A constant temperature metal bath nonlinear temperature calibration method and system fusing machine learning
Patent Information
- Application Number
- CN202610986085.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-07-03
- Publication Date
- 2026-09-29
AI Technical Summary
这种受多工况参数联合驱动的非线性温漂行为，使得传统校准方法难以建立起能够覆盖全量程全工况的准确补偿关系
本发明公开了一种融合机器学习的恒温金属浴非线性温度校准方法，通过提取设定温度、环境温度、升温速率和降温速率等多维工况特征，利用随机森林算法筛选出对温度偏差影响最显著的核心特征，并结合实际温度偏差构建训练样本集，经过异常值剔除处理后，采用支持向量回归算法建立多维工况与温度偏差之间的非线性映射模型。在实际运行过程中，本发明根据环境温度是否超出阈值来决定是否对实时工况特征进行归一化处理，确保输入特征的合理性，再通过已建立的非线性映射模型预测当前工况下的温度偏差，生成精确的补偿数值并实施动态校准。该方法有效解决了传统线性校准方法难以应对复杂工况变化的局限性，显著提升了恒温金属浴在全工况范围内的温度控制精度和稳定性。
Smart Images
Figure CN122835595A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of information technology, and in particular to a nonlinear temperature calibration method and system for isothermal metal baths that integrates machine learning. Background Technology
[0002] As a core piece of equipment in the field of temperature metrology and calibration, the temperature accuracy of constant-temperature metal baths directly affects the reliability of industrial production process control and scientific research experiments. In critical industries such as petrochemicals, pharmaceuticals, and food testing, even a temperature deviation of only a few tenths of a degree can lead to product quality control failure or distorted experimental results. Therefore, the precise calibration of constant-temperature metal baths is a fundamental step in ensuring the accuracy of temperature measurement transmission.
[0003] However, existing calibration methods generally employ fixed compensation values or piecewise linear corrections, assuming that the temperature deviation remains constant or changes linearly within a specific range. In actual operation, the temperature deviation of the metal bath is affected by a combination of factors, including the set temperature, ambient temperature fluctuations, and differences in heating and cooling rates. These operating parameters are not independent but rather interact with each other to influence temperature stability. When the metal bath rapidly rises from room temperature to its high operating point, the dynamic changes in heating power, heat conduction rate, and heat dissipation equilibrium point trigger nonlinear drift in the temperature response. This drift characteristic exhibits significant differences across different temperature ranges and heating rates.
[0004] More complexly, the nonlinear characteristics of temperature drift cannot be accurately described by simple mathematical functions. For example, in heating a metal bath from room temperature to 150 degrees Celsius, if a rapid heating mode is used, the equipment may exhibit positive overshoot during the initial stabilization phase after reaching the set temperature, followed by negative drift during the establishment of thermal equilibrium. Conversely, the same set temperature may exhibit a completely different drift trajectory under a slow heating mode. This nonlinear temperature drift behavior, driven by multiple operating parameters, makes it difficult for traditional calibration methods to establish accurate compensation relationships that cover the entire range and all operating conditions.
[0005] Therefore, how to extract the complex nonlinear mapping law between multi-dimensional operating parameters such as set temperature, environmental conditions, heating and cooling rates and actual temperature deviation from the massive calibration data of the metal bath's historical operation, and generate accurate temperature compensation values in real time under different operating scenarios, has become the key issue to break through the bottleneck of the calibration accuracy of constant temperature metal baths. Summary of the Invention
[0006] The purpose of this invention is to address the above-mentioned problems by proposing a nonlinear temperature calibration method and system for isothermal metal baths that integrates machine learning.
