Machine learning-based preheating state evaluation method, device and equipment for a tobacco cutter

Abstract

The application discloses a method, device and equipment for judging the preheating state of a cut tobacco dryer based on machine learning, which comprises the following steps: acquiring multi-source time sequence data in the preheating process of the cut tobacco dryer, dynamically calculating and obtaining a feature parameter group based on a preset time interval; inputting the calculated feature parameter group as an input vector into a trained machine learning classification model to output a binary classification result of the current preheating state; the model is built based on a feature attention mechanism and an improved support vector machine; judging the stability of the output result of the model; when the model continuously outputs a classification result indicating sufficient preheating for a preset number of times, it is determined that the cut tobacco dryer is preheated sufficiently; otherwise, it is determined that the cut tobacco dryer is not preheated sufficiently. The application overcomes the hysteresis and inaccuracy of traditional single parameter judgment by fusing multiple sensing parameters and machine learning classification, and accurately judges the overall thermal balance state of the equipment.

Description

Technical Field

[0001] This invention relates to the field of tobacco processing equipment technology, and more specifically to a method, apparatus and equipment for judging the preheating status of a tobacco drying machine based on machine learning. Background Technology

[0002] The airflow drying machine is a key piece of equipment in the tobacco processing production line. It uses a high-temperature process airflow to dry the tobacco shreds, removing excess moisture and improving their sensory quality. Before production begins, the drying machine must be fully preheated to ensure that the internal components (including piping and the combustion furnace) reach a stable and uniform thermodynamic state. Insufficient preheating before production can lead to:

[0003] 1. Unstable product quality: The initial batch of tobacco shreds was not dried evenly, and key indicators such as moisture and color fluctuated greatly, resulting in product downgrading or scrapping.

[0004] 2. Energy waste: Cold equipment absorbs a lot of heat. In the early stage of preheating, high energy input is often used to quickly raise the temperature. If the end point of preheating cannot be accurately determined, it will lead to overheating and waste of gas.

[0005] 3. Process timing is difficult to control: It relies on the operator's experience and judgment, and the feeding of materials is often done too early or too late, which is not conducive to automated production.

[0006] Currently, common methods for judging preheating mainly include: 1. Time-based control method: Set a fixed preheating time. This method ignores the influence of factors such as ambient temperature, initial equipment condition, and energy pressure fluctuations, resulting in poor reliability.

[0007] 2. Single-point temperature method: This method monitors a single temperature point at the outlet or return air duct of the wire drying machine. Preheating is considered complete when the temperature at that point reaches a set value. This method has a serious flaw: reaching the set value at a single temperature point only indicates localized heating in that area, but it cannot determine whether the entire equipment's vast heat capacity has reached thermal equilibrium (i.e., sufficient preheating). Internal components such as pipes may not yet have reached their operating temperature, leading to a sudden drop in overall temperature after material input and process failure.

[0008] Therefore, providing a technical solution that can more comprehensively, accurately, and automatically determine whether an airflow-type wire drying machine is sufficiently preheated is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0009] In view of the above problems, the present invention is proposed to provide a machine learning-based method, apparatus and equipment for judging the preheating state of a wire drying machine, which overcomes or at least partially solves the above problems. By analyzing the comprehensive state of return air temperature, process gas temperature and their derived parameters (temperature difference, temperature rise rate), the invention achieves accurate judgment of the overall thermal balance state of the equipment.

[0010] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, embodiments of the present invention provide a method for evaluating the preheating status of a yarn drying machine based on machine learning, including: S1. Real-time acquisition of multi-source time-series data during the preheating process of the wire drying machine, dynamic calculation and deriving of a set of characteristic parameters based on a preset time interval; S2. The feature parameter set obtained in real time is used as the input vector and input into the trained machine learning classification model to output the binary classification result of the current warm-up state; the model is constructed based on feature attention mechanism and improved support vector machine; S3: Perform a stability judgment on the output results of the model; when the model continuously outputs classification results indicating sufficient preheating for a preset number of times, it is determined that the preheating of the wire drying machine is sufficient; otherwise, it is determined that the preheating of the wire drying machine is insufficient.

[0011] Furthermore, in step S1, the multi-source time-series data includes: process gas temperature. Return air temperature of the drying machine and real-time heat flux of the wire drying machine .

