Intelligent control method and system for processing temperature of refractory air duct

Abstract

This invention relates to the field of temperature control technology in refractory duct processing, and discloses an intelligent temperature control method and system for refractory duct processing. The method includes the acquisition and analysis of real-time temperature data and external environmental fluctuation data to achieve precise temperature control. The system analyzes temperature trends based on an initial temperature distribution map combined with a fuzzy rule base, identifies overheated areas, and adjusts the heating power. Power allocation is optimized through a multi-physics coupling model and historical processing data to generate a temperature equilibrium model. Further, by combining a temperature prediction model and feedback data, the power allocation is iteratively corrected to ensure a stable heating process and eliminate residual temperature differences, ultimately outputting a balanced heat distribution, improving material performance consistency and processing quality. This method can adjust power in real time to cope with dynamic changes in the external environment and material properties, thereby improving processing quality.

Description

Technical Field

[0001] This invention relates to the field of temperature control technology in the processing of refractory ducts, and in particular to an intelligent temperature control method and system for the processing of refractory ducts. Background Technology

[0002] In the process of refractory duct processing, especially in high-temperature processing environments, how to accurately obtain the temperature status during the processing through data acquisition and control methods, and thereby achieve stable and controllable temperature regulation, has become a core issue that urgently needs to be addressed in the field of modern building materials manufacturing.

[0003] Existing temperature control methods suffer from numerous problems when dealing with dynamic changes in the production environment, particularly under conditions of external environmental variations and fluctuations in material thermal response characteristics. These methods fail to provide sufficient response speed and accuracy, leading to uneven temperature distribution and frequent instances of localized overheating or underheating. Current technologies cannot effectively handle the interaction of multiple factors, making it difficult to accurately control temperature changes and impacting processing results. They remain inadequate in terms of precise temperature control, power adjustment, and thermal response compensation, and cannot cope with real-time changes in complex production environments. Therefore, a more precise and intelligent control method is urgently needed.

[0004] Therefore, existing technologies suffer from the problem that due to the lag in temperature response, they cannot adjust the power in real time to cope with the dynamic changes in the external environment and material properties. Summary of the Invention

[0005] This invention provides a method and system for intelligent temperature control in the processing of refractory ducts, which can adjust the power in real time to cope with dynamic changes in the external environment and material properties, thereby improving the processing quality.

[0006] Firstly, in order to solve the above-mentioned technical problems, the present invention provides an intelligent temperature control method for refractory duct processing, comprising:

[0007] Real-time temperature data and external environmental fluctuation data are acquired during the processing of refractory ducts, and the real-time temperature data is preprocessed to obtain an initial temperature distribution map. Based on the initial temperature distribution map, the temperature change trend is analyzed by combining the preset fuzzy rule base to obtain the local overheating probability distribution map and calculate the geometric centroid to determine the coordinates of the overheating area. Based on the coordinates of the overheated area and the real-time temperature data, differential calculation is performed to obtain the material temperature rise rate data. After filtering out power adjustment samples from the preset historical processing database, vector synthesis calculation is performed to obtain the power correction vector. The cumulative temperature rise rate deviation is calculated based on the power correction vector and the real-time temperature data, and a heating power allocation matrix is ​​constructed by combining the initial temperature distribution map and the power correction vector. Based on the mapping relationship between the heating power allocation matrix and the external environmental fluctuation data, a feedforward compensation gain value is generated, and the heating power allocation matrix is ​​corrected by combining a preset closed-loop state observer to obtain a temperature equilibrium model. Based on the temperature equilibrium model, temperature prediction is performed in conjunction with a preset temperature prediction model to obtain a predicted temperature sequence. The predicted temperature sequence is then substituted back into the temperature equilibrium model to output a stable process path. Based on the stable process path, the two-dimensional difference of the preset ideal uniform temperature field is calculated and a gradient deviation matrix is ​​generated. Based on the gradient deviation matrix, the local temperature difference abnormal region is located and the corresponding iterative power correction vector is generated. The iterative power correction vector is input into a preset multiphysics coupling model to obtain a material performance consistency index. Based on the material performance consistency index, a thermal diffusion equilibrium network is constructed and residual temperature difference is eliminated to obtain a balanced heat distribution state.

[0008] Secondly, the present invention provides an intelligent control system for the processing temperature of refractory ducts, comprising: The data processing module is used to acquire real-time temperature data and external environmental fluctuation data during the processing of refractory ducts, and to preprocess the real-time temperature data to obtain an initial temperature distribution map. The location coordinate module is used to analyze the temperature change trend based on the initial temperature distribution map and a preset fuzzy rule base, obtain a local overheating probability distribution map, and calculate the geometric centroid to determine the location coordinates of the overheating zone. The power correction module is used to perform differential calculations based on the coordinates of the overheated area and the real-time temperature data to obtain the material temperature rise rate data, and then perform vector synthesis calculations after filtering out power adjustment samples from a preset historical processing database to obtain the power correction vector. The power allocation module is used to calculate the cumulative temperature rise rate deviation based on the power correction vector and the real-time temperature data, and to construct a heating power allocation matrix by combining the initial temperature distribution map and the power correction vector. The temperature equalization module is used to generate a feedforward compensation gain value based on the mapping relationship between the heating power distribution matrix and the external environment fluctuation data, and to perform correction on the heating power distribution matrix in conjunction with a preset closed-loop state observer to obtain a temperature equalization model. The process curve module is used to perform temperature prediction based on the temperature equalization model and combined with the preset temperature prediction model to obtain the predicted temperature sequence, and then substitute the predicted temperature sequence back into the temperature equalization model to output a stable process path. The iterative power module is used to calculate the two-dimensional difference of the preset ideal uniform temperature field and generate a gradient deviation matrix according to the stable process path, locate the local temperature difference abnormal region according to the gradient deviation matrix and generate the corresponding iterative power correction vector. The state distribution module is used to input the iterative power correction vector into a preset multiphysics coupling model to obtain a material performance consistency index. Based on the material performance consistency index, a thermal diffusion equilibrium network is constructed and residual temperature difference is eliminated to obtain a uniform heat distribution state.

[0009] Compared with the prior art, the present invention has the following beneficial effects: (1) This invention obtains real-time temperature data and external environmental fluctuation data, and analyzes the temperature change trend in combination with a fuzzy rule base, which can predict local overheating areas and thus determine the coordinates of the overheating area. The material temperature rise rate data is obtained through differential operation and matched with the historical processing database to select the most suitable power adjustment sample, thereby realizing accurate power correction vector calculation, effectively avoiding the problem of excessive temperature difference during processing, and improving the stability and consistency of the refractory air duct processing process.

[0010] (2) In this invention, by introducing the derivation of instantaneous energy dissipation rate and the identification of thermal response lag time, and combining the feedforward compensation gain numerical correction of the heating power allocation matrix, the thermal response lag effect in the processing process can be reduced, the power correction vector can be adjusted in real time during the heat treatment process, the dynamic stability of the temperature equilibrium model can be further enhanced, and the external environmental fluctuations can be effectively compensated. Through the application of virtual heat treatment calculation and heat diffusion balance network, the energy utilization efficiency in the heat treatment process is optimized.

[0011] (3) The present invention combines the stable process path with the feedback temperature signal of the sensor array, which can dynamically adjust and update the power correction vector, eliminate the abnormal area of ​​local temperature difference, and ensure the uniformity of temperature distribution and the stability of the processing through the Kriging space interpolation algorithm and the iterative correction control strategy. Finally, by combining the material performance consistency index and the heat diffusion balance network, the residual temperature difference is eliminated and a balanced heat distribution state is achieved. Attached Figure Description

[0012] Figure 1 This is a schematic diagram of the intelligent temperature control method for refractory duct processing provided in the first embodiment of the present invention; Figure 2 This is a schematic diagram of the intelligent control system for refractory duct processing temperature provided in the second embodiment of the present invention. Detailed Implementation

[0013] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0014] Reference Figure 1 The first embodiment of the present invention provides a method for intelligent control of the processing temperature of refractory air ducts, including the following steps: S11, acquire real-time temperature data and external environmental fluctuation data during the processing of refractory ducts, and preprocess the real-time temperature data to obtain an initial temperature distribution map; S12, Based on the initial temperature distribution map, the temperature change trend is analyzed by combining the preset fuzzy rule base to obtain the local overheating probability distribution map and calculate the geometric centroid to determine the coordinates of the overheating area; S13. Based on the coordinates of the overheated area and the real-time temperature data, perform differential calculation to obtain the material temperature rise rate data, and combine it with the preset historical processing database to filter out power adjustment samples and then perform vector synthesis calculation to obtain the power correction vector. S14, calculate the cumulative temperature rise rate deviation based on the power correction vector and the real-time temperature data, and construct the heating power allocation matrix by combining the initial temperature distribution map and the power correction vector; S15. Based on the mapping relationship between the heating power distribution matrix and the external environment fluctuation data, a feedforward compensation gain value is generated, and the heating power distribution matrix is ​​corrected in conjunction with a preset closed-loop state observer to obtain a temperature equilibrium model. S16. Based on the temperature equilibrium model, temperature prediction is performed in combination with the preset temperature prediction model to obtain a predicted temperature sequence. The predicted temperature sequence is then substituted back into the temperature equilibrium model to output a stable process path. S17. Based on the stable process path, calculate the two-dimensional difference of the preset ideal uniform temperature field and generate a gradient deviation matrix. Based on the gradient deviation matrix, locate the local temperature difference abnormal region and generate the corresponding iterative power correction vector. S18, the iterative power correction vector is input into a preset multiphysics coupling model to obtain a material performance consistency index. Based on the material performance consistency index, a thermal diffusion balance network is constructed and a balanced heat distribution state is obtained after eliminating residual temperature difference.