[0007] The technical solution of this invention is: This invention provides a nonlinear temperature calibration method for isothermal metal baths that integrates machine learning, the method comprising: S1. Obtain historical operating data and extract the set temperature, ambient temperature, heating rate and cooling rate as the first operating condition feature set. S2. Based on the first working condition feature set, the random forest algorithm is used to evaluate the feature weights to obtain the second working condition feature set containing the core features. S3. Obtain the actual temperature deviation corresponding to the feature set of the second working condition, construct an initial training sample set containing features and deviations, and perform outlier removal processing on the initial training sample set to obtain a standard training sample set. S4. By performing nonlinear fitting on the standard training sample set using the support vector regression algorithm, a nonlinear mapping model between multidimensional operating conditions and temperature deviation is obtained. S5. Obtain the set temperature, ambient temperature, and heating rate under the current real-time operating status to form a real-time operating condition feature vector; S6. If the ambient temperature in the real-time operating condition feature vector exceeds the preset ambient temperature threshold, the real-time operating condition feature vector is normalized to obtain the standard operating condition feature vector. S7. If the ambient temperature in the real-time operating condition feature vector does not exceed the preset ambient temperature threshold, then the real-time operating condition feature vector is directly determined as the standard operating condition feature vector. S8. Process the characteristic vector of the standard operating condition through a nonlinear mapping model to determine the predicted value of the target temperature deviation under the current operating condition; S9. Generate the corresponding compensation value based on the predicted value of the target temperature deviation to obtain the final compensation value of the constant temperature metal bath.
[0008] Furthermore, S1 includes: Historical operation data is acquired, and a sliding window algorithm is used to extract continuous operation cycles from the historical operation data to obtain a time series containing timestamps. Extract the set temperature and ambient temperature from the time series, calculate the temperature difference between the set temperature and the ambient temperature, and obtain a temperature difference sequence. The temperature change gradient is calculated using a differential algorithm based on the temperature difference sequence and the continuous operating cycle to obtain the heating rate and cooling rate. Determine whether the heating rate is greater than a preset threshold. If the heating rate is greater than the preset threshold, then extract features from the cooling rate and determine the set temperature, the ambient temperature, the heating rate, and the cooling rate as a first operating condition feature set.
[0009] Furthermore, S2 includes: An initial classification model was obtained by sampling the feature set of the first working condition using the random forest algorithm. The change in the Gini index is calculated to obtain the feature weight values, which are derived from the initial classification model. The feature weight sequence is determined by sorting the feature weight values. The cumulative weight value is obtained by summing the feature weight sequence. If the cumulative weight value is greater than the truncation threshold, then the features before the truncation threshold are extracted to obtain a second set of working condition features containing the core features.
[0010] Furthermore, S3 includes: Obtain an initial training sample set containing the feature set of the second operating condition and the actual temperature deviation; An anomaly mapping matrix is generated for the initial training sample set, which is obtained by calculating the anomaly score value for each sample; If the abnormal score value in the deviation mapping matrix is greater than the preset rejection threshold, the corresponding sample is extracted and stored in the sample cleaning queue. The samples in the sample cleaning queue are removed from the initial training sample set to remove outliers, thus obtaining the standard training sample set.
[0011] Furthermore, S4 includes: Obtain an initial running dataset, which contains multi-dimensional operating condition features and temperature deviation values. Perform anomaly removal on the initial running dataset to obtain a standard training sample set. Obtain a dimensionality-reduced multi-dimensional operating condition feature vector based on the standard training sample set. The initial mapping relationship is obtained by fitting the reduced multidimensional working condition feature vectors using a support vector regression algorithm. If the error of the initial mapping relationship is greater than the threshold, the initial mapping relationship is refitted to obtain a nonlinear mapping model of multidimensional working conditions and temperature deviation.
[0012] Furthermore, S5 includes: The difference between the first set temperature and the first ambient temperature is calculated to obtain the first temperature deviation. Based on the first temperature deviation, the first heating rate is obtained and the first thermodynamic feature is extracted. The first thermodynamic feature is spatially mapped to determine the first state matrix, and the first state matrix is dimensionality reduced to obtain the first fused feature. If the dimension of the first fused feature is greater than a preset threshold, the first fused feature is truncated to obtain the second fused feature, and a real-time operating condition feature vector is formed based on the second fused feature.
[0013] Furthermore, S8 includes: Acquire operational status data and reduce the dimensionality of the operational status data to obtain an initial operating condition feature vector; The initial operating condition feature vector is normalized to obtain the standard operating condition feature vector; The feature mapping matrix is obtained by processing the feature vectors of the standard operating condition using a nonlinear mapping model. The deviation compensation value is obtained by operating the temperature fluctuation sequence using the feature mapping matrix, and the deviation compensation value is used to add to the target temperature reference. The target temperature deviation prediction value under the current operating state is determined based on the summation result.
