A temperature monitoring system for a box furnace

Abstract

This invention discloses a temperature monitoring system for a box furnace, relating to the field of industrial furnace temperature control technology. The system includes a data acquisition module, an anomaly identification module, an evaluation and analysis module, a strategy control module, and an instruction execution module. The data acquisition module acquires real-time temperature data streams from multiple zones of the box furnace, operational status data of key furnace components, and power quality data from the external power grid. The anomaly identification module uses an improved abnormal temperature point identification algorithm, combined with the spatiotemporal correlation of the furnace temperature field, to dynamically adjust the anomaly judgment threshold, accurately identifying potential abnormal temperature fluctuation patterns. The evaluation and analysis module integrates multiple types of data to generate a furnace health status assessment result. The strategy control module calculates optimized parameters for heating power, fan speed, and process duration. The instruction execution module issues control commands. This solution achieves accurate identification of the box furnace's temperature operating conditions, completes collaborative closed-loop monitoring of temperature and energy consumption, and improves the adaptability and overall performance of furnace operation control.

Description

Technical Field

[0001] This invention belongs to the field of industrial furnace temperature control technology, specifically a temperature monitoring system for box furnaces. Background Technology

[0002] Conventional box-type furnace temperature monitoring often adopts a single-point temperature measurement mode, only collecting temperature information at local points inside the furnace. It does not deploy multi-area temperature sensor arrays to collect temperature data streams across the entire area, nor does it simultaneously incorporate data related to the operating status of key furnace components and the quality of external power grid power supply for monitoring and analysis. Traditional temperature control architectures have a single data acquisition dimension, focusing only on surface temperature parameters inside the furnace, lacking support from equipment operation and power supply-side related data.

[0003] Current methods for detecting temperature anomalies in box furnaces generally employ fixed threshold judgment mechanisms. Once the threshold parameters are set, they cannot adjust autonomously according to changes in furnace operating conditions, failing to consider the inherent spatiotemporal correlation characteristics of the furnace's temperature field. Fixed judgment standards are ill-suited to the temperature distribution changes under different production stages and loads in box furnaces, easily leading to missed or false detections of temperature anomalies and failing to accurately capture hidden temperature anomalies. Conventional monitoring modes only analyze temperature data independently, lacking a mechanism for fusing and analyzing multi-source heterogeneous data. Control methods are mostly based on individual adjustments of single equipment parameters, failing to link furnace temperature status with energy consumption, resulting in insufficient overall monitoring adaptability and linkage.

[0004] A multi-dimensional integrated data acquisition system needs to be built, relying on the inherent spatiotemporal correlation characteristics of the temperature field to achieve adaptive adjustment of the anomaly judgment threshold. At the same time, a multi-source data fusion analysis mechanism should be established, coupled with multi-device parameter collaborative control logic, to make up for the shortcomings of traditional technology such as insufficient data dimensions, rigid anomaly judgment, and isolated control mode. Summary of the Invention

[0005] This invention aims to solve at least one of the technical problems existing in the prior art; Therefore, the present invention proposes a temperature monitoring system for a box furnace, comprising: The data acquisition module acquires real-time temperature data streams collected by the multi-zone temperature sensor array inside the box furnace, operating status data of key furnace components, and power quality data of the external power grid. The anomaly identification module performs an improved anomaly temperature point identification algorithm on the real-time temperature data stream to identify potential abnormal temperature fluctuation patterns. The improved anomaly temperature point identification algorithm dynamically adjusts the anomaly judgment threshold based on the spatiotemporal correlation of the temperature field inside the furnace. The assessment and analysis module integrates and analyzes the identified potential abnormal temperature fluctuation patterns, the operating status data, and the power supply quality data to generate a comprehensive furnace health status assessment result. Based on the comprehensive furnace health status assessment results, the strategy control module calculates a set of optimized control parameters for adjusting the power of heating elements, the speed of circulating fans, and the duration of process stages through the energy-saving control strategy engine. The instruction execution module converts the optimized control parameter set into control instructions and sends them to the heating actuator, fan driver and process controller of the box furnace to achieve coordinated monitoring of the temperature and energy consumption of the box furnace.

[0006] Furthermore, an improved abnormal temperature point identification algorithm is applied to the real-time temperature data stream to identify potential abnormal temperature fluctuation patterns, including: The real-time temperature data stream collected by the multi-region temperature sensor array is sampled using a sliding window to form a series of temperature time series windows of equal length; For the data within each temperature time series window, its statistical characteristics are calculated, including mean, standard deviation, slope and curvature, and the spatial gradient characteristics of temperature change are calculated using the data between sliding windows. The statistical features and spatial gradient features are input into the improved abnormal temperature point identification algorithm; The improved abnormal temperature point identification algorithm is trained based on temperature data under historical normal operating conditions to obtain a dynamic threshold model. The dynamic threshold model dynamically outputs the normal range threshold of each temperature sensor data at the current moment according to the current process stage, the average temperature in the furnace and the temperature change trend. The real-time temperature data within the current temperature time series window is compared with the normal range threshold, and data segments that continuously exceed the threshold range and conform to the preset fluctuation pattern are marked as the potential abnormal temperature fluctuation pattern.

[0007] Furthermore, the process of fusing and analyzing the identified potential abnormal temperature fluctuation patterns, the operating status data, and the power supply quality data to generate a comprehensive furnace health status assessment result includes: Feature extraction is performed on the potential abnormal temperature fluctuation patterns to obtain the fluctuation amplitude, fluctuation frequency, spatial distribution range, and spatial correlation characteristics with the heating element, thus forming a temperature anomaly feature vector; The operating status data, including the current and voltage of the heating element, the vibration and noise of the circulating fan, and the furnace pressure data, are encoded into an equipment status feature vector. The power supply quality data is processed to extract voltage fluctuation, harmonic content, and power factor features, forming a power supply quality feature vector. The temperature anomaly feature vector, equipment status feature vector, and power supply quality feature vector are input into a pre-trained furnace condition evaluation neural network model. The furnace condition assessment neural network model outputs a multi-dimensional assessment vector, where each dimension of the multi-dimensional assessment vector corresponds to an aspect score of the furnace condition health status, and the assessment vectors together constitute the comprehensive furnace condition health status assessment result.

