat loss power is normalized to obtain the normalized surface radiative heat loss, which is then used as input to calculate the reducing agent correction coefficient.

4. The intelligent adjustment and control method for ferroalloy preparation and operation according to claim 1, characterized in that, The specific steps for identifying the piercing point and generating spatial cloth instructions include: The three-dimensional composite feature map is scanned in real time, and the connected component analysis algorithm is used to merge adjacent pixels with temperatures exceeding the preset high temperature threshold into high temperature connected regions. When the highest temperature of the high-temperature connected area in multiple consecutive thermal images exceeds the preset ignition determination threshold, and its projected area exceeds the proportion threshold of the total area of ​​the material surface, it is determined to be an ignition point. Extract the average coordinates of the geometric center of the ignition point in the actual plane coordinates of the furnace, and convert them into polar coordinates with the geometric center of the furnace as the origin to obtain the radial distance and azimuth angle, which are used as the target position of the material distributor. Based on the difference between the highest temperature at the ignition point and the ignition judgment threshold, the thickness of the material to be covered is dynamically determined, and combined with the projected area of ​​the ignition point, the total volume of the material to be covered and the covering time are calculated.

5. The intelligent regulation and control method for ferroalloy preparation and operation according to claim 4, characterized in that, The specific steps for generating space cloth instructions include: Extract the material level height at the coordinates of the ignition point from the three-dimensional composite feature map; Based on the fixed vertical distance from the rotation center of the feeder chute to the bottom of the furnace, the radial distance of the ignition point, and the material level height, the inclination angle of the chute is calculated through geometric relationships to ensure that the material falls accurately to the target point. The azimuth angle, radial distance, chute inclination angle, material placement duration, and preset unloading flow rate of the target location are combined to generate a complete spatial material placement command.

6. The intelligent regulation and control method for ferroalloy preparation and operation according to claim 1, characterized in that, It also includes trend prediction and advance adjustment steps based on long short-term memory neural networks, specifically: Extract continuous thermal field evolution data from historical databases to construct a training dataset. Each sample contains a sequence of feature vectors corresponding to the thermal field distribution maps collected at multiple past time steps, as well as the actual thermal field change trend in the future period as a label. Construct and train a long short-term memory neural network to learn the temporal patterns of thermal field evolution and output a prediction of future thermal field trends; Real-time data on the evolution of the thermal field during the current period is collected and input into a trained long short-term memory neural network to obtain a prediction of the material surface temperature trend over a future period. If the prediction results show that the temperature in the center area of ​​the material surface has a continuous downward trend, then while the current furnace condition is still within the normal range, the reducing agent correction coefficient should be increased in advance by the predetermined amount.

7. The intelligent regulation and control method for ferroalloy preparation and operation according to claim 1, characterized in that, It also includes electrode adjustment linkage steps: Based on the characteristic regions divided by the polar coordinate grid, extract the average temperature or standard deviation of the characteristic regions corresponding to the electrode positions of each phase; When the average temperature or the standard deviation of temperature exceeds the preset reaction intensity threshold, it is determined that the reaction intensity below the electrode of that phase is too high. While simultaneously adjusting the amount of reducing agent and operating the feeder, a command is sent to the electrode lifting mechanism to raise the phase electrode upward by a preset amount in order to reduce the intensity of the local reaction.

8. A smart adjustment and control system for the preparation and operation of ferroalloys, characterized in that, The system for implementing the method according to any one of claims 1 to 7, the system comprising: The multi-dimensional situational awareness unit is used to obtain a panoramic thermal field distribution map and three-dimensional geometric shape surface of the material surface inside the electric arc furnace through the deep fusion of infrared imaging and microwave sensing, and to construct a three-dimensional composite feature map containing temperature dimension and spatial geometric dimension. The regional thermal field quantization analysis unit is used to divide the material surface in the three-dimensional composite feature map into polar coordinate grids, forming multiple feature regions, and to calculate the temperature standard deviation of each feature region and the offset of the central thermal field relative to the geometric center of the furnace. The thermal balance compensation model unit has a built-in thermodynamic database based on the principle of minimizing Gibbs free energy. It is used to calculate the reducing agent correction coefficient for adjusting the amount of reducing agent added, based on the real-time collected iron tapping temperature and alloy composition, combined with the furnace condition index output by the regional thermal field quantitative analysis unit. The local anomaly point identification and cloth placement instruction generation unit is used to identify the piercing point based on the three-dimensional composite feature map and automatically calculate the spatial cloth placement instruction, which includes cloth placement position, cloth placement duration, and chute tilt angle. The collaborative closed-loop control unit is used to synchronously execute the adjustment of the reducing agent dosage corresponding to the reducing agent correction coefficient and the action of the fabric distributor corresponding to the space fabric distribution command.

Images