[0015] In step S11, the real-time temperature data needs to be preprocessed to obtain an initial temperature distribution map, including: The real-time temperature data is parsed into multi-point discrete temperature values; Retrieve the physical space coordinate matrix of the sensor array, map the multi-point discrete temperature values ​​to the physical space coordinate matrix, and establish a discrete temperature point set; The Kriging interpolation algorithm is used to spatially interpolate the discrete temperature point set to generate continuous thermal field data. The continuous thermal field data is linearly normalized according to a preset temperature range to generate a grayscale pixel value matrix. The grayscale pixel value matrix is ​​mapped to an image pixel array according to two-dimensional grid coordinates, and the initial temperature distribution map of the refractory duct processing process is output.

[0016] During the fabrication of refractory ducts, real-time temperature data is first collected using a sensor array. This array is located at multiple positions within the heating zone, enabling the acquisition of temperature values ​​at each point. For example, eight thermocouple sensors are used, distributed across different locations within the duct's heating zone. The collected data includes voltage signals at each sensor point, such as 3.8mV and 5.2mV. Using a pre-defined thermocouple calibration table, these voltage signals are converted into corresponding temperature values. For instance, according to the thermocouple calibration table, 3.8mV corresponds to 320℃, and 5.2mV corresponds to 415℃. This step then analyzes the temperature values ​​measured by each sensor to obtain discrete temperature values ​​for each point within the heating zone.

[0017] It should be noted that the construction of the thermocouple calibration table is based on experimental data and the standard characteristics of thermocouple types. Specifically, the method involves experimentally measuring the voltage output of thermocouples at different temperatures, and then using interpolation to generate a mapping relationship between voltage and temperature based on the measured voltage value and temperature. Standard thermocouples, such as type K and type J, are typically used, and calibration is performed according to international standards such as IEC and ASTM. By measuring voltage values ​​at multiple known temperature points, a linear or non-linear voltage-temperature correspondence is established and converted into a calibration table. In this way, the voltage signal measured by the sensor can be directly used to calculate the corresponding temperature value through table lookup or interpolation, ensuring the accuracy and reliability of the temperature data.

[0018] In this step, the physical space coordinate matrix of the sensor array is first retrieved. This matrix stores the two-dimensional position coordinates of each sensor on the heating surface of the duct, for example, along the length and width of the duct. Assuming one end of the duct is the origin, the coordinates of sensor 1 are (0, 10) cm, sensor 2 is (30, 10) cm, and sensor 3 is (60, 15) cm. These two-dimensional coordinates are combined with the corresponding discrete temperature values ​​to establish a discrete temperature point set, where each point consists of position coordinates and a temperature value.

[0019] The Kriging interpolation algorithm is used to spatially interpolate the discrete temperature point set. This algorithm, based on the spatial autocorrelation of temperature, smoothly transitions discrete temperature points into continuous thermal field data. In implementation, the spatial distance between each discrete point is first calculated, and the spatial autocorrelation of temperature is modeled using a variogram function. Through interpolation, the Kriging algorithm can estimate the temperature value at each unsampled location, generating continuous thermal field data covering the entire heating area. This method can effectively fill in areas that sensors cannot cover, generating smooth thermal field data.

[0020] Based on the generated continuous thermal field data, the system first determines the display range for temperature visualization and sets color mapping thresholds. These thresholds include a lower display temperature threshold and an upper display temperature threshold, used to limit the effective temperature range for grayscale mapping calculations. In this embodiment, the lower display temperature threshold is set to 320℃, and the upper display temperature threshold is set to 580℃. Temperatures below 320℃ in the continuous thermal field data are uniformly treated as 320℃, and temperatures above 580℃ are uniformly treated as 580℃.

[0021] The temperature values ​​within this range are then linearly normalized, mapping 320℃ to a grayscale value of 0, 580℃ to a grayscale value of 255, and other temperatures proportionally. For example, a node temperature of 480℃ corresponds to a grayscale value of approximately 157 within the 320℃ to 580℃ range. By setting this color mapping threshold, the impact of local extreme values ​​on the overall image contrast can be suppressed, avoiding overall grayscale compression due to individual hot spots, thus ensuring a stable display resolution for the initial temperature distribution map.

[0022] In this embodiment, the color mapping threshold is determined based on the process safety range and historical statistical data of the refractory duct material. Specifically, before formal operation, statistical analysis is performed on nearly 100 batches of qualified processing data to calculate the statistical distribution range of steady-state temperature during the processing stage. The statistical results show that the material temperature under normal process conditions is mainly distributed between 330℃ and 570℃, with 95% of the effective processing data falling within this range. Meanwhile, according to the material thermal stability test report, the refractory material has not yet entered the effective sintering stage below 300℃, while above 600℃ there is a risk of abnormal grain growth. Therefore, the lower limit is set to 320℃, and the upper limit is set to 580℃. This range covers the actual effective processing temperature range and reserves a safety margin of approximately ±10℃.

[0023] Based on the constructed grayscale pixel value matrix, the matrix is ​​mapped to an image pixel array according to the two-dimensional grid coordinate order to generate an initial temperature distribution map. In specific implementation, the row and column indices of the grayscale pixel value matrix are mapped to the vertical and horizontal coordinates of the image pixels. For example, when the heating area is divided into 100×60 spatial grids, the grayscale pixel value matrix has a size of 100 rows and 60 columns. The system establishes an image buffer of the same size, and each matrix element is written into the image buffer as the grayscale value of the corresponding pixel.

[0024] During image generation, an 8-bit grayscale image format is used, with each pixel's grayscale value ranging from 0 to 255. First, the system stores the grayscale value of each grid point in a grayscale buffer, with each element in the buffer corresponding to a pixel in the image. Then, the data in the grayscale buffer is mapped to a two-dimensional image through an image processing interface, generating a grayscale heatmap according to the set color levels. The rendered image is the initial temperature distribution map, and the grayscale changes in the image visually reflect the temperature distribution of the heated area.

[0025] The final output initial temperature distribution map is a two-dimensional temperature heat map with an image resolution consistent with the grid division, such as 100×60 pixels, and a refresh cycle set to 1 second to achieve real-time monitoring.

[0026] In step S12, based on the initial temperature distribution map and combined with a preset fuzzy rule base, it is necessary to analyze the temperature change trend, obtain a local overheating probability distribution map, and calculate the geometric centroid to determine the coordinates of the overheating region, including: The initial temperature distribution map is convolved using a preset spatial gradient operator to obtain a temperature gradient feature matrix; By combining the preset fuzzy rule base, the temperature gradient feature matrix is ​​input into the preset fuzzy inference model, and a local overheating probability distribution map is output. When the pixel probability value corresponding to the local overheating probability distribution map is greater than the preset probability threshold, the connected component labeling algorithm is used to aggregate the pixel points corresponding to the pixel probability value that is greater than the preset probability threshold to obtain the connected component of the suspected overheating area. Calculate the geometric centroid of the connected domain of the suspected overheated area, and convert the geometric centroid into the physical spatial coordinates of the heating surface of the fire-resistant air duct based on a preset spatial mapping relationship matrix to obtain the coordinates of the overheated area.

[0027] First, a pre-defined spatial gradient operator is used to convolve the initial temperature distribution map to extract edge features of temperature changes. The pre-defined spatial gradient operator is the Sobel operator, which convolves the image in both the horizontal and vertical directions. For each pixel, a 3×3 neighborhood window centered on it is taken. The horizontal Sobel template is convolved with the gray values ​​within this window; that is, the gray values ​​of each neighboring pixel are multiplied by their corresponding weights in the template and then summed to obtain the horizontal gradient value. Similarly, a vertical Sobel template is used for convolution to obtain the vertical gradient value. The weights of the horizontal Sobel template are: first row [-1,0,1]; second row [-2,0,2]; third row [-1,0,1]. The weights of the vertical Sobel template are: first row [-1,-2,-1]; second row [0,0,0]; third row [1,2,1]. Through the above operations, the horizontal and vertical gradient values ​​of each pixel are obtained, and then the temperature gradient amplitude is synthesized to form a temperature gradient feature matrix with the same size as the initial temperature distribution map.

[0028] For example, suppose a pixel has a grayscale value of 180, corresponding to a temperature of 460°C, and the adjacent pixel to its right has a grayscale value of 140, corresponding to 380°C. Then the horizontal gradient is -40 grayscale units, representing the rate of temperature change at that point. Similarly, the temperature change in the vertical direction is also calculated. Combining the horizontal and vertical gradients yields the temperature gradient feature matrix for that region, reflecting the variations in the temperature field across different areas.