[0008] Furthermore, based on the comprehensive furnace health status assessment results, an optimized set of control parameters for adjusting the heating element power, circulating fan speed, and process stage duration is calculated through the energy-saving control strategy engine, including: The energy-saving control strategy engine has a multi-objective optimization function embedded in it. The multi-objective optimization function takes temperature uniformity, energy consumption minimization and process completion time as optimization objectives, and takes equipment safe operating range and process temperature curve requirements as constraints. The comprehensive furnace health status assessment results are used as input parameters for the multi-objective optimization function. The scores of different dimensions in the assessment results are used to adjust the weights of the objective function and the boundaries of the constraints. Solve the multi-objective optimization function to obtain the Pareto optimal solution set under the current furnace condition evaluation results; From the Pareto optimal solution set, an optimal solution is selected according to a preset priority strategy, and the heating element power adjustment, circulating fan speed setting, and duration adjustment of each process stage corresponding to the optimal solution are used as the optimized control parameter set.

[0009] Furthermore, by solving the multi-objective optimization function, a Pareto optimal solution set is obtained under the current furnace condition evaluation results, including: An improved non-dominated sorting genetic algorithm is used to solve the multi-objective optimization function, and a population composed of multiple control parameters is initialized. Calculate the objective function values ​​for temperature uniformity, total energy consumption, and estimated process completion time for each individual in the population; Based on the objective function value, non-dominated ranking and crowding degree calculation are performed on individuals in the population to simulate natural selection; Perform selection, crossover, and mutation operations to generate a new generation of population; During the evolution process, the specific dimension scores in the comprehensive furnace health status assessment results are used as adaptive parameters to dynamically adjust the crossover probability and mutation probability of the improved non-dominated sorting genetic algorithm. When the evolution reaches the preset number of generations, the set of individuals that represent the last generation as the non-dominant frontier will be output as the Pareto optimal solution set.

[0010] Furthermore, the optimized control parameter set is converted into control commands and sent to the heating actuators, fan drivers, and process controllers of the box furnace, including: The optimized control parameter set is analyzed to obtain the power adjustment amount, the target speed value of the circulating fan, and the target duration adjustment amount for each process stage for each heating circuit. According to the communication protocol of the heating actuator, the power adjustment is converted into the corresponding analog output value or digital control word to form a heating control command; Based on the control mode of the wind turbine driver, the target speed value is converted into a frequency command or a set value of the speed feedback closed loop to form a wind turbine control command. Based on the timing logic of the process controller, the end time of each process stage is recalculated to form updated process stage timing instructions. The heating control command, the fan control command, and the updated process stage timing command are sent to the corresponding heating actuator, fan driver, and process controller according to their respective control cycles.

[0011] Furthermore, the improved abnormal temperature point identification algorithm is trained based on temperature data under historical normal operating conditions to obtain a dynamic threshold model, including: Collect complete temperature data from a multi-zone temperature sensor array during various process stages of a box furnace during its historical normal and stable operation, and use it as a training dataset. For each temperature sensor in the training dataset, the data is divided according to the process stage, and the same statistical features and spatial gradient features as the current time window are extracted. For each process stage and each temperature sensor's corresponding feature dataset, the kernel density estimation method is used to fit the probability distribution of its feature values. Based on the probability distribution obtained by fitting, a confidence interval covering the vast majority of normal data points is set for each feature, and the boundary values ​​of the confidence interval constitute the basic threshold of the feature under the corresponding process stage. The dynamic threshold model stores the basic thresholds corresponding to all process stages, all temperature sensors, and all features, and performs linear interpolation correction on the basic thresholds based on the current average temperature inside the furnace to generate the normal range thresholds.

[0012] Further, the step of linearly interpolating and correcting the base threshold based on the current average temperature inside the furnace to generate the normal range threshold includes: The average value of all temperature sensor readings inside the box furnace is calculated in real time and used as the current average temperature inside the furnace. Determine the current process stage and obtain the basic upper and lower thresholds of each feature of the current temperature sensor under the current process stage; Query the historical data to find the feature thresholds corresponding to the two reference average temperature points adjacent to the current process stage; Based on the current average temperature inside the furnace, the correction values ​​of each characteristic threshold are calculated using linear interpolation when the temperature is between two reference average temperature points. The corrected value obtained by linear interpolation is used as the normal range threshold.

[0013] Furthermore, the step of using an improved non-dominated sorting genetic algorithm to solve the multi-objective optimization function and initializing a population composed of multiple control parameters includes: Within the defined range of the control parameters, a preset number of initial individuals are generated uniformly and randomly, and each individual is represented by a chromosome encoded by a real number. The chromosome contains multiple gene loci, and each gene locus corresponds to a control parameter that needs to be optimized. The control parameters include heating element power, circulating fan speed, and process stage duration. Boundary checks are performed on the real-number encoded chromosomes, and gene position values ​​that exceed the defined domain are adjusted to the closest valid boundary values. Calculate the fitness values ​​of all initial individuals on three objectives: temperature uniformity, total energy consumption, and expected process completion time; The chromosomes and corresponding fitness values ​​of all initial individuals that have completed fitness evaluation are used together to form the initial population for solving the multi-objective optimization function.

[0014] Furthermore, during the evolutionary process, the specific dimension scores in the comprehensive furnace health status assessment results are used as adaptive parameters to dynamically adjust the crossover and mutation probabilities of the improved non-dominated sorting genetic algorithm, including: From the comprehensive furnace health status assessment results, scores reflecting the uniformity of furnace temperature and scores reflecting the stability of equipment operation are extracted; The product of the temperature uniformity score and the equipment operating stability score is calculated and used as an adaptive adjustment factor. In each generation of the improved non-dominated sorting genetic algorithm, the preset base crossover probability is multiplied by the adaptive adjustment factor to obtain the actual crossover probability of the current generation; Multiply the preset base mutation probability by the adaptive adjustment factor to obtain the actual mutation probability of the current generation; The actual crossover probability and the actual mutation probability are used to guide the selection, crossover, and mutation operations of the current generation.

[0015] Compared with the prior art, the beneficial effects of the present invention are: An abnormal temperature point identification algorithm adapted to the operating conditions of a box furnace is used to process real-time temperature data streams. The algorithm adjusts the anomaly judgment threshold parameters in real time based on the spatiotemporal correlation attributes of the temperature field within the furnace. It abandons the single judgment method with fixed thresholds; the threshold parameters can adaptively and dynamically change according to the temperature distribution and temporal operating status within the furnace, conforming to the natural evolution law of the temperature field inside the box furnace. Subtle fluctuations hidden in the temperature data can be fully analyzed, and potential trends of abnormal temperature fluctuations can be completely identified. This reduces the identification bias caused by fixed judgment standards, adapts to the temperature detection needs of different production and operation stages of the box furnace, and ensures that the temperature anomaly identification logic conforms to the dynamic changes of the actual operating conditions inside the furnace.