[0014] Furthermore, S9 includes: The initial temperature deviation value is calculated by acquiring real-time and set temperature data of the constant temperature metal bath, and the target temperature deviation prediction value is derived based on the initial temperature deviation value. The basic compensation value is calculated based on the predicted value of the target temperature deviation; The basic compensation value that is greater than a preset threshold is filtered to obtain a smooth compensation value. The corresponding compensation value is generated based on the smooth compensation value to obtain the final compensation value of the constant temperature metal bath.
[0015] This invention provides a nonlinear temperature calibration system for a constant-temperature metal bath that integrates machine learning, the system comprising: The first operating condition feature set acquisition module is used to acquire historical operating data and extract the set temperature, ambient temperature, heating rate and cooling rate as the first operating condition feature set. The second working condition feature set acquisition module is used to evaluate the feature weights using the random forest algorithm based on the first working condition feature set, and obtain a second working condition feature set containing core features. The standard training sample set construction module is used to obtain the actual temperature deviation corresponding to the feature set of the second working condition, construct an initial training sample set containing features and deviations, and perform outlier removal processing on the initial training sample set to obtain the standard training sample set. The nonlinear mapping model training module is used to perform nonlinear fitting on the standard training sample set using the support vector regression algorithm to obtain a nonlinear mapping model between multidimensional operating conditions and temperature deviation. The real-time operating condition feature vector acquisition module is used to acquire the set temperature, ambient temperature and heating rate under the current real-time operating status, and form a real-time operating condition feature vector. The first judgment module is used to normalize the real-time operating condition feature vector to obtain a standard operating condition feature vector if the ambient temperature in the real-time operating condition feature vector exceeds the preset ambient temperature threshold. The second judgment module is used to directly determine the real-time operating condition feature vector as the standard operating condition feature vector if the ambient temperature in the real-time operating condition feature vector does not exceed the preset ambient temperature threshold. The target temperature deviation prediction module is used to process the feature vector of standard operating conditions through a nonlinear mapping model to determine the predicted value of the target temperature deviation under the current operating state. The final compensation value acquisition module is used to generate the corresponding compensation value based on the predicted value of the target temperature deviation, and obtain the final compensation value of the constant temperature metal bath.
[0016] The present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed, implements the calibration method.
[0017] The beneficial effects of this invention are: This invention discloses a nonlinear temperature calibration method for isothermal metal baths that integrates machine learning. By extracting multi-dimensional operating condition features such as set temperature, ambient temperature, heating rate, and cooling rate, a random forest algorithm is used to select the core features that have the most significant impact on temperature deviation. A training sample set is constructed based on the actual temperature deviation. After outlier removal, a support vector regression algorithm is used to establish a nonlinear mapping model between the multi-dimensional operating conditions and the temperature deviation. During actual operation, this invention determines whether to normalize the real-time operating condition features based on whether the ambient temperature exceeds a threshold, ensuring the rationality of the input features. Then, the established nonlinear mapping model is used to predict the temperature deviation under the current operating conditions, generating accurate compensation values and implementing dynamic calibration. This method effectively solves the limitation of traditional linear calibration methods in handling complex operating condition changes, significantly improving the temperature control accuracy and stability of the isothermal metal bath across the entire operating range.
[0018] Other features and advantages of the present invention will be described in detail in the following detailed description section. Attached Figure Description
[0019] The above and other objects, features and advantages of the present invention will become more apparent from the more detailed description of exemplary embodiments of the invention in conjunction with the accompanying drawings, wherein the same reference numerals generally represent the same components in the exemplary embodiments of the invention.
[0020] Figure 1 This is a flowchart of the isothermal metal bath nonlinear temperature calibration method integrating machine learning according to the present invention.
[0021] Figure 2 This is a schematic diagram of the nonlinear temperature calibration system for a constant-temperature metal bath that integrates machine learning, as described in this invention. Detailed Implementation
[0022] Preferred embodiments of the invention will now be described in more detail with reference to the accompanying drawings. While preferred embodiments of the invention are shown in the drawings, it should be understood that the invention can be implemented in various forms and should not be limited to the embodiments set forth herein.
