Kiln temperature automatic control method, system and medium based on deep learning

Abstract

The present application relates to the technical field of kiln temperature automatic control, and discloses a kiln temperature automatic control method, system and medium based on deep learning, which comprises the following steps: acquiring temperature time sequence data, process stage identification sequence and finished product magnetic property data of ferrite material, and obtaining a standardized historical data set through preprocessing; training a deep learning model based on the data set, using the model to divide the current sintering temperature section and calculate the influence weight of each section on the magnetic property deviation prediction value; identifying the high sensitive temperature section according to the weight, generating corresponding temperature control constraint parameters for different sensitive level sections; and dynamically executing temperature control according to the constraint parameters called according to the current section in real-time sintering. The present application does not need to change the kiln hardware, can accurately control the temperature disturbance of the sensitive temperature zone, improves the magnetic property consistency and batch stability of the ferrite material, and is suitable for ferrite sintering industrial production scenes.

Description

Technical Field

[0001] This invention relates to the field of automatic kiln temperature control technology, specifically to a method, system, and medium for automatic kiln temperature control based on deep learning. Background Technology

[0002] In the industrial production of ferrite magnetic materials such as magnetic cores and rings, the sintering process is a crucial step determining the magnetic properties of the product. During ferrite sintering, the 800–1100℃ range is a highly sensitive region for oxygen partial pressure, within which the valence state equilibrium of iron (Fe) is crucial. 2+ / Fe 3+ The ratio directly affects the stability of the material's crystal structure, magnetic domain behavior, and final magnetic properties, specifically manifested in the stability of core parameters such as permeability, loss tangent, and Curie temperature.

[0003] In industrial production, kilns can only detect and control temperature parameters, but cannot accurately detect and regulate the oxygen partial pressure inside the furnace in real time. Within this sensitive temperature range, even small fluctuations in the heating rate or differences in the amplitude of temperature fluctuations can alter local oxygen reaction kinetics and diffusion behavior, leading to an imbalance in the distribution of iron ion valence states. This ultimately results in problems such as deviations in product magnetic properties, poor batch-to-batch consistency, and large performance dispersion.

[0004] Existing temperature control methods employ a uniform control strategy, failing to differentiate the varying impacts of different temperature ranges on magnetic properties. They focus solely on minimizing temperature error, making it difficult to adapt to the nonlinear characteristics of ferrite sintering. Traditional control models and process experience cannot quantify the complex effects of temperature history on magnetic properties, nor can they specifically suppress temperature disturbances in sensitive temperature zones, thus failing to meet the high requirements for magnetic property stability of ferrite materials in industrial production. Summary of the Invention

[0005] The purpose of this invention is to provide a method, system, and medium for automatic control of kiln temperature based on deep learning, so as to solve the problems mentioned in the background art.

[0006] According to one aspect of this application, a method for automatic temperature control of a kiln based on deep learning is provided, comprising the following steps: Historical production data is acquired and preprocessed to obtain a standardized historical dataset; the historical production data includes temperature time series data of ferrite materials, process stage identification sequences, and corresponding finished product magnetic property data; Based on the trained deep learning model, the complete temperature-time history of the current sintering process is divided into temperature segments, and the influence weight of each temperature segment on the predicted value of magnetic performance deviation is calculated. The deep learning model is trained based on the standardized historical dataset. The input of the deep learning model is the temperature time series data and the process stage identifier sequence, and the output is the corresponding predicted value of magnetic performance deviation. Based on the influence weight, high-sensitivity temperature segments are identified from all the divided temperature segments, and corresponding temperature control constraint parameters are generated for the high-sensitivity temperature segments and low-sensitivity temperature segments respectively. During real-time sintering, the corresponding temperature control constraint parameters are invoked according to the current temperature range to dynamically execute temperature control.

[0007] Preferably, the deep learning model includes a feature extraction layer and an attention layer; the feature extraction layer is a long short-term memory network layer, used to extract the temporal features of the temperature time series data; the attention layer is used to assign weights to the temporal features.

[0008] Preferably, the influence weight of each temperature range on the predicted value of the magnetic performance deviation is calculated as follows: The complete temperature time series containing the temperature segment division information is input into the trained deep learning model to obtain the first magnetic property deviation prediction value. The temperature data within the selected temperature range is randomly perturbed, and the perturbed temperature time series is input into the deep learning model to obtain the second magnetic property deviation prediction value. Based on the difference between the first and second predicted magnetic property deviation values, the initial influence weight of the selected temperature range is calculated. Repeat the random perturbation operation at least once, and take the average value of the initial influence weights obtained from multiple calculations as the final influence weight of the selected temperature range.

[0009] Preferably, highly sensitive temperature zones are identified from all the divided temperature zones, specifically: Calculate the statistical characteristic values ​​of the influence weights for all temperature ranges; Temperature ranges whose influence weight is greater than or equal to the judgment threshold set based on the statistical feature value are marked as the high-sensitivity temperature ranges.

[0010] Preferably, the temperature control constraint parameters generated for the high-sensitivity temperature range include an allowable temperature change rate threshold and an allowable temperature fluctuation amplitude threshold; the allowable temperature change rate threshold and the allowable temperature fluctuation amplitude threshold are determined based on statistical analysis of actual temperature data of historical stable production batches within the corresponding high-sensitivity temperature range.

[0011] Preferably, when the temperature is determined to be in the high-sensitivity temperature range at the current moment, dynamic temperature control is performed, specifically including: Obtain the current real-time temperature value, the preset target temperature value, and the temperature control constraint parameters corresponding to the high-sensitivity temperature range; Based on current and historical temperature data, predict the temperature change trend for a preset duration in the future; Construct and solve a constrained optimization problem with the objective of minimizing temperature tracking deviation and satisfying the allowable temperature change rate threshold and the allowable temperature fluctuation amplitude threshold, to obtain the heating power adjustment amount at the current moment; Temperature control is performed based on the heating power adjustment amount.

[0012] In another aspect, this application also provides a deep learning-based automatic kiln temperature control system, comprising: The preprocessing module is used to acquire historical production data and preprocess the historical production data to obtain a standardized historical dataset. The historical production data includes temperature time series data of ferrite materials, process stage identification sequences, and corresponding finished product magnetic property data. The weight calculation module is used to divide the complete temperature time history of the current sintering process into temperature segments based on the trained deep learning model, and calculate the influence weight of each temperature segment on the predicted value of magnetic property deviation. The deep learning model is trained based on the standardized historical dataset. The input of the deep learning model is the temperature time series data and the process stage identification sequence, and the output is the corresponding predicted value of magnetic property deviation. The parameter generation module is used to identify the high-sensitivity temperature segment from all the divided temperature segments according to the influence weight, and generate corresponding temperature control constraint parameters for the high-sensitivity temperature segment and the low-sensitivity temperature segment respectively. The temperature control module is used to dynamically execute temperature control during real-time sintering by calling the corresponding temperature control constraint parameters according to the current temperature range.

