Intelligent optimization method and system for operation parameters of industrial electric furnace based on energy consumption modeling

Abstract

The application discloses an industrial electric furnace operation parameter intelligent optimization method and system based on energy consumption modeling, and relates to the technical field of industrial electric furnace operation control. The industrial electric furnace operation parameter intelligent optimization method based on energy consumption modeling obtains geometric position data of a workpiece after charging through non-contact scanning, and constructs a radiation heat exchange space relationship by matching the geometric position data with fixed coordinates of heating elements and furnace wall linings. A Monte Carlo method is used to solve an angular coefficient distribution matrix of two radiation heat exchange paths for discrete surface units, and the matrix is modified by weighting and correcting the difference between emissivity and reflectivity. The matrix is used as a radiation boundary condition of a heat conduction equation, and a sample entropy constraint multivariate nonlinear regression is established to build an energy consumption representation model of an output heat storage state. The application obtains the geometric position of the workpiece through scanning, constructs a radiation heat exchange space relationship, solves an angular coefficient matrix to quantify the efficiency difference of the heat transfer path, builds an energy consumption model of the output heat storage state, and embeds the model into a predictive control framework. The power optimization strategy is solved under multiple constraints, and closed-loop regulation and control are performed.

Description

Technical Field

[0001] This invention relates to the field of industrial electric furnace operation control technology, specifically to an intelligent optimization method and system for industrial electric furnace operating parameters based on energy consumption modeling. Background Technology

[0002] Industrial electric furnaces are core thermal equipment for heating metal workpieces. The operation and control of their high-temperature holding stage are directly related to the heat treatment quality of the workpieces and the overall energy consumption level of the equipment. In intermittent heat treatment production mode, workpieces are put into the furnace in different batches. The geometric shape, placement posture and loading position of each batch of workpieces are different, which causes the spatial relative relationship between the heating element surface, the workpiece surface and the furnace wall lining surface to change with each batch.

[0003] Currently, industrial electric furnaces generally employ a closed-loop control method during the high-temperature holding stage, with the furnace gas temperature as the controlled variable. This involves real-time acquisition of gas temperature signals via thermocouples installed at specific locations within the furnace, comparing these signals with the process setpoint, and adjusting the input power of the heating elements based on the deviation. A hysteresis compensation mechanism is also introduced to address the inherent response delay of the large-inertia thermal object. Under this control logic, the control system determines whether the overall thermal state of the furnace meets process requirements based solely on the gas temperature measured at a single point or a limited number of points by temperature sensors, and executes power output commands accordingly.

[0004] At the level of heat transfer physics, after an industrial electric furnace enters the high-temperature holding stage, heat transfer within the furnace is primarily radiative. Part of the heat generated by the heating element is directly projected onto the workpiece surface via radiation and absorbed, while the other part is projected onto the furnace lining surface, absorbed and stored by the lining, and then released into the furnace space via radiation and convection. Due to changes in the workpiece's geometry, the radiative heat transfer path efficiency of different parts of the heating element surface relative to the workpiece surface and the furnace lining surface is not uniformly distributed. The radiation angle coefficient varies at different spatial locations, and this difference dynamically changes depending on the furnace loading conditions.

[0005] In actual operation, operators usually preset the heat preservation power based on experience, or the control system executes a step-by-step power reduction strategy according to a fixed program to balance the uniform temperature requirements of the workpiece and prevent furnace overshoot. The recording of relevant operating data and energy consumption statistics are mostly based on the accumulation of active power, and there is a lack of means to distinguish between the part of the input power that actually participates in the effective heating of the workpiece and the part that is converted into furnace wall heat storage redundancy during the heat preservation stage.

[0006] The limitations of existing technologies include at least the following problems: In the operation control of the high-temperature holding stage of industrial electric furnaces, existing solutions focus on closed-loop regulation and lag compensation of furnace gas temperature. However, they lack real-time quantitative characterization methods for the differences in radiative heat transfer path efficiency between the heating element and the workpiece surface and the furnace wall lining caused by changes in the geometric position of the workpiece inside the furnace. This makes it difficult for the control system to identify the actual heat storage state inside the furnace caused by the offset of the radiation angle coefficient distribution. It is easy to see the phenomenon of energy distribution imbalance when the furnace wall has accumulated redundant heat energy in a local area while the core of the workpiece is still in a state of under-uniform temperature, based solely on temperature feedback to determine that the process has met the standard. This causes repeated input power redundancy and temperature overshoot tendency during the holding stage to compensate for the lag in the thermal response of the workpiece. As a result, the operation parameter optimization strategy based on this control logic is difficult to accurately match the actual thermal efficiency requirements under variable load conditions, and it is also difficult to maintain the batch energy consumption stability of intermittent heat treatment equipment under micro-overshoot constraints. Summary of the Invention

[0007] To address the shortcomings of existing technologies, this invention provides an intelligent optimization method and system for the operating parameters of industrial electric furnaces based on energy consumption modeling, which solves the problem of energy distribution imbalance caused by the difficulty in quantifying the differences in radiative heat transfer paths during the heat preservation stage of industrial electric furnaces in existing technologies.

[0008] To achieve the above objectives, this invention provides the following technical solution: an intelligent optimization method for industrial electric furnace operating parameters based on energy consumption modeling, comprising the following steps: performing non-contact three-dimensional scanning of the workpiece inside the furnace after loading to obtain the workpiece's three-dimensional geometric position data; constructing the radiative heat transfer spatial relationship between the heating element surface, the workpiece surface, and the furnace wall lining surface based on the workpiece's three-dimensional geometric position data, the fixed spatial coordinates of the heating element inside the furnace, and the fixed spatial coordinates of the furnace wall lining; discretizing the radiative heat transfer spatial relationship; and using the Monte Carlo method to solve for the radiation angle coefficient for each pair of discrete surface units to form a radiation angle coefficient distribution matrix of the radiative heat transfer path from the heating element to the workpiece surface and the radiative heat transfer path from the heating element to the furnace wall lining surface; and, during the solution process, the workpiece surface... The unit and the furnace wall lining surface unit are assigned weight factors based on the difference between surface emissivity and surface reflectivity; the radiation angle coefficient distribution matrix is ​​used as the boundary condition for the radiation heat transfer term of the heat conduction equation, and a multivariate nonlinear regression constrained by sample entropy is used to construct an energy consumption characterization model. The output of this energy consumption characterization model is the actual heat storage state inside the furnace caused by the deviation of the radiation angle coefficient distribution; the energy consumption characterization model is embedded into the model predictive control framework, and under the constraints of the input power sequence, temperature overshoot amplitude, and workpiece core temperature uniformity time during the heat preservation stage, the dynamic time warping algorithm is used to match the heat balance threshold as the optimization termination condition to solve the optimization strategy for the heating element operating parameters; according to the optimization strategy for the heating element operating parameters, the input power of the heating element during the high-temperature heat preservation stage of the industrial electric furnace is adjusted.

[0009] Furthermore, the specific steps for constructing the spatial relationship of radiative heat transfer between the heating element surface, the workpiece surface, and the furnace wall lining surface include: using a structured light scanner to scan the workpiece inside the furnace from multiple angles after loading to obtain point cloud data of the workpiece surface; denoising and registering the point cloud data of the workpiece surface to generate three-dimensional geometric position data of the workpiece; and spatially registering the three-dimensional geometric position data of the workpiece with the fixed spatial coordinates of the heating element inside the furnace and the fixed spatial coordinates of the furnace wall lining pre-stored in the control system to construct the spatial relationship of radiative heat transfer.

[0010] Furthermore, the specific steps for forming the radiation angle coefficient distribution matrix of the radiative heat transfer path from the heating element to the workpiece surface and to the furnace wall lining surface include: discretizing the heating element, workpiece, and furnace wall lining surface in the radiative heat transfer spatial relationship into several surface unit grids; for each discrete unit on the surface of the heating element, randomly emitting an energy beam into space using the Monte Carlo method, and statistically analyzing the proportion of the energy beam reaching the discrete units on the workpiece and furnace wall lining surfaces, which is used as the radiation angle coefficient of that pair of discrete units; arranging the radiation angle coefficients of all discrete unit pairs with the discrete units on the surface of the heating element as rows and the discrete units on the surface of the workpiece and furnace wall lining as columns to form a radiation angle coefficient distribution matrix.