[0029] To analyze the risk of localized overheating, the temperature gradient feature matrix is ​​input into a pre-defined fuzzy inference model, which performs inference based on a pre-generated fuzzy rule base. First, two input variables are defined: temperature gradient magnitude and directional consistency. The temperature gradient magnitude is obtained by combining the aforementioned horizontal and vertical gradient values ​​through energy calculation, i.e., calculating the sum of squares and then taking the square root; its value reflects the severity of temperature changes in the region. Directional consistency is measured by statistically analyzing the average deviation of the gradient directions of the target pixel and its eight neighboring pixels. The smaller the deviation, the higher the directional consistency, indicating a more consistent temperature change trend in the region; the larger the deviation, the lower the directional consistency, suggesting the possible presence of chaotic heat flow disturbances.

[0030] The specific calculation method for the directional consistency index is as follows: First, the gradient direction angle of each pixel is calculated using the arctangent function. Then, the average of the absolute differences between the gradient direction angles of the target pixel and its eight neighboring pixels is calculated. This average is then normalized by dividing by 180° to obtain a directional consistency value ranging from 0 to 1. Analysis of actual processing data shows that the directional consistency in the normal heating region is generally greater than 0.65, while in the overheating abrupt change region it is usually lower than 0.4. Based on this, the directional consistency is divided into three fuzzy intervals, described using triangular membership functions: the membership function parameters for the low consistency interval are [0, 0.2, 0.45]; the parameters for the medium consistency interval are [0.4, 0.525, 0.7]; and the parameters for the high consistency interval are [0.6, 0.825, 1]. Adjacent intervals form an overlapping region with a width of 0.05 between 0.4-0.45 and 0.6-0.7 to ensure smooth fuzzy transitions.

[0031] To establish a fuzzy rule base, the consistency of temperature gradient amplitude and direction was first quantified and classified. The temperature gradient amplitude was derived from the temperature gradient feature matrix calculated by the spatial gradient operator. Its numerical range was determined through statistical analysis of 50 batches of qualified processing data. 95% of the gradient amplitudes were distributed between 0 and 200 grayscale units, with a very small number of peaks exceeding 200 treated as 200. According to the statistical results, approximately 70% of the gradient values ​​were distributed between 0 and 60, approximately 25% were distributed between 60 and 150, and less than 5% exceeded 150. Therefore, the gradient amplitude was divided into three fuzzy intervals, described using triangular membership functions: the membership function parameters for the low gradient interval were [0, 30, 65]; for the medium gradient interval, the parameters were [55, 105, 155]; and for the high gradient interval, the parameters were [145, 155, 200]. Adjacent intervals formed an overlapping region of 10 width between 55-65 and 145-155 to ensure smooth fuzzy transitions.

[0032] Based on the aforementioned membership function intervals, the fuzzy rule base contains nine rules. Each rule assigns an overheating probability level based on a combination of gradient magnitude and direction consistency. The output variable overheating probability also uses a triangular membership function, dividing it into three fuzzy sets: low, medium, and high, with a universe of discourse ranging from 0 to 1. The low probability interval is 0 to 0.3, with membership function parameters [0, 0.15, 0.35]; the medium probability interval is 0.3 to 0.7, with parameters [0.3, 0.5, 0.75]; and the high probability interval is 0.7 to 1, with parameters [0.65, 0.85, 1]. Adjacent intervals form an overlapping region with a width of 0.05 between 0.3-0.35 and 0.65-0.75.

[0033] The rule base is as follows: if the gradient magnitude is high and the directional consistency is low, then the overheating probability is high; If the gradient magnitude is high and the directional consistency is medium, the overheating probability is high; if the gradient magnitude is high and the directional consistency is high, the overheating probability is medium; if the gradient magnitude is medium and the directional consistency is low, the overheating probability is medium; if the gradient magnitude is medium and the directional consistency is medium, the overheating probability is medium; if the gradient magnitude is medium and the directional consistency is high, the overheating probability is low; if the gradient magnitude is low and the directional consistency is low, the overheating probability is medium; if the gradient magnitude is low and the directional consistency is medium, the overheating probability is low; if the gradient magnitude is low and the directional consistency is high, the overheating probability is low.

[0034] The inference process employs a Mamdani-type min-max composition method, and the defuzzification uses the centroid method, calculating for each pixel point by point to ultimately generate a local overheating probability distribution map with the same size as the original image. When the probability value of a pixel in the local overheating probability distribution map exceeds a preset probability threshold (0.75 in this embodiment), a connected component labeling algorithm is used to aggregate these pixels into independent suspected overheating region connected components. Connected component labeling can be based on four-neighbor or eight-neighbor domains; this embodiment uses eight-neighbor domains to ensure region integrity. For each suspected overheating region connected component, its geometric centroid is calculated, which is the average of the coordinates of all pixels. For example, if a connected component contains pixel coordinates (45,30), (46,30), and (47,31), then the geometric centroid is (46,30,33).

[0035] To convert the geometric centroid in the image coordinate system into the physical coordinates of the actual heating surface, a pre-established spatial mapping matrix needs to be created. This matrix is ​​obtained through offline calibration: during equipment installation, at least 12 physical calibration points are selected on the heating surface of the refractory duct, and their physical coordinates (in cm) are accurately recorded using measuring tools. Simultaneously, the pixel coordinates of the corresponding points are read from the initial temperature distribution map. Based on these corresponding point pairs, a two-dimensional affine transformation model (including translation, scaling, and rotation parameters) is used for fitting, and the transformation coefficients are solved using the least squares method to obtain the pre-established spatial mapping matrix. This matrix is ​​used consistently during system operation. Subsequently, when the pixel coordinates of the geometric centroid are calculated, they are directly calculated with this matrix to convert them into the corresponding physical spatial coordinates, i.e., the coordinates of the overheated area. The calibration process error is controlled within ±0.5 cm, meeting the accuracy requirements for engineering applications.

[0036] In step S13, differential calculations are performed based on the coordinates of the overheated zone and the real-time temperature data to obtain material temperature rise rate data. Then, after filtering power adjustment samples from a preset historical processing database, vector synthesis calculations are performed to obtain a power correction vector, including: The temperature rise rate data is obtained by performing a differential operation based on the coordinates of the overheated zone and the real-time temperature data. The material temperature rise rate data is input into a preset historical processing database to calculate the Euclidean distance. The historical data with the smallest Euclidean distance is selected from the temperature rise rate data as the power adjustment sample. Extract the historical adjustment values ​​from the power adjustment sample and calculate the heat input suppression amplitude by combining them with the overheating zone location coordinates; The power control matrix of a multi-channel system is constructed using the heat input suppression amplitude and vector synthesis is performed to obtain the power correction vector.

[0037] Based on the determined coordinates of the overheated zones, temperature information at the corresponding locations is extracted from real-time temperature data, and a differential operation is performed to calculate the material's temperature rise rate. The differential operation calculates the temperature change rate by comparing the temperatures at adjacent time points. For example, if the temperature at a certain location is 450°C one minute and 455°C the next, the temperature rise rate at that location is (455°C - 450°C) / 1 min = 5°C / min. This operation is performed on all spatial points corresponding to the overheated zone coordinates to obtain a material temperature rise rate vector composed of the temperature rise rates at each point.

[0038] The material temperature rise rate vector is input into a preset historical processing database for Euclidean distance calculation. This database stores historical material temperature rise rate vectors under different processing conditions and their corresponding heating power adjustment records. The database is built based on data from multiple processing experiments, and each record also includes environmental data such as ambient temperature, humidity, and wind speed. By calculating the Euclidean distance between the current material temperature rise rate vector and each historical record vector in the database, the record with the smallest distance is selected as the power adjustment sample, and the historical power adjustment values ​​(usually expressed as percentages, such as a 20% decrease for channel 1 and a 15% increase for channel 2) are extracted from this sample.

[0039] To determine the required heat input suppression amplitude for the current heating zone, it is necessary to consider the real-time temperature rise rate deviation, the spatial heat dissipation coefficient, and the power adjustment values ​​from historical samples. First, the temperature rise rate deviation is calculated, which is the difference between the standard process temperature rise rate and the actual temperature rise rate in the current zone. For example, if the standard temperature rise rate is 2.0°C / min and the current zone's temperature rise rate is 3.5°C / min, then the rate deviation is 1.5°C / min. According to the system's preset rules, such as a 15% power adjustment ratio corresponding to every 1°C / min rate deviation, the basic suppression ratio is calculated: 1.5°C / min corresponds to 22.5%.

[0040] Secondly, the shortest distance from the overheated area to the edge of the duct is determined based on the coordinates of the overheated area, and the heat dissipation coefficient is calculated accordingly. If the distance is less than 5cm, the heat dissipation coefficient is taken as 1.2; if the distance is greater than 20cm, the heat dissipation coefficient is taken as 1.0; if the distance is between 5cm and 20cm, the heat dissipation coefficient is calculated by linear interpolation, for example, the coefficient corresponding to distance d is 1.2-(d-5)×(0.2 / 15). This coefficient is used to correct for the impact of edge heat exchange on power demand.