[0016] By integrating and analyzing potential abnormal temperature fluctuation patterns, operational status data of key furnace components, and external power grid quality data, the energy-saving control strategy engine generates optimized control parameters for heating element power, circulating fan speed, and process stage duration. The fusion of multi-source heterogeneous data supplements the dimensions of operational condition analysis, overcoming the limitations of single-dimensional temperature data analysis and fully reconstructing the true operational status of the entire box furnace. The generated control parameters can synchronously match the furnace's internal temperature variation patterns, actual power grid conditions, and furnace equipment operating status. After the control parameters are converted into control commands and transmitted to each execution structure, the furnace's temperature and energy consumption status can be synchronously adjusted. This allows the various operating parameters of the box furnace to adapt and balance each other, achieving an integrated and coordinated operation mode for temperature control and energy consumption regulation. Attached Figure Description

[0017] Figure 1 This is a timing diagram of a temperature monitoring system for a box furnace according to the present invention; Figure 2 A flowchart illustrating the improved algorithm for identifying abnormal temperature points; Figure 3 A flowchart for calculating and optimizing control parameters for the energy-saving control strategy engine. Detailed Implementation

[0018] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. 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.

[0019] See Figure 1The present invention discloses a temperature monitoring system for a box furnace, the overall implementation of which is as follows: The system includes a data acquisition module, an anomaly identification module, an evaluation and analysis module, a strategy control module, and an instruction execution module. The data acquisition module acquires real-time temperature data streams collected by a multi-zone temperature sensor array within the box furnace, operational status data of key furnace components, and power quality data from the external power grid. The anomaly identification module processes the real-time temperature data stream using an improved anomaly point identification algorithm to identify potential abnormal temperature fluctuation patterns. The improved anomaly point identification algorithm dynamically adjusts the anomaly judgment threshold based on the spatiotemporal correlation of the furnace temperature field. The evaluation and analysis module performs a fusion analysis on the identified potential abnormal temperature fluctuation patterns, the operational status data, and the power quality data to generate a comprehensive furnace health status assessment result. Based on the comprehensive furnace health status assessment result, the strategy control module calculates a set of optimized control parameters for adjusting the heating element power, circulating fan speed, and process stage duration using an energy-saving control strategy engine. The instruction execution module converts the optimized control parameter set into control instructions and sends them to the heating actuator, fan driver and process controller of the box furnace to achieve coordinated monitoring of the temperature and energy consumption of the box furnace.

[0020] In one embodiment of the present invention, when performing an improved abnormal temperature point identification algorithm on the real-time temperature data stream to identify potential abnormal temperature fluctuation patterns, refer to... Figure 2 The real-time temperature data stream collected by the multi-region temperature sensor array is sampled using a sliding window method to form a series of temperature time series windows of equal length. For the data within each temperature time series window, its statistical characteristics are calculated, including mean, standard deviation, slope, and curvature. The spatial gradient characteristics of temperature change are then calculated using the data between the sliding windows. These statistical characteristics and spatial gradient characteristics are input into the improved abnormal temperature point identification algorithm. The improved abnormal temperature point identification algorithm is trained based on historical temperature data under normal operating conditions to obtain a dynamic threshold model. This dynamic threshold model dynamically outputs the normal range threshold for each temperature sensor data at the current moment, based on the current process stage, the average temperature inside the furnace, and the temperature change trend. The real-time temperature data within the current temperature time series window is compared with the normal range threshold. Data segments that continuously exceed the threshold range and conform to a preset fluctuation pattern are marked as potential abnormal temperature fluctuation patterns.

[0021] When fusing and analyzing the identified potential abnormal temperature fluctuation patterns, the operating status data, and the power supply quality data to generate a comprehensive furnace health status assessment result, feature extraction is performed on the potential abnormal temperature fluctuation patterns to obtain fluctuation amplitude, fluctuation frequency, spatial distribution range, and spatial correlation characteristics with heating elements, forming a temperature anomaly feature vector. The operating status data, including the current and voltage of heating elements, vibration and noise of the circulating fan, and furnace pressure data, is encoded into an equipment status feature vector.

[0022] The power supply quality data is processed to extract features such as voltage fluctuation, harmonic content, and power factor, forming a power supply quality feature vector. The temperature anomaly feature vector, equipment status feature vector, and power supply quality feature vector are then input into a pre-trained furnace condition assessment neural network model. The model outputs a multi-dimensional assessment vector, where each dimension corresponds to an aspect score of the furnace condition health status. These assessment vectors collectively constitute the comprehensive furnace condition health status assessment result.

[0023] In practical implementation, taking a box furnace used for metal quenching as an example, a multi-region temperature sensor array consisting of nine K-type thermocouples arranged on the top and side walls of the furnace is used. The data acquisition module acquires the real-time temperature data stream of each thermocouple at a sampling rate of 10 times per second. When processing the real-time temperature data stream using the improved abnormal temperature point identification algorithm, a sliding window sampling is first performed on the real-time temperature data stream of each thermocouple. The window length is set to 60 sampling points, and the sliding step size is 30 sampling points, forming a series of temperature time series windows of equal length. For the data within each temperature time series window, its statistical characteristics are calculated. These statistical characteristics include the arithmetic mean, standard deviation, slope of the temperature value versus time curve fitted using the least squares method, and curvature of the second derivative of the temperature value curve. Simultaneously, the spatial gradient characteristics of temperature change are calculated by dividing the temperature difference between the thermocouples at the same position in the current window and the previous sliding window by the time interval between the center points of the two windows.

[0024] In some embodiments, the statistical features and spatial gradient features are input into the improved abnormal temperature point identification algorithm, which is trained on a dynamic threshold model based on temperature data under historical normal operating conditions. Complete temperature data of nine thermocouples at various process stages in multiple normal quenching processes of the box furnace are collected as a training dataset. For each process stage and each thermocouple's corresponding feature dataset, a kernel density estimation method is used to fit the probability distribution of each feature, and a confidence interval boundary covering 99.7% of normal data points is set as the basic threshold for that feature. The dynamic threshold model stores the basic thresholds corresponding to all process stages, all thermocouples, and all features. During real-time operation, it dynamically outputs the normal range thresholds for each thermocouple data at the current moment based on the current process stage, the average temperature inside the furnace, and the current temperature change trend. The real-time temperature data within the current temperature time series window is compared with the normal range thresholds, and data segments that continuously exceed the threshold range and conform to a preset fluctuation pattern are marked as the potential abnormal temperature fluctuation pattern.