[0011] Furthermore, when solving for the radiation angle coefficient, the specific steps for assigning weight factors based on the difference between surface emissivity and surface reflectivity to the workpiece surface unit and the furnace wall lining surface unit include: obtaining the surface emissivity of the workpiece surface material and the surface reflectivity of the furnace wall lining surface material; setting the weight factor of each workpiece surface discrete unit to be positively correlated with its surface emissivity; setting the weight factor of each furnace wall lining surface discrete unit to be negatively correlated with its surface reflectivity; and when using the Monte Carlo method to solve for the radiation angle coefficient, multiplying the proportion of energy beam reaching the corresponding discrete unit by its weight factor and including it in the radiation angle coefficient statistics.

[0012] Furthermore, the specific steps for constructing an energy consumption characterization model using sample entropy-constrained multivariate nonlinear regression include: establishing the internal heat conduction equation of the furnace, substituting the radiation angle coefficient distribution matrix into its radiation heat transfer term as the boundary conditions for the heating element, workpiece, and furnace wall lining surface; using the furnace gas temperature, heating element input power sequence, and radiation angle coefficient distribution matrix from historical operating data during the heat preservation stage as inputs, and the measured value of the internal heat storage state of the furnace as outputs, to construct a multivariate nonlinear regression model; introducing a sample entropy regularization term into the loss function of the multivariate nonlinear regression model, determining the regularization term coefficient through cross-validation, obtaining the sample entropy-constrained multivariate nonlinear regression model, and determining it as the energy consumption characterization model.

[0013] Furthermore, the steps for calculating sample entropy include: obtaining the time series of furnace gas temperature during the heat preservation stage; reconstructing the phase space of the time series and setting the embedding dimension and similarity tolerance; calculating the number of vector pairs that satisfy the similarity tolerance condition to obtain the sample entropy value; and introducing the sample entropy value as the sample entropy regularization term into the loss function of the multivariate nonlinear regression model.

[0014] Furthermore, the specific steps for solving the optimization strategy for the heating element operating parameters include: using the energy consumption characterization model as the prediction model in the model predictive control framework; setting constraints on the amplitude and rate of change of the input power sequence during the heat preservation stage, the amplitude constraint on temperature overshoot, and the time constraint on the temperature uniformity of the workpiece core; within each control cycle, using the input power sequence during the heat preservation stage as the decision variable, calling the energy consumption characterization model to predict the heat storage state inside the furnace, and calculating the furnace gas temperature response curve in the prediction time domain; using the dynamic time warping algorithm to calculate the similarity between the response curve and the preset thermal equilibrium temperature curve, terminating the optimization iteration when the similarity reaches the thermal equilibrium threshold; and outputting the input power sequence that satisfies the constraints and reaches the optimization termination condition as the optimization strategy for the heating element operating parameters.

[0015] Furthermore, the specific steps for calculating similarity using the dynamic time warping algorithm include: obtaining the time series of the furnace gas temperature response curve and the preset thermal equilibrium temperature curve in the predicted time domain; constructing a distance matrix between the two time series, where the matrix elements are the absolute values ​​of the differences between the corresponding time point data; searching for the warping path with the smallest cumulative distance in the distance matrix, which is the dynamic time warping distance between the two time series; and using the dynamic time warping distance as a similarity measure between the two.

[0016] Furthermore, the specific steps for regulating the input power of the heating element during the high-temperature heat preservation stage of the industrial electric furnace include: converting the input power sequence of the heat preservation stage in the heating element operating parameter optimization strategy into a control signal for the heating element power supply circuit; adjusting the conduction time or amplitude of the heating element power supply circuit according to the control signal during each control cycle of the high-temperature heat preservation stage; and monitoring the furnace gas temperature and the actual input power of the heating element and feeding it back to the model predictive control framework for the next control cycle.

[0017] An intelligent optimization system for industrial electric furnace operating parameters based on energy consumption modeling includes: a spatial relationship construction module, used to perform non-contact 3D scanning of the workpiece inside the furnace after loading, acquiring the 3D geometric position data of the workpiece, and constructing the radiative heat transfer spatial relationship between the heating element surface, the workpiece surface, and the furnace wall lining surface based on the 3D geometric position data of the workpiece and the fixed spatial coordinates of the heating element and the furnace wall lining; and an angle coefficient matrix formation module, used to discretize the radiative heat transfer spatial relationship, and solve the radiation angle coefficient for each pair of discrete surface elements using the Monte Carlo method, forming a radiation angle coefficient distribution matrix for the radiative heat transfer path from the heating element to the workpiece surface and the radiative heat transfer path from the heating element to the furnace wall lining surface. During the solution process, the workpiece surface elements and the furnace wall lining surface elements are assigned surface-based... The module includes: a weighting factor for the difference between emissivity and surface reflectivity; an energy consumption model construction module, which uses the radiation angle coefficient distribution matrix as the boundary condition for the radiation heat transfer term in the heat conduction equation, and constructs an energy consumption characterization model using multivariate nonlinear regression constrained by sample entropy. The output of this energy consumption characterization model is the actual heat storage state inside the furnace caused by the shift in the radiation angle coefficient distribution; an optimization strategy solution module, which embeds the energy consumption characterization model into the model predictive control framework, and solves the optimization strategy for heating element operating parameters under the constraints of the input power sequence, temperature overshoot amplitude, and workpiece core temperature equalization time during the heat preservation stage, using a dynamic time warping algorithm to match the thermal equilibrium threshold as the optimization termination condition; and a power regulation module, which regulates the input power of the heating element during the high-temperature heat preservation stage of the industrial electric furnace according to the optimization strategy for the heating element operating parameters.

[0018] The present invention has the following beneficial effects:

[0019] (1) The intelligent optimization method for the operating parameters of industrial electric furnace based on energy consumption modeling is to perform non-contact three-dimensional scanning of the workpiece after loading into the furnace, and to uniformly register the geometric position data of the workpiece with the fixed spatial coordinates of the heating element and the furnace wall lining to construct a radiation heat transfer spatial relationship covering the complete spatial posture of the three; and to perform mesh discretization on each heat transfer surface, and to solve the radiation angle coefficient element by element using the Monte Carlo method. The calculation results are weighted and corrected by introducing a weight factor based on the difference between surface emissivity and reflectivity, so as to obtain a radiation angle coefficient distribution matrix that can accurately distinguish the heat transfer efficiency of the two radiation paths of the heating element and the workpiece, and the heating element and the furnace wall lining. Thus, the radiation heat transfer bias problem implied by the geometric layout of the furnace is transformed into quantifiable and comparable matrix data.

[0020] (2) The intelligent optimization method for the operating parameters of industrial electric furnace based on energy consumption modeling uses the radiation angle coefficient distribution matrix as the boundary condition of the radiation heat transfer term in the heat conduction equation and embeds it into the furnace heat transfer calculation system. The furnace gas temperature, heating element input power sequence and radiation angle coefficient distribution matrix in the historical operating data of the heat preservation stage are used as model inputs, and the measured value of the heat storage state inside the furnace is used as the output. A multivariate nonlinear regression energy consumption characterization model with sample entropy regularization term is constructed. This model can directly output the actual heat storage state of the furnace caused by the radiation angle coefficient distribution shift, rather than simply relying on the temperature sensor collection value. It can penetrate the lag characterization of the furnace gas temperature and identify the redundant heat storage in the furnace wall area and the energy imbalance state of the under-uniform temperature in the core of the workpiece in advance.