[0041] Finally, the basic suppression ratio is corrected using the power adjustment values ​​of historical samples. The k most similar historical samples to the current operating conditions are selected (k=3 in this embodiment). The similarity weight is calculated based on the Euclidean distance between each sample and the current material temperature rise rate data, using the formula w_i=1 / (d_i+ε), where d_i is the Euclidean distance and ε is a small constant. The historical adjustment values ​​of the selected samples are weighted and averaged to obtain the historical correction value. The basic suppression ratio and the historical correction value are then weighted and fused according to a preset fusion weight, for example, 50% each. This weight can be calibrated through process experiments. The result is then multiplied by the spatial heat dissipation coefficient to obtain the final heat input suppression amplitude.

[0042] The small constant is set based on the numerical order of magnitude. According to the typical order of magnitude of Euclidean distance in the preset historical processing database, ε is set to a value one order of magnitude lower than the minimum non-zero Euclidean distance. For example, if the minimum Euclidean distance between historical samples is about 0.01, then ε can be set to 0.001 to ensure that it does not affect the relative size of the weights, but only serves to avoid division by zero.

[0043] The heat input suppression amplitude is used as a regional correction coefficient, and a power influence coefficient matrix is ​​constructed by combining the correspondence between heating channels and physical space. The system pre-divides the heating surface into an 8×8 spatial grid, and obtains the proportional coefficient of each grid affected by each heating channel through calibration, i.e., the channel influence coefficient table. The calibration method is to measure the temperature rise contribution rate of each grid under single-channel full power output conditions, and form coefficients after normalization. When an overheated area is identified, the corresponding channel influence coefficient is read according to the grid covered by the area to form a channel influence vector.

[0044] For example, assuming that after heat dissipation coefficient correction and historical correction, the final heat input suppression amplitude of the overheated area is 25%, based on the channel influence coefficients of the grid covering this area (e.g., channel 1 contributes 40%, channel 2 contributes 30%, channel 3 contributes 20%, and channel 4 contributes 10%), the power adjustment amount for each channel is calculated: channel 1 increases by 25% × 40% = 10%, channel 2 increases by 25% × 30% = 7.5%, channel 3 increases by 25% × 20% = 5%, and channel 4 increases by 25% × 10% = 2.5%, thus obtaining the power correction vector for this area [+10%, +7.5%, +5%, +2.5%]. If multiple overheated areas exist, the power correction vectors of each area are superimposed to form a multi-channel power control matrix. Finally, the updated power control vector, i.e., the final power correction vector, is obtained through matrix operations. This power correction vector is used to adjust the power output of each heating channel to ensure uniform temperature distribution throughout the heating area and avoid local overheating or unevenness.

[0045] In step S14, the temperature rise rate data needs to be calculated based on the power correction vector and the real-time temperature data to obtain the cumulative temperature rise rate deviation and the heating power allocation matrix. This is then combined with the initial temperature distribution map and the power correction vector to construct a spatially weighted power distribution matrix, resulting in the heating power allocation matrix, which includes: The power correction vector and the real-time temperature data are input into a preset long short-term memory network model to generate a predicted temperature evolution trajectory sequence. Perform differential calculations between the predicted temperature evolution trajectory sequence and the preset standard process reference curve to obtain the temperature rise rate deviation sequence; The cumulative temperature rise rate deviation is obtained by integrating the temperature rise rate deviation sequence. If the cumulative temperature rise rate deviation exceeds the preset allowable fluctuation threshold, the cumulative temperature rise rate deviation is arranged in a gradient to obtain the deviation distribution gradient. Based on the deviation distribution gradient, the initial temperature distribution map is retrieved to obtain the temperature fluctuation amplitude, and the node energy weight coefficient is calculated based on the temperature fluctuation amplitude. The power correction vector is multiplied by matrix multiplication using the node energy weight coefficients to construct a spatially weighted power distribution matrix as the heating power allocation matrix.

[0046] The power correction vector and real-time temperature data are input into a pre-defined long short-term memory (LSTM) network model for processing. This model learns temperature evolution patterns based on historical time-series data to predict future temperature changes. The input data is organized by time step, with each time step containing the real-time temperature values ​​of each temperature measurement point in the heating zone and the power correction values ​​of each heating channel. In this embodiment, there are 4 temperature measurement points and 4 heating channels, so the input for each time step is an 8-dimensional vector, i.e., 4 temperature values ​​and 4 power correction values. The sampling period is set to 30 seconds, and the input sequence length is 20 time steps, covering the historical data of the most recent 10 minutes. To ensure the stability of model training, all input data are min-max normalized, mapping the temperature values ​​and power correction values ​​to the [0,1] interval, respectively.

[0047] The network structure uses a single-layer LSTM with 50 hidden units, followed by a fully connected layer to output the predicted temperature values ​​of each temperature measurement point at the next time step; therefore, the output layer has 4 nodes. The hyperbolic tangent function is used as the activation function, and the output layer uses linear activation. The Adam optimization algorithm is used with an initial learning rate of 0.001, a batch size of 64, and a maximum of 1000 training epochs. An early stopping mechanism is introduced, stopping training when the validation set loss does not decrease for 50 consecutive epochs. The training data comes from continuously recorded processing data from an actual production line, accumulating to no less than 300 hours, with a sampling frequency of 30 seconds, resulting in approximately 36,000 time-series samples. The data is divided into training, validation, and test sets in an 8:1:1 ratio, and the root mean square error of the model on the test set after training is controlled within 0.8°C.

[0048] After training, the model is deployed to the control system. During real-time operation, the data from the most recent 20 time steps are input into the model, which outputs the predicted temperature for the next moment. Then, new predicted values ​​are added to the input sequence in a rolling manner, the earliest time is removed, and the temperature evolution trajectory for the next 10 minutes is iteratively generated to obtain the predicted temperature evolution trajectory sequence, which contains the temperature values ​​of all temperature measurement points at each predicted time point.

[0049] The predicted temperature evolution trajectory sequence is compared with a preset standard process reference curve. The preset standard process reference curve is determined based on material heat treatment experimental data and reflects the ideal temperature rise process, typically with a linear temperature increase rate of 0.5°C per minute (the temperature remains constant if a holding period is included). Using the current starting temperature as a reference, a target temperature sequence is generated according to the sampling period. The difference between the predicted temperature and the standard temperature is calculated at each time point to obtain a temperature deviation sequence, i.e., a temperature rise rate deviation sequence. This sequence is then integrated, i.e., the deviation value at each time point is multiplied by the sampling period (30 seconds = 0.5 minutes) and accumulated to obtain the cumulative temperature rise rate deviation, with dimensions in °C·min. This cumulative value reflects the overall degree of temperature rise deviation.

[0050] If the cumulative temperature rise rate deviation exceeds the preset allowable fluctuation threshold (set to 30°C·min in this embodiment based on material thermal stress experiments), temperature control adjustment is triggered. At this point, it is necessary to further locate local anomaly regions based on the temperature rise deviation of each spatial grid. Therefore, the temperature deviation sequence of each spatial grid is first integrated to obtain the cumulative temperature rise rate deviation of each grid, forming a cumulative deviation distribution matrix. Then, the deviation distribution gradient matrix is ​​calculated based on this matrix, specifically by spatially differentiating the cumulative deviation using the finite difference method. The gradient value of each grid node reflects the degree of drastic change in the cumulative deviation in that region. The gradient values ​​are sorted from largest to smallest, and the top 20% of nodes are selected as high-gradient regions, i.e., key correction areas.

[0051] Each high-gradient region corresponds to a spatial grid. Historical temperature values ​​from the three most recent sampling periods are extracted from the initial temperature distribution map for that grid, and the temperature fluctuation amplitude is calculated, i.e., the difference between the highest and lowest temperatures within that time window. The fluctuation amplitudes of all high-gradient regions are normalized to obtain the node energy weight coefficient for each region. For example, if the fluctuation amplitudes of the three high-gradient regions are 4°C, 3°C, and 3°C, then the weights are 0.4, 0.3, and 0.3, respectively.

[0052] The node energy weight coefficient characterizes the sensitivity of each region to temperature fluctuations and needs to be mapped to the heating channel level to generate a spatially weighted power distribution matrix, i.e., a heating power allocation matrix. A coverage ratio matrix (channel influence coefficient table) between heating channels and spatial grids is pre-established, recording the proportion of each grid affected by each channel. For each high-gradient region, the corresponding channel influence coefficient is read based on the grid it covers, and the region weight is allocated to each channel according to the influence coefficient, obtaining the contribution weight of each channel to that region. The channel contribution weights of all high-gradient regions are superimposed channel by channel to obtain the comprehensive weight coefficient of each channel. For example, if the comprehensive weight of channel 1 is 0.4, channel 2 is 0.3, channel 3 is 0.2, and channel 4 is 0.1, then the channel weight vector is [0.4, 0.3, 0.2, 0.1].