[0025] Optionally, feature extraction is performed on the identified potential abnormal temperature fluctuation patterns, calculating the fluctuation amplitude, fluctuation frequency, spatial distribution range, and spatial correlation characteristics with the heating element. These four feature values ​​are then arranged in sequence to form a temperature anomaly feature vector. The operating status data, including the current and voltage of the heating element, the vibration and noise of the circulating fan, and the furnace pressure data, are encoded. The current and voltage fluctuation rate, the average vibration amplitude, the noise sound pressure level, and the furnace pressure deviation are arranged in sequence and encoded as an equipment status feature vector. The power supply quality data is processed, extracting voltage fluctuation, harmonic content, and power factor features. These three feature values ​​form a power supply quality feature vector. The temperature anomaly feature vector, the equipment status feature vector, and the power supply quality feature vector are concatenated and input into a pre-trained furnace condition assessment neural network model. The furnace condition assessment neural network model outputs a five-dimensional assessment vector, with each dimension corresponding to an aspect score of the furnace condition health status. These five-dimensional assessment vectors together constitute the comprehensive furnace condition health status assessment result. It can be understood that the dynamic threshold model performs linear interpolation correction on the base threshold based on the current average furnace temperature to generate the normal range threshold. Assume the current average temperature inside the furnace is The two adjacent reference average temperature points are respectively and The corresponding basic threshold upper limits are respectively and The upper limit of the normal range threshold at the current moment The calculation formula is: in: This is the upper limit of the normal range threshold after linear interpolation correction. and Temperature points and The corresponding upper limit of the basic threshold, The current average temperature inside the furnace is calculated in real time. and These are the values ​​from two reference average temperature points adjacent to the current process stage in historical data. Similarly, the correction for the lower limit of the normal range threshold uses the same linear interpolation formula.

[0026] In one embodiment of the present invention, based on the comprehensive furnace health status assessment results, when the energy-saving control strategy engine calculates the set of optimized control parameters for adjusting the heating element power, circulating fan speed, and process stage duration, the energy-saving control strategy engine embeds a multi-objective optimization function. This multi-objective optimization function takes temperature uniformity, energy consumption minimization, and process completion time as optimization objectives, and uses the equipment safe operating range and process temperature curve requirements as constraints. (See also...) Figure 3 The comprehensive furnace health status assessment results are used as input parameters for the multi-objective optimization function. Scores from different dimensions in the assessment results are used to adjust the weights of the objective function and the boundaries of the constraints. The multi-objective optimization function is solved to obtain a Pareto optimal solution set under the current furnace condition assessment results. From the Pareto optimal solution set, an optimal solution is selected according to a preset priority strategy. The heating element power adjustment, circulating fan speed setpoint, and duration adjustment of each process stage corresponding to the optimal solution are used as the set of optimized control parameters.

[0027] To obtain the Pareto optimal solution set under the current furnace condition assessment, an improved non-dominated sorting genetic algorithm is used to solve the multi-objective optimization function, initializing a population composed of multiple control parameters. The objective function values ​​for temperature uniformity, total energy consumption, and expected process completion time are calculated for each individual in the population. Based on these objective function values, non-dominated sorting and crowding calculations are performed on the individuals in the population to simulate natural selection. Selection, crossover, and mutation operations are executed to generate a new generation of the population. During the evolutionary process, specific dimension scores from the comprehensive furnace condition health status assessment are used as adaptive parameters to dynamically adjust the crossover and mutation probabilities of the improved non-dominated sorting genetic algorithm. After the evolution reaches a preset number of generations, the set of individuals representing the non-dominated frontier in the last generation is output as the Pareto optimal solution set.

[0028] In specific implementation, based on the comprehensive furnace health status assessment results, an energy-saving control strategy engine calculates a set of optimized control parameters for adjusting the heating element power, circulating fan speed, and process stage duration. The energy-saving control strategy engine embeds a multi-objective optimization function, which takes temperature uniformity, energy consumption minimization, and process completion time as optimization objectives, and equipment safe operating range and process temperature curve requirements as constraints. The comprehensive furnace health status assessment results are used as input parameters for the multi-objective optimization function, and the scores of different dimensions in the assessment results are used to adjust the weights of the objective function and the boundaries of the constraints.

[0029] In some embodiments, the multi-objective optimization function is solved to obtain a Pareto optimal solution set under the current furnace health status assessment results. An improved non-dominated sorting genetic algorithm is used to solve the multi-objective optimization function. A population composed of multiple control parameters is initialized. The objective function values ​​of temperature uniformity, total energy consumption, and expected process completion time for each individual in the population are calculated. Based on the objective function values, the individuals in the population are non-dominated sorted and crowding is calculated. Selection, crossover, and mutation operations are performed to generate a new generation of population. During the evolution process, the specific dimension scores in the comprehensive furnace health status assessment results are used as adaptive parameters to dynamically adjust the crossover and mutation probabilities of the improved non-dominated sorting genetic algorithm. When the evolution reaches a preset number of generations, the set of individuals that represent the non-dominated frontier in the last generation is output as the Pareto optimal solution set. An optimal solution is selected from the Pareto optimal solution set according to a preset priority strategy. The heating element power adjustment, circulating fan speed setting, and duration adjustment of each process stage corresponding to the optimal solution are used as the set of optimized control parameters.

[0030] Optionally, when dynamically adjusting the crossover and mutation probabilities during the evolution process, scores reflecting furnace temperature uniformity and equipment operational stability are extracted from the comprehensive furnace health status assessment results. The product of the temperature uniformity score and the equipment operational stability score is calculated as an adaptive adjustment factor. In each generation of the improved non-dominated sorting genetic algorithm, the preset basic crossover probability is multiplied by the adaptive adjustment factor to obtain the actual crossover probability of the current generation, and the preset basic mutation probability is multiplied by the adaptive adjustment factor to obtain the actual mutation probability of the current generation. The actual crossover and mutation probabilities are used to guide the selection, crossover, and mutation operations of the current generation. It can be understood that the adaptive adjustment factor... The calculation formula is: in: As an adaptive adjustment factor, This is a score reflecting the uniformity of furnace temperature, extracted from the comprehensive furnace health status assessment results. This is a score reflecting the operational stability of the equipment, extracted from the comprehensive furnace health status assessment results.

[0031] In some embodiments, when using an improved non-dominated sorting genetic algorithm to solve the multi-objective optimization function and initialize the population, a preset number of initial individuals are uniformly and randomly generated within the domain of the control parameters. Each individual is represented by a chromosome with a real-number encoding. The chromosome contains multiple gene loci, each corresponding to a control parameter to be optimized. The control parameters include heating element power, circulating fan speed, and process stage duration. Boundary checks are performed on the real-number encoded chromosomes, and gene locus values ​​exceeding the domain are adjusted to the closest valid boundary values. The fitness values ​​of all initial individuals on the three objectives of temperature uniformity, total energy consumption, and expected process completion time are calculated. The chromosomes of all initial individuals that have completed fitness evaluation and the corresponding fitness value set together constitute the initial population for solving the multi-objective optimization function. It can be understood that the score reflecting the equipment operating stability is extracted from the comprehensive furnace health status assessment results. Take the arithmetic mean of the heating element health score, fan operation efficiency score, and power supply compatibility score.

