Parameter prediction method, system and equipment applied to graphite spheroidization production line and medium

Abstract

The invention discloses a parameter prediction method, system and equipment applied to a graphite spheroidization production line and a medium, and relates to the field of intelligent manufacturing, and the method comprises the steps: obtaining the feature data of a target graphite raw material and the target data of a finished product; inputting the characteristic data of the target graphite raw material and the target data of the finished product into a parameter prediction model, and determining spheroidization production line parameters; the parameter prediction model establishment method comprises the steps of obtaining an input data set and an output data set; expanding the input data set by adopting a small sample data expansion and enhancement method to obtain an expanded data set; training an XGBoost model by adopting the expansion data set and the output data set, and determining the XGBoost model as a parameter prediction initial model; and optimizing and adjusting parameters of the parameter prediction initial model by adopting a grid search method and a beetle antennae search algorithm, and determining the optimized and adjusted parameter prediction initial model as a parameter prediction model. According to the invention, the cost of sample acquisition is reduced and the prediction precision of the model is improved.

Description

Technical Field

[0001] The present application relates to the field of intelligent manufacturing, and in particular to parameter prediction methods, systems, equipment and media applied to graphite spheroidization production lines. Background Art

[0002] In the process of natural graphite spheroidization, data such as the particle size, tap density and specific surface area of ​​the finished product are important indicators for evaluating the quality and market value of spherical graphite products. Controlling these data and ensuring that the finished product meets the quality requirements is the core goal of the natural graphite spheroidization production line operation process. Generally, the data in the natural graphite spheroidization process is affected by the characteristics of the raw material flake graphite (such as grade, fineness, etc.), equipment parameters (such as host frequency, feed frequency, etc.) and other external environmental factors (such as humidity, temperature, etc.), and is highly nonlinear and complex.

[0003] In order to accurately control these data, the natural graphite spheroidization processing industry usually relies on a large amount of experimental data, and achieves the ideal processing effect through repeated experiments and adjustment of data. However, this traditional method that relies on a large amount of data has many shortcomings, which are mainly manifested in the following aspects:

[0004] (1) Traditional methods require a large amount of experimental data to train the model. In actual industrial applications, the indicators of flake graphite, equipment configuration, external environment, etc. will affect the stability of experimental conditions. Obtaining these data usually involves time-consuming experimental processes and tedious parameter adjustments. Especially when the characteristics of natural graphite and the process flow change, the data is often insufficient. It is difficult to obtain sufficient sample data, which greatly increases the cost.

[0005] (2) The natural graphite spheroidization production line involves a variety of equipment frequencies, raw material characteristics and processing requirements. There are nonlinear and interdependent relationships between the parameters, which brings great challenges to model design and parameter optimization. Most existing methods rely on large sample data for model training, but in scenarios with insufficient data, the accuracy and robustness of the model are difficult to guarantee. For small sample data, traditional machine learning models are difficult to achieve good prediction results, and although deep learning methods have strong learning capabilities, they are prone to overfitting, resulting in poor model generalization ability. Simple linear regression and traditional machine learning methods often have difficulty handling these complex relationships, resulting in low model prediction accuracy.

[0006] Based on the above background, a new method for predicting graphite spheroidization parameters is needed to reduce the cost of sample acquisition and improve the prediction accuracy of the model. Summary of the invention

[0007] The purpose of this application is to provide a parameter prediction method, system, equipment and medium applied to a graphite spheroidization production line, which can reduce the cost of sample acquisition and improve the prediction accuracy of the model.