[0012] Furthermore, in step S1, the set of characteristic parameters for the preset time interval includes: real-time temperature difference. Process gas temperature rise rate Rate of temperature rise of return air in the wire drying machine and the real-time heat flux value of the wire drying machine The formula is expressed as follows:

[0013] The real-time temperature difference of the wire drying machine is expressed by the formula:

[0014] The process gas temperature rise rate is expressed by the formula:

[0015] The rate of temperature rise of the return air in the wire drying machine is expressed by the following formula:

[0016] in, Represents a set of characteristic parameters. This indicates the current temperature of the process gas. This indicates the current return air temperature of the wire drying machine. This indicates the preset time interval.

[0017] Furthermore, in step S2, the data processing procedure of the machine learning classification model includes: Z-score standardization is applied to the input vector to eliminate dimensional differences; A feature attention mechanism is introduced to perform weighted processing on the standardized features to obtain a weighted feature vector. The weighted feature vector is input into an improved support vector machine for nonlinear mapping, and the corresponding binary classification result is output. The improved support vector machine employs an improved radial basis function kernel, introducing a dynamic factor of process gas temperature into the original radial basis function kernel. This allows the kernel width to adapt to different temperature ranges. The improved radial basis kernel function is expressed by the following formula:

[0018] in, The input feature vector represents the thermal state at different times. Indicates the dynamic factor of process gas temperature. Represents kernel function parameters. This represents the squared Euclidean distance between two eigenvectors.

[0019] Furthermore, the feature attention mechanism assigns dynamic weights to each input feature and performs weighted processing; the dynamic weights are jointly determined by the feature importance score and the stability of real-time parameters. The feature importance score is calculated by quantifying the correlation between each feature and the preheated label using mutual information to obtain the basic weight; the real-time parameter stability is obtained by introducing a parameter fluctuation coefficient to correct the basic weight and obtain the dynamic weight. The feature importance score is expressed by the formula: =

[0020] The dynamic weight is expressed by the formula:

[0021] Among them, among them, Indicates the first Dynamic weights of each feature Indicates the first The basic weights of each feature Indicates the first The fluctuation coefficient of each characteristic This represents the maximum value of all characteristic stability indices.

[0022] Furthermore, in step S2, the model training process of the machine learning classification model includes: Collect historical preheating process data of the wire dryer covering different seasons, equipment operating conditions, and initial states. Each data point is a time series, containing full data on process gas temperature, wire dryer return air temperature, and real-time heat flux of the wire dryer. Calculate and convert each time series data into a set of characteristic parameter sequences. The feature parameter sequence groups at each time step are labeled with tags indicating whether the preheating is insufficient or sufficient, forming a labeled sample set; Based on the labeled sample set, the machine learning classification model is trained; the improved support vector machine adopts an improved radial basis kernel function, and the key hyperparameters are optimized using grid search and cross-validation techniques. After optimization, the trained machine learning classification model is obtained.

[0023] Furthermore, the key hyperparameters include: the penalty factor C, the kernel function parameter γ, and the attention regularization parameter. .

[0024] Secondly, embodiments of the present invention provide a machine learning-based device for evaluating the preheating status of a yarn drying machine, comprising: Data acquisition module: used to acquire multi-source time-series data during the preheating process of the wire drying machine, dynamically calculate and derive a set of characteristic parameters based on a preset time interval; The intelligent evaluation module is used to input the calculated feature parameter set as an input vector into the trained machine learning classification model and output the binary classification result of the current preheating state; the model is based on the feature attention mechanism and the improved support vector machine. Decision output module: used to judge the stability of the output results of the model; when the model continuously outputs classification results indicating sufficient preheating for a preset number of times, it is determined that the wire drying machine is sufficiently preheated; otherwise, it is determined that the wire drying machine is insufficiently preheated.

[0025] Thirdly, embodiments of the present invention provide a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, when the processor executes the program, it implements the machine learning-based method for judging the preheating status of a wire drying machine as described in any one of the first aspects.

[0026] Fourthly, embodiments of the present invention provide a computer-readable storage medium having a computer program stored thereon, characterized in that, when the program is executed by a processor, it implements the machine learning-based method for judging the preheating state of a wire drying machine as described in any one of the first aspects.