[0032] In one embodiment of the present invention, when the optimized control parameter set is converted into control commands and sent to the heating actuator, fan driver, and process controller of the box furnace, the optimized control parameter set is first parsed to obtain the power adjustment amount for each heating loop, the target speed value of the circulating fan, and the target duration adjustment amount for each process stage. According to the communication protocol of the heating actuator, the power adjustment amount is converted into a corresponding analog output value or digital control word to form a heating control command. According to the control mode of the fan driver, the target speed value is converted into a frequency command or a speed feedback closed-loop set value to form a fan control command. According to the timing logic of the process controller, the end time point of each process stage is recalculated to form an updated process stage timing command. The heating control command, fan control command, and updated process stage timing command are then sent to the corresponding heating actuator, fan driver, and process controller according to their respective control cycles.

[0033] In practical implementation, taking a box furnace used for metal quenching as an example, this box furnace is equipped with three independent heating circuits forming heating actuators, a variable frequency drive circulating fan as a fan driver, and a programmable logic controller as a process controller. The optimized control parameter set includes a power adjustment of +5% for the left heating circuit, a power adjustment of -3% for the middle heating circuit, a power adjustment of +2% for the right heating circuit, a target speed of 1350 rpm for the circulating fan, a heating stage duration adjustment of +120 seconds, and a holding stage duration adjustment of -60 seconds. When converting the optimized control parameter set into control commands and issuing them, the optimized control parameter set is first parsed to obtain the power adjustment for each heating circuit, the target speed of the circulating fan, and the target duration adjustment for each process stage.

[0034] In some embodiments, according to the communication protocol of the heating actuator, the heating actuator uses a 4 to 20 mA analog signal to control the thyristor power regulator. The power adjustment is superimposed with the current power setpoint to obtain a new power percentage. The power percentage is then converted into a corresponding analog output value through linear transformation to form a heating control command. According to the control mode of the fan driver, the fan driver uses a frequency converter and operates in a speed feedback closed-loop mode. The target speed value of 1350 rpm is directly converted into the speed setpoint of the frequency converter to form a fan control command. Optionally, according to the timing logic of the process controller, the process controller internally uses a timer to manage the remaining time of each process stage. The original duration of the heating stage (1800 seconds) is added to the duration adjustment + 120 seconds, and the original duration of the heat preservation stage (3600 seconds) is added to the duration adjustment - 60 seconds. The end time of each process stage is recalculated to form an updated process stage timing command. The heating control command, fan control command, and updated process stage timing command are then sent to the corresponding heating actuator, fan driver, and process controller according to their respective control cycles.

[0035] It is understandable that the input signal range of the heating actuator is assumed to be... to The corresponding power percentage range is to The current power percentage after adjustment is: Then the analog output value The calculation formula is: in: This is the analog output value corresponding to the heating control command. The lower limit (4 mA) of the input signal to the heating actuator. This is the upper limit (20 mA) of the input signal to the heating actuator. This represents the percentage of heating circuit power after the power adjustment is added together. The minimum power percentage (0%) allowed for the heating actuator. This represents the maximum permissible power percentage (100%) for the heating actuator. In some embodiments, Table 1 shows the correspondence between the control parameters of the three heating loops before and after the conversion.

[0036] Table 1: Conversion Table of Control Parameters for Heating Circuit In one embodiment of the present invention, when the improved abnormal temperature point identification algorithm trains a dynamic threshold model based on temperature data under historical normal operating conditions, it collects complete temperature data from a multi-region temperature sensor array during each process stage of the box furnace during its historical normal and stable operation, using this data as a training dataset. For each temperature sensor in the training dataset, the data is divided according to the process stage, and statistical and spatial gradient features identical to those in the current time window are extracted. For each process stage and each temperature sensor's corresponding feature dataset, a kernel density estimation method is used to fit the probability distribution of its feature values. Based on the fitted probability distribution, a confidence interval covering the vast majority of normal data points is set for each feature, and the boundary values ​​of the confidence interval constitute the basic threshold for the feature under the corresponding process stage. The dynamic threshold model stores the basic thresholds corresponding to all process stages, all temperature sensors, and all features, and performs linear interpolation correction on the basic thresholds based on the current average furnace temperature to generate the normal range threshold.

[0037] When generating the normal range threshold by linear interpolating the base threshold based on the current average furnace temperature, the average value of all temperature sensor readings in the box furnace is calculated in real time as the current average furnace temperature. The current process stage is determined, and the upper and lower limits of the base thresholds for each feature of the current temperature sensor under this stage are obtained. The feature thresholds corresponding to two reference average temperature points adjacent to the current process stage are retrieved from historical data. Based on the current average furnace temperature, the correction values ​​for each feature threshold when it falls between the two reference average temperature points are calculated using linear interpolation. The correction values ​​obtained from linear interpolation are used as the normal range thresholds.

[0038] In specific implementation, taking a box furnace used for metal quenching as an example, a multi-region temperature sensor array is formed by 9 K-type thermocouples arranged on the top and side walls of the furnace. The improved abnormal temperature point identification algorithm is trained based on temperature data under historical normal operating conditions to obtain a dynamic threshold model. First, complete temperature data of the 9 thermocouples are collected during each process stage (heating stage, holding stage, cooling stage) of the box furnace during its historical normal and stable operation, as a training dataset. The temperature data of each thermocouple in the training dataset is divided according to the process stage, and the same statistical features as the current time window are extracted. The system uses mean, standard deviation, slope, curvature, and spatial gradient features. For each process stage and each thermocouple, a kernel density estimation method is used to fit the probability distribution of each feature value. Based on the fitted probability distribution, a confidence interval covering 99.7% of normal data points is set for each feature. The boundary values ​​of the confidence interval constitute the basic threshold of the feature under the corresponding process stage. The dynamic threshold model stores the basic thresholds corresponding to all process stages, all thermocouples, and all features. The basic thresholds are linearly interpolated and corrected according to the current average temperature in the furnace to generate the normal range threshold.