[0013] This application also provides a computer device comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the deep learning-based automatic kiln temperature control method as described above.

[0014] In another aspect, this application provides a computer-readable storage medium having stored thereon computer program instructions that can be executed by a processor to implement the deep learning-based automatic kiln temperature control method described above.

[0015] Another aspect of this application provides a computer program product, including a computer program that, when executed by a processor, implements the deep learning-based automatic kiln temperature control method described above.

[0016] This application utilizes a deep learning model to accurately identify highly sensitive temperature ranges, sets strict temperature control constraints for these ranges, and employs a model predictive control algorithm to achieve refined temperature control. Simultaneously, conventional PID control is used for low-sensitivity ranges to balance efficiency, significantly improving temperature stability within the oxygen-sensitive temperature range and effectively ensuring the valence equilibrium of the ferrite material. Combined with a data update mechanism, the model prediction accuracy and control parameter adaptability are continuously optimized, ultimately achieving a significant improvement in the consistency and batch stability of the magnetic properties of the ferrite material. This meets the high requirements of industrial production for magnetic property stability and demonstrates good practicality and economic efficiency. Attached Figure Description

[0017] Figure 1 A schematic diagram of a deep learning-based automatic kiln temperature control method provided in an embodiment of this disclosure; Figure 2 This is a schematic diagram of the deep learning model structure provided in the embodiments of this disclosure; Figure 3 This is a schematic diagram of the influence weight calculation process provided in the embodiments of this disclosure; Figure 4 This is a schematic diagram of the high-sensitivity temperature range identification process provided in the embodiments of this disclosure; Figure 5 This is a schematic diagram of the dynamic control flow provided in the embodiments of this disclosure; Figure 6 A schematic diagram of a kiln temperature automatic control system based on deep learning provided in an embodiment of this disclosure; Figure 7 This is a schematic diagram of the structure of a computer device provided in an embodiment of this disclosure. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] It should be noted that all user information (including but not limited to user device information, user personal information, object information corresponding to device usage data, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, device usage data, etc.) involved in all embodiments of this disclosure are information and data authorized by the user or fully authorized by all parties.

[0020] This method is applicable to industrial kilns for preparing magnetic cores, magnetic rings, and other ferrite magnetic materials, including but not limited to pusher kilns, roller kilns, or intermittent sintering furnaces. It is particularly suitable for industrial production scenarios of ferrite materials where batch-to-batch stability of magnetic properties is critical. Implementing this method requires no changes to the existing kiln hardware structure, only relying on the existing control system. This control system may include an industrial control computing unit, a temperature acquisition module, and a power adjustment unit for performing heating control. The industrial control computing unit may have the capabilities for data storage, model training, and control logic operations. The temperature acquisition module may acquire temperature data during the sintering process at a preset sampling frequency and transmit it to the industrial control computing unit. The power adjustment unit may adjust the heating power in response to instructions from the industrial control computing unit. Specific embodiments are described in detail below.

[0021] like Figure 1 As shown in the figure, an embodiment of the present invention discloses an automatic kiln temperature control method based on deep learning, including the following method steps: S1: Acquire historical production data and preprocess the historical production data to obtain a standardized historical dataset; the historical production data includes temperature time series data of ferrite materials, process stage identification sequences and corresponding finished product magnetic property data; S2: Based on the trained deep learning model, the complete temperature-time history of the current sintering process is divided into temperature segments, and the influence weight of each temperature segment on the predicted value of magnetic performance deviation is calculated; wherein, the deep learning model is trained based on the standardized historical dataset; the input of the deep learning model is the temperature time series data and the process stage identifier sequence, and the output is the corresponding predicted value of magnetic performance deviation; S3: Based on the influence weight, identify the high-sensitivity temperature segment from all the divided temperature segments, and generate corresponding temperature control constraint parameters for the high-sensitivity temperature segment and the low-sensitivity temperature segment respectively; S4: During real-time sintering, the corresponding temperature control constraint parameters are called according to the current temperature range to dynamically execute temperature control.

[0022] In some embodiments, the purpose of step S1 is to obtain standardized and valid historical production data, eliminate data noise and format differences through preprocessing operations, and obtain a standardized historical dataset for subsequent training of deep learning models.

[0023] In one embodiment, the industrial control computing unit of the kiln control system accesses the historical production database to read historical production data related to the sintering process of multiple batches of ferrite materials. The historical production data specifically includes three types of data: temperature time series data, process stage identifier sequences, and corresponding finished product magnetic property data.

[0024] The temperature time-series data originates from the kiln's temperature acquisition module and consists of continuously arranged temperature samples arranged in chronological order. The data structure is a one-dimensional time-series array, recording the temperature changes within the kiln over time throughout the sintering process. The process stage identifier sequence indicates the process stage at each time point, including the heating stage, holding stage, and cooling stage. Its data structure corresponds one-to-one with the temperature time-series data; each temperature sample corresponds to a process stage identifier. The finished product magnetic property data originates from the finished product quality inspection process and includes conventional magnetic property parameters in this field, such as permeability, loss tangent, and Curie temperature. The data is in numerical form, and each batch of products corresponds to a complete set of magnetic property data.

[0025] In one embodiment, after acquiring the aforementioned data, the industrial control computing unit performs unified formatting preprocessing on the data. The processing includes three stages: time alignment, outlier removal, and standardization, as detailed below: Time alignment is designed to ensure consistency in the time dimension of data across different batches, facilitating feature extraction and comparison during subsequent model training. Specifically, based on a preset standard sintering time, the temperature time series data and process stage identifier sequences of each batch are normalized in terms of time scale. For batch data exceeding the standard sintering time, the portion corresponding to the standard time is truncated; for batch data shorter than the standard sintering time, missing time point data is supplemented using linear interpolation to ensure consistency in time length across all batches.

[0026] Outlier removal is used to eliminate the impact of outliers on model training accuracy. For temperature time series data, data with missing sampling points or abrupt changes are removed by setting reasonable anomaly detection thresholds. For example, when the temperature difference between a sampling point and its adjacent sampling points exceeds a preset threshold, the sampling point is determined to be an abrupt change, and the average temperature value of the adjacent sampling points is used to replace the outlier data. For multiple consecutive missing sampling points, linear interpolation is used to complete the data. For finished product magnetic property data, if the magnetic property data of a batch exceeds a preset reasonable range, the batch of data is determined to be outlier and removed from the historical production data.