[0008] To achieve the above objectives, this application provides the following solutions:

[0009] In a first aspect, the present application provides a parameter prediction method applied to a graphite spheroidization production line, comprising:

[0010] Acquire characteristic data of target graphite raw material and target finished product data of target graphite raw material; the target finished product data includes particle size distribution data, tap density data and specific surface area of ​​graphite raw material during spheroidization;

[0011] Inputting characteristic data of the target graphite raw material and target finished product data of the target graphite raw material into a parameter prediction model to determine the spheroidization production line parameters of the target graphite raw material; the spheroidization production line parameters include the feeding frequency of the spheroidization production line equipment during the spheroidization process of the graphite raw material;

[0012] The method for establishing the parameter prediction model specifically includes:

[0013] Acquire an input data set and an output data set; the input data set includes: characteristic data of sample graphite raw materials and finished product target data; the output data set includes: spheroidization production line parameters of sample graphite raw materials;

[0014] The input data set is expanded using a small sample data expansion and enhancement method to obtain an expanded data set;

[0015] The expanded data set and the output data set are used to train an XGBoost model, and the trained XGBoost model is determined as an initial model for parameter prediction;

[0016] The parameters in the parameter prediction initial model are optimized and adjusted by using a grid search method and a longicorn beetle whisker search algorithm, and the optimized and adjusted parameter prediction initial model is determined as the parameter prediction model.

[0017] Optionally, the input data set is expanded using a small sample data expansion and enhancement method to obtain an expanded data set, specifically including:

[0018] Performing missing value processing, data cleaning and standardization on the input data set to obtain a processed input data set;

[0019] The processed input data set is expanded using a small sample data expansion and enhancement method to obtain an expanded data set.

[0020] Optionally, the parameters in the parameter prediction initial model are optimized and adjusted by using a grid search method and a longicorn beard search algorithm, and the optimized and adjusted parameter prediction initial model is determined as the parameter prediction model, specifically including:

[0021] The parameters in the parameter prediction initial model are preliminarily optimized by using a grid search method to obtain the parameter prediction preliminary optimization model;

[0022] The parameters in the parameter prediction preliminary optimization model are finally optimized by using a longicorn beard search algorithm to obtain the parameter prediction model.

[0023] Optionally, the input data set is subjected to missing value processing, data cleaning and standardization processing to obtain a processed input data set, specifically including:

[0024] Detecting and marking the characteristic data of the sample graphite raw material that is missing in the input data set due to the lack of corresponding finished product target sample data; and treating the missing characteristic data of the sample graphite raw material as the missing characteristic data;

[0025] Filling the missing feature data using Lagrange interpolation method to obtain filled feature data;

[0026] Performing statistical analysis on the filled feature data to ensure that the filled feature data conforms to the actual physical properties, and obtaining the feature data after statistical analysis; and using the feature data after statistical analysis and the input data set as a complete input data set;

[0027] The complete input data set is standardized so that the data in the complete input data set are scaled to the same magnitude, thereby obtaining a processed input data set.

[0028] Optionally, a small sample data augmentation method is used to augment the processed input data set to obtain an augmented data set, specifically including:

[0029] adding uniform noises of different noise amplitudes to the characteristic data of the sample graphite raw material and the finished product target data in the processed input data set respectively, to obtain the disturbed input data set;

[0030] The combined enhancement method is used to perform sample expansion on the perturbed input data set to obtain an expanded data set.

[0031] Optionally, using the expanded data set and the output data set to train an XGBoost model, and determining the trained XGBoost model as an initial model for parameter prediction, specifically includes:

[0032] According to the model training mode, the XGBoost model is trained using the expanded data set and the output data set, and the trained XGBoost model is determined as the parameter prediction initial model;

[0033] The model training modes include: by device mode, by module mode and by machine mode;

[0034] The equipment-based mode is used to train each set of spherical production line equipment as a separate model;

[0035] The module-based mode is used to train each module of the spherical production line equipment classified by functional modules as a separate model;

[0036] The machine-based mode is used to train each type of spherical production line equipment classified by machine type as a separate model.