[0039] In some embodiments, when generating the normal range threshold by linear interpolation correction of the base threshold based on the current furnace average temperature, the arithmetic mean of all nine thermocouple readings in the box furnace is calculated in real time as the current furnace average temperature. The current process stage (e.g., the heat preservation stage) is determined, and the upper and lower limits of the base thresholds for each characteristic of the current thermocouple under the current process stage are obtained. Two reference average temperature points adjacent to the current process stage in historical data (denoted as...) are then queried. and The corresponding feature thresholds are calculated using linear interpolation based on the current average temperature inside the furnace, with correction values ​​for each feature threshold calculated when it falls between two reference average temperature points. These correction values ​​are then used as the normal range thresholds. For the mean feature of the current thermocouple, a reference average temperature point is set... The upper limit of the mean feature base threshold corresponding to 750 degrees Celsius is 785 degrees Celsius, referring to the average temperature point. The upper limit of the basic threshold for the mean feature corresponding to 850 degrees Celsius is 810 degrees Celsius. The current average temperature inside the furnace is 800 degrees Celsius. Therefore, the upper limit of the normal range threshold for the current mean feature is calculated to be 797.5 degrees Celsius through linear interpolation.

[0040] Optionally, the dynamic threshold model dynamically outputs the normal range threshold based on the current process stage, the average temperature inside the furnace, and the temperature change trend during real-time operation. The temperature change trend is characterized by the average temperature slope of the most recent 10 sampling windows, used to adjust the width coefficient of the confidence interval. When the temperature change trend is positive and its absolute value is greater than 0.5 degrees Celsius per second, the confidence interval boundary is expanded outward by 5%. The table below shows the basic threshold values ​​for the central thermocouple at different reference average temperature points during the insulation stage. Refer to Table 2, Basic Threshold Values ​​for the Central Thermocouple During the Insulation Stage.

[0041] Table 2: Basic Thresholds for Central Thermocouple Characteristics During the Insulation Stage It is understandable that we assume the current average temperature inside the furnace is... The two adjacent reference average temperature points are respectively and and The corresponding upper limits of the mean feature basic thresholds are respectively and Then the upper limit of the normal range threshold for the current mean feature. The linear interpolation formula is: in: This represents the upper limit of the normal range threshold for the current mean feature after linear interpolation correction. Reference average temperature point The corresponding upper limit of the basic threshold for the mean feature, Reference average temperature point The corresponding upper limit of the basic threshold for the mean feature, The current average temperature inside the furnace is calculated in real time. and The values ​​are taken from two reference average temperature points adjacent to the current process stage in historical data. The lower limit of the mean characteristic threshold and the normal range thresholds of other characteristics (standard deviation, slope, curvature) are calculated separately using the same linear interpolation formula.

[0042] In some embodiments, when the current average temperature inside the furnace When the average temperature is not equal to any stored reference average temperature point, a linear interpolation method is used to calculate the dynamic threshold for each feature. If the current average temperature inside the furnace is lower than the stored minimum reference average temperature point or higher than the maximum reference average temperature point, the basic threshold corresponding to the nearest reference average temperature point is directly used as the normal range threshold, and no extrapolation interpolation is performed. For example, when the current average temperature inside the furnace is 730 degrees Celsius, the threshold is directly used. This is the base threshold corresponding to 750 degrees Celsius. It can be understood that the dynamic threshold model updates the current process stage, average furnace temperature, and temperature change trend every 500 milliseconds, and recalculates the normal range threshold for real-time temperature data comparison by the anomaly identification module.

[0043] In one embodiment of the present invention, when using an improved non-dominated sorting genetic algorithm to solve the multi-objective optimization function, and initializing a population composed of multiple control parameters, a preset number of initial individuals are uniformly and randomly generated within the domain of the control parameters. Each individual is represented by a chromosome encoded by a real number. The chromosome contains multiple gene loci, each corresponding to a control parameter to be optimized. The control parameters include heating element power, circulating fan speed, and process stage duration. Boundary checks are performed on the real-number encoded chromosomes, and gene locus values ​​exceeding the domain are adjusted to the closest valid boundary values. The fitness values ​​of all initial individuals are calculated for the three objectives of temperature uniformity, total energy consumption, and expected process completion time. The chromosomes of all initial individuals that have completed fitness evaluation and their corresponding fitness value sets together constitute the initialization population for solving the multi-objective optimization function.

[0044] During the evolutionary process, specific dimension scores from the comprehensive furnace health status assessment results are used as adaptive parameters. When dynamically adjusting the crossover and mutation probabilities of the improved non-dominated sorting genetic algorithm, scores reflecting furnace temperature uniformity and equipment operational stability are extracted from the comprehensive furnace health status assessment results. The product of the temperature uniformity score and the equipment operational stability score is calculated as an adaptive adjustment factor. In each generation of the improved non-dominated sorting genetic algorithm, the preset base crossover probability is multiplied by the adaptive adjustment factor to obtain the actual crossover probability of the current generation. The preset base mutation probability is multiplied by the adaptive adjustment factor to obtain the actual mutation probability of the current generation. The actual crossover and mutation probabilities are used to guide the selection, crossover, and mutation operations of the current generation.

[0045] In a specific implementation, taking a box furnace used for metal quenching as an example, the improved non-dominated sorting genetic algorithm needs to perform population initialization when solving the multi-objective optimization function. When using the improved non-dominated sorting genetic algorithm to solve the multi-objective optimization function, a preset number (e.g., 100) of initial individuals are uniformly and randomly generated within the domain of the control parameters. Each individual is represented by a chromosome encoded by a real number. The chromosome contains multiple gene positions, and each gene position corresponds to a control parameter that needs to be optimized. The control parameters include six gene positions: power percentage of the left heating element, power percentage of the middle heating element, power percentage of the right heating element, circulating fan speed percentage, heating stage duration percentage, and holding stage duration percentage. The domain of each gene position is the power percentage range [30%, 100%], the speed percentage range [20%, 100%], and the duration percentage range [80%, 120%].

[0046] Boundary checks are performed on the real-number encoded chromosomes, adjusting gene loci values ​​outside the defined domain to the closest valid boundary values. For example, if the power percentage gene locus value is 105%, it is adjusted to 100%; if the duration percentage gene locus value is 75%, it is adjusted to 80%. The fitness values ​​of all initial individuals are calculated for three objectives: temperature uniformity (standard deviation of the temperature values ​​of the nine thermocouples in the furnace), total energy consumption (sum of heating element power consumption and fan power consumption), and expected process completion time (sum of heating stage duration and holding stage duration). The fitness values ​​are the reciprocals of the three objective function values, respectively. The chromosomes of all initial individuals that have completed fitness evaluation, along with their corresponding fitness value sets, constitute the initial population for solving the multi-objective optimization function.