[0021] (3) The intelligent optimization method for the operating parameters of industrial electric furnace based on energy consumption modeling embeds the energy consumption characterization model into the model prediction control framework and sets multi-dimensional constraints such as the amplitude and rate of change of input power, the amplitude of temperature overshoot, and the temperature uniformity time of the core of the workpiece. The dynamic time warping algorithm is used to quantify the similarity between the predicted temperature response curve and the preset thermal equilibrium temperature curve, and the similarity reaching the thermal equilibrium threshold is used as the termination condition for optimization iteration. The traditional empirical and conservative heat preservation power adjustment is transformed into a rolling optimization strategy that takes into account process constraints. The output power of the heating element can avoid redundant loading to compensate for thermal lag, and will not cause the temperature uniformity time to be prolonged due to premature power reduction. While ensuring the heat treatment quality of the workpiece, it significantly suppresses temperature overshoot and reduces ineffective energy consumption, and achieves simultaneous improvement of energy consumption stability and thermal efficiency under variable load conditions.

[0022] (4) The intelligent optimization system for the operating parameters of industrial electric furnace based on energy consumption modeling achieves fully automated operation from the perception of the spatial attitude inside the furnace, the modeling of radiation heat exchange, the prediction of energy consumption status to the closed-loop control of power through the coordinated cooperation of spatial relationship construction, angle coefficient matrix formation, energy consumption model construction, optimization strategy solution and power control module. It continuously and stably outputs the optimal operating parameters without human intervention, effectively solving problems such as energy distribution imbalance, power redundancy, temperature overshoot and batch energy consumption fluctuation during the high temperature heat preservation stage of industrial electric furnace.

[0023] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0024] Figure 1 This is a flowchart of the intelligent optimization method for industrial electric furnace operating parameters based on energy consumption modeling, as described in this invention.

[0025] Figure 2 This is a flowchart illustrating the specific steps involved in constructing the spatial relationship of radiative heat transfer between the surface of the heating element, the surface of the workpiece, and the surface of the furnace lining in the intelligent optimization method for industrial electric furnace operating parameters based on energy consumption modeling, as described in this invention.

[0026] Figure 3 This is a block diagram of the intelligent optimization system for industrial electric furnace operating parameters based on energy consumption modeling, as described in this invention. Detailed Implementation

[0027] Please see Figure 1 This invention provides a technical solution: an intelligent optimization method for industrial electric furnace operating parameters based on energy consumption modeling, comprising the following steps: performing non-contact three-dimensional scanning of the workpiece inside the furnace after loading to obtain the workpiece's three-dimensional geometric position data; constructing the radiative heat transfer spatial relationship between the heating element surface, the workpiece surface, and the furnace wall lining surface based on the workpiece's three-dimensional geometric position data and the fixed spatial coordinates of the heating element and the furnace wall lining; discretizing the radiative heat transfer spatial relationship; and using the Monte Carlo method to solve the radiation angle coefficient for each pair of discrete surface units to form a radiation angle coefficient distribution matrix of the radiative heat transfer path from the heating element to the workpiece surface and the radiative heat transfer path from the heating element to the furnace wall lining surface; and, during the solution process, adjusting the relationship between the workpiece surface unit and the furnace wall lining surface. The inner lining surface units are assigned weighting factors based on the difference between surface emissivity and surface reflectivity. The radiation angle coefficient distribution matrix is ​​used as the boundary condition for the radiation heat transfer term in the heat conduction equation. A multivariate nonlinear regression constrained by sample entropy is used to construct an energy consumption characterization model. The output of this energy consumption characterization model is the actual heat storage state inside the furnace caused by the shift in the radiation angle coefficient distribution. The energy consumption characterization model is embedded into the model predictive control framework. Under the constraints of the input power sequence, temperature overshoot amplitude, and workpiece core temperature equalization time during the heat preservation stage, the dynamic time warping algorithm is used to match the heat balance threshold as the optimization termination condition to solve the heating element operating parameter optimization strategy. Based on the heating element operating parameter optimization strategy, the input power of the heating element during the high-temperature heat preservation stage of the industrial electric furnace is adjusted.

[0028] Before performing the above steps, the structured light scanner is pre-calibrated to determine core parameters such as scanning resolution and scanning angle. The scanning error is calculated and controlled by angle error and scanning distance to ensure that the scanning error after calibration does not exceed 0.01mm.

[0029] The number of energy beams emitted and the discrete element division criteria for the Monte Carlo method are preset, with the preset number of energy beams being 10. 6 The side length of the discrete unit is preset to 8cm.

[0030] The pre-training of the multivariate nonlinear regression model constrained by sample entropy is performed in the following steps:

[0031] 1. Construction of pre-training dataset: A pre-training sample set was constructed using historical operating data of industrial electric furnaces over the past 6 months. The sample set contains 5,000 valid samples. Each sample set covers the furnace gas temperature, heating element input power, radiation angle coefficient distribution matrix, and measured values ​​of furnace heat storage status. The samples were normalized (all input and output parameters were mapped to the 0-1 range), and outlier samples (samples with a deviation exceeding 3 times the standard deviation) were removed. The dataset was divided into a training set (80%) and a validation set (20%).

[0032] 2. Model Initialization: Initialize the regression coefficients (a0-a9) of the multivariate nonlinear regression model, set the initial learning rate to 0.001, the maximum number of iterations to 1000, and preset the prediction error threshold to no more than 5%;

[0033] 3. Iterative pre-training: Using the training set as input, the model parameters are optimized through the gradient descent algorithm. The mean square error between the model's predicted value and the measured value of the heat storage state is used as the loss function. The validation set error is calculated every 100 iterations. If the validation set error increases for two consecutive times, the learning rate is adjusted using a learning rate decay strategy (decay coefficient 0.8).

[0034] 4. Pre-training termination and parameter saving: When the model validation set error is lower than the preset threshold (≤5%) or the upper limit of the number of iterations is reached, the pre-training is terminated and the optimal model parameters (regression coefficients a0-a9) are saved for real-time prediction of subsequent energy consumption characterization models; at the same time, the fixed spatial coordinates of the heating elements in the furnace and the furnace wall lining are pre-stored, and the preset heat balance threshold is not lower than 0.95.

[0035] Specifically, such as Figure 2 As shown, the specific steps for establishing the spatial relationship of radiative heat transfer between the heating element surface, the workpiece surface, and the furnace wall lining surface include:

[0036] A structured light scanner was used to scan the workpieces inside the furnace from multiple angles after loading, obtaining point cloud data of the workpiece surface. Specifically:

[0037] A line structured light scanner is used to scan the workpiece after it is loaded into the furnace from three angles: left, right and top. The scanning resolution is preset to 0.3mm and the scanning frequency is 10Hz. The scanning range completely covers all parts of the workpiece and avoids scanning blind spots.

[0038] During the scanning process, the scanning angle (0°≤θ≤180°) and scanning distance (0.5m≤d≤2.0m) are collected simultaneously. A spatial rectangular coordinate system is established with the geometric center of the furnace as the origin. The two-dimensional scanning data is converted into three-dimensional coordinates through coordinate transformation to form the initial point cloud data of the workpiece surface.

[0039] For example, when the scanning angle is 60°, the scanning distance is 1.0m, and the scanning height is 1.0m, the corresponding point cloud coordinates are (0.5m, approximately 0.866m, 1.0m).

[0040] The point cloud data of the workpiece surface is denoised and registered to generate three-dimensional geometric shape and position data of the workpiece, specifically as follows:

[0041] The Gaussian filtering algorithm is used to denoise the initial point cloud data. The filtering parameters are set and the data is removed.

[0042] Point cloud registration is performed using the Iterative Closest Point (ICP) algorithm. The point cloud scanned from the top angle is used as the reference point cloud, and the point clouds scanned from the left and right sides are used as the point clouds to be registered. The rotation matrix and translation vector are iteratively optimized to make the point cloud to be registered and the reference point cloud accurately aligned. The iteration terminates when the number of iterations reaches 50 or the registration error is less than 0.05 mm.