[0023] like Figure 1 As shown, this embodiment of the invention provides a nonlinear temperature calibration method for a constant-temperature metal bath that integrates machine learning. The method includes: Step S1: Obtain historical operating data and extract the set temperature, ambient temperature, heating rate, and cooling rate as the first operating condition feature set.
[0024] Specifically, historical operation records are read from the data storage unit of the constant-temperature metal bath. These records include timestamps, temperature sensor values, and control command sequences. The historical operation records are segmented chronologically, with each segment corresponding to a complete heating or cooling process. The set temperature value is extracted from each segment as the target temperature parameter. The ambient temperature sensor value is read from the start time of each segment and used as the ambient temperature parameter. Simultaneously, the ratio of temperature change to time interval within the segment is calculated to obtain the heating rate and cooling rate. The extracted set temperature, ambient temperature, heating rate, and cooling rate are combined into a feature vector. The feature vectors of all segments are then aggregated to form the first operating condition feature set.
[0025] Step S2: Based on the first set of working conditions features, use the random forest algorithm to evaluate the feature weights and obtain the second set of working conditions features containing the core features.
[0026] Specifically, each sample in the first working condition feature set is randomly selected, with the selection ratio set to 63%. The selected samples are used to construct a decision tree, and the unselected samples are used as out-of-bag data for verification.
[0027] When splitting at each decision tree node for the extracted samples, a subset of features is randomly selected from all features in the first working condition feature set. The number of feature subsets is set to the square root of the total number of features, rounded down. For each feature in the feature subset, the change in Gini index before and after the split is calculated. The change in Gini index is obtained by calculating the difference between the Gini impurity of the node before the split and the weighted Gini impurity of the child nodes after the split. Gini impurity is defined as one minus the sum of the squares of the proportions of samples in each category.
[0028] Select the feature with the largest change in Gini index as the splitting feature of the current node, and divide the sample into left and right child nodes according to the threshold of this feature. Repeat this process until the preset tree depth is reached or the number of node samples is lower than the threshold.
[0029] Multiple decision trees are constructed to form a random forest. The change in Gini index for each feature in each decision tree is summed, and the summation results of all decision trees are averaged to obtain the average change in Gini index for each feature as the feature weight value.
[0030] Sort all features in the first working condition feature set in descending order of feature weight value to form a feature weight sequence. Starting from the feature with the largest weight, add the feature weight values one by one to obtain the cumulative weight value.
[0031] The cumulative weight truncation threshold is set to 90% of the total weight. When the cumulative weight value exceeds the truncation threshold for the first time, all features before the truncation threshold are extracted as core features, and the core features constitute the second working condition feature set.
[0032] Step S3: Obtain the actual temperature deviation corresponding to the feature set of the second working condition, construct an initial training sample set containing features and deviations, and perform outlier removal processing on the initial training sample set to obtain a standard training sample set.
[0033] Specifically, the standard thermometer measurement value and the metal bath display temperature value corresponding to the second operating condition feature set are read from historical operating data, and the difference between the two is calculated to obtain the actual temperature deviation. Each feature vector in the second operating condition feature set is combined with the corresponding actual temperature deviation to form a sample pair containing input features and output labels. All sample pairs are summarized to form the initial training sample set. The absolute value of the difference between the actual temperature deviation of each sample in the initial training sample set and the mean of the temperature deviation of all samples is calculated. This absolute value is divided by the standard deviation of the temperature deviation of all samples to obtain the abnormal score value of the sample. A removal threshold of three standard deviations is set. Samples with abnormal scores greater than the removal threshold are marked as abnormal samples. All abnormal samples are removed from the initial training sample set, and the remaining samples constitute the standard training sample set.
[0034] Step S4: Nonlinearly fit the standard training sample set using the support vector regression algorithm to obtain a nonlinear mapping model of multidimensional operating conditions and temperature deviation.
[0035] Specifically, the standard training sample set is divided into a training subset and a validation subset in an 8:2 ratio. The training subset is used for model parameter learning, and the validation subset is used for model performance evaluation. A radial basis function (RBF) is chosen as the kernel function for support vector regression, mapping the original feature space to a higher-dimensional space. The penalty coefficient and kernel width parameters are initialized; the penalty coefficient balances model complexity and training error, and the kernel width controls the range of the RBF, with initial values set to 1 and 0.1, respectively. A sequential minimum optimization algorithm is used to solve the dual problem of support vector regression, finding the set of support vectors that minimizes the training error by iteratively updating the Lagrange multipliers.