[0047] In some embodiments, during the evolution of the improved non-dominated sorting genetic algorithm, the crossover probability and mutation probability need to be dynamically adjusted. A score reflecting the uniformity of furnace temperature is extracted from the comprehensive furnace health status assessment results. And a score reflecting the stability of equipment operation ,in The temperature uniformity dimension is directly taken from the comprehensive furnace health status assessment results. The arithmetic mean of the heating element health score, fan operating efficiency score, and power supply adaptability score in the comprehensive furnace condition health status assessment results is taken. The product of the temperature uniformity score and the equipment operating stability score is calculated as the adaptive adjustment factor. ,Right now In each generation of the improved non-dominated sorting genetic algorithm, a preset base crossover probability is used. With the adaptive adjustment factor Multiply to obtain the actual crossover probability of the current generation. The preset base mutation probability With the adaptive adjustment factor Multiply to obtain the actual mutation probability of the current generation. Using the actual crossover probability. Compared with the actual mutation probability It guides the selection, crossover, and mutation operations of the current generation.

[0048] It is understood that the adaptive adjustment factor The calculation formula is: in: As an adaptive adjustment factor, This is a score extracted from the comprehensive furnace health status assessment results, reflecting the uniformity of furnace temperature. To assess the health of heating elements in the overall furnace health status evaluation results, To score the operating efficiency of the wind turbine, To obtain the power supply adaptability score, the sum of the three factors is divided by 3 to get the equipment operational stability score. Preset base crossover probability The value is 0.9, representing the preset base mutation probability. The value is 0.1. When the calculated... When it is greater than 1.0, The amplitude is limited to 1.0, when When it is less than 0.2, Adjust it to 0.2 to ensure the population has basic search capabilities.

[0049] Optionally, during the initialization of the population, the chromosome of each initial individual consists of 6 real-valued gene loci, generated using a uniformly random method. For each gene locus, a random number is drawn uniformly within its corresponding domain interval. After boundary verification, the chromosome of each individual is stored as a floating-point array of length 6. When calculating the fitness value, the specific values ​​of temperature uniformity, total energy consumption, and expected process completion time are first calculated based on the power percentage, rotation speed percentage, and time percentage in the chromosome. Since the three objective functions have different dimensions, each objective function value needs to be normalized. The minimum value of each objective in the current population is mapped to 0, and the maximum value is mapped to 1. The reciprocal is then taken to obtain the fitness value. The normalized fitness value is stored in the individual structure, which, together with the chromosome, constitutes the initialized population.

[0050] In some embodiments, the actual crossover probability in each generation of the improved non-dominated sorting genetic algorithm and actual mutation probability With adaptive adjustment factor Changes generation by generation. The temperature uniformity score in the comprehensive furnace health status assessment results... When the score is low (e.g., below 0.5) or the equipment's operational stability score is low, A smaller value leads to a lower actual crossover probability and an lower actual mutation probability, thereby reducing drastic disturbances to the population and preserving currently superior individuals; when and When both are relatively high, The value is close to 1.0, and the actual crossover probability and actual mutation probability are close to the base value, maintaining a normal search intensity. During each generation of evolution, when performing selection, crossover, and mutation operations, for each pair of parent individuals, the following values ​​are used: The probability of performing a simulated binary crossover operation is given, for each gene position, with... The probability of performing a polynomial mutation operation is [a certain value].

[0051] It is understood that the process of initializing the population and dynamically adjusting the crossover and mutation probabilities described above is repeated in each run of the improved non-dominated sorting genetic algorithm until a preset number of generations (e.g., 200 generations) is reached. The final Pareto optimal solution set is used to generate the set of optimized control parameters.

[0052] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A temperature monitoring system for a box furnace, characterized in that, The system includes: The data acquisition module acquires real-time temperature data streams collected by the multi-zone temperature sensor array inside the box furnace, operating status data of key furnace components, and power quality data of the external power grid. The anomaly identification module performs an improved anomaly temperature point identification algorithm on the real-time temperature data stream to identify potential abnormal temperature fluctuation patterns. The improved anomaly temperature point identification algorithm dynamically adjusts the anomaly judgment threshold based on the spatiotemporal correlation of the temperature field inside the furnace. The assessment and analysis module integrates and analyzes the identified potential abnormal temperature fluctuation patterns, the operating status data, and the power supply quality data to generate a comprehensive furnace health status assessment result. Based on the comprehensive furnace health status assessment results, the strategy control module calculates a set of optimized control parameters for adjusting the power of heating elements, the speed of circulating fans, and the duration of process stages through the energy-saving control strategy engine. The instruction execution module converts the optimized control parameter set into control instructions and sends them to the heating actuator, fan driver and process controller of the box furnace to achieve coordinated monitoring of the temperature and energy consumption of the box furnace.

2. A temperature monitoring system for a box furnace as claimed in claim 1, wherein, An improved abnormal temperature point identification algorithm is applied to the real-time temperature data stream to identify potential abnormal temperature fluctuation patterns, including: The real-time temperature data stream collected by the multi-region temperature sensor array is sampled using a sliding window to form a series of temperature time series windows of equal length; For the data within each temperature time series window, its statistical characteristics are calculated, including mean, standard deviation, slope and curvature, and the spatial gradient characteristics of temperature change are calculated using the data between sliding windows. The statistical features and spatial gradient features are input into the improved abnormal temperature point identification algorithm; The improved abnormal temperature point identification algorithm is trained based on temperature data under historical normal operating conditions to obtain a dynamic threshold model. The dynamic threshold model dynamically outputs the normal range threshold of each temperature sensor data at the current moment according to the current process stage, the average temperature in the furnace and the temperature change trend. The real-time temperature data within the current temperature time series window is compared with the normal range threshold, and data segments that continuously exceed the threshold range and conform to the preset fluctuation pattern are marked as the potential abnormal temperature fluctuation pattern.

3. The temperature monitoring system for a box furnace according to claim 1, characterized in that, The process involves fusing and analyzing the identified potential abnormal temperature fluctuation patterns, the operating status data, and the power supply quality data to generate a comprehensive furnace health status assessment result, including: Feature extraction is performed on the potential abnormal temperature fluctuation patterns to obtain the fluctuation amplitude, fluctuation frequency, spatial distribution range, and spatial correlation characteristics with the heating element, thus forming a temperature anomaly feature vector; The operating status data, including the current and voltage of the heating element, the vibration and noise of the circulating fan, and the furnace pressure data, are encoded into an equipment status feature vector. The power supply quality data is processed to extract voltage fluctuation, harmonic content, and power factor features, forming a power supply quality feature vector. The temperature anomaly feature vector, equipment status feature vector, and power supply quality feature vector are input into a pre-trained furnace condition evaluation neural network model. The furnace condition assessment neural network model outputs a multi-dimensional assessment vector, where each dimension of the multi-dimensional assessment vector corresponds to an aspect score of the furnace condition health status, and the assessment vectors together constitute the comprehensive furnace condition health status assessment result.