[0043] After registration, complete three-dimensional geometric shape and position data of the workpiece are obtained, and the total number of point clouds after registration is approximately 85%-95% of the total number of initial point clouds.

[0044] The three-dimensional geometric shape and position data of the workpiece are spatially registered with the fixed spatial coordinates of the heating elements inside the furnace and the fixed spatial coordinates of the furnace wall lining, which are pre-stored in the control system, to construct the spatial relationship of radiative heat transfer. Specifically:

[0045] A unified spatial rectangular coordinate system is established with the geometric center of the furnace as the origin. The pre-stored spatial coordinates of the heating elements, the fixed spatial coordinates of the furnace wall lining, and the registered three-dimensional geometric position data of the workpiece are uniformly transformed into this coordinate system (using translation transformation).

[0046] Spatial geometric calculations are used to determine the spatial distances between any point on the surface of the heating element and any point on the surface of the workpiece and the inner lining of the furnace wall, and the normal vectors of each surface are calculated (obtained by the cross product of two non-collinear vectors of the surface).

[0047] By combining spatial distance and normal vector, the radiative heat transfer path and spatial attitude relationship between any two surfaces can be clearly defined.

[0048] In this implementation scheme, a structured light scanner is used to acquire point clouds of the workpiece surface from multiple angles. After filtering and registration, the three-dimensional geometric position data of the workpiece is generated. Then, it is uniformly registered with the fixed spatial coordinates of the heating element and the furnace wall lining. This constructs a spatial relationship of radiative heat transfer that covers the relative positions and orientations of the three, so that the radiative heat transfer geometry that changes due to different workpiece placement postures in each batch can be accurately reproduced. This avoids the blind control that previously relied solely on furnace gas temperature feedback and ignored the specific shading effect of the workpiece. By clarifying the spatial distance and normal vector relationship between the heating element surface, the workpiece surface, and the furnace wall lining surface, it helps to reduce power redundancy and maintain stable batch energy consumption during the heat preservation stage.

[0049] Specifically, the steps for forming the radiation angle coefficient distribution matrix of the radiative heat transfer path from the heating element to the workpiece surface and to the furnace wall lining surface include:

[0050] The heating element, workpiece, and furnace wall lining surfaces in the spatial relationship of radiative heat transfer are discretized into several surface element meshes, specifically:

[0051] The quadrilateral meshing method is used to discretize the surface of the heating element, the surface of the workpiece, and the surface of the furnace wall lining. After discretization, each surface unit mesh is a square with a side length of 8cm (the area of ​​a single unit is 64cm²≤100cm², taking into account both the accuracy and efficiency of the calculation of radiative heat transfer in industrial electric furnaces).

[0052] Discretized elements are obtained on the surface of the heating element. ( This represents the total number of discrete units of the heating element, based on the number and size of the heating element. (Value range is 80-120) Discrete elements on the workpiece surface ( (The total number of discrete units in the workpiece, ranging from 150 to 250), and discrete units on the inner lining surface of the furnace wall. ( (This represents the total number of discrete units in the furnace wall lining, ranging from 200 to 300). The center coordinates of each discrete unit are as follows: ;

[0053] For each discrete unit on the surface of the heating element, an energy beam is randomly emitted into space using the Monte Carlo method. The proportion of energy beams reaching the discrete units on the workpiece and furnace wall lining surfaces is statistically analyzed and used as the radiation angle coefficient for that pair of discrete units. Specifically:

[0054] For each discrete unit of the heating element, according to the preset 10 6 A beam of energy is randomly emitted into space. The direction of the energy beam is uniformly distributed, and the direction angle (the angle between the beam and the element normal vector) and the azimuth angle are both within a reasonable range.

[0055] The propagation path of the energy beam is a straight line. The ray tracing method is used to determine whether the energy beam is blocked by other discrete units (the criterion for blocking is: there are discrete units blocking the energy beam path, and the distance from the intersection point to the transmitting unit is less than the distance to the target unit).

[0056] The number of energy beams that are not blocked and reach the discrete unit of the workpiece and the inner surface of the furnace wall are counted. The number of reaching energy beams is divided by the total number of emitted energy beams to obtain the radiation angle coefficient of the corresponding discrete unit pair.

[0057] For example: when the discrete unit H1 of the heating element emits 10 6 A beam of energy; of which 25,000 beams reach the discrete unit W1 of the workpiece, the radiation angle coefficient of both is 0.025;

[0058] The radiation angle coefficients of all discrete element pairs are arranged with the discrete elements on the heating element surface as rows and the discrete elements on the workpiece and furnace wall lining surface as columns, forming a radiation angle coefficient distribution matrix, which is as follows:

[0059] Radiation angle coefficient distribution matrix for The matrix is ​​of order 1, and the row indices of the matrix correspond to the discrete units of the heating element. ( ), before column index Column corresponding to the discrete unit of the workpiece surface ( ),back The column corresponds to the discrete unit of the furnace wall lining surface. ( );

[0060] Matrix elements ( )correspond Matrix elements ( )correspond The matrix expression is:

[0061] ;

[0062] The range of values ​​for each element in the matrix is: And the sum of the elements in each row satisfies (Consider energy loss caused by energy beam blocking).

[0063] The specific steps for assigning weighting factors based on the difference between surface emissivity and surface reflectivity to the workpiece surface elements and the furnace wall lining surface elements when solving for the radiation angle coefficient include:

[0064] The surface emissivity of the workpiece surface material and the surface reflectivity of the furnace wall lining surface material are obtained as follows:

[0065] The surface emissivity of the workpiece surface material was measured using an infrared emissivity meter. The test temperature was the typical temperature during the heat preservation stage of an industrial electric furnace (800-1200℃), and the test accuracy was ±0.01. The value range is 0.75-0.85;

[0066] The surface reflectance of the furnace wall lining material was measured using a reflectance meter. The test wavelength was 2-10μm (the main wavelength range for radiative heat transfer in industrial electric furnaces), and the results were obtained. The value range is 0.15-0.25;

[0067] Discrete unit on each workpiece surface Surface emissivity Average emissivity of workpiece surface The deviation does not exceed ±0.02, and each discrete unit of the furnace wall lining surface Surface reflectivity Average reflectivity of furnace wall lining The deviation shall not exceed ±0.02;

[0068] The weighting factor for each discrete element on the workpiece surface is set to be positively correlated with its surface emissivity, specifically as follows:

[0069] The weighting factor of the discrete unit on the workpiece surface is calculated using a linear positive correlation method. A proportionality coefficient and a constant term are set to ensure that the weighting factor ranges from 0.875 to 0.925. The higher the emissivity, the greater the contribution of radiative heat transfer, and the larger the weighting factor.

[0070] For example, when the surface emissivity of discrete unit W1 on the workpiece surface is 0.8, its weighting factor is 0.9;

[0071] The weighting factor for each discrete element of the furnace wall lining surface is set to be negatively correlated with its surface reflectivity, specifically as follows:

[0072] The weighting factor of the discrete unit on the inner surface of the furnace wall lining is calculated using a linear negative correlation method. A proportionality coefficient and a constant term are set to ensure that the weighting factor ranges from 0.975 to 1.025. The higher the reflectivity, the smaller the contribution of radiative heat transfer, and the smaller the weighting factor.

[0073] For example, when the surface reflectance of discrete unit B1 on the inner lining surface of the furnace wall is 0.2, its weighting factor is 1.0;

[0074] When using the Monte Carlo method to solve for the radiation angle coefficient, the proportion of the energy beam reaching the corresponding discrete element is multiplied by its weighting factor and included in the radiation angle coefficient statistics. Specifically:

[0075] The original radiation angle coefficients are multiplied by the weighting factor of the corresponding discrete element to obtain the corrected radiation angle coefficients. The distribution matrix of the corrected radiation angle coefficients replaces the original matrix.

[0076] For example: if the original radiation angle coefficient is 0.025 and the weighting factor of the discrete element W1 of the workpiece is 0.9, then the corrected radiation angle coefficient is 0.0225.