[0036] The root mean square error (RMSE) of the deviation between the model's predicted values and the actual temperature is calculated on the validation subset. If the RMSE is greater than the preset threshold of 0.2 degrees Celsius, the penalty coefficient and kernel width parameters are adjusted by grid search, the support vector regression model is retrained, and the RMSE of the validation subset is calculated. The parameter combination that minimizes the RMSE is selected.
[0037] The support vector regression model is retrained on the entire standard training sample set using the optimal parameter combination. The resulting model parameters include the set of support vectors, the corresponding Lagrange multipliers, and the bias term. These model parameters together determine the nonlinear mapping model between multidimensional operating conditions and temperature deviation.
[0038] Step S5: Obtain the set temperature, ambient temperature, and heating rate under the current real-time operating status to form a real-time operating condition feature vector.
[0039] Specifically, the current set temperature command value is read from the metal bath controller, the current ambient temperature measurement value is read from the ambient temperature sensor, and the current heating rate is obtained by calculating the ratio of the temperature change in the last ten seconds to the time interval.
[0040] The current set temperature, ambient temperature, and heating rate are arranged in the same feature order as the second operating condition feature set to form a real-time operating condition feature vector.
[0041] Step S6: If the ambient temperature in the real-time operating condition feature vector exceeds the preset ambient temperature threshold, the real-time operating condition feature vector is normalized to obtain the standard operating condition feature vector.
[0042] Step S7: If the ambient temperature in the real-time operating condition feature vector does not exceed the preset ambient temperature threshold, then the real-time operating condition feature vector is directly determined as the standard operating condition feature vector.
[0043] Step S8: Process the standard operating condition feature vector using a nonlinear mapping model to determine the target temperature deviation prediction value under the current operating state.
[0044] Specifically, the standard operating condition feature vector is input into the nonlinear mapping model, and the radial basis function kernel value between the standard operating condition feature vector and each support vector is calculated. Each kernel value is multiplied by the Lagrange multiplier of the corresponding support vector, and the sum of all products is added to the model's bias term to obtain the predicted value of the target temperature deviation.
[0045] Step S9: Generate the corresponding compensation value based on the predicted value of the target temperature deviation to obtain the final compensation value of the constant temperature metal bath.
[0046] Specifically, the predicted value of the target temperature deviation is set to a negative value as the initial compensation value. If the predicted value of the target temperature deviation is positive, it means that the actual temperature is higher than the set temperature. In this case, the compensation value is negative and used to reduce the control temperature.
[0047] The initial compensation value is processed by moving average filtering, and the average of the five most recent compensation values is taken as the smooth compensation value to avoid the temperature control instability caused by drastic fluctuations in the compensation value.
[0048] The smoothing compensation value is superimposed on the current set temperature to obtain the final compensation value of the constant temperature metal bath. The final compensation value is then output to the temperature controller to adjust the heating power.
[0049] like Figure 2 As shown, this embodiment of the invention also provides a nonlinear temperature calibration system for a constant-temperature metal bath that integrates machine learning. The system includes modules that correspond one-to-one with the steps of the above method, used to implement all or part of the process of the method. Each module can be implemented by hardware, software, or a combination of both. The system specifically includes: The first operating condition feature set acquisition module is used to acquire historical operating data and extract the set temperature, ambient temperature, heating rate and cooling rate as the first operating condition feature set. The second working condition feature set acquisition module is used to evaluate the feature weights using the random forest algorithm based on the first working condition feature set, and obtain a second working condition feature set containing core features. The standard training sample set construction module is used to obtain the actual temperature deviation corresponding to the feature set of the second working condition, construct an initial training sample set containing features and deviations, and perform outlier removal processing on the initial training sample set to obtain the standard training sample set. The nonlinear mapping model training module is used to perform nonlinear fitting on the standard training sample set using the support vector regression algorithm to obtain a nonlinear mapping model between multidimensional operating conditions and temperature deviation. The real-time operating condition feature vector acquisition module is used to acquire the set temperature, ambient temperature and heating rate under the current real-time operating status, and form a real-time operating condition feature vector. The first judgment module is used to normalize the real-time operating condition feature vector to obtain a standard operating condition feature vector if the ambient temperature in the real-time operating condition feature vector exceeds the preset ambient temperature threshold. The second judgment module is used to directly determine the real-time operating condition feature vector as the standard operating condition feature vector if the ambient temperature in the real-time operating condition feature vector does not exceed the preset ambient temperature threshold. The target temperature deviation prediction module is used to process the feature vector of standard operating conditions through a nonlinear mapping model to determine the predicted value of the target temperature deviation under the current operating state. The final compensation value acquisition module is used to generate the corresponding compensation value based on the predicted value of the target temperature deviation, and obtain the final compensation value of the constant temperature metal bath.