4. A temperature monitoring system for a box furnace according to claim 1, characterized in that, Based on the comprehensive furnace health status assessment results, an optimized set of control parameters for adjusting heating element power, circulating fan speed, and process stage duration is calculated by the energy-saving control strategy engine, including: The energy-saving control strategy engine has a multi-objective optimization function embedded in it. The multi-objective optimization function takes temperature uniformity, energy consumption minimization and process completion time as optimization objectives, and takes equipment safe operating range and process temperature curve requirements as constraints. The comprehensive furnace health status assessment results are used as input parameters for the multi-objective optimization function. The scores of different dimensions in the assessment results are used to adjust the weights of the objective function and the boundaries of the constraints. Solve the multi-objective optimization function to obtain the Pareto optimal solution set under the current furnace condition evaluation results; From the Pareto optimal solution set, an optimal solution is selected according to a preset priority strategy, and the heating element power adjustment, circulating fan speed setting, and duration adjustment of each process stage corresponding to the optimal solution are used as the optimized control parameter set.

5. A temperature monitoring system for a box furnace according to claim 4, characterized in that, Solving the multi-objective optimization function yields a Pareto optimal solution set under the current furnace condition assessment, including: An improved non-dominated sorting genetic algorithm is used to solve the multi-objective optimization function, and a population composed of multiple control parameters is initialized. Calculate the objective function values ​​for temperature uniformity, total energy consumption, and estimated process completion time for each individual in the population; Based on the objective function value, non-dominated ranking and crowding degree calculation are performed on individuals in the population to simulate natural selection; Perform selection, crossover, and mutation operations to generate a new generation of population; During the evolution process, the specific dimension scores in the comprehensive furnace health status assessment results are used as adaptive parameters to dynamically adjust the crossover probability and mutation probability of the improved non-dominated sorting genetic algorithm. When the evolution reaches the preset number of generations, the set of individuals that represent the last generation as the non-dominant frontier will be output as the Pareto optimal solution set.

6. A temperature monitoring system for a box furnace according to claim 1, characterized in that, The optimized control parameter set is converted into control commands and sent to the heating actuators, fan drivers, and process controllers of the box furnace, including: The optimized control parameter set is analyzed to obtain the power adjustment amount, the target speed value of the circulating fan, and the target duration adjustment amount for each process stage for each heating circuit. According to the communication protocol of the heating actuator, the power adjustment is converted into the corresponding analog output value or digital control word to form a heating control command; Based on the control mode of the wind turbine driver, the target speed value is converted into a frequency command or a set value of the speed feedback closed loop to form a wind turbine control command. Based on the timing logic of the process controller, the end time of each process stage is recalculated to form updated process stage timing instructions. The heating control command, the fan control command, and the updated process stage timing command are sent to the corresponding heating actuator, fan driver, and process controller according to their respective control cycles.

7. A temperature monitoring system for a box furnace according to claim 2, characterized in that, The improved abnormal temperature point identification algorithm is based on a dynamic threshold model trained using historical temperature data under normal operating conditions, including: Collect complete temperature data from a multi-zone temperature sensor array during various process stages of a box furnace during its historical normal and stable operation, and use it as a training dataset. For each temperature sensor in the training dataset, the data is divided according to the process stage, and the same statistical features and spatial gradient features as the current time window are extracted. For each process stage and each temperature sensor's corresponding feature dataset, the kernel density estimation method is used to fit the probability distribution of its feature values. Based on the probability distribution obtained by fitting, a confidence interval covering the vast majority of normal data points is set for each feature, and the boundary values ​​of the confidence interval constitute the basic threshold of the feature under the corresponding process stage. The dynamic threshold model stores the basic thresholds corresponding to all process stages, all temperature sensors, and all features, and performs linear interpolation correction on the basic thresholds based on the current average temperature inside the furnace to generate the normal range thresholds.

8. A temperature monitoring system for a box furnace according to claim 7, characterized in that, The step of linearly interpolating and correcting the base threshold based on the current average temperature inside the furnace to generate the normal range threshold includes: The average value of all temperature sensor readings inside the box furnace is calculated in real time and used as the current average temperature inside the furnace. Determine the current process stage and obtain the basic upper and lower thresholds of each feature of the current temperature sensor under the current process stage; Query the historical data to find the feature thresholds corresponding to the two reference average temperature points adjacent to the current process stage; Based on the current average temperature inside the furnace, the correction values ​​of each characteristic threshold are calculated using linear interpolation when the temperature is between two reference average temperature points. The corrected value obtained by linear interpolation is used as the normal range threshold.

9. A temperature monitoring system for a box furnace according to claim 5, characterized in that, The improved non-dominated sorting genetic algorithm is used to solve the multi-objective optimization function, and a population composed of multiple control parameters is initialized, including: Within the defined range of the control parameters, a preset number of initial individuals are generated uniformly and randomly, and each individual is represented by a chromosome encoded by a real number. The chromosome contains multiple gene loci, and each gene locus corresponds to a control parameter that needs to be optimized. The control parameters include heating element power, circulating fan speed, and process stage duration. Boundary checks are performed on the real-number encoded chromosomes, and gene position values ​​that exceed the defined domain are adjusted to the closest valid boundary values. Calculate the fitness values ​​of all initial individuals on three objectives: temperature uniformity, total energy consumption, and expected process completion time; The chromosomes and corresponding fitness values ​​of all initial individuals that have completed fitness evaluation are used together to form the initial population for solving the multi-objective optimization function.

10. A temperature monitoring system for a box furnace according to claim 5, characterized in that, During the evolutionary process, specific dimension scores from the comprehensive furnace health status assessment results are used as adaptive parameters to dynamically adjust the crossover and mutation probabilities of the improved non-dominated sorting genetic algorithm, including: From the comprehensive furnace health status assessment results, scores reflecting the uniformity of furnace temperature and scores reflecting the stability of equipment operation are extracted; The product of the temperature uniformity score and the equipment operating stability score is calculated and used as an adaptive adjustment factor. In each generation of the improved non-dominated sorting genetic algorithm, the preset base crossover probability is multiplied by the adaptive adjustment factor to obtain the actual crossover probability of the current generation; Multiply the preset base mutation probability by the adaptive adjustment factor to obtain the actual mutation probability of the current generation; The actual crossover probability and the actual mutation probability are used to guide the selection, crossover, and mutation operations of the current generation.

Images