[0037] Optionally, the parameters in the preliminary optimization model for parameter prediction are finally optimized using a longicorn beard search algorithm to obtain the parameter prediction model, specifically comprising:

[0038] Setting a search range based on the parameters in the parameter prediction preliminary optimization model;

[0039] Determine the hyperparameter value of the centroid of the longhorn beetle at the current number of iterations; wherein the parameters in the parameter prediction preliminary optimization model are used as the hyperparameter value of the centroid of the longhorn beetle at the initial iteration;

[0040] Determine the hyperparameter values ​​of the left whisker direction and the right whisker direction of the longicorn beetle at the current number of iterations according to the hyperparameter value of the longicorn beetle's centroid at the current number of iterations and the search range;

[0041] Calculate the fitness function values ​​of the hyperparameter values ​​of the left whisker direction and the right whisker direction of the longhorn beetle under the current number of iterations, and determine the hyperparameter value of the direction where the fitness function value is the smallest under the current number of iterations;

[0042] Dynamically decay the step size at the current iteration number according to the current iteration number; the calculation formula of the step size is:

[0043]

[0044] Among them, s (t) is the step size at the current iteration number t, s max is the maximum setting step size, s min is the minimum set step size, T is the maximum number of iterations set;

[0045] Update the hyperparameter value of the centroid of the longicorn at the current number of iterations according to the hyperparameter value of the direction where the fitness function value is the smallest at the current number of iterations and the step size at the current number of iterations;

[0046] When the termination condition is reached, the hyperparameter value of the longhorn beetle mass center updated at the current iteration number is used as the final optimization parameter of the parameter prediction preliminary optimization model to obtain the parameter prediction model; otherwise, the hyperparameter value of the longhorn beetle mass center updated at the current iteration number is used as the hyperparameter value of the longhorn beetle mass center at the next iteration number, and after updating the iteration number, the step of determining the hyperparameter values ​​of the longhorn beetle left whisker direction and the longhorn beetle right whisker direction at the current iteration number according to the hyperparameter value of the longhorn beetle mass center at the current iteration number and the search range is returned;

[0047] Among them, the termination conditions include: the current number of iterations reaches the set maximum number of iterations, the change range between the fitness function value of the hyperparameter value of the left whisker direction of the longhorn beetle at the current number of iterations and the fitness function value of the hyperparameter value of the center of mass of the longhorn beetle after the current number of iterations is less than the set threshold, or the change range between the fitness function value of the hyperparameter value of the right whisker direction of the longhorn beetle at the current number of iterations and the fitness function value of the hyperparameter value of the center of mass of the longhorn beetle after the current number of iterations is less than the set threshold.

[0048] In a second aspect, the present application provides a parameter prediction system applied to a graphite spheroidization production line, comprising:

[0049] The test data acquisition module is used to acquire characteristic data of the target graphite raw material and finished product target data of the target graphite raw material; the finished product target data includes particle size distribution data, tap density data and specific surface area during the spheroidization process of the graphite raw material;

[0050] A model determination module, used for determining a parameter prediction model;

[0051] A processing parameter prediction module is used to input characteristic data of the target graphite raw material and target finished product data of the target graphite raw material into a parameter prediction model to determine the spheroidization production line parameters of the target graphite raw material; the spheroidization production line parameters include the feeding frequency of the spheroidization production line equipment during the spheroidization process of the graphite raw material;

[0052] Wherein, the model determination module specifically includes:

[0053] A sample data set acquisition unit is used to acquire an input data set and an output data set; the input data set includes: characteristic data of the sample graphite raw material and finished product target data; the output data set includes: spheroidization production line parameters of the sample graphite raw material;

[0054] A sample data expansion unit, used to expand the input data set by using a small sample data expansion and enhancement method to obtain an expanded data set;

[0055] A model training unit, used to train an XGBoost model using the expanded data set and the output data set, and determine the trained XGBoost model as an initial model for parameter prediction;

[0056] The model parameter optimization unit is used to optimize and adjust the parameters in the parameter prediction initial model by using a grid search method and a longicorn beetle whisker search algorithm, and determine the optimized and adjusted parameter prediction initial model as the parameter prediction model.

[0057] In a third aspect, the present application provides a computer device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of any one of the above-mentioned parameter prediction methods applied to a graphite spheroidization production line.

[0058] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any one of the above-mentioned parameter prediction methods applied to a graphite spheroidization production line.