[0077] Even after correction, the sum of the elements in each row still satisfies the constraint that it is no greater than 1.

[0078] In this implementation scheme, by discretizing the heating element surface, workpiece surface, and furnace wall lining surface into unit grids and using the Monte Carlo method to track the energy beam arrival ratio, a radiation angle coefficient distribution matrix is ​​obtained that can distinguish the efficiency of the two paths of heat transfer from the heating element to the workpiece and heat transfer to the furnace wall lining. Based on this, a weighting factor based on the difference between surface emissivity and surface reflectivity is introduced to correct the calculation results. This takes into account the tendency of the workpiece surface to enhance radiation absorption due to its higher emissivity and the tendency of the furnace wall lining to weaken heat retention due to its higher reflectivity. The resulting matrix mathematically solidifies the radiation heat transfer bias that dynamically shifts inside the furnace due to changes in the geometric position of the workpiece, which helps to avoid energy waste caused by improper power distribution during the heat preservation stage.

[0079] Specifically, the steps for constructing an energy consumption characterization model using multivariate nonlinear regression constrained by sample entropy include:

[0080] A heat conduction equation for the furnace interior is established, and the radiation angle coefficient distribution matrix is ​​substituted into its radiation heat transfer term as the boundary conditions for the heating element, workpiece, and furnace wall lining surface. Specifically:

[0081] The heat conduction equation inside the furnace adopts a three-dimensional unsteady-state heat conduction equation, the formula of which is:

[0082] ;

[0083] in, The density of the medium inside the furnace (kg / m³). is the specific heat capacity of the medium (J / (kg·℃)). For time (s), The thermal conductivity of the medium is (W / (m·℃)). The temperature (°C) is the temperature at a certain point inside the furnace. The intensity of the internal heat source is (W / m³).

[0084] The boundary conditions for the radiative heat transfer term are:

[0085] ;

[0086] in, The Stefan-Boltzmann constant ( ), Discrete unit for heating element Temperature (°C) Discrete unit of workpiece Temperature (°C) Discrete unit for furnace wall lining Temperature (°C);

[0087] The corrected radiation angle coefficient distribution matrix Substitute the boundary conditions;

[0088] Using the furnace gas temperature, heating element input power sequence, and radiation angle coefficient distribution matrix from historical operating data during the heat preservation stage as inputs, and the measured value of the internal heat storage state of the furnace as output, a multivariate nonlinear regression model is constructed, specifically as follows:

[0089] There are three input variables: furnace gas temperature, heating element input power sequence, and radiation angle coefficient distribution matrix.

[0090] The output variable is the actual heat storage state inside the furnace. The measured value of the heat storage state is calculated from the temperature, density, volume and specific heat capacity of each discrete unit inside the furnace.

[0091] The multivariate nonlinear regression model uses a quadratic polynomial fitting, which includes the first term, the second term, and the cross term of the input variables. The regression coefficients are optimized to obtain initial values ​​through a pre-training step. The overall expression of the model is that the output variable is equal to the nonlinear fitting function of the input variables plus the random error term (the random error term follows a normal distribution).

[0092] A sample entropy regularization term is introduced into the loss function of the multivariate nonlinear regression model. The coefficient of the regularization term is determined through cross-validation, resulting in a sample entropy-constrained multivariate nonlinear regression model, which is then identified as an energy consumption characterization model. Specifically:

[0093] Construct the loss function of the unconstrained regression model (mean square error between model predictions and measured values ​​of thermal storage).

[0094] By introducing a sample entropy regularization term (the product of the regularization coefficient and the sample entropy value), a constrained loss function is constructed.

[0095] The optimal regularization term coefficient (ranging from 0.001 to 0.1) was determined using 5-fold cross-validation, and the coefficient with the smallest error on the test set was selected as the optimal value.

[0096] By substituting the optimal regularization coefficient into the loss function and optimizing the regression coefficient through the gradient descent algorithm, a multivariate nonlinear regression model constrained by sample entropy (i.e., energy consumption characterization model) is obtained. The prediction error of this model does not exceed 5%, and it can accurately output the actual heat storage state inside the furnace.

[0097] The steps for calculating sample entropy include:

[0098] The time series of furnace gas temperature during the heat preservation stage is obtained as follows: During the heat preservation stage of the industrial electric furnace, furnace gas temperature data is collected every 1 second using a temperature sensor inside the furnace. The collection time is the total heat preservation stage duration (assumed to be 3600s), and the temperature time series is obtained.

[0099] The time series needs to be preprocessed to remove outliers (values ​​whose difference from the average temperature of the five adjacent time points is greater than 5℃ are outliers) and replace outliers with linear interpolation.

[0100] For example: if the temperature at time 100 is 1010℃, and the average temperature of the next 5 times is 1003℃, the difference is 7℃ (greater than 5℃), then the temperature at that time should be replaced with the average temperature of the next two times.

[0101] For the phase space reconstruction of this time series, the embedding dimension and similarity tolerance are set as follows:

[0102] The CC method was used to determine the embedding dimension and time delay of phase space reconstruction. By calculating the autocorrelation function and mutual information function of the time series, the optimal embedding dimension was determined to be 3 and the optimal time delay to be 5.

[0103] The phase space is reconstructed into a 3-dimensional vector, and the number of vectors after reconstruction is calculated based on the length of the original time series and the time delay.

[0104] The similarity tolerance is set at 0.15 times the standard deviation of the time series, which is calculated from the original time series.

[0105] For example, if the average value of the time series is 1000℃ and the standard deviation is 10℃, then the similarity tolerance is 1.5℃.

[0106] The number of vector pairs satisfying the similarity tolerance condition is calculated to obtain the sample entropy value, which is as follows:

[0107] Calculate the Chebyshev distance between each 3D vector and all other vectors (take the maximum value of the difference in the corresponding dimension);

[0108] Count the number of vector pairs that meet the similarity tolerance (distance not greater than a set value), and calculate their proportion of the total number of vector pairs;

[0109] Increase the embedding dimension to 4 and repeat the above steps to obtain a new scale value;

[0110] The sample entropy value is calculated by the logarithmic ratio of the two scale values;

[0111] The smaller the sample entropy value, the stronger the regularity of the time series and the better the stability of the model. In this embodiment, the sample entropy value ranges from 0.5 to 1.5.

[0112] The sample entropy value is used as the sample entropy regularization term and introduced into the loss function of the multivariate nonlinear regression model, specifically as follows:

[0113] The calculated sample entropy value Substitute the regular expression terms:

[0114] ;

[0115] in, These are the coefficients of the optimal regularization term determined through cross-validation;

[0116] Adding the regularization term to the unconstrained loss function yields the constrained loss function:

[0117] ;

[0118] The purpose of introducing the sample entropy regularization term is to suppress model overfitting and ensure the model's generalization ability under different operating conditions. When the sample entropy value increases (the regularity of the time series deteriorates), the weight of the regularization term increases, constraining the changes in model parameters.

[0119] In this implementation scheme, the radiation angle coefficient distribution matrix is ​​substituted into the furnace heat transfer calculation framework as the radiation heat transfer boundary condition of the heat conduction equation. The furnace gas temperature, input power sequence and angle coefficient matrix accumulated in historical operation are used as inputs, and the measured value of heat storage state is used as output. A preliminary mapping relationship is constructed through multivariate nonlinear regression, and the sample entropy obtained by phase space reconstruction based on furnace temperature time series is introduced as the regularization term of the loss function. This makes the model constrained by the time series complexity while fitting the training data, avoiding the problem of reduced adaptability to the working conditions of new batches of workpieces due to overfitting historical samples. The final energy consumption characterization model can output the actual heat storage state inside the furnace caused by the angle coefficient offset with high accuracy.

[0120] Specifically, the steps for solving the optimization strategy for the operating parameters of the heating element include:

[0121] The energy consumption characterization model is used as the prediction model in the model prediction control framework, specifically as follows:

[0122] The Model Predictive Control (MPC) framework includes a prediction module, an optimization module, and a feedback correction module. The completed energy consumption characterization model is embedded in the prediction module as a model to predict the actual heat storage state inside the furnace.