[0027] Standardization is used to ensure that different types of magnetic property data fall within a uniform dimension range, preventing model training from being biased towards one type of parameter due to differences in data dimensions. Common standardization methods include Min-Max standardization and Z-score standardization. For example, when using the Z-score standardization method, the formula... The magnetic performance data is processed, including For standardized magnetic property data, This is the raw magnetic property data. This represents the mean of the magnetic properties data for this type. This represents the standard deviation of this type of magnetic property data.

[0028] After completing the above preprocessing operations, a standardized historical dataset is obtained. This dataset contains standardized temperature time series data of multiple batches of ferrite materials, corresponding process stage identification sequences, and standardized finished product magnetic property data, providing a high-quality data foundation for the subsequent training of deep learning models.

[0029] In some embodiments, for step S2, a mapping relationship between temperature time series data, process stage identification sequence and magnetic performance deviation prediction value is established by constructing and training a deep learning model with a specific structure. Then, based on the model, temperature segments are divided and the influence weight of each segment is calculated. This solves the technical problem that traditional control methods cannot quantify the degree of influence of temperature segments on magnetic performance, and realizes accurate assessment of the sensitivity of temperature segments, providing a basis for the subsequent identification of highly sensitive temperature segments.

[0030] Please see Figure 2 , Figure 2 This is a schematic diagram of the deep learning model structure provided in an embodiment of the present disclosure. The deep learning model includes five parts: an input layer, a feature extraction layer, an attention layer, a fully connected layer, and an output layer. Each layer works together to realize the mapping from input data to predicted values ​​of magnetic property deviation.

[0031] Specifically, the input layer receives preprocessed temperature time series data and process stage identifier sequences, and converts them into tensor forms that the model can process. The process stage identifier sequences need to be encoded first, for example, the heating stage is encoded as 0, the holding stage is encoded as 1, and the cooling stage is encoded as 2, so as to convert them into numerical features, which are convenient for the model to calculate.

[0032] The feature extraction layer is a Long Short-Term Memory (LSTM) network layer, whose core function is to extract temporal features from temperature time series data and capture the long-term and short-term dependencies of temperature changes. The LSTM network effectively solves the gradient vanishing or gradient exploding problems that occur in traditional Recurrent Neural Networks (RNNs) when processing long-sequence data through gating mechanisms (input gate, forget gate, output gate), and can fully explore the hidden dynamic change patterns in temperature time series. For example, the number of hidden layers in the LSTM network can be configured to 2-4 layers, the number of hidden units in each layer can be configured to 64-256, and the dropout rate can be set to 0.1-0.3 to avoid model overfitting; the activation function of the LSTM network can be the tanh function, and the output activation function can be the sigmoid function.

[0033] The attention layer is used to assign weights to the temporal features output by the feature extraction layer, making the model pay more attention to the feature information that has a greater impact on magnetic performance deviation, thereby improving the model's prediction accuracy. According to an embodiment of this disclosure, the attention layer adopts an additive attention mechanism, and its specific calculation process is as follows: First, the temporal features output by the feature extraction layer... ( , (Dimensions of temporal features) and learnable query vectors Perform matching calculations to obtain the initial attention score. The calculation formula is:

[0034] in, For attention weight vectors, and The weight matrix is ​​a learnable matrix. This represents the matrix transpose operation.

[0035] Subsequently, the initial attention scores are normalized using the softmax function to obtain the normalized attention weights. The calculation formula is:

[0036] Finally, the normalized attention weights are summed with the corresponding temporal features to obtain the fused features processed by the attention mechanism. The calculation formula is:

[0037] Fully connected layers are used to perform non-linear transformations on the fused features processed by the attention mechanism, further fusing feature information and improving the model's expressive power. One to two fully connected layers can be set, and the number of neurons in each layer can be configured according to actual needs. For example, the first layer can have 128 neurons, and the second layer can have 64 neurons. The ReLU activation function is used.

[0038] The output layer is a regression layer used to output the predicted magnetic performance deviation value. Its activation function is a linear activation function to ensure that the predicted value covers a reasonable range of magnetic performance deviation. The predicted magnetic performance deviation value is the difference between the actual magnetic performance data and the target magnetic performance data, where the target magnetic performance data is the preset ideal magnetic performance parameter value of the ferrite material.

[0039] In one embodiment, the training also includes deep learning. Specifically, the deep learning model is trained based on the standardized historical dataset obtained in step S1. The specific training process includes data partitioning, loss function definition, optimizer selection, training iteration, and model validation, as detailed below: In the data partitioning stage, the standardized historical dataset is divided into a training set, a validation set, and a test set according to a preset ratio. For example, a 7:2:1 ratio can be used, with the training set used for learning model parameters, the validation set used to monitor the model training process, adjust model hyperparameters in a timely manner to avoid overfitting, and the test set used to evaluate the model's generalization performance.

[0040] The choice of loss function affects the training performance of the model. Since this model is a regression prediction model, the mean squared error (MSE) can be chosen as the loss function, and its formula is as follows:

[0041] in, This is the loss value. For the sample size, For the first The actual magnetic property deviation value of each sample For the first The predicted magnetic property deviation value for each sample. The mean square error effectively measures the mean of the squared difference between the predicted and actual values, enabling the model to converge quickly to the optimal solution.

[0042] The optimizer can employ the Adaptive Moment Estimation (Adam) optimizer, which combines the advantages of momentum gradient descent and RMSProp, and can adaptively adjust the learning rate to accelerate model training and improve training performance.

[0043] During training iterations, training data is input into the model in batches, with batch sizes ranging from 32 to 128. The model's predicted output is calculated via forward propagation, and the loss function is used to calculate the loss between the predicted and actual outputs. Then, the model's weight parameters are updated using backpropagation to continuously reduce the loss. During training, after a certain number of iterations (e.g., 10-50), the model is validated using validation set data. The loss value of the validation set is calculated. If the loss value of the validation set does not decrease for several consecutive iterations (e.g., 3-5 times), it indicates that the model may be overfitting. In this case, measures such as adjusting the dropout rate, reducing the learning rate, or stopping training early can be taken. The number of training iterations can be set from 100 to 500, and can be adjusted based on the model's convergence performance.