[0053] The final heating power allocation adjustment is obtained by element-wise multiplying the channel weight vector with the current power correction vector. For example, if the current power correction vector is [+30%, +20%, +10%, +25%], then the weighted adjustment for each channel is: Channel 1 increases by 30% × 0.4 = 12%, Channel 2 increases by 20% × 0.3 = 6%, Channel 3 increases by 10% × 0.2 = 2%, and Channel 4 increases by 25% × 0.1 = 2.5%, resulting in a new power correction vector of [+12%, +6%, +2%, +2.5%]. This vector is used to update the heating power allocation matrix, enabling priority power compensation for temperature-sensitive areas and ensuring a uniform and stable heating process. This updated matrix is ​​the spatially weighted power distribution matrix, which serves as the final heating power allocation matrix.

[0054] In step S15, a feedforward compensation gain value needs to be generated based on the mapping relationship between the heating power allocation matrix and the external environmental fluctuation data. This value is then used in conjunction with a preset closed-loop state observer to correct the heating power allocation matrix, resulting in a temperature equilibrium model, including: Establish the mapping relationship between the heating power distribution matrix and the external environmental fluctuation data, and derive the instantaneous energy dissipation rate; The thermal response lag time is identified based on the instantaneous energy dissipation rate, and a feedforward compensation gain value matching the thermal response lag time is generated. Using the feedforward compensation gain value, superposition correction is performed on the heating power distribution matrix to generate environmental adaptive correction control variables; The environmental adaptive correction control variables are input into a preset closed-loop state observer to obtain a temperature equilibrium model.

[0055] First, a quantitative mapping relationship between the heating power allocation matrix and external environmental fluctuation data is established to calculate the instantaneous energy dissipation rate of each area. During the equipment commissioning phase, an environmental impact calibration experiment is conducted. Under fixed heating power conditions, such as a total power of 20kW, and with the allocation matrix kept constant, the ambient temperature and wind speed are varied, and the temperature drop rate of the target area per unit time is recorded. The experiment sets the ambient temperature to 20°C, 15°C, and 10°C (based on 20°C), and the wind speed to 0m / s, 0.5m / s, and 1.0m / s (based on 0m / s). Each set of conditions is run stably for 5 minutes, with a sampling period of 30 seconds. The environmental impact coefficients are obtained statistically. For example, the calibration yields the following relationship: for every 1°C decrease in ambient temperature (relative to the baseline 20°C), the heat loss per unit area increases by approximately 0.6%; for every 0.5m / s increase in wind speed (relative to the baseline 0m / s), the heat loss per unit area increases by approximately 1.0%. These values ​​are derived from the average results of at least 30 sets of operating condition data, and the environmental impact coefficient table is obtained through linear regression fitting.

[0056] During operation, ambient temperature and wind speed data are collected in real time, and the current environmental parameters are substituted into the influence coefficient table. For example, if the current ambient temperature is 12°C, which is 8°C lower than the baseline of 20°C, then the temperature influence coefficient is 8 × 0.6% = 4.8%; if the current wind speed is 1.0 m / s, which is 1.0 m / s higher than the baseline of 0 m / s (equivalent to two 0.5 m / s), then the wind speed influence coefficient is 2 × 1.0% = 2.0%; the comprehensive environmental influence coefficient is 4.8% + 2.0% = 6.8%. Then, combining the power values ​​of each grid in the current heating power distribution matrix, the instantaneous energy dissipation rate of each region is calculated. The power value of each grid is multiplied by the comprehensive environmental influence coefficient to obtain the additional heat loss power of that grid. For example, if the current power of a grid is 500W, then the additional heat loss is 500W × 6.8% = 34W. The heat loss values ​​of all grids are then used to form an energy dissipation rate matrix.

[0057] The thermal response lag time was obtained through a step power experiment. During the equipment commissioning phase, a typical heating area was selected, and environmental parameters were kept constant, such as an ambient temperature of 20°C, a wind speed of 1.0 m / s, and an initial stable temperature controlled at 460°C. A fixed-amplitude power step was then applied to this area, for example, from 500W to 600W, with a step amplitude of 100W. Simultaneously, the temperature change curve was continuously recorded with a sampling period of 1 second. When the rate of temperature change was less than 0.02°C / s for 30 consecutive seconds, the temperature was considered to have entered a new steady-state range. The system recorded the time elapsed from the moment of power adjustment to entering the steady-state range; this time was the thermal response lag time under that operating condition. For example, if the recorded result showed that the rate of temperature change dropped to the stable standard at 118 seconds, the lag time was recorded as 118 seconds. At least five step experiments were repeated under the same environmental conditions, and the average value was taken as the lag time for that environmental condition. To establish a lag model under environmental influences, the above experiment was repeated at different wind speeds (0 m / s, 0.5 m / s, 1.0 m / s) and different power step amplitudes (50 W, 100 W, 150 W) (with the ambient temperature kept constant at 20°C), and a lookup table for lag time versus wind speed and step amplitude was established. During system operation, the corresponding lag time value is retrieved based on the current wind speed. If the ambient temperature changes significantly, the effect of temperature on the lag time can be calibrated separately; however, this embodiment assumes that the temperature effect is negligible.

[0058] After determining the lag time, the feedforward compensation gain is calculated according to a quantitative relationship. First, the current target heating rate is determined based on the material and process requirements, for example, 0.5°C per minute, which is approximately 0.0083°C per second. If the experimentally calibrated lag time is 120 seconds, the theoretical temperature rise during the lag period should be approximately 120 × 0.0083 ≈ 1°C, and this 1°C is the compensation amount for the temperature difference caused by the lag. Then, the calibrated power-temperature rise sensitivity coefficient is used. For example, in the 460°C range, a 50W power change corresponds to a 0.8°C temperature rise through a perturbation experiment, so the sensitivity is 0.016°C / W. The required compensation power is 1 ÷ 0.016 ≈ 62W. If the current power is 500W, the feedforward compensation gain is (500 + 62) ÷ 500 ≈ 1.12. Therefore, a lag time of 120 seconds corresponds to a gain of approximately 1.12. The longer the lag time, the greater the compensation temperature difference, and the gain is automatically calculated according to the above relationship.

[0059] The feedforward compensation gain is applied to the heating power allocation matrix to correct the power settings of each channel. For example, suppose the current power allocation scheme is: Channel 1 power adjustment +30%, Channel 2 +20%, Channel 3 +10%, Channel 4 +25%. Based on the feedforward compensation gain of 1.12, the power adjustment value of each channel is multiplied by this gain coefficient to obtain the corrected power adjustment values: Channel 1 +33.6%, Channel 2 +22.4%, Channel 3 +11.2%, Channel 4 +28%. The corrected power allocation matrix is ​​the environmental adaptive correction control variable.

[0060] The environmental adaptive correction control variable is input into a preset closed-loop state observer. In this embodiment, the closed-loop control system employs a discrete PI controller combined with a state observer, with a sampling period set to 1 second. The state observer estimates the actual system temperature and temperature rise rate through weighted fusion based on the sensor's measured temperature and model predictions, providing the PI controller with more accurate state information. The PI controller performs feedback control based on the error between the current temperature and the target temperature, with a proportional gain of 0.6 and an integral gain of 0.02. These parameters are tuned through step response experiments, adhering to the principles of no oscillation, overshoot less than 3%, and settling time less than 150 seconds. For example, when the actual temperature is 465°C and the target temperature is 460°C, the temperature difference is 5°C, resulting in a 3W correction for the proportional term and an additional correction for the integral term based on the accumulated error. The feedback correction output from the PI controller is superimposed on the environmental adaptive correction control variable to form the final power control signal. The power control strategy corrected by this closed-loop state observer is the temperature equalization model.

[0061] In step S16, temperature prediction needs to be performed based on the temperature equalization model and a preset temperature prediction model to obtain a predicted temperature sequence. The predicted temperature sequence is then substituted back into the temperature equalization model to output a stable process path, including: Extract the set of control variables from the temperature equilibrium model and map the set of control variables to the initial model parameter population; The initial model parameter population is input into a preset temperature prediction model to obtain the predicted temperature sequence; Calculate the Euclidean distance between the predicted temperature sequence and the preset standard target temperature rise trajectory, and generate a trajectory deviation vector; The preset thermal inertia weight is corrected based on the trajectory deviation vector to obtain the equilibrium coefficient matrix; Substitute the equilibrium coefficient matrix back into the temperature equilibrium model to output a stable process path.

[0062] In the temperature equalization model, the control variables include the power distribution ratio of the heating channels, the feedforward compensation gain, and the feedback control parameters (such as the proportional and integral coefficients of the PI controller). These variables constitute the set of control variables at the current moment. For example, if the power distribution ratio of the four heating channels is [0.30, 0.20, 0.15, 0.25], the feedforward compensation gain is 1.12, and the PI parameters are [0.6, 0.02], then the current set of control variables can be expressed as [0.30, 0.20, 0.15, 0.25, 1.12, 0.6, 0.02], reflecting the current heating strategy.