[0059] According to the specific embodiments provided in this application, this application has the following technical effects:

[0060] The present application provides a parameter prediction method, system, equipment and medium applied to a graphite spheroidization production line, which uses a small sample data expansion and enhancement method to expand the input data set to simulate different working conditions of natural graphite characteristics and equipment parameters, prevent model overfitting, improve the generalization ability of the model under small sample conditions, increase the number of samples and reduce the sample acquisition cost. The XGBoost model is used for training, and the nonlinear relationship between input features is learned to adapt to multi-dimensional and multi-parameter industrial data to improve the prediction accuracy of the model. The XGBoost model parameters are optimized using the grid search method. The model can achieve higher accuracy and adaptability under limited sample data conditions, ensure the effect of the XGBoost algorithm on the prediction of graphite spheroidization production line parameters, and optimize the XGBoost model parameters through the longicorn beard search algorithm to further improve the prediction accuracy of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0062] Figure 1A schematic diagram of a training process of a parameter prediction method applied to a graphite spheroidization production line in an embodiment of the present application;

[0063] Figure 2 A schematic diagram of kernel density distribution of an input data set before and after processing provided by an embodiment of the present application;

[0064] Figure 3 A schematic diagram of the overall operation flow of production line parameter prediction provided in an embodiment of the present application;

[0065] Figure 4 A schematic diagram of the overall design steps provided for an embodiment of the present application;

[0066] Figure 5 A schematic diagram of the structure of a computer device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0067] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0068] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0069] In an exemplary embodiment, Figure 1 As shown, a parameter prediction method applied to a graphite spheroidization production line is provided, characterized in that the parameter prediction method applied to a graphite spheroidization production line includes:

[0070] Acquire characteristic data of target graphite raw material and target finished product data of target graphite raw material; the target finished product data includes particle size distribution data, tap density data and specific surface area during the spheroidization process of graphite raw material; the characteristic data includes: fineness data of target graphite raw material (such as natural graphite raw material: flake graphite).

[0071] The characteristic data of the target graphite raw material and the target finished product data of the target graphite raw material are input into the parameter prediction model to determine the spheroidization production line parameters of the target graphite raw material; the spheroidization production line parameters include: the frequency parameters and feeding frequency of each device of the spheroidization production line during the spheroidization of the graphite raw material.

[0072] The method for establishing the parameter prediction model specifically includes:

[0073] Step 100: Obtain input data set and output data set.

[0074] The input data set X includes: characteristic data of sample graphite raw materials and target data of finished products; the output data set Y includes: spheroidization production line parameters of sample graphite raw materials.

[0075] As an optional implementation, characteristic data of raw material flake graphite, including fineness data (i.e., vibration screening test data), is extracted from a designated file (a file after data aggregation of production line production), and particle size distribution data is extracted according to the screening interval, which involves the material properties after screening and is used to indicate the distribution of raw materials in different particle size intervals. The screening interval parameters include 100 mesh above the sieve (%), 100 mesh-200 mesh (%), 200 mesh-325 mesh (%) and under 325 mesh (%), a total of 4 raw material screening data items.

[0076] Extract the finished product target data from the specified file, including the particle size distribution (D 10 , D 50 , D 90 ) (test instrument: Better Laser Particle Size Analyzer), tap density data (test instrument is tap density analyzer, the number of times is 3000) and specific surface area data (test instrument is nitrogen adsorption specific surface area analyzer), a total of 3 types of finished product target data indicators, set the reasonable range of particle size indicators to D 10 :0-20μm,D 50 :5-40μm,D 90 : 10-50μm, vibration: 0.2-1.2g / cm 3 , ratio: 1-10m 2 / g; these data are used to describe the basic physical properties of the finished product.

[0077] The above five finished product target data items and four raw material screening data items are used as input data X i (i=0,1,...,8), construct the input data set.