[0123] The inputs to the prediction module are the current furnace gas temperature, the current heating element input power, and the current radiation angle coefficient distribution matrix. The output is a prediction sequence of the furnace internal heat storage state for the next 20 control cycles (each control cycle is 10s).

[0124] The deviation between the output of the prediction model and the actual heat storage state is corrected by the feedback correction module, with the correction coefficient set to 0.1 to ensure prediction accuracy;

[0125] The following constraints are set for the amplitude and rate of change of the input power sequence during the heat preservation stage, the amplitude constraint of temperature overshoot, and the temperature uniformity time constraint of the workpiece core:

[0126] The constraints are set based on the actual operating capacity of the industrial electric furnace, as follows:

[0127] 1. Input power sequence amplitude constraint: The input power of the heating element is controlled between 50kW (minimum operating power) and 150kW (maximum rated power);

[0128] 2. Input power change rate constraint: The power change between adjacent control cycles shall not exceed 10kW to avoid sudden power changes that cause drastic fluctuations in furnace temperature;

[0129] 3. Temperature overshoot amplitude constraint: The furnace gas temperature overshoot amplitude (the difference between the actual temperature and the set insulation temperature) shall not exceed 5℃;

[0130] 4. Workpiece core temperature uniformity time constraint: The temperature difference between the workpiece core and the surface temperature shall not exceed 3℃, and the duration of this state shall not be less than 300s;

[0131] Within each control cycle, using the input power sequence during the heat preservation stage as the decision variable, the energy consumption characterization model is invoked to predict the internal heat storage state of the furnace, and the furnace gas temperature response curve in the prediction time domain is calculated, specifically as follows:

[0132] Within each control cycle (10s), the input power sequence of the next 20 control cycles is used as the decision variable;

[0133] Call the energy consumption characterization model, input the current furnace gas temperature, input power sequence and radiation angle coefficient distribution matrix, and predict the furnace internal heat storage state sequence within the next 200 seconds;

[0134] Based on the correlation between the heat storage state and the furnace gas temperature (obtained by fitting historical data), the predicted sequence of the heat storage state is converted into a furnace gas temperature response curve, which reflects the trend of furnace gas temperature change in the future prediction time domain.

[0135] For example: If the heat storage state at a certain moment is 2×10 7J corresponds to a furnace gas temperature of 1000℃;

[0136] The similarity between the response curve and the preset thermal equilibrium temperature curve is calculated using a dynamic time warping algorithm. The optimization iteration terminates when the similarity reaches the thermal equilibrium threshold. Specifically:

[0137] The preset thermal equilibrium temperature curve is a fixed value sequence (a total of 20 data points, consistent with the prediction time domain length, and the fixed value is the heat preservation setting temperature).

[0138] The similarity between the two curves is calculated using the Dynamic Time Warping (DTW) algorithm, and the DTW distance is obtained. The similarity is then calculated by normalizing the DTW distance.

[0139] The maximum possible DTW distance is set to 50℃ based on the temperature range.

[0140] The preset thermal equilibrium threshold is 0.95. When the similarity is not lower than this threshold, the furnace is determined to have reached thermal equilibrium and the optimization iteration is terminated.

[0141] If the similarity does not reach the threshold, the input power sequence is adjusted, and prediction and calculation are performed again until the condition is met.

[0142] The input power sequence that satisfies the constraints and reaches the optimization termination condition is output as the heating element operating parameter optimization strategy, specifically:

[0143] After the optimization iteration terminates, the current input power sequence That is, the optimal input power sequence that satisfies all constraints (power amplitude, power change rate, temperature overshoot, and temperature equalization time) and reaches the thermal equilibrium threshold;

[0144] The sequence is organized into an optimization strategy for the operating parameters of the heating element, and the input power value, power adjustment timing and adjustment range corresponding to each control cycle (10s) are defined and output to the industrial electric furnace control system.

[0145] The optimization strategy output format is in tabular form, as shown in Table 1, with an example as follows:

[0146] Table 1. Example of optimization strategy output format (taking 5 control cycles as an example)

[0147] Control cycle Input power (kW) Power change (kW) Corresponding predicted temperature (°C) 1 100 0 (Initial Power) 1000 2 105 +5 1002 3 108 +3 1003 4 108 0 1003 5 106 -2 1002

[0148] The specific steps for calculating similarity using the dynamic time warping algorithm include:

[0149] The time series of the furnace gas temperature response curve and the preset thermal equilibrium temperature curve within the predicted time domain are obtained as follows:

[0150] The time series of the furnace gas temperature response curve in the prediction time domain is as follows: ( (The predicted furnace gas temperature for the k-th control cycle, in °C).

[0151] The preset thermal equilibrium temperature curve time series is as follows ( The set temperature for heat preservation is 1000℃, which is... );

[0152] Both time series have a length of 20, which is consistent with the number of control cycles in the prediction time domain;

[0153] For example: if the first 5 data points of the predicted response curve time series are [998, 1001, 1003, 1002, 1001], then the first 5 data points of the heat balance time series are [1000, 1000, 1000, 1000, 1000];

[0154] Construct a distance matrix between the two time series, where each element represents the absolute difference between the data points at corresponding time points. Specifically:

[0155] Build The distance matrix D of order, matrix elements The absolute difference between the i-th data point in the response curve time series and the j-th data point in the thermal equilibrium time series is expressed by the following formula:

[0156] ;

[0157] The expression for the distance matrix is:

[0158] ;

[0159] For example: if the first data point of the response curve is 998℃ and the first data point of the heat balance curve is 1000℃, then the absolute difference between the two is 2.

[0160] If the second data point of the response curve is 1001℃ and the first data point of the thermal equilibrium curve is 1000℃, then the absolute difference between the two is 1. By analogy, a complete distance matrix is ​​constructed, and the matrix elements take values ​​ranging from 0 to 5℃ (in accordance with the temperature overshoot constraint).

[0161] The search in the distance matrix yields the normalized path with the minimum cumulative distance, which is the dynamic time normalized distance between the two time series. Specifically:

[0162] Define a cumulative distance matrix. The cumulative distance is calculated by adding the minimum cumulative distance of the three adjacent directions (right, down, and down-right) to the current distance. Set initial conditions.

[0163] The search for a regular path starts from the top left corner of the matrix and ends at the bottom right corner. Each step can only move in the adjacent right, down, or bottom right directions.

[0164] The cumulative distance at each step during the search process is recorded, and the cumulative distance when the bottom right corner is reached is the dynamic time warped distance between the two time series.

[0165] For example: if the last element of the cumulative distance matrix is ​​18, then the DTW distance is 18;

[0166] The dynamic time-warped distance is used as a similarity measure between the two, specifically as follows:

[0167] Similarity was normalized.

[0168] The maximum possible DTW distance is set to 50 (i.e., the maximum temperature deviation per cycle is 2.5℃, which meets the temperature overshoot constraint).

[0169] The similarity value ranges from 0 to 1. The closer the similarity is to 1, the higher the similarity between the two time series and the closer the furnace is to thermal equilibrium.

[0170] When the similarity is not less than 0.95 (thermal balance threshold), the matching degree of the two curves is determined to meet the requirements, and the optimization iteration terminates.

[0171] In this implementation scheme, the energy consumption characterization model is embedded into the model predictive control framework, and the input power sequence during the heat preservation stage is used as the decision variable. Rolling optimization is performed under multiple constraints such as power amplitude and rate of change, temperature overshoot amplitude, and the temperature uniformity time of the workpiece core. At the same time, a dynamic time warping algorithm is used to measure the similarity between the predicted temperature response curve and the preset thermal equilibrium temperature curve and use it as the termination condition for optimization iteration. This means that the output command of the heating element in each control cycle no longer depends solely on the lag feedback of temperature deviation, but rather assesses the comprehensive impact of different power combinations on the furnace heat storage state and temperature response trend in advance in the prediction time domain. Thus, redundant power is actively reduced while meeting process quality requirements. By using the similarity target as the criterion for stopping optimization, energy loss caused by repeated power adjustments due to excessive pursuit of precise temperature control is avoided, and the waiting time required to reach stable thermal equilibrium during the heat preservation stage is also shortened.