[0063] To evaluate whether the current control strategy can ensure stable temperature operation along the desired trajectory, forward simulation using a pre-defined temperature prediction model is necessary. The temperature prediction model employs a Long Short-Term Memory (LSTM) network structure. Its input consists of historical temperature time series (temperature values ​​at each measurement point) for N consecutive sampling periods, the current set of control variables, and external environmental parameters (such as ambient temperature and wind speed). In the model input layer, the temperature value at each sampling moment is concatenated with the current set of control variables (repeated as global features at that moment) and environmental parameters to form the feature vector for that moment, thus constructing the input sequence for each time step. The model is trained based on historical processed data (including synchronous records of temperature, control parameters, and environmental parameters) and can learn the mapping relationship between control variables and temperature evolution. Before training, all input data is normalized, and the output is the predicted temperature sequence for each measurement point within a preset future time period, with the sampling interval consistent with historical data (e.g., 1 minute). For example, if the current time is t0, the model takes the temperature sequence from t-10 to t0, a total of 11 times, with the temperature values ​​of 4 temperature measurement points and the current set of control variables as input, and predicts the temperature values ​​from t+1 to t+30, a total of 30 times, to form a predicted temperature sequence.

[0064] After obtaining the predicted temperature sequence, it is compared with a preset standard target temperature rise trajectory. The standard target temperature rise trajectory is set according to material process requirements. For example, with an initial temperature of 460°C as the baseline, and a rise of 0.5°C per minute, the target sequence is [460.0, 460.5, 461.0, 461.5, 462.0, ..., 475.0]. The difference between the predicted sequence and the target sequence at each time point is calculated to obtain the trajectory deviation vector. At the same time, the Euclidean distance (in °C) between the two sequences is calculated to quantify the overall degree of deviation. For example, if the predicted sequence is [460, 463, 465, 467, 468, ...], and the point-by-point difference is [0, 2.5, 4, 5.5, 6, ...], the square root of the sum of squares yields an Euclidean distance of 9.4°C.

[0065] If the Euclidean distance is less than a preset first threshold (e.g., 5°C, determined based on process requirements and historical data statistics), the predicted trajectory is considered to match the target trajectory well, and the current control strategy is acceptable, requiring no adjustment. If the Euclidean distance is greater than a preset second threshold (e.g., 8°C), it indicates a large overall deviation, requiring correction of the power allocation in the control strategy to suppress the deviation. In this case, a thermal inertia weight parameter is introduced to characterize the response sensitivity of each heating region to temperature changes. The initial value of the thermal inertia weight can be set to be equal, for example, 0.25 for all four heating regions. The thermal inertia weight is adjusted based on the average deviation of the temperature measurement points corresponding to each region in the trajectory deviation vector (i.e., the average deviation of all predicted time points in that region). The new weight is equal to the original weight multiplied by 1 plus the adjustment coefficient multiplied by the average deviation, where the adjustment coefficient can be set to 0.02. The adjusted weights are normalized to make the sum of the weights of each region equal to 1, resulting in an equilibrium coefficient matrix. This matrix reflects the required power compensation ratio for each region.

[0066] Next, based on the preset region-channel influence coefficient table (obtained through offline calibration, recording the proportion of each heating region affected by each heating channel), the weight of each region in the equilibrium coefficient matrix is ​​converted into the compensation coefficient of each heating channel, resulting in a channel compensation coefficient vector. This vector is then multiplied element-wise with the current power allocation value of each heating channel to generate a corrected power control vector. This vector, along with the current temperature state and environmental parameters, is fed back into the temperature prediction model for forward simulation to generate a new predicted temperature sequence. This iterative process is repeated until the maximum temperature difference between the predicted and target sequences is less than ±1°C, and there is no continuous overshoot exceeding 2 minutes. Overshoot is defined as the predicted temperature continuously exceeding the target temperature by more than 0.5°C. The corresponding power control strategy at this point is the stable process path.

[0067] It should be noted that the current power allocation value for each heating channel refers to the actual power applied at the current moment, which may be derived from the real-time calculation results or initial settings of the previous moment (such as equal allocation to each channel [0.25, 0.25, 0.25, 0.25]). During the iterative correction process, this value is continuously updated, thereby gradually approaching the optimal control strategy. The final output stable process path is a set of power control commands that vary over time, ensuring that the refractory duct can be stably heated according to the target temperature rise trajectory during subsequent processing.

[0068] In step S17, it is necessary to calculate the two-dimensional difference of the preset ideal uniform temperature field and generate a gradient deviation matrix based on the stable process path, locate the local temperature difference abnormal region based on the gradient deviation matrix, and generate the corresponding iterative power correction vector, including: The sensor array is triggered according to the stable process path to obtain discrete feedback temperature signals; The discrete feedback temperature signal is reconstructed into a continuous real-time heat distribution map after being processed by the Kriging space interpolation algorithm. Calculate the two-dimensional difference between the continuous real-time heat distribution map and the preset ideal uniform temperature field to generate a gradient deviation matrix; If the maximum eigenvalue of the gradient deviation matrix exceeds a preset eigenvalue threshold, a feature vector is extracted from the gradient deviation matrix to locate the local temperature difference abnormality region. Based on the aforementioned local temperature difference abnormality region, the energy compensation amplitude is calculated; The energy compensation amplitude is used to update the preset gain weight to generate an iterative power correction vector.

[0069] According to the stable process path, a sensor array is triggered to acquire discrete feedback temperature signals. The sensor array consists of multiple sensors, evenly distributed on the heating surface of the refractory duct. Each sensor is pre-calibrated to ensure measurement accuracy, with a sampling period of 1 second, and the acquired temperature range covers 300℃ to 500℃. The acquired discrete feedback temperature signals are reconstructed into a continuous real-time heat distribution map using a Kriging spatial interpolation algorithm. This algorithm employs an exponential variogram model, with a spatial autocorrelation range set to 5 meters, and maps the discrete point temperature to a continuous temperature field covering the entire heating area through interpolation.

[0070] Subsequently, a preset ideal temperature uniformity field is invoked. This field contains a preset standard target temperature rise trajectory in the time dimension and a target temperature distribution matrix in the spatial dimension. The preset standard target temperature rise trajectory is obtained by time alignment and statistical averaging of multiple batches of qualified processing batches. For example, with the starting time as zero, the average temperature is recorded once per minute, forming a sequence of [460.0, 460.5, 461.0, ...]. The spatial target temperature distribution matrix is ​​obtained by statistically analyzing the temperature of each grid point at the same time. For example, at the 10th minute, the temperature of each grid point is concentrated in the range of 450℃±2℃, forming a two-dimensional ideal temperature matrix at that time. The continuous real-time heat distribution map at the current time is then subjected to point-by-point difference operation with the spatial target temperature distribution matrix at that time to generate a gradient deviation matrix. This matrix reflects the difference between the actual temperature and the ideal temperature at each location on the heated surface.

[0071] Singular value decomposition (SVD) is performed on the gradient deviation matrix to obtain a series of singular values ​​and their corresponding left and right singular vectors. The maximum singular value characterizes the main degree of variation in the temperature deviation field, with its dimension being the square of the temperature. If the maximum singular value exceeds a preset singular value threshold (e.g., set to 2.5℃² based on historical data statistics), a significant local temperature anomaly is determined to exist on the heated surface. In this case, based on the left and right singular vectors corresponding to the maximum singular value, the spatial region with the largest deviation contribution can be identified. Combining the left and right singular vectors yields the weight distribution corresponding to the original mesh. Mesh points with larger absolute weight values ​​constitute the local temperature anomaly region.

[0072] For each identified local temperature aberration region, the energy compensation amplitude is calculated. First, the average temperature difference ΔT of each grid point within the region is extracted, which is the average difference between the actual temperature and the ideal temperature. The heat dissipation coefficient k is determined based on the spatial location of the region; for example, the heat dissipation coefficient is 1.2 for the edge region and 1.0 for the central region. Then, according to a preset temperature difference-power mapping relationship, such as a 5% base power adjustment for every 1℃ temperature difference, the base compensation ratio P_base = 5% × ΔT is calculated. The energy compensation amplitude C = P_base × k represents the proportion of heating power that needs to be increased for this region. If ΔT is negative, then C is negative, indicating that the power needs to be reduced.

[0073] Converting the energy compensation amplitude of each region into the power correction amount of the heating channel requires using a preset region-channel influence coefficient table. This coefficient table is obtained through offline calibration and records the proportion of each heating region affected by each heating channel. For example, region 1 has an influence coefficient of 0.4 from channel A and 0.3 from channel B. For each heating channel, the compensation amplitude of all regions it covers is multiplied by the corresponding influence coefficient and then summed to obtain the total compensation ratio of that channel. For example, if channel A covers region 1 (compensation amplitude +30%, influence coefficient 0.4) and region 2 (compensation amplitude +20%, influence coefficient 0.6), then the total compensation ratio of channel A is 30% × 0.4 + 20% × 0.6 = 24%. After calculating for all channels, the power adjustment ratio of each channel is obtained, forming an iterative power correction vector, such as [+24%, +18%, +12%, +8%]. This vector is used to update the heating power distribution to achieve accurate compensation for areas with abnormal local temperature differences, gradually eliminate residual temperature differences, and ensure the temperature uniformity of the entire heating process.