[0027] As can be seen from the above technical solutions, compared with the prior art, the present invention discloses a method, apparatus, and equipment for evaluating the preheating state of a yarn drying machine based on machine learning, which has the following beneficial effects: 1. Comprehensive evaluation dimensions: It integrates information from multiple dimensions such as temperature, temperature difference, rate, and energy.

[0028] 2. Precise and intelligent judgment: It uses machine learning models to learn the complex nonlinear relationship between multiple parameters and the preheating state, resulting in high accuracy.

[0029] 3. Strong adaptability: It has a stronger adaptability to different environmental conditions and equipment operating conditions.

[0030] 4. Achieve true automation: The preheating process is controlled in a closed loop.

[0031] 5. Energy saving and quality assurance: Accurate determination of the preheating endpoint avoids energy waste and ensures product quality stability. Attached Figure Description

[0032] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0033] Figure 1 This is a flowchart of the machine learning-based method for evaluating the preheating status of a wire drying machine provided in this embodiment of the invention. Figure 2 This is a structural block diagram of the preheating status evaluation device for a wire drying machine based on machine learning provided in an embodiment of the present invention. Detailed Implementation

[0034] 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.

[0035] Example 1 This invention discloses a machine learning-based method for evaluating the preheating status of a wire drying machine, referring to... Figure 1 As shown, it includes: S1. Acquire multi-source time-series data during the preheating process of the wire drying machine, dynamically calculate and derive a set of characteristic parameters based on a preset time interval; S2. The calculated feature parameter set is used as an input vector and input into the trained machine learning classification model to output the binary classification result of the current preheating state; the model is based on feature attention mechanism and improved support vector machine construction; S3: Determine the stability of the model's output results; when the model continuously outputs classification results indicating sufficient preheating for a preset number of times, it is determined that the wire drying machine is sufficiently preheated; otherwise, it is determined that the wire drying machine is insufficiently preheated.

[0036] This embodiment takes a certain type of airflow-type wire dryer as an example. First, hardware deployment and data acquisition are carried out. By installing sensors on the process gas pipeline, return air pipeline, and heating element of the wire dryer, the process gas temperature, return air temperature, and heat flux data are collected in real time. The real-time temperature difference, process gas temperature rise rate, and return air temperature rise rate are calculated. Second, a machine learning model is constructed, using the above-mentioned multi-dimensional feature parameter set as input, to determine the current preheating state of the wire dryer. Finally, the stability of the current preheating state is judged to determine whether the preheating is sufficient. This embodiment, through the fusion of multi-sensor information and an improved machine learning model, achieves accurate, adaptive, and automated evaluation of the preheating state of the airflow-type wire dryer, significantly improving product quality consistency and reducing energy waste.

[0037] The implementation steps of this embodiment are described in detail below.

[0038] The first step is to deploy the hardware and collect data.

[0039] Regarding hardware deployment, in this embodiment, a temperature sensor is installed at a stable and representative location within the process gas pipeline of the airflow-type wire drying machine to ensure accurate measurement of the process gas temperature. In this embodiment, a temperature sensor with a range of 0-300℃ and an accuracy of ±0.5℃ is installed at one-third of the pipeline length from the heating element outlet.

[0040] A temperature sensor is installed inside the return air duct near the equipment's air outlet to monitor the return air temperature in real time. The airflow is relatively stable here, which can better reflect the overall temperature of the return air.

[0041] In this embodiment, a heat flux sensor is installed near the heating element, either on the surface of the heating element or on the outside of the insulation layer adjacent to the heating element, to ensure accurate monitoring of the heat flux transferred from the heating element to the interior of the equipment. The heat flux sensor has a measurement range of 0-250 kW / m².

[0042] Regarding data acquisition, in this embodiment, all sensor signals are connected to an industrial computer (IPC). The IPC is configured with a data acquisition card to synchronously acquire all sensor data at a sampling frequency of 1 second.

[0043] All sensor data are multi-source time-series data, including: process gas temperature. Return air temperature of the drying machine and real-time heat flux of the wire drying machine Outlier handling is performed on all sensor data.

[0044] The "3σ principle" is used to eliminate outlier data, such as data caused by sensor malfunctions. And use "linear interpolation" to fill in missing values, such as individual missing points within a 1-second sampling interval.