[0172] Specifically, the steps for regulating the input power of the heating element during the high-temperature holding stage of an industrial electric furnace include:

[0173] The input power sequence during the heat preservation stage in the heating element operating parameter optimization strategy is converted into a control signal for the heating element power supply circuit, specifically as follows:

[0174] The heating element power supply circuit adopts the thyristor rectification power supply method, and the control signal is the pulse width modulation (PWM) signal. The duty cycle of the PWM signal is linearly positively correlated with the input power. By setting the proportional coefficient and the reference duty cycle, the input power (50-150kW) of each control cycle is converted into the corresponding PWM signal duty cycle.

[0175] For example, when the input power is 100kW, the duty cycle is 1.0;

[0176] When the input power is 50kW, the duty cycle is 0.666;

[0177] The frequency of the converted control signal is set to 1kHz to ensure stable operation of the power supply circuit.

[0178] During each control cycle of the high-temperature insulation stage, the conduction time or amplitude of the heating element power supply circuit is adjusted according to the control signal, specifically as follows:

[0179] Within each control cycle (10s), the industrial electric furnace control system receives the converted PWM control signal and adjusts the conduction time of the power supply circuit (i.e., the high-level duration of the PWM signal) through the thyristor trigger circuit, thereby adjusting the actual input power of the heating element.

[0180] The conduction time is the product of the duty cycle and the control cycle;

[0181] For example, when the duty cycle is 0.8, the on-time is 8s and the off-time is 2s, ensuring that the input power of the heating element is stable at the target value;

[0182] During the adjustment process, the voltage (380V±5V) and current of the power supply circuit are monitored in real time. The duty cycle is corrected through a PID control algorithm. The PID parameters are set (proportional coefficient 0.5, integral coefficient 0.1, derivative coefficient 0.05) to ensure that the deviation between the actual input power and the target power in the optimization strategy does not exceed 2%.

[0183] The model predictive control framework, which monitors the furnace gas temperature and the actual input power of the heating elements and feeds this feedback to the next control cycle, is as follows:

[0184] One second before the end of each control cycle, the current furnace gas temperature is collected by the thermocouple sensor in the furnace, and the actual input power is collected by the current sensor and voltage sensor in the power supply circuit.

[0185] Calculate the temperature deviation (the difference between the actual temperature and the set insulation temperature) and the power deviation (the difference between the actual power and the target power), and feed the deviation data back to the feedback correction module of the model predictive control framework.

[0186] The feedback correction module adjusts the prediction model parameters for the next control cycle based on the deviation data, sets the power deviation conversion coefficient, ensures the prediction accuracy for the next control cycle, and achieves closed-loop control.

[0187] In this implementation scheme, the input power sequence of the heat preservation stage output by the heating element operating parameter optimization strategy is converted into a pulse width modulation signal duty cycle, and the power supply circuit conduction time is adjusted by the thyristor trigger circuit. At the same time, the furnace gas temperature and the actual input power of the heating element are monitored in real time during each control cycle, and the deviation is fed back to the feedback correction module of the model predictive control framework. This ensures that the power command calculated in the optimization strategy can be accurately transmitted to the execution end of the heating element, rather than just remaining at the algorithm level. Voltage and current monitoring and proportional-integral-derivative correction during the power adjustment process jointly ensure the degree of fit between the actual input power and the target value. After the temperature deviation and power deviation are fed back to the predictive control framework to form a closed loop, the predictive model parameters of the next control cycle can be dynamically corrected according to the actual operating conditions. This avoids the phenomenon that the control effect gradually deviates from the expectation due to the accumulation of model errors, and realizes the continuous matching between the energy input and the thermal state of the furnace during the heat preservation stage.

[0188] Please see Figure 3 This invention provides a technical solution: an intelligent optimization system for industrial electric furnace operating parameters based on energy consumption modeling, comprising: a spatial relationship construction module, used to perform non-contact three-dimensional scanning of the workpiece inside the furnace after loading, to obtain the three-dimensional geometric position data of the workpiece, and to construct the radiative heat transfer spatial relationship between the surface of the heating element, the surface of the workpiece, and the surface of the furnace wall lining based on the three-dimensional geometric position data of the workpiece and the fixed spatial coordinates of the heating element inside the furnace and the fixed spatial coordinates of the furnace wall lining; and an angle coefficient matrix formation module, used to discretize the radiative heat transfer spatial relationship, and to solve the radiation angle coefficient for each pair of discrete surface units using the Monte Carlo method, forming a radiation angle coefficient distribution matrix of the radiative heat transfer path from the heating element to the workpiece surface and the radiative heat transfer path from the heating element to the furnace wall lining surface, wherein the solution process involves the workpiece surface unit and the furnace wall lining surface unit... The module assigns weighting factors based on the difference between surface emissivity and surface reflectivity; the energy consumption model construction module uses the radiation angle coefficient distribution matrix as the boundary condition for the radiation heat transfer term in the heat conduction equation, and constructs an energy consumption characterization model using multivariate nonlinear regression constrained by sample entropy. The output of this energy consumption characterization model is the actual heat storage state inside the furnace caused by the shift in the radiation angle coefficient distribution; the optimization strategy solution module embeds the energy consumption characterization model into the model predictive control framework, and solves the optimization strategy for heating element operating parameters under the constraints of the input power sequence, temperature overshoot amplitude, and workpiece core temperature equalization time during the heat preservation stage, using a dynamic time warping algorithm to match the thermal equilibrium threshold as the optimization termination condition; the power regulation module regulates the input power of the heating element during the high-temperature heat preservation stage of the industrial electric furnace according to the optimization strategy for the heating element operating parameters.

[0189] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0190] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for intelligent optimization of operating parameters of industrial electric furnaces based on energy consumption modeling, characterized in that, Includes the following steps: Non-contact 3D scanning is performed on the workpiece inside the furnace after loading to obtain the 3D geometric shape and position data of the workpiece. Based on the 3D geometric shape and position data of the workpiece and the fixed spatial coordinates of the heating element in the furnace and the fixed spatial coordinates of the furnace wall lining, the spatial relationship of radiative heat transfer between the surface of the heating element, the surface of the workpiece and the surface of the furnace wall lining is constructed. The spatial relationship of radiative heat transfer is discretized, and the radiation angle coefficient is solved by Monte Carlo method for each pair of discrete surface elements to form the radiation angle coefficient distribution matrix of the radiative heat transfer path from the heating element to the workpiece surface and the radiative heat transfer path from the heating element to the furnace wall lining surface. During the solution process, the workpiece surface element and the furnace wall lining surface element are assigned a weight factor based on the difference between surface emissivity and surface reflectivity. The radiation angle coefficient distribution matrix is ​​used as the boundary condition for the radiation heat transfer term in the heat conduction equation. A multivariate nonlinear regression constrained by sample entropy is used to construct an energy consumption characterization model. The output of this energy consumption characterization model is the actual heat storage state inside the furnace caused by the offset of the radiation angle coefficient distribution. The energy consumption characterization model is embedded into the model predictive control framework. Under the constraints of the input power sequence, temperature overshoot amplitude, and workpiece core temperature equalization time during the heat preservation stage, the dynamic time warping algorithm is used to match the thermal equilibrium threshold as the optimization termination condition to solve the optimization strategy of the heating element operating parameters. Based on the optimization strategy for heating element operating parameters, the input power of the heating element during the high-temperature heat preservation stage of the industrial electric furnace is adjusted.