[0074] In step S18, the iterative power correction vector needs to be input into a preset multiphysics coupling model to obtain a material performance consistency index. Based on the material performance consistency index, a thermal diffusion equilibrium network is constructed, and after eliminating residual temperature differences, a uniform heat distribution state is obtained, including: The iterative power correction vector is input into a preset multiphysics coupling model to perform virtual heat processing calculations, thereby obtaining lattice arrangement structure prediction data. The standard deviation of the predicted lattice arrangement structure is calculated to obtain the predicted standard deviation; If the predicted standard deviation is within the preset convergence interval, a material performance consistency index is generated based on the predicted standard deviation, and the material is deemed qualified based on the material performance consistency index. A duty cycle signal is generated based on the aforementioned material performance consistency index; The duty cycle signal is used to drive a solid-state relay array to generate dynamic heat flux density. A thermal diffusion equilibrium network is constructed using the dynamic heat flux density, and a balanced heat distribution state is obtained after eliminating residual temperature differences through the thermal diffusion equilibrium network.

[0075] The iterative power correction vector is input into a preset multiphysics coupling model for virtual heat treatment calculation. This model employs finite element thermal-structural coupled numerical simulation. A two-dimensional cross-section of a fire-resistant duct is used as the analysis object, with a wall thickness of 8mm and a length of 1m, divided into 5mm×5mm mesh elements, totaling approximately 3200 elements. The heating channels are divided into multiple boundary regions (e.g., 4) based on their physical location, with each region receiving a corresponding power value. Assuming the iterative power correction vector is [+30%, +20%, +15%, +25%], and the rated power of a single channel is 8kW, the power applied to each channel is 10.4kW, 9.6kW, 9.2kW, and 10kW, respectively. The applied power is converted into heat flux density and applied as a thermal boundary condition to the corresponding element. For example, if the heating area of ​​a channel is 0.4m², the heat flux density is the applied power divided by the area (10.4kW corresponds to 26000W / m²). Material parameters are based on actual measured values, such as density 7800 kg / m³, specific heat capacity 500 J / (kg·K), thermal conductivity 45 W / (m·K), and coefficient of thermal expansion 1.2 × 10⁻⁶. -5 / K. The model uses transient solution with a time step of 1s and a total simulation time of 600s. It calculates the temperature field and thermal stress field sequentially, and predicts the grain size of each unit based on the empirical grain growth model based on temperature-time integration. Finally, it outputs the transient temperature field, thermal stress distribution and lattice arrangement structure prediction data, i.e., the grain size distribution matrix.

[0076] The standard deviation of the predicted lattice arrangement structure data is calculated to obtain the predicted standard deviation, which is used to evaluate the consistency of the material's microstructure. The smaller the standard deviation, the smaller the grain size fluctuation and the better the material performance consistency. For example, if the predicted grain size of a batch is [10μm, 12μm, 11.5μm, 9.8μm], the calculated standard deviation is 0.5μm. The preset convergence interval is determined through performance comparison experiments based on the material's process requirements, for example, set to 0.4μm to 0.8μm. If the predicted standard deviation is within this interval, such as 0.5μm, the material performance consistency is considered qualified, and a corresponding material performance consistency index, such as a "qualified" label, is generated; if it exceeds the interval, it is considered unqualified, and the iteration power correction vector needs to be corrected or the process parameters adjusted.

[0077] A duty cycle signal is generated based on material performance consistency indicators. This signal controls the on / off state of the solid-state relay array to adjust the average power output of each heating channel. Specifically, a baseline duty cycle value is determined based on the material performance consistency indicators. For example, when the indicator is "qualified," the duty cycle is set to 70%, and when the indicator is "unqualified," the duty cycle is increased to 80% to enhance regulation. The duty cycle signal can also be fine-tuned based on the current temperature control error, but it is primarily set according to the consistency indicators. The generated duty cycle signal drives the solid-state relay array, causing each heating channel to output power according to its duty cycle within a cycle, thereby generating dynamic heat flux density.

[0078] A thermal diffusion equilibrium network is constructed using the dynamic heat flux density output from a solid-state relay array to eliminate residual temperature differences. The thermal diffusion equilibrium network consists of multiple nodes and edges. Each node corresponds to a temperature control unit within the heating region, similar to a spatial grid, and its temperature state is updated by fusing real-time sensor data with the output of a preset temperature prediction model. Each edge represents a heat transfer path between adjacent nodes, and the transfer intensity depends on the physical distance between nodes, the material's thermal conductivity, and the temperature gradient. The dynamic heat flux density serves as input, driving heat exchange among the nodes in the network. The system monitors node temperature differences in real time and automatically adjusts the duty cycle of each channel according to the thermal diffusion law, reducing heat input to high-temperature nodes and increasing heat input to low-temperature nodes. This process is iteratively repeated until the temperature deviation of all nodes from the target value is less than the allowable range, ultimately achieving temperature equilibrium across the entire heating region and obtaining a balanced heat distribution.

[0079] In summary, this invention provides a method and system for intelligent temperature control in the processing of refractory ducts, which can adjust the power in real time to cope with dynamic changes in the external environment and material properties, thereby improving processing quality.

[0080] Reference Figure 2 The second embodiment of the present invention provides an intelligent control system for the processing temperature of refractory ducts, comprising: The data processing module is used to acquire real-time temperature data and external environmental fluctuation data during the processing of refractory ducts, and to preprocess the real-time temperature data to obtain an initial temperature distribution map. The location coordinate module is used to analyze the temperature change trend based on the initial temperature distribution map and a preset fuzzy rule base, obtain a local overheating probability distribution map, and calculate the geometric centroid to determine the location coordinates of the overheating zone. The power correction module is used to perform differential calculations based on the coordinates of the overheated area and the real-time temperature data to obtain the material temperature rise rate data, and then perform vector synthesis calculations after filtering out power adjustment samples from a preset historical processing database to obtain the power correction vector. The power allocation module is used to calculate the cumulative temperature rise rate deviation based on the power correction vector and the real-time temperature data, and to construct a heating power allocation matrix by combining the initial temperature distribution map and the power correction vector. The temperature equalization module is used to generate a feedforward compensation gain value based on the mapping relationship between the heating power distribution matrix and the external environment fluctuation data, and to perform correction on the heating power distribution matrix in conjunction with a preset closed-loop state observer to obtain a temperature equalization model. The process curve module is used to perform temperature prediction based on the temperature equalization model and combined with the preset temperature prediction model to obtain the predicted temperature sequence, and then substitute the predicted temperature sequence back into the temperature equalization model to output a stable process path. The iterative power module is used to calculate the two-dimensional difference of the preset ideal uniform temperature field and generate a gradient deviation matrix according to the stable process path, locate the local temperature difference abnormal region according to the gradient deviation matrix and generate the corresponding iterative power correction vector. The state distribution module is used to input the iterative power correction vector into a preset multiphysics coupling model to obtain a material performance consistency index. Based on the material performance consistency index, a thermal diffusion equilibrium network is constructed and residual temperature difference is eliminated to obtain a uniform heat distribution state.

[0081] It should be noted that the intelligent control system for refractory duct processing temperature provided in this embodiment of the invention is used to execute all the process steps of the intelligent control method for refractory duct processing temperature in the above embodiment. The working principle and beneficial effects of the two are one-to-one, so they will not be described again.

[0082] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0083] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A method for intelligent temperature control in the processing of refractory ducts, characterized in that, include: Real-time temperature data and external environmental fluctuation data are acquired during the processing of refractory ducts, and the real-time temperature data is preprocessed to obtain an initial temperature distribution map. Based on the initial temperature distribution map, the temperature change trend is analyzed by combining the preset fuzzy rule base to obtain the local overheating probability distribution map and calculate the geometric centroid to determine the coordinates of the overheating area. Based on the coordinates of the overheated area and the real-time temperature data, differential calculation is performed to obtain the material temperature rise rate data. After filtering out power adjustment samples from the preset historical processing database, vector synthesis calculation is performed to obtain the power correction vector. The cumulative temperature rise rate deviation is calculated based on the power correction vector and the real-time temperature data, and a heating power allocation matrix is ​​constructed by combining the initial temperature distribution map and the power correction vector. Based on the mapping relationship between the heating power allocation matrix and the external environmental fluctuation data, a feedforward compensation gain value is generated, and the heating power allocation matrix is ​​corrected by combining a preset closed-loop state observer to obtain a temperature equilibrium model. Based on the temperature equilibrium model, temperature prediction is performed in conjunction with a preset temperature prediction model to obtain a predicted temperature sequence. The predicted temperature sequence is then substituted back into the temperature equilibrium model to output a stable process path. Based on the stable process path, the two-dimensional difference of the preset ideal uniform temperature field is calculated and a gradient deviation matrix is ​​generated. Based on the gradient deviation matrix, the local temperature difference abnormal region is located and the corresponding iterative power correction vector is generated. The iterative power correction vector is input into a preset multiphysics coupling model to obtain a material performance consistency index. Based on the material performance consistency index, a thermal diffusion equilibrium network is constructed and residual temperature difference is eliminated to obtain a balanced heat distribution state.