[0045] Specifically, this embodiment calculates the historical average of each parameter. with standard deviation If a sample value exceeds Values ​​suspected of being outliers are marked as such. For these suspected outliers, the gradient between adjacent sampled values ​​is further analyzed. Since the maximum temperature change rate during the preheating stage of the wire drying machine is 0.08℃ / s, if the gradient exceeds 5℃ / s, it is considered an outlier (e.g., sensor malfunction), and is filled using the "moving average of three adjacent normal sampled values". The original data after removing outliers is then filtered using a "5-point moving average filter", the formula of which is:

[0046] After smoothing, The fluctuation range can be reduced from ±0.5℃ to ±0.1℃, avoiding deviations in rate calculation caused by instantaneous fluctuations.

[0047] Next, based on the above data, a set of dynamically changing characteristic parameters will be calculated. These parameters collectively characterize the thermal dynamics of the system: Real-time temperature difference of the wire drying machine: ; Process gas temperature rise rate: ; Rate of temperature rise of return air in the wire drying machine: ; Real-time heat flux value of the wire drying machine: Take the latest or short-term average value of the sensor reading directly.

[0048] in, Represents a set of characteristic parameters. This indicates the current temperature of the process gas. This indicates the current return air temperature of the wire drying machine. This indicates the preset time interval.

[0049] In this implementation's optimization design regarding feature parameter calculation, the sliding window size is determined based on the "warm-up phase". The minimum change period is used to determine a 10-second window, based on historical data statistics. The minimum change period from 5℃ / min to 0.3℃ / min is 8 seconds; a 10-second window can capture the entire change. The changing trend, if the window is too small (e.g., 5 seconds), will lead to... The calculation fluctuates greatly; if the window is too large (such as 15 seconds), it will delay the feature response (missing the best time for judgment when the preheating is sufficient).

[0050] heat flux Q Supplement to the mean calculation: 5 minutes before preheating , Q The weight increases linearly over time (from 0.5 to 1.0) because the initial heat loss of the heating element is large, while the heat output stabilizes later; after 5 minutes of preheating... Using equal-weighted average, the formula is: 1. When the process gas temperature (5 minutes before preheating):

[0051] 2. When the process gas temperature (After preheating for 5 minutes):

[0052] in, This represents the average value of the heat flux, used to characterize the average level of heat output during the preheating stage; This represents the summation of all terms from 0 to 9. This design enables... Q The average value better reflects the actual effective heat output of the heating element, avoiding the problems caused by initial heat loss. Q underestimate.

[0053] The threshold values ​​and physical meanings of each parameter in the feature parameter group obtained in this embodiment are shown in Table 1 below: Table 1. Thresholds and Physical Meanings of Each Parameter in the Feature Parameter Group

[0054] The second step is to build a machine learning model, using a multi-dimensional set of feature parameters as input, to determine the current preheating status of the wire drying machine.

[0055] Regarding the construction of the machine learning model, the machine learning model in this embodiment includes: an input layer, a feature attention layer, an improved RBF kernel SVM classification layer, and an output layer; The input layer is used to process the input vector. Characteristic parameters are standardized; Z-score standardization is used to eliminate dimensional differences and avoid heat flux. Q (Unit: W / m²) and rate of temperature rise Differences in magnitude (unit: °C / min) affect model training. The standardized formula used in this embodiment is:

[0056] in, The first i Historical mean and standard deviation of each feature Represents the first element in the input vector. i One characteristic, Represents the standardized first i One characteristic, These correspond to four features. The historical mean and standard deviation in this embodiment are obtained from statistical analysis of 300 preheating historical data. of .

[0057] The feature attention layer, based on a parameterized attention mechanism with dynamic weight allocation, assigns dynamic weights to each input feature. And perform weighted processing; weight It is determined by both the feature importance score and the stability of real-time parameters.

[0058] Regarding the calculation of feature importance scores, the correlation between each feature and the fully warmed-up label is quantified using mutual information to calculate the basic weights; the formula is expressed as follows:

[0059] in, Y For label 0 or 1, Features With tags Y The joint probability; Indicates the first The set of values ​​for each feature; This represents the set of tags indicating the preheating status. This indicates insufficient preheating. This indicates that the preheating is sufficient; Representation of features With tags The joint probability; Representation of features The marginal probability; Indicates label The marginal probability.

[0060] The calculation in this embodiment is MI value: according to MI Normalization of proportions to preliminarily determine basic weights .