[0044] In the model validation phase, test set data is input into the trained model, and evaluation metrics such as the mean absolute error (MAE) and root mean square error (RMSE) of the test set are calculated to assess the model's generalization performance. The formula for mean absolute error is:

[0045] The formula for root mean square error is:

[0046] If the evaluation metrics of the test set meet the preset requirements, such as the mean absolute error being less than a certain threshold or the root mean square error being less than a certain threshold, it indicates that the model has good generalization performance and training is complete. If the preset requirements are not met, the model structure or hyperparameters need to be readjusted and training should be performed again until the model performance meets the standards.

[0047] In one embodiment, after the deep learning model is trained, the complete temperature-time history of the current sintering process is divided into temperature segments based on the model. The principle of segment division is to ensure that the temperature change characteristics within each segment have a certain consistency, while also ensuring that the granularity of the segment division is fine enough to accurately capture local temperature changes that have a significant impact on magnetic properties.

[0048] The specific division method can employ either a fixed time length division method or a dynamic time length division method. In this embodiment, the dynamic time length division method is preferred. This method can adaptively adjust the segment length according to the actual temperature changes, better reflecting the temperature change characteristics of the sintering process. The specific implementation of the dynamic time length division method is as follows: First, set a temperature change rate threshold, for example, 0.5℃ / min-2℃ / min. Then, iterate through the temperature time series of the current sintering process in chronological order, and calculate the temperature change rate between two adjacent sampling points. The formula for calculating the temperature change rate is:

[0049] in, For the rate of temperature change, and These are the temperature values ​​from two adjacent sampling points. This represents the sampling time interval.

[0050] When the temperature change rate at a certain location exceeds a preset threshold, that location is used as the dividing point for segmentation, thereby dividing the entire sintering temperature time history into multiple continuous temperature segments. Furthermore, auxiliary segmentation can be achieved by combining process stage identifiers, dynamically dividing the heating, holding, and cooling stages to ensure that the segmentation within each process stage accurately reflects the impact of temperature changes on magnetic properties.

[0051] Alternatively, a fixed time length division method can be used, for example, setting the time length of each segment to 5min-30min, and dividing the entire sintering temperature time history evenly into multiple temperature segments according to the time sequence. This method is simple to implement, easy to operate, and suitable for sintering processes with relatively stable temperature changes.

[0052] In one embodiment, see Figure 3 , Figure 3 This is a schematic diagram of the influence weight calculation process provided in the embodiments of this disclosure. The influence weight of each temperature range is calculated through the following steps, which can accurately quantify the degree of influence of each temperature range on the predicted value of magnetic performance deviation.

[0053] In step S301, the predicted value of the first magnetic performance deviation is obtained. The complete temperature time series containing temperature segmentation information and the corresponding process stage identifier sequence are input into the trained deep learning model. Through forward propagation of the model, the predicted value of the first magnetic performance deviation is obtained. The first magnetic performance deviation prediction value is the magnetic performance deviation prediction result output by the model when no temperature range is disturbed, which can reflect the comprehensive influence of the original temperature time history on the magnetic performance.

[0054] In step S302, for each predefined temperature segment, the temperature data within that segment is randomly perturbed to generate a new temperature time series, while the temperature data for the remaining segments remain unchanged. The purpose of this random perturbation is to disrupt the original time-series characteristics of the temperature data within that segment, thereby observing the impact of those segment characteristics on the model's prediction results. For example, for a given temperature segment... Each temperature sample value is perturbed by randomly shuffling its order to obtain a perturbed temperature subsequence. This subsequence is then used to replace the corresponding segment in the original temperature time series to form a complete perturbed temperature time series.

[0055] In step S303, the predicted value of the second magnetic property deviation is obtained. The perturbed temperature time series and the corresponding process stage identifier sequence are input into the deep learning model, and the predicted value of the second magnetic property deviation is obtained through forward propagation calculation. The second magnetic performance deviation prediction value is the magnetic performance deviation prediction result output by the model after the selected temperature range is disturbed. The difference between it and the first magnetic performance deviation prediction value can reflect the influence of the selected temperature range on the magnetic performance deviation prediction. It can be understood that the selected temperature range here is the temperature range that is currently being randomly disturbed (the data of other temperature ranges remain unchanged).

[0056] In S304, based on the difference between the predicted values ​​of the first and second magnetic property deviations, the initial influence weight of the selected temperature range is calculated. The initial influence weight can be calculated using either absolute or relative error. For example, when using absolute error, the initial influence weight... The calculation formula is:

[0057] When using relative error, the initial influence weights The calculation formula is:

[0058] in, This is to avoid the case where the denominator is zero.

[0059] In step S305, the average value of the initial influence weights obtained from multiple calculations is taken. To improve the reliability of the influence weight calculation and avoid the randomness caused by a single random perturbation, the above random perturbation operation is repeated at least once, for example, 5-10 times. After each perturbation, the corresponding initial influence weight is calculated, and then the average value of the initial influence weights obtained from multiple calculations is taken as the final influence weight for the selected temperature range. The calculation formula is as follows:

[0060] in, For the number of perturbations, For the first The initial influence weights are calculated after the first perturbation.

[0061] Following the steps outlined above, the final influence weights for all divided temperature segments are calculated sequentially. Each temperature segment corresponds to a unique final influence weight; the larger the weight value, the more significant the impact of that temperature segment on the predicted magnetic property deviation, and the higher its sensitivity to magnetic properties. This calculation method, by comparing the model prediction results before and after the perturbation, intuitively and accurately quantifies the degree of influence of each temperature segment, solving the technical problem that traditional methods cannot accurately assess the sensitivity of temperature segments, and providing a scientific and reliable basis for the subsequent identification of highly sensitive temperature segments.

[0062] In some embodiments, for step S3, highly sensitive temperature segments are identified by setting a judgment threshold, and differentiated temperature control constraint parameters are generated for segments with different sensitivity levels, thereby realizing personalized customization of the control strategy and providing a clear execution standard for subsequent precise control.

[0063] Please see Figure 4 , Figure 4 This is a schematic diagram of the high-sensitivity temperature range identification process provided in this embodiment of the disclosure. According to this embodiment, the identification of high-sensitivity temperature ranges is based on the final influence weight of each temperature range and a preset judgment threshold. The specific implementation steps are as follows: In S401, calculate the statistical characteristic value of the final influence weight for all temperature ranges. The statistical characteristic value includes the mean of all influence weights. , median Standard deviation etc., where the mean The calculation formula is:

[0064] in, This represents the total number of temperature ranges. For the first The final influence weight of each temperature range.