2. The intelligent optimization method for industrial electric furnace operating parameters based on energy consumption modeling according to claim 1, characterized in that, The specific steps for establishing the spatial relationship of radiative heat transfer between the heating element surface, the workpiece surface, and the furnace wall lining surface include: A structured light scanner was used to scan the workpiece inside the furnace from multiple angles after it was loaded into the furnace, and point cloud data of the workpiece surface was obtained. Denoise and register the point cloud data of the workpiece surface to generate three-dimensional geometric shape and position data of the workpiece; The three-dimensional geometric position data of the workpiece is spatially registered with the fixed spatial coordinates of the heating element in the furnace and the fixed spatial coordinates of the furnace wall lining, which are pre-stored in the control system, to construct the spatial relationship of radiation heat transfer.

3. The intelligent optimization method for industrial electric furnace operating parameters based on energy consumption modeling according to claim 1, characterized in that, The specific steps for forming the radiation angle coefficient distribution matrix of the radiative heat transfer path from the heating element to the workpiece surface and to the furnace wall lining surface include: The heating element, workpiece, and furnace wall lining surface in the spatial relationship of radiative heat transfer are discretized into several surface unit grids; For each discrete unit on the surface of the heating element, an energy beam is randomly emitted into space using the Monte Carlo method. The proportion of the energy beam reaching the discrete units on the workpiece and the inner lining of the furnace wall is statistically analyzed and used as the radiation angle coefficient of that pair of discrete units. The radiation angle coefficients of all discrete unit pairs are arranged in rows by the discrete units on the surface of the heating element and columns by the discrete units on the surface of the workpiece and the inner lining of the furnace wall, forming a radiation angle coefficient distribution matrix.

4. The intelligent optimization method for industrial electric furnace operating parameters based on energy consumption modeling according to claim 3, characterized in that, The specific steps for assigning weighting factors based on the difference between surface emissivity and surface reflectivity to the workpiece surface elements and the furnace wall lining surface elements when solving for the radiation angle coefficient include: To obtain the surface emissivity of the workpiece surface material and the surface reflectivity of the furnace wall lining material; The weighting factor for each discrete element on the workpiece surface is set to be positively correlated with its surface emissivity; The weighting factor for each discrete unit of the furnace wall lining surface is set to be negatively correlated with its surface reflectivity; When using the Monte Carlo method to solve for the radiation angle coefficient, the proportion of the energy beam reaching the corresponding discrete unit is multiplied by its weighting factor and included in the radiation angle coefficient statistics.

5. The intelligent optimization method for industrial electric furnace operating parameters based on energy consumption modeling according to claim 1, characterized in that, The specific steps for constructing an energy consumption characterization model using multivariate nonlinear regression constrained by sample entropy include: The heat conduction equation inside the furnace is established, and the radiation angle coefficient distribution matrix is ​​substituted into its radiation heat transfer term as the boundary conditions for the heating element, workpiece and furnace wall lining surface. Using the furnace gas temperature, heating element input power sequence and radiation angle coefficient distribution matrix from the historical operating data of the heat preservation stage as inputs, and the measured value of the heat storage state inside the furnace as outputs, a multivariate nonlinear regression model is constructed. A sample entropy regularization term is introduced into the loss function of the multivariate nonlinear regression model. The coefficient of the regularization term is determined by cross-validation to obtain a multivariate nonlinear regression model constrained by sample entropy, which is then identified as an energy consumption characterization model.

6. The intelligent optimization method for industrial electric furnace operating parameters based on energy consumption modeling according to claim 5, characterized in that, The steps for calculating sample entropy include: Obtain the time series of furnace gas temperature during the heat preservation stage; The phase space of the time series is reconstructed, and the embedding dimension and similarity tolerance are set. Calculate the number of vector pairs that satisfy the similarity tolerance condition to obtain the sample entropy value; The sample entropy value is used as the sample entropy regularization term and introduced into the loss function of the multivariate nonlinear regression model.

7. The intelligent optimization method for industrial electric furnace operating parameters based on energy consumption modeling according to claim 1, characterized in that, The specific steps for solving the optimization strategy for the operating parameters of the heating element include: The energy consumption characterization model is used as the prediction model in the model prediction control framework; Set constraints on the amplitude and rate of change of the input power sequence during the heat preservation stage, the amplitude constraint of temperature overshoot, and the temperature uniformity time constraint of the workpiece core; Within each control cycle, the input power sequence during the heat preservation stage is used as the decision variable. The energy consumption characterization model is called to predict the heat storage state inside the furnace and the furnace gas temperature response curve in the prediction time domain is calculated. The similarity between the response curve and the preset thermal equilibrium temperature curve is calculated using a dynamic time warping algorithm. The optimization iteration is terminated when the similarity reaches the thermal equilibrium threshold. The input power sequence that satisfies the constraints and reaches the optimization termination condition is output as the heating element operating parameter optimization strategy.

8. The intelligent optimization method for industrial electric furnace operating parameters based on energy consumption modeling according to claim 7, characterized in that, The specific steps for calculating similarity using the dynamic time warping algorithm include: Obtain the time series of the furnace gas temperature response curve and the preset thermal equilibrium temperature curve within the predicted time domain; Construct a distance matrix between the two time series, where each element represents the absolute difference between the data at the corresponding time points. Search the distance matrix for the normalized path with the smallest cumulative distance. This cumulative distance is the dynamic time normalized distance between the two time series. The dynamic time-warped distance is used as a similarity measure between the two.

9. The intelligent optimization method for industrial electric furnace operating parameters based on energy consumption modeling according to claim 1, characterized in that, The specific steps for regulating the input power of heating elements during the high-temperature holding stage of an industrial electric furnace include: The input power sequence of the heat preservation stage in the heating element operating parameter optimization strategy is converted into a control signal for the heating element power supply circuit. During each control cycle of the high-temperature heat preservation stage, the conduction time or amplitude of the power supply circuit of the heating element is adjusted according to the control signal. A model predictive control framework that monitors furnace gas temperature and actual input power of heating elements and feeds it back to the next control cycle.

10. An intelligent optimization system for industrial electric furnace operating parameters based on energy consumption modeling, employing the intelligent optimization method for industrial electric furnace operating parameters based on energy consumption modeling as described in any one of claims 1-9, characterized in that, include: The spatial relationship construction module is used to perform non-contact three-dimensional scanning of the workpiece in the furnace after loading, obtain the three-dimensional geometric shape and position data of the workpiece, and construct the radiative heat transfer spatial relationship between the surface of the heating element, the surface of the workpiece and the surface of the furnace wall lining based on the three-dimensional geometric shape and position data of the workpiece and the fixed spatial coordinates of the heating element in the furnace and the fixed spatial coordinates of the furnace wall lining. The angle coefficient matrix forming module is used to discretize the spatial relationship of radiation heat transfer. For each pair of discrete surface elements, the Monte Carlo method is used to solve the radiation angle coefficient, forming the radiation angle coefficient distribution matrix of the radiation heat transfer path from the heating element to the workpiece surface and the radiation heat transfer path from the heating element to the furnace wall lining surface. During the solution process, the workpiece surface elements and the furnace wall lining surface elements are assigned weight factors based on the difference between surface emissivity and surface reflectivity. The energy consumption model construction module is used to construct an energy consumption characterization model by using the radiation angle coefficient distribution matrix as the boundary condition of the radiation heat transfer term in the heat conduction equation and employing multivariate nonlinear regression constrained by sample entropy. The output of this energy consumption characterization model is the actual heat storage state inside the furnace caused by the shift in the radiation angle coefficient distribution. The optimization strategy solution module is used to embed the energy consumption characterization model into the model predictive control framework. Under the constraints of the input power sequence, temperature overshoot amplitude, and workpiece core temperature equalization time during the heat preservation stage, the module uses the dynamic time warping algorithm to match the thermal equilibrium threshold as the optimization termination condition to solve the optimization strategy for the heating element operating parameters. The power control module is used to regulate the input power of the heating element during the high-temperature heat preservation stage of the industrial electric furnace based on the optimization strategy of the heating element operating parameters.

Images