[0061] Regarding real-time parameter stability correction, this embodiment introduces a parameter fluctuation coefficient. Adjust the weights; the parameter fluctuation coefficient is the ratio of the standard deviation to the mean of the characteristic over 10 seconds; if the parameter fluctuation coefficient is large, such as the temperature difference fluctuation coefficient... If the value is not stable, its weight should be reduced; otherwise, its weight should be increased. The corrected formula is:

[0062] The final weights satisfy The weighted feature vector is

[0063] An improved RBF kernel SVM classification layer is used for non-linear classification of weighted feature vectors; the traditional RBF kernel function is... ,in To maintain a fixed core width, this embodiment introduces a dynamic factor for process gas temperature. To adapt the kernel function to the characteristic sensitivity of different temperature ranges: the process gas temperature dynamic factor is expressed by the formula:

[0064] The improved kernel function is as follows:

[0065] The input feature vector represents the thermal state at different times. The input feature vector at the current moment (or a certain time window) includes dynamically calculated parameters such as process gas temperature, return air temperature, and real-time heat flux, which are used to reflect the thermal state of the system at this moment. That is with The corresponding feature vector at another moment is used to represent the feature data at a certain historical moment or within a specific time interval; This represents the dynamic factor of process gas temperature, which varies with... Adjust the kernel function to improve its sensitivity to features; This represents the kernel function parameter, which is one of the hyperparameters, and its optimal value is 0.1. This represents the squared Euclidean distance between two eigenvectors.

[0066] This embodiment uses an improved RBF kernel SVM classification layer for nonlinear classification, and the resulting binary classification result reaches the output layer; the output is "0" (insufficient warm-up) or "1" (preliminary achievement at the current time); the judgment logic is: if Output "1" if the condition is met, otherwise output "0". Additional output: Calculate the classification confidence score. When P>0.9, the reliability of the "1" class result is higher. This value can be used for dynamic adjustment of the subsequent stable number threshold. For example, when the confidence level is high, N can be reduced to 8 times.

[0067] The training process for improving the RBF kernel SVM classification layer includes: Construct an objective function to minimize classification error and model complexity, expressed by the following formula:

[0068] Constraints:

[0069]

[0070] As a slack variable, it allows for a small number of misclassifications and avoids overfitting; C The penalty factor controls the degree of overfitting, with an optimal value of 10. This is the model weight vector; For model bias terms; The total number of samples.

[0071] This embodiment uses "grid search plus 5-fold hierarchical cross-validation" for hyperparameter optimization. The hyperparameter grid settings for this embodiment are shown in Table 2 below: Table 2 Hyperparameter Reference Table

[0072] Training termination condition: If the harmonic mean of precision and recall on the cross-validation set, F1-score, does not improve for 5 consecutive iterations, the training will terminate to avoid overfitting.

[0073] Regarding the training process of the machine learning model, historical data is first acquired. In this embodiment, a large amount of historical preheating process data of the wire drying machine is collected, covering different seasons, different equipment operating conditions, and different initial states. Each data point is a time series, containing... The entire data process is then processed. Next, following the dynamic feature parameter extraction method described above, each time series data point is converted into a feature parameter sequence. {X}=[ T, Subsequently, experts in the field were hired to label the characteristic parameter data at each moment as "0" (insufficient) or "1" (sufficient) based on whether the quality indicators (such as moisture content and color) of the first batch of tobacco produced in the corresponding batch were consistently qualified, or based on the historical operating experience of the equipment. This ultimately formed a labeled sample set. ,in , M The total number of samples.

[0074] This embodiment selects Support Vector Machine (SVM) as the classifier, which excels at handling small-sample, nonlinear, high-dimensional pattern recognition problems. The labeled sample set is randomly divided into training and test sets at a certain ratio, for example, 7:3. The SVM model is trained using the training set data. A Radial Basis Function (RBF) kernel function is employed, and grid search and cross-validation techniques are used to optimize key hyperparameters (such as the penalty factor C and kernel parameter γ) to find the model that performs best on the training set and has the strongest generalization ability.