[0050] The various embodiments of the present invention have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments.
Claims
1. A nonlinear temperature calibration method for a isothermal metal bath integrating machine learning, characterized in that, The method includes: S1. Obtain historical operating data and extract the set temperature, ambient temperature, heating rate and cooling rate as the first operating condition feature set. S2. Based on the first working condition feature set, the random forest algorithm is used to evaluate the feature weights to obtain the second working condition feature set containing the core features. S3. Obtain the actual temperature deviation corresponding to the feature set of the second working condition, construct an initial training sample set containing features and deviations, and perform outlier removal processing on the initial training sample set to obtain a standard training sample set. S4. By performing nonlinear fitting on the standard training sample set using the support vector regression algorithm, a nonlinear mapping model between multidimensional operating conditions and temperature deviation is obtained. S5. Obtain the set temperature, ambient temperature, and heating rate under the current real-time operating status to form a real-time operating condition feature vector; S6. If the ambient temperature in the real-time operating condition feature vector exceeds the preset ambient temperature threshold, the real-time operating condition feature vector is normalized to obtain the standard operating condition feature vector. S7. If the ambient temperature in the real-time operating condition feature vector does not exceed the preset ambient temperature threshold, then the real-time operating condition feature vector is directly determined as the standard operating condition feature vector. S8. Process the characteristic vector of the standard operating condition through a nonlinear mapping model to determine the predicted value of the target temperature deviation under the current operating condition; S9. Generate the corresponding compensation value based on the predicted value of the target temperature deviation to obtain the final compensation value of the constant temperature metal bath.
2. The isothermal metal bath nonlinear temperature calibration method integrating machine learning according to claim 1, characterized in that... S1 includes: Historical operation data is acquired, and a sliding window algorithm is used to extract continuous operation cycles from the historical operation data to obtain a time series containing timestamps. Extract the set temperature and ambient temperature from the time series, calculate the temperature difference between the set temperature and the ambient temperature, and obtain a temperature difference sequence. The temperature change gradient is calculated using a differential algorithm based on the temperature difference sequence and the continuous operating cycle to obtain the heating rate and cooling rate. Determine whether the heating rate is greater than a preset threshold. If the heating rate is greater than the preset threshold, then extract features from the cooling rate and determine the set temperature, the ambient temperature, the heating rate, and the cooling rate as a first operating condition feature set.
3. The isothermal metal bath nonlinear temperature calibration method integrating machine learning according to claim 1, characterized in that, S2 include: An initial classification model was obtained by sampling the feature set of the first working condition using the random forest algorithm. The change in the Gini index is calculated to obtain the feature weight values, which are derived from the initial classification model. The feature weight sequence is determined by sorting the feature weight values. The cumulative weight value is obtained by summing the feature weight sequence. If the cumulative weight value is greater than the truncation threshold, then the features before the truncation threshold are extracted to obtain a second set of working condition features containing the core features.
4. The isothermal metal bath nonlinear temperature calibration method integrating machine learning according to claim 1, characterized in that, S3 include: Obtain an initial training sample set containing the feature set of the second operating condition and the actual temperature deviation; An anomaly mapping matrix is generated for the initial training sample set, which is obtained by calculating the anomaly score value for each sample; If the abnormal score value in the deviation mapping matrix is greater than the preset rejection threshold, the corresponding sample is extracted and stored in the sample cleaning queue. The samples in the sample cleaning queue are removed from the initial training sample set to remove outliers, thus obtaining the standard training sample set.