2. The intelligent temperature control method for refractory duct processing according to claim 1, characterized in that, The preprocessing of the real-time temperature data to obtain an initial temperature distribution map includes: The real-time temperature data is parsed into multi-point discrete temperature values; Retrieve the physical space coordinate matrix of the sensor array, map the multi-point discrete temperature values ​​to the physical space coordinate matrix, and establish a discrete temperature point set; The Kriging interpolation algorithm is used to spatially interpolate the discrete temperature point set to generate continuous thermal field data. The continuous thermal field data is linearly normalized according to a preset temperature range to generate a grayscale pixel value matrix. The grayscale pixel value matrix is ​​mapped to an image pixel array according to two-dimensional grid coordinates, and the initial temperature distribution map of the refractory duct processing process is output.

3. The intelligent temperature control method for refractory duct processing according to claim 1, characterized in that, The step of analyzing temperature change trends based on the initial temperature distribution map and a preset fuzzy rule base to obtain a local overheating probability distribution map and calculating the geometric centroid to determine the coordinates of the overheating region includes: The initial temperature distribution map is convolved using a preset spatial gradient operator to obtain a temperature gradient feature matrix; By combining the preset fuzzy rule base, the temperature gradient feature matrix is ​​input into the preset fuzzy inference model, and a local overheating probability distribution map is output. When the pixel probability value corresponding to the local overheating probability distribution map is greater than the preset probability threshold, the connected component labeling algorithm is used to aggregate the pixel points corresponding to the pixel probability value that is greater than the preset probability threshold to obtain the connected component of the suspected overheating area. Calculate the geometric centroid of the connected domain of the suspected overheated area, and convert the geometric centroid into the physical spatial coordinates of the heating surface of the fire-resistant air duct based on a preset spatial mapping relationship matrix to obtain the coordinates of the overheated area.

4. The intelligent temperature control method for refractory duct processing according to claim 1, characterized in that, The process involves performing a differential operation based on the coordinates of the overheated zone and the real-time temperature data to obtain material temperature rise rate data. This data is then combined with power adjustment samples selected from a pre-set historical processing database and subjected to vector synthesis to obtain a power correction vector, including: Based on the coordinates of the overheated zone and the real-time temperature data, a differential operation is performed to obtain the material temperature rise rate data. The material temperature rise rate data is input into a preset historical processing database to calculate the Euclidean distance. The historical data with the smallest Euclidean distance is selected from the temperature rise rate data as the power adjustment sample. Extract the historical adjustment values ​​from the power adjustment sample and calculate the heat input suppression amplitude by combining them with the overheating zone location coordinates; The power control matrix of a multi-channel system is constructed using the heat input suppression amplitude and vector synthesis is performed to obtain the power correction vector.

5. The intelligent temperature control method for refractory duct processing according to claim 1, characterized in that, The calculation based on the power correction vector and the real-time temperature data to obtain the cumulative temperature rise rate deviation, and the construction of a heating power allocation matrix by combining the initial temperature distribution map and the power correction vector, includes: The power correction vector and the real-time temperature data are input into a preset long short-term memory network model to generate a predicted temperature evolution trajectory sequence. Perform differential calculations between the predicted temperature evolution trajectory sequence and the preset standard process reference curve to obtain the temperature rise rate deviation sequence; The cumulative temperature rise rate deviation is obtained by integrating the temperature rise rate deviation sequence. If the cumulative temperature rise rate deviation exceeds the preset allowable fluctuation threshold, the cumulative temperature rise rate deviation is arranged in a gradient to obtain the deviation distribution gradient. Based on the deviation distribution gradient, the initial temperature distribution map is retrieved to obtain the temperature fluctuation amplitude, and the node energy weight coefficient is calculated based on the temperature fluctuation amplitude. The power correction vector is multiplied by matrix multiplication using the node energy weight coefficients to construct a spatially weighted power distribution matrix as the heating power allocation matrix.

6. The intelligent temperature control method for refractory duct processing according to claim 1, characterized in that, The step involves generating a feedforward compensation gain value based on the mapping relationship between the heating power allocation matrix and the external environmental fluctuation data, and then modifying the heating power allocation matrix using a preset closed-loop state observer to obtain a temperature equilibrium model, including: Establish the mapping relationship between the heating power distribution matrix and the external environmental fluctuation data, and derive the instantaneous energy dissipation rate; The thermal response lag time is identified based on the instantaneous energy dissipation rate, and a feedforward compensation gain value matching the thermal response lag time is generated. Using the feedforward compensation gain value, superposition correction is performed on the heating power distribution matrix to generate environmental adaptive correction control variables; The environmental adaptive correction control variables are input into a preset closed-loop state observer to obtain a temperature equilibrium model.

7. The intelligent temperature control method for refractory duct processing according to claim 1, characterized in that, The process of performing temperature prediction based on the temperature equalization model and a preset temperature prediction model to obtain a predicted temperature sequence, and then substituting the predicted temperature sequence back into the temperature equalization model to output a stable process path includes: Extract the set of control variables from the temperature equilibrium model and map the set of control variables to the initial model parameter population; The initial model parameter population is input into a preset temperature prediction model to obtain the predicted temperature sequence; Calculate the Euclidean distance between the predicted temperature sequence and the preset standard target temperature rise trajectory, and generate a trajectory deviation vector; The preset thermal inertia weight is corrected based on the trajectory deviation vector to obtain the equilibrium coefficient matrix; Substitute the equilibrium coefficient matrix back into the temperature equilibrium model to output a stable process path.

8. The intelligent temperature control method for refractory duct processing according to claim 1, characterized in that, The step of calculating the two-dimensional difference of the preset ideal uniform temperature field and generating a gradient deviation matrix based on the stable process path, and locating local temperature difference abnormal regions based on the gradient deviation matrix and generating corresponding iterative power correction vectors includes: The sensor array is triggered to acquire discrete feedback temperature signals according to the stable process path; The discrete feedback temperature signal is reconstructed into a continuous real-time heat distribution map after being processed by the Kriging space interpolation algorithm. Calculate the two-dimensional difference between the continuous real-time heat distribution map and the preset ideal uniform temperature field to generate a gradient deviation matrix; If the maximum eigenvalue of the gradient deviation matrix exceeds a preset eigenvalue threshold, a feature vector is extracted from the gradient deviation matrix to locate the local temperature difference abnormality region. Based on the aforementioned local temperature difference abnormality region, the energy compensation amplitude is calculated; The energy compensation amplitude is used to update the preset gain weight to obtain the updated gain weight; An iterative power correction vector is generated based on the updated gain weights.

9. The intelligent temperature control method for refractory duct processing according to claim 1, characterized in that, The step of inputting the iterative power correction vector into a preset multiphysics coupling model to obtain a material performance consistency index, constructing a thermal diffusion equilibrium network based on the material performance consistency index, and eliminating residual temperature differences to obtain a balanced heat distribution state includes: The iterative power correction vector is input into a preset multiphysics coupling model to perform virtual heat processing calculations, thereby obtaining lattice arrangement structure prediction data. The standard deviation of the predicted lattice arrangement structure is calculated to obtain the predicted standard deviation; If the predicted standard deviation is within the preset convergence interval, a material performance consistency index is generated based on the predicted standard deviation, and the material is deemed qualified based on the material performance consistency index. A duty cycle signal is generated based on the aforementioned material performance consistency index; The duty cycle signal is used to drive a solid-state relay array to generate dynamic heat flux density. A thermal diffusion equilibrium network is constructed using the dynamic heat flux density, and a balanced heat distribution state is obtained after eliminating residual temperature differences through the thermal diffusion equilibrium network.

10. A method and system for intelligent temperature control in the processing of refractory ducts, characterized in that, include: The data processing module is used to acquire real-time temperature data and external environmental fluctuation data during the processing of refractory ducts, and to preprocess the real-time temperature data to obtain an initial temperature distribution map. The location coordinate module is used to analyze the temperature change trend based on the initial temperature distribution map and a preset fuzzy rule base, obtain a local overheating probability distribution map, and calculate the geometric centroid to determine the location coordinates of the overheating zone. The power correction module is used to perform differential calculations based on the coordinates of the overheated area and the real-time temperature data to obtain the material temperature rise rate data, and then perform vector synthesis calculations after filtering out power adjustment samples from a preset historical processing database to obtain the power correction vector. The power allocation module is used to calculate the cumulative temperature rise rate deviation based on the power correction vector and the real-time temperature data, and to construct a heating power allocation matrix by combining the initial temperature distribution map and the power correction vector. The temperature equalization module is used to generate a feedforward compensation gain value based on the mapping relationship between the heating power distribution matrix and the external environment fluctuation data, and to perform correction on the heating power distribution matrix in conjunction with a preset closed-loop state observer to obtain a temperature equalization model. The process curve module is used to perform temperature prediction based on the temperature equalization model and combined with the preset temperature prediction model to obtain the predicted temperature sequence, and then substitute the predicted temperature sequence back into the temperature equalization model to output a stable process path. The iterative power module is used to calculate the two-dimensional difference of the preset ideal uniform temperature field and generate a gradient deviation matrix according to the stable process path, locate the local temperature difference abnormal region according to the gradient deviation matrix and generate the corresponding iterative power correction vector. The state distribution module is used to input the iterative power correction vector into a preset multiphysics coupling model to obtain a material performance consistency index. Based on the material performance consistency index, a thermal diffusion equilibrium network is constructed and residual temperature difference is eliminated to obtain a uniform heat distribution state.

Images