[0075] Because the sample size of the preheating data from the wire drying machine is limited and the feature dimension is low (only 4 core parameters), deep learning models are prone to overfitting. SVM, on the other hand, excels at small-sample, high-dimensional nonlinear classification. By improving the kernel function and attention mechanism, it can accurately capture the nonlinear mapping relationship between "features" and "preheating state." This embodiment introduces a feature attention layer to address the shortcomings of traditional SVM in "equally weighting" all features. The effect of T on thermal balance is far greater than Higher weights are needed; Improved Radial Basis Function (RBF): Incorporating process gas temperature The dynamic adjustment factor allows the kernel function to adapt to the characteristic sensitivity of different temperature ranges, such as the low-temperature stage. At temperatures below 100°C, the nuclear width decreases to enhance [the nuclear structure / function]. Feature discrimination.

[0076] The third step is to assess the stability of the current preheating state and determine whether the preheating is sufficient.

[0077] The model's output condition "sufficient preheating" must meet the model's output stability constraints to avoid misjudgments caused by instantaneous data fluctuations. The specific rules are as follows: Single model output: When the input feature parameters meet the above quantization threshold, the model outputs a binary classification result of "1", representing "the current preheating state is initially up to standard"; if any parameter exceeds the threshold, it outputs "0", representing "insufficient preheating". Continuous stability verification: In this embodiment, the consecutive number threshold N=10, corresponding to 100 seconds. Since the feature parameters are calculated every 10 seconds, "sufficient preheating" is only determined when the model outputs "1" 10 consecutive times; this filters out instantaneous sensor interference, such as airflow fluctuations. A brief exceedance of the limit ensures the reliability of the judgment.

[0078] After the final determination is generated, this embodiment automatically sends a "allow feeding" command to the main control system of the wire drying machine.

[0079] Meanwhile, this embodiment also establishes a safety monitoring mechanism: if the total preheating time exceeds the preset maximum safety time limit, such as 40 minutes, and the "allow feeding" instruction is still not received, the alarm system will be automatically triggered to prompt the operator to check the equipment.

[0080] Example 2 Based on the same inventive concept, this embodiment of the invention also provides a machine learning-based device for evaluating the preheating status of a wire drying machine. Since the principle of solving the problem by these devices and clients is similar to the aforementioned machine learning-based method for evaluating the preheating status of a wire drying machine, the implementation of these devices and clients can refer to the implementation of the aforementioned method, and the repeated parts will not be described again.

[0081] This invention discloses a machine learning-based device for evaluating the preheating status of a wire drying machine, with reference to... Figure 2 As shown, it includes: Data acquisition module: used to acquire multi-source time-series data during the preheating process of the wire drying machine, dynamically calculate and derive a set of characteristic parameters based on a preset time interval; The intelligent evaluation module is used to input the calculated feature parameter set as an input vector into the trained machine learning classification model and output the binary classification result of the current preheating state; the model is based on the feature attention mechanism and the improved support vector machine. Decision output module: used to judge the stability of the output results of the model; when the model continuously outputs classification results indicating sufficient preheating for a preset number of times, it is determined that the wire drying machine is sufficiently preheated; otherwise, it is determined that the wire drying machine is insufficiently preheated.

[0082] This embodiment acquires and calculates multi-source time-series data in real time during the preheating process of the wire drying machine through a data acquisition module. The intelligent evaluation module uses a machine learning classification model to evaluate the preheating process. The decision output module then judges the stability of the results. Through multi-parameter fusion and intelligent algorithms, the lag and inaccuracy of traditional single-parameter judgment are overcome. It can accurately and adaptively judge the overall thermal balance state of the equipment, thereby ensuring product quality consistency, reducing energy waste, and realizing intelligent closed-loop control of the preheating process.

[0083] Example 3 This invention discloses a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, it implements the machine learning-based method for judging the preheating status of a wire drying machine as described in any one of Embodiments 1.

[0084] Example 4 This invention discloses a computer-readable storage medium storing a computer program that, when executed by a processor, implements the machine learning-based method for judging the preheating status of a wire drying machine as described in any one of Embodiments 1.

[0085] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0086] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A machine learning-based method for evaluating the preheating status of a wire drying machine, characterized in that, include: S1. Acquire multi-source time-series data during the preheating process of the wire drying machine, dynamically calculate and derive a set of characteristic parameters based on a preset time interval; S2. The calculated feature parameter set is used as an input vector and input into the trained machine learning classification model to output the binary classification result of the current preheating state; the model is based on feature attention mechanism and improved support vector machine construction; S3: Determine the stability of the output results of the model; When the model continuously outputs classification results indicating sufficient preheating for a preset number of times, it is determined that the wire drying machine is sufficiently preheated; otherwise, it is determined that the wire drying machine is insufficiently preheated.