[0065] In step S402, a judgment threshold is set based on the aforementioned statistical characteristic values. The setting of the judgment threshold must balance the accuracy and practicality of identification, ensuring accurate screening of temperature ranges that significantly affect magnetic properties, while avoiding marking too many ranges as highly sensitive temperature ranges, which would lead to excessive control difficulty. For example, the judgment threshold can be set to 1.5-3 times the average value, i.e. ,in This is a proportionality coefficient, which can be adjusted according to the requirements for magnetic stability in actual production. A larger proportionality coefficient can be selected when higher magnetic stability is required, and a smaller proportionality coefficient can be selected when the requirements for magnetic stability are relatively lower. Alternatively, the judgment threshold can be set as the sum of the median and twice the standard deviation. This method can effectively avoid the influence of extreme values ​​on the threshold setting, making the identification of highly sensitive temperature ranges more robust.

[0066] In S403, the final influence weight of each temperature range is compared with the judgment threshold. If the final influence weight of a certain temperature range is... If the temperature range is high, then that temperature range is marked as a highly sensitive temperature range; if the final influence weight of a certain temperature range is... If so, then the temperature range is marked as a low-sensitivity temperature range.

[0067] After identifying the high-sensitivity temperature range, the system generates and stores identification information for both high-sensitivity and low-sensitivity temperature ranges. This identification information includes the range's start and end times, length, and corresponding final impact weight value, providing a clear basis for generating subsequent temperature control constraint parameters. This identification process achieves precise positioning of the high-sensitivity temperature range through quantified weight comparison, solving the problem that traditional control methods cannot distinguish the differences in the impact of different temperature ranges on magnetic properties, thus laying the foundation for subsequent refined segmented control.

[0068] In one embodiment, temperature control constraint parameters are generated separately for high-sensitivity temperature ranges and low-sensitivity temperature ranges. The control constraint parameters for the high-sensitivity temperature range include an allowable temperature change rate threshold and an allowable temperature fluctuation amplitude threshold. The control constraint parameters for the low-sensitivity temperature range use conventional temperature control parameters, as detailed below: (1) Generation of control constraint parameters for high-sensitive temperature zones. The core principle of setting control constraint parameters for high-sensitive temperature zones is to minimize the rate of temperature change and the amplitude of temperature fluctuations while meeting the requirements of the production process, suppressing the impact of temperature disturbances on oxygen reaction kinetics, and ensuring Fe 2+ / Fe 3+ The ratio is balanced, thereby improving the stability and consistency of magnetic properties. This parameter is determined based on statistical analysis of actual temperature data from historical stable production batches within the corresponding high-sensitivity temperature range. The specific setting steps are as follows: The permissible temperature change rate threshold refers to the maximum allowable temperature change per unit time within a highly sensitive temperature range. The setting steps are as follows: First, select production batches from the historical production database that meet the magnetic performance requirements and have good batch stability, i.e., historically stable production batches; Second, extract the temperature time series data of these historically stable production batches within each highly sensitive temperature range; Third, calculate the actual temperature change rate of each historically stable production batch within the corresponding highly sensitive temperature range. The actual temperature change rate is obtained by calculating the ratio of the temperature difference between adjacent sampling points to the time difference, i.e.:

[0069] in, For the first The rate of temperature change over time intervals and These are the temperature values ​​from two adjacent sampling points. The sampling time interval is defined as follows: Fourth, statistical analysis is performed on all calculated actual temperature change rates to calculate their maximum values. ,average value and standard deviation Fifth, set the allowable temperature change rate threshold based on statistical results. For example, the allowable rate of temperature change threshold can be set to 0.7-0.9 times the statistical maximum value, i.e. Or 1.0-1.2 times the statistical average, that is This ensures that the temperature change rate of most historically stable production batches meets the constraint requirement, while leaving a certain margin to cope with slight fluctuations in the production process; alternatively, it can also be customized according to the specific type of ferrite material and process requirements. For example, for ferrite materials whose magnetic properties are more sensitive to temperature changes, the allowable temperature change rate threshold can be set to a smaller value.

[0070] The allowable temperature fluctuation threshold refers to the maximum permissible range of temperature fluctuation per unit time within a highly sensitive temperature range. The setting steps are as follows: First, extract the temperature time series data of historical stable production batches within each highly sensitive temperature range; second, set the time window length, for example, configurable from 1 minute to 10 minutes. The time window length setting must balance the accuracy of temperature fluctuation capture and computational efficiency; third, using a sliding window method, calculate the difference between the maximum and minimum temperature values ​​within each time window. This difference represents the temperature fluctuation amplitude within that time window. The fourth step is to perform statistical analysis on the temperature fluctuation amplitude across all time windows and calculate its maximum value. ,average value and standard deviation Fifth, set the allowable temperature fluctuation threshold based on the statistical results. For example, the allowable temperature fluctuation threshold can be set to 0.6-0.8 times the statistical maximum value, i.e. Or 1.0-1.3 times the statistical average, that is This ensures that temperature fluctuations are controlled within a reasonable range, preventing excessive temperature fluctuations from affecting the valence equilibrium of ferrite materials.

[0071] In addition, a cumulative temperature fluctuation limit can be set, meaning the total temperature fluctuation within the entire highly sensitive temperature range must not exceed a preset value, further ensuring overall temperature stability. For example, the cumulative temperature fluctuation limit can be set to 5-10℃, and adjusted according to the length of the highly sensitive temperature range; the longer the range, the greater the cumulative temperature fluctuation limit can be.

[0072] (2) Generation of control constraint parameters for the low-sensitivity temperature range. The low-sensitivity temperature range has a relatively small impact on magnetic properties, so its control constraint parameters can adopt conventional temperature control parameters without additional strict restrictions, in order to balance control efficiency and production costs. Conventional temperature control parameters include conventional temperature change rate and conventional temperature fluctuation range. Their settings can refer to the control parameters in the existing production process or be set based on the statistical results of historical production data. For example, the conventional temperature change rate can be set to 2-5℃ / min, and the conventional temperature fluctuation range can be set to 3-5℃. The specific values ​​can be adjusted according to the type, performance, and production process requirements of the kiln to ensure that the basic temperature control requirements are met and the sintering process proceeds smoothly.

[0073] After completing the above parameter settings, the system generates a complete set of temperature control constraint parameters. This set includes the identification information of each temperature range, the corresponding control constraint parameter types, and specific parameter values, providing a clear basis for subsequent real-time temperature control execution. This parameter generation process, through differentiated setting methods, achieves refined constraints on highly sensitive temperature ranges, solving the problem of traditional control methods having a single control target and being unable to specifically suppress temperature disturbances in key areas. This provides a strong guarantee for improving the stability of magnetic properties.