5. The isothermal metal bath nonlinear temperature calibration method incorporating machine learning according to claim 1, characterized in that, S4 includes: Obtain an initial running dataset, which contains multi-dimensional operating condition features and temperature deviation values. Perform anomaly removal on the initial running dataset to obtain a standard training sample set. Obtain a dimensionality-reduced multi-dimensional operating condition feature vector based on the standard training sample set. The initial mapping relationship is obtained by fitting the reduced multidimensional working condition feature vectors using a support vector regression algorithm. If the error of the initial mapping relationship is greater than the threshold, the initial mapping relationship is refitted to obtain a nonlinear mapping model of multidimensional working conditions and temperature deviation.
6. The isothermal metal bath nonlinear temperature calibration method integrating machine learning according to claim 1, characterized in that, S5 include: The difference between the first set temperature and the first ambient temperature is calculated to obtain the first temperature deviation. Based on the first temperature deviation, the first heating rate is obtained and the first thermodynamic feature is extracted. The first thermodynamic feature is spatially mapped to determine the first state matrix, and the first state matrix is dimensionality reduced to obtain the first fused feature. If the dimension of the first fused feature is greater than a preset threshold, the first fused feature is truncated to obtain the second fused feature, and a real-time operating condition feature vector is formed based on the second fused feature.
7. The isothermal metal bath nonlinear temperature calibration method incorporating machine learning according to claim 1, characterized in that, S8 includes: Acquire operational status data and reduce the dimensionality of the operational status data to obtain an initial operating condition feature vector; The initial operating condition feature vector is normalized to obtain the standard operating condition feature vector; The feature mapping matrix is obtained by processing the feature vectors of the standard operating condition using a nonlinear mapping model. The deviation compensation value is obtained by operating the temperature fluctuation sequence using the feature mapping matrix, and the deviation compensation value is used to add to the target temperature reference. The target temperature deviation prediction value under the current operating state is determined based on the summation result.
8. The isothermal metal bath nonlinear temperature calibration method incorporating machine learning according to claim 1, characterized in that, S9 includes: The initial temperature deviation value is calculated by acquiring real-time and set temperature data of the constant temperature metal bath, and the target temperature deviation prediction value is derived based on the initial temperature deviation value. The basic compensation value is calculated based on the predicted value of the target temperature deviation; The basic compensation value that is greater than a preset threshold is filtered to obtain a smooth compensation value. The corresponding compensation value is generated based on the smooth compensation value to obtain the final compensation value of the constant temperature metal bath.
9. A nonlinear temperature calibration system for a constant-temperature metal bath integrating machine learning, characterized in that, The system includes: The first operating condition feature set acquisition module is used to acquire historical operating data and extract the set temperature, ambient temperature, heating rate and cooling rate as the first operating condition feature set. The second working condition feature set acquisition module is used to evaluate the feature weights using the random forest algorithm based on the first working condition feature set, and obtain a second working condition feature set containing core features. The standard training sample set construction module is used to obtain the actual temperature deviation corresponding to the feature set of the second working condition, construct an initial training sample set containing features and deviations, and perform outlier removal processing on the initial training sample set to obtain the standard training sample set. The nonlinear mapping model training module is used to perform nonlinear fitting on the standard training sample set using the support vector regression algorithm to obtain a nonlinear mapping model between multidimensional operating conditions and temperature deviation. The real-time operating condition feature vector acquisition module is used to acquire the set temperature, ambient temperature and heating rate under the current real-time operating status, and form a real-time operating condition feature vector. The first judgment module is used to normalize the real-time operating condition feature vector to obtain a standard operating condition feature vector if the ambient temperature in the real-time operating condition feature vector exceeds the preset ambient temperature threshold. The second judgment module is used to directly determine the real-time operating condition feature vector as the standard operating condition feature vector if the ambient temperature in the real-time operating condition feature vector does not exceed the preset ambient temperature threshold. The target temperature deviation prediction module is used to process the feature vector of standard operating conditions through a nonlinear mapping model to determine the predicted value of the target temperature deviation under the current operating state. The final compensation value acquisition module is used to generate the corresponding compensation value based on the predicted value of the target temperature deviation, and obtain the final compensation value of the constant temperature metal bath.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that... When the program is executed, the calibration method as described in any one of claims 1-8 is implemented.