2. The method as described in claim 1, characterized in that, In step S1, the multi-source time-series data includes: process gas temperature. Return air temperature of the drying machine and real-time heat flux of the wire drying machine .

3. The method as described in claim 2, characterized in that, In step S1, the set of characteristic parameters for the preset time interval includes: real-time temperature difference. Process gas temperature rise rate Rate of temperature rise of return air in the wire drying machine and the real-time heat flux value of the wire drying machine The formula is expressed as follows: The real-time temperature difference of the wire drying machine is expressed by the formula: The process gas temperature rise rate is expressed by the formula: The rate of temperature rise of the return air in the wire drying machine is expressed by the following formula: in, Represents a set of characteristic parameters. This indicates the current temperature of the process gas. This indicates the current return air temperature of the wire drying machine. This indicates the preset time interval.

4. The method as described in claim 1, characterized in that, In step S2, the data processing procedure of the machine learning classification model includes: Z-score standardization is applied to the input vector to eliminate dimensional differences; A feature attention mechanism is introduced to perform weighted processing on the standardized features to obtain a weighted feature vector. The weighted feature vector is input into an improved support vector machine for nonlinear mapping, and the corresponding binary classification result is output. The improved support vector machine adopts an improved radial basis kernel function, which introduces a dynamic factor of process gas temperature into the original radial basis kernel function, so that the kernel width adapts to the temperature range. The improved radial basis kernel function is expressed by the following formula: in, The input feature vector represents the thermal state at different times. Indicates the dynamic factor of process gas temperature. Represents kernel function parameters. This represents the squared Euclidean distance between two eigenvectors.

5. The method as described in claim 4, characterized in that, The feature attention mechanism assigns dynamic weights to each input feature and performs weighted processing; the dynamic weights are jointly determined by the feature importance score and the stability of real-time parameters. The feature importance score is calculated by quantifying the correlation between each feature and the preheated label using mutual information to obtain the basic weight; the real-time parameter stability is obtained by introducing a parameter fluctuation coefficient to correct the basic weight and obtain the dynamic weight. The feature importance score is expressed by the formula: = The dynamic weight is expressed by the formula: in, Indicates the first Dynamic weights of each feature Indicates the first The basic weights of each feature Indicates the first The fluctuation coefficient of each characteristic This represents the maximum value of all characteristic stability indices.

6. The method as described in claim 1, characterized in that, In step S2, the model training process of the machine learning classification model includes: Collect historical preheating process data of the wire dryer covering different seasons, equipment operating conditions, and initial states. Each data point is a time series, containing full data on process gas temperature, wire dryer return air temperature, and real-time heat flux of the wire dryer. Calculate and convert each time series data into a set of characteristic parameter sequences. The feature parameter sequence groups at each time step are labeled with tags indicating whether the preheating is insufficient or sufficient, forming a labeled sample set; Based on the labeled sample set, the machine learning classification model is trained; the improved support vector machine adopts an improved radial basis kernel function, and the key hyperparameters are optimized using grid search and cross-validation techniques. After optimization, the trained machine learning classification model is obtained.

7. The method as described in claim 6, characterized in that, The key hyperparameters include: the penalty factor C, the kernel function parameter γ, and the attention regularization parameter. .

8. A machine learning-based device for evaluating the preheating status of a wire drying machine, characterized in that, include: Data acquisition module: used to acquire multi-source time-series data during the preheating process of the wire drying machine, dynamically calculate and derive a set of characteristic parameters based on a preset time interval; Intelligent evaluation module: used to take the calculated feature parameter set as the input vector, input it into the trained machine learning classification model, and output the binary classification result of the current preheating state; The model is constructed based on a feature attention mechanism and an improved support vector machine. Decision output module: used to determine the stability of the model's output results; When the model continuously outputs classification results indicating sufficient preheating for a preset number of times, it is determined that the wire drying machine is sufficiently preheated. Otherwise, it is determined that the preheating of the wire drying machine is insufficient.

9. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the machine learning-based method for judging the preheating status of a wire drying machine as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the machine learning-based method for judging the preheating status of a wire drying machine as described in any one of claims 1 to 7.

Images