[0074] In some embodiments, for step S4, the corresponding temperature control constraint parameters are dynamically invoked according to the temperature range at the current moment, and a differentiated control strategy is adopted to perform temperature control, so as to ensure the temperature stability of the high-sensitivity temperature range and take into account the control efficiency of the low-sensitivity temperature range.

[0075] Specifically, during the real-time sintering production process, the temperature acquisition module of the kiln control system collects temperature data inside the kiln in real time according to a preset sampling frequency. The sampling frequency can be set to 1 time / 10s to 1 time / 1min to ensure timely capture of dynamic temperature changes. The collected temperature data is transmitted to the industrial control computing unit in real time, and the industrial control computing unit records the current time information to form real-time temperature time series data.

[0076] The industrial control computing unit determines whether the current time falls within a high-sensitivity temperature range based on the real-time recorded time information and the high-sensitivity temperature range identifier information (start time, end time) generated in step S3. The specific determination logic is as follows: First, extract the current time. Then, iterate through the identification information of all high-sensitivity temperature zones, and for each high-sensitivity temperature zone, obtain its start time. and end time ( , (the number of highly sensitive temperature zones), if satisfied If the current time does not fall within the time range of any high-sensitive temperature zone, then the current temperature zone is determined to be low-sensitive.

[0077] To ensure the accuracy of zone selection and avoid drastic temperature fluctuations caused by abrupt changes in control strategy during zone switching, a buffer time can be set. For example, 1-3 minutes before the start time and 1-3 minutes after the end time of a high-sensitivity temperature zone, the control constraint parameters are adjusted gradually. When entering a high-sensitivity zone, the allowable temperature change rate is gradually adjusted from the normal value of the low-sensitivity zone to the limit value of the high-sensitivity zone; when exiting a high-sensitivity zone, the allowable temperature change rate is gradually adjusted from the limit value of the high-sensitivity zone to the normal value of the low-sensitivity zone, ensuring temperature stability through a smooth transition.

[0078] In one embodiment, when it is determined that the current temperature is in a high-sensitivity zone, a model predictive control (MPC)-based control algorithm is used to perform temperature control. This algorithm combines predicted temperature values ​​over a future period with control constraint parameters to optimize the current heating power adjustment, ensuring that temperature changes meet the requirements of the allowable temperature change rate threshold and the allowable temperature fluctuation amplitude threshold. (See also...) Figure 5 , Figure 5 This is a schematic diagram of the dynamic control flow provided in an embodiment of the present disclosure. The specific control execution steps are as follows: In S501, the current real-time temperature value, the preset target temperature value, and temperature control constraint parameters are obtained. The industrial control computing unit obtains the current real-time temperature value. Preset target temperature value And the temperature control constraint parameters corresponding to this highly sensitive temperature range, namely the allowable temperature change rate threshold. Permissible temperature fluctuation threshold The target temperature value is determined based on the sintering process of the ferrite material, and serves as a temperature setting curve for the entire sintering process, which changes over time.

[0079] In S502, based on current temperature values ​​and historical temperature data, a simple time-series forecasting model is used to predict the temperature change trend over a preset period of time, resulting in a predicted temperature sequence. The time-series forecasting model can be an ARIMA model or a moving average model, and the preset future period can be set to 5-15 minutes to ensure that the forecast results provide a valid reference for control decisions. For example, when using a moving average model, the predicted temperature sequence is obtained by calculating past... The average temperature of each sampling point is used as the temperature value for the first prediction step, and this is used to calculate the predicted temperature sequence for the next preset time period. ,in To predict the number of steps.

[0080] In S503, based on the allowable temperature change rate threshold of the high-sensitivity temperature zone... The maximum permissible temperature change for each prediction step is calculated using the following formula:

[0081] in, For the first The maximum allowable temperature change of the step. To predict the time interval corresponding to the step size, .

[0082] At the same time, according to the allowable temperature fluctuation range threshold Determine the allowable temperature fluctuation range within each forecast time window, i.e.:

[0083] in, For the first The actual temperature value of the step.

[0084] In S504, an optimization objective function for model predictive control is constructed. The primary objective of this function is to minimize the deviation between the actual and target temperatures, while also considering the smoothness of the temperature change rate to ensure that the temperature change meets the constraints. The expression for the optimization objective function is:

[0085] in, To optimize the objective function value, and These are weighting coefficients used to adjust the importance of temperature deviation and temperature change rate deviation. Prioritizing temperature tracking accuracy; For the first The target temperature value for the step; The desired rate of temperature change can be set to 0.8-0.9 times the allowable rate of temperature change threshold to ensure that the temperature change proceeds smoothly within the constraints.

[0086] The constraints on the objective function include constraints on the rate of temperature change and the magnitude of temperature fluctuations, as detailed below: Rate of temperature change constraint:

[0087] Temperature fluctuation range constraint:

[0088] In addition, the physical constraints of heating power must be considered, namely, the adjustment of heating power must not exceed the maximum output range of the power regulation unit to ensure the executability of control commands.

[0089] In S505, by solving the above constrained optimization problem, the heating power adjustment amount at the current moment is obtained. The optimization problem can be solved using numerical optimization algorithms such as the interior-point method or sequential quadratic programming. The industrial control computing unit can quickly obtain the optimal heating power adjustment by calling the preset optimization algorithm program.

[0090] In S506, temperature control is achieved by adjusting the heating power based on the heating power adjustment amount. The industrial control computing unit sends the calculated heating power adjustment amount to the power regulation unit, which adjusts the heating power of the heating device according to the adjustment amount to achieve temperature control. During control execution, the industrial control computing unit monitors temperature changes in real time. If it finds that the rate of temperature change exceeds the allowable temperature change rate threshold, or the temperature fluctuation amplitude exceeds the allowable temperature fluctuation amplitude threshold, it immediately re-solves the optimization problem and adjusts the heating power adjustment amount to ensure that the temperature always remains within the constraints.

[0091] In one embodiment, when it is determined that the current temperature is in a low-sensitivity temperature range, a proportional-integral-derivative (PID) control algorithm is used to perform temperature control. This algorithm has a simple structure, strong robustness, and can meet the temperature control requirements of the low-sensitivity temperature range, ensuring that the temperature quickly tracks the target temperature value.

[0092] The control logic of the PID control algorithm is as follows: First, calculate the current temperature value. With target temperature value deviation The calculation formula is:

[0093] Then, the heating power adjustment is calculated based on the PID control formula. The calculation formula is:

[0094] in, This is a proportionality coefficient used for rapid response to temperature deviations. The larger the proportionality coefficient, the faster the temperature response, but an excessively large coefficient may cause system oscillation. The integral coefficient is used to eliminate steady-state error. The larger the integral coefficient, the faster the steady-state error is eliminated, but too large a coefficient may lead to an increase in overshoot. These are differential coefficients used to predict the trend of temperature deviation, suppress system oscillations, and improve system stability.

[0095] The above PID parameters ( , , The integral and derivative coefficients can be determined through engineering tuning methods, such as the critical proportional coefficient method, the decay curve method, or the empirical trial-and-error method. For example, when using the empirical trial-and-error method, the integral and derivative coefficients are first set to zero, and the proportional coefficient is gradually increased until the system exhibits critical oscillation. Then, based on the critical oscillation frequency and the critical proportional coefficient, the integral and derivative coefficients are calculated according to a preset formula. Finally, fine-tuning is performed through actual debugging to ensure that the PID controller has good dynamic response characteristics and steady-state accuracy.

[0096] The industrial control computing unit sends the calculated heating power adjustment to the power regulation unit, which then adjusts the heating power of the heating device to ensure the temperature quickly tracks the target temperature value, guaranteeing the smooth progress of the sintering process. In the low-sensitivity temperature range, there is no need to strictly limit the rate of temperature change and the amplitude of fluctuations; it is only necessary to ensure that the temperature remains stable near the target temperature to improve control efficiency and reduce energy consumption.

[0097] In one embodiment, the method further includes updating historical production data by supplementing the historical database with production data from new batches. This allows the deep learning model to continuously learn new production data features, thereby continuously improving the model's prediction accuracy and segment identification accuracy. The specific execution steps are as follows: After the sintering process is completed, the finished ferrite material is subjected to magnetic performance testing. The testing items include conventional magnetic performance parameters in this field such as magnetic permeability, loss tangent, and Curie temperature, to obtain the actual magnetic performance data of the new batch of products.

[0098] The production data of the new batch is added to the historical production database. The production data of the new batch includes the temperature time series data, process stage identification sequence, finished product magnetic property data, and corresponding control constraint parameters and control execution results.

[0099] The industrial control computing unit updates the input data set of the model, supplementing the standardized historical dataset with the new batch of standardized data, providing richer historical data support for subsequent deep learning model training, temperature zone division, influence weight calculation, and control constraint parameter generation.

[0100] This update step enables the method to form a complete optimization mechanism. As production batches increase and historical production data becomes richer, the performance of the deep learning model continues to improve, the temperature control constraint parameters are set more precisely, and the temperature control effect is continuously optimized, ensuring that the method can adapt to changes in the production process in the long term and maintain the stability and consistency of the magnetic properties of ferrite materials.

[0101] Therefore, this method can effectively solve the problem of Fe loss caused by minute temperature changes in the oxygen partial pressure sensitive zone (800-1100℃) during ferrite sintering without altering the existing kiln hardware structure. 2+ / Fe 3+ This paper addresses the technical challenges of proportional imbalance and discrete magnetic properties. By employing a deep learning model to accurately identify highly sensitive temperature ranges and setting strict temperature control constraints, a model predictive control algorithm is used to achieve refined temperature control. Simultaneously, conventional PID control is used in low-sensitivity ranges to balance efficiency, significantly improving temperature stability within the oxygen-sensitive temperature zone and effectively ensuring the valence state balance of the ferrite material. Combined with a historical data update mechanism, the model prediction accuracy and control parameter adaptability are continuously optimized, ultimately achieving a significant improvement in the consistency and batch stability of the magnetic properties of the ferrite material. This meets the high requirements of industrial production for magnetic property stability and demonstrates good practicality and economic efficiency.

[0102] Please see Figure 6 , Figure 6 This invention provides a deep learning-based automatic temperature control system 600 for kilns. This system embodiment is related to... Figure 1 Corresponding to the illustrated method embodiments, this system can be specifically applied to various computer devices. The system specifically includes: The preprocessing module 601 is used to acquire historical production data and preprocess the historical production data to obtain a standardized historical dataset. The historical production data includes temperature time series data of ferrite materials, process stage identification sequence and corresponding finished product magnetic property data. The weight calculation module 602 is used to divide the complete temperature time history of the current sintering process into temperature segments based on the trained deep learning model, and calculate the influence weight of each temperature segment on the predicted value of magnetic performance deviation. The deep learning model is trained based on the standardized historical dataset. The input of the deep learning model is the temperature time series data and the process stage identifier sequence, and the output is the corresponding predicted value of magnetic performance deviation. The parameter generation module 603 is used to identify the high-sensitivity temperature segment from all the divided temperature segments according to the influence weight, and generate corresponding temperature control constraint parameters for the high-sensitivity temperature segment and the low-sensitivity temperature segment respectively. The temperature control module 604 is used to dynamically execute temperature control by calling the corresponding temperature control constraint parameters according to the current temperature range during the real-time sintering process.

[0103] Based on the same inventive concept, this application also provides a computer device, the method corresponding to which can be the method in the foregoing embodiments, and the principle of solving the problem is similar to that method. The computer device provided in this application includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to execute the methods and / or technical solutions of the foregoing embodiments of this application.

[0104] The computer device can be a user device, or a device formed by integrating user devices and network devices through a network, or it can be an application running on the aforementioned devices. The user device includes, but is not limited to, various terminal devices such as computers, mobile phones, tablets, smartwatches, and smart bands. The network device includes, but is not limited to, network hosts, single network servers, multiple network server sets, or cloud computing-based computer sets, and can be used to implement some processing functions when setting an alarm clock. Here, the cloud consists of a large number of hosts or network servers based on cloud computing. Cloud computing is a type of distributed computing, consisting of a virtual computer composed of a group of loosely coupled computer sets.

[0105] Figure 7 The diagram illustrates the structure of an apparatus suitable for implementing the methods and / or technical solutions in the embodiments of this application. The apparatus 700 includes a central processing unit (CPU) 701, which can perform various appropriate actions and processes based on a program stored in a read-only memory (ROM) 702 or a program loaded from a storage portion 708 into a random access memory (RAM) 703. The RAM 703 also stores various programs and data required for system operation. The CPU 701, ROM 702, and RAM 703 are interconnected via a bus 704. An input / output (I / O) interface 705 is also connected to the bus 704.

[0106] The following components are connected to I / O interface 705: an input section 706 including a keyboard, mouse, touchscreen, microphone, infrared sensor, etc.; an output section 707 including a cathode ray tube (CRT), liquid crystal display (LCD), LED display, OLED display, etc., and speakers, etc.; a storage section 708 including one or more computer-readable media such as hard disk, optical disk, magnetic disk, semiconductor memory, etc.; and a communication section 709 including a network interface card such as a LAN (local area network) card, modem, etc. The communication section 709 performs communication processing via a network such as the Internet.

[0107] In particular, the methods and / or embodiments in this application can be implemented as computer software programs. For example, the embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowchart. When the computer program is executed by the central processing unit (CPU) 701, it performs the functions defined in the methods of this application.

[0108] Another embodiment of this application provides a computer-readable storage medium having computer program instructions stored thereon, which can be executed by a processor to implement the methods and / or technical solutions of any one or more embodiments of this application described above.

[0109] Program code contained on a computer-readable medium may be transmitted using any suitable medium, including but not limited to wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0110] The flowcharts or block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of devices, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-specific system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0111] Furthermore, the inclusion of a single word does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices recited in a system claim may also be implemented by a single unit or device through software or hardware. Terms such as "first," "second," etc., are used to indicate names and do not indicate any particular order.

Claims

1. A method for automatic temperature control of a kiln based on deep learning, characterized in that, Includes the following steps: Historical production data is acquired and preprocessed to obtain a standardized historical dataset; the historical production data includes temperature time series data of ferrite materials, process stage identification sequences, and corresponding finished product magnetic property data; Based on the trained deep learning model, the complete temperature-time history of the current sintering process is divided into temperature segments, and the influence weight of each temperature segment on the predicted magnetic performance deviation is calculated. Specifically, the deep learning model is trained based on the standardized historical dataset. The input of the deep learning model is the temperature time series data and the process stage identifier sequence, and the output is the corresponding predicted magnetic performance deviation value. The calculation of the influence weight of each temperature segment on the predicted magnetic performance deviation value specifically involves: inputting the complete temperature time series containing the temperature segment division information into the trained deep learning model to obtain a first predicted magnetic performance deviation value; randomly perturbing the temperature data within the selected temperature segment, and inputting the perturbed temperature time series into the deep learning model to obtain a second predicted magnetic performance deviation value; calculating the initial influence weight of the selected temperature segment based on the difference between the first and second predicted magnetic performance deviation values; repeating the random perturbation operation at least once, and averaging the calculated initial influence weights to obtain the final influence weight of the selected temperature segment. Based on the influence weight, high-sensitivity temperature segments are identified from all the divided temperature segments, and corresponding temperature control constraint parameters are generated for the high-sensitivity temperature segments and low-sensitivity temperature segments respectively. During real-time sintering, the corresponding temperature control constraint parameters are invoked according to the current temperature range to dynamically execute temperature control.

2. The method for automatic kiln temperature control based on deep learning according to claim 1, characterized in that, The deep learning model includes a feature extraction layer and an attention layer; the feature extraction layer is a long short-term memory network layer, used to extract the temporal features of the temperature time series data; The attention layer is used to assign weights to the temporal features.

3. The method for automatic kiln temperature control based on deep learning according to claim 2, characterized in that, The highly sensitive temperature ranges were identified from all the divided temperature ranges, specifically: Calculate the statistical characteristic values ​​of the influence weights for all temperature ranges; Temperature ranges whose influence weight is greater than or equal to the judgment threshold set based on the statistical feature value are marked as the high-sensitivity temperature ranges.

4. The method for automatic kiln temperature control based on deep learning according to claim 1, characterized in that, The temperature control constraint parameters generated for the high-sensitivity temperature range include an allowable temperature change rate threshold and an allowable temperature fluctuation amplitude threshold; the allowable temperature change rate threshold and the allowable temperature fluctuation amplitude threshold are determined based on statistical analysis of actual temperature data of historical stable production batches within the corresponding high-sensitivity temperature range.

5. The method for automatic kiln temperature control based on deep learning according to claim 4, characterized in that, When the current moment is determined to be within the high-sensitivity temperature range, dynamic temperature control is implemented, specifically including: Obtain the current real-time temperature value, the preset target temperature value, and the temperature control constraint parameters corresponding to the high-sensitivity temperature range; Based on current and historical temperature data, predict the temperature change trend for a preset duration in the future; Construct and solve a constrained optimization problem with the objective of minimizing temperature tracking deviation and satisfying the allowable temperature change rate threshold and the allowable temperature fluctuation amplitude threshold, to obtain the heating power adjustment amount at the current moment; Temperature control is performed based on the heating power adjustment amount.

6. A deep learning-based automatic temperature control system for kilns, characterized in that, include: The preprocessing module is used to acquire historical production data and preprocess the historical production data to obtain a standardized historical dataset. The historical production data includes temperature time series data of ferrite materials, process stage identification sequences, and corresponding finished product magnetic property data. The weight calculation module is used to divide the complete temperature-time history of the current sintering process into temperature segments based on a trained deep learning model, and calculate the influence weight of each temperature segment on the predicted magnetic performance deviation. The deep learning model is trained based on the standardized historical dataset. The input of the deep learning model is the temperature time series data and the process stage identifier sequence, and the output is the corresponding predicted magnetic performance deviation value. Specifically, calculating the influence weight of each temperature segment on the predicted magnetic performance deviation value involves: inputting the complete temperature time series containing the temperature segment division information into the trained deep learning model to obtain a first predicted magnetic performance deviation value; randomly perturbing the temperature data within the selected temperature segment, and inputting the perturbed temperature time series into the deep learning model to obtain a second predicted magnetic performance deviation value; calculating the initial influence weight of the selected temperature segment based on the difference between the first and second predicted magnetic performance deviation values; repeating the random perturbation operation at least once, and averaging the calculated initial influence weights to obtain the final influence weight of the selected temperature segment. The parameter generation module is used to identify the high-sensitivity temperature segment from all the divided temperature segments according to the influence weight, and generate corresponding temperature control constraint parameters for the high-sensitivity temperature segment and the low-sensitivity temperature segment respectively. The temperature control module is used to dynamically execute temperature control during real-time sintering by calling the corresponding temperature control constraint parameters according to the current temperature range.

7. A computer device, wherein the computer device is characterized in that, include: At least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-5.

8. A computer-readable medium having computer program instructions stored thereon, characterized in that, The computer program instructions can be executed by a processor to implement the method as described in any one of claims 1-5.

9. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1-5.

Images