[0078] Read the frequency parameters and feed frequency of all spheroidization production line equipment from the specified file as the output data set. The equipment models include 4 sets of 60 machines, 6 sets of 50 machines, and 4 sets of 30 machines, a total of 14 equipment. The processing parameters of each set of graphite spheroidization production line involve the four major processing modules of airflow vortex micro powder machine host, internal classifier, external classifier, and fan. In this way, there are a total of 56 frequency parameters. The above 56 frequency parameters are used as output data Y i , all data are concatenated in sequence to obtain a complete output data set Y.

[0079] The input and output data are spliced ​​in sequence to establish a complete input-output mapping relationship, provide data support for model training, and achieve accurate prediction of the spheroidization production line parameters of the target graphite raw materials.

[0080] Step 200: Expand the input data set using a small sample data expansion and enhancement method to obtain an expanded data set.

[0081] Step 300: Use the expanded data set and the output data set to train the XGBoost model, and determine the trained XGBoost model as the initial model for parameter prediction.

[0082] After model training, the performance of the trained XGBoost model can also be evaluated through cross-validation: K-fold cross-validation (such as 3-fold, 5-fold, and 10-fold cross-validation) is used to evaluate model performance to ensure that the model has strong generalization ability and avoid overfitting.

[0083] Step 400: Optimizing and adjusting the parameters in the parameter prediction initial model using a grid search method and a longicorn beetle whisker search algorithm, and determining the optimized and adjusted parameter prediction initial model as the parameter prediction model.

[0084] In another exemplary embodiment of the present application, step 200 specifically includes:

[0085] Step 201: Perform missing value processing, data cleaning and standardization on the input data set to obtain a processed input data set.

[0086] Step 202: Expand the processed input data set using a small sample data expansion and enhancement method to obtain an expanded data set.

[0087] In another exemplary embodiment of the present application, the grid search method and the longicorn beard search algorithm are used to optimize the XGBoost model parameters to further improve the prediction accuracy of the model.

[0088] Step 400 specifically includes:

[0089] Step 401: Preliminary optimization of parameters in the parameter prediction initial model is performed using a grid search method to obtain a preliminary optimization model for parameter prediction.

[0090] Step 402: finally optimizing the parameters in the parameter prediction preliminary optimization model using the longicorn beard search algorithm to obtain a parameter prediction model. The parameters in the parameter prediction model include: the number of trees, the maximum tree depth, the learning rate, the L1 regularization term, the L2 regularization term, and the random sampling ratio.

[0091] The specific steps for preliminary optimization of model parameters using the grid search method are as follows:

[0092] Step 1: Divide the input data set into a training set (80%) and a test set (20%) for model training and testing, or divide the input data set into multiple subsets, take one of the subsets as the test set each time, and the rest as the training set for model training and testing.

[0093] Step 2: Perform grid search and cross-validation. Set the parameter combination table, traverse the parameters in the 6 parameter prediction models, and evaluate the performance of the parameter combination through 3-fold, 5-fold, and 10-fold cross-validation. Repeat the training-validation process multiple times to evaluate the performance of the parameter combination, and finally select the parameter combination with the highest cross-validation score for modeling to ensure the stability of the model and the optimal prediction performance.

[0094] This application first uses a combination of grid search and cross validation to preliminarily optimize the model parameters. Use grid search to list the parameters required by the parameter prediction model into a grid, and traverse the parameter combination. This application uses 6 parameters. At the same time, cross validation (30% off, 50% off, or 10% off) is used to score the parameter combination, and the parameter combination with the highest scoring mechanism is selected.

[0095] Through grid search and cross-validation optimization, the model can achieve higher accuracy and adaptability under limited sample data conditions, ensuring the effectiveness of the XGBoost algorithm in predicting graphite spheroidization production line parameters.

[0096] In another exemplary embodiment of the present application, step 201 specifically includes:

[0097] Step 2011: Detect and mark the feature data of the sample graphite raw material that is missing in the input data set due to the lack of corresponding finished product target sample data; and use the missing feature data of the sample graphite raw material as the missing feature data.

[0098] Step 2012: Use Lagrange interpolation method to fill in the missing feature data to obtain filled feature data.

[0099] Step 2013: Perform statistical analysis on the filled feature data to ensure that the filled feature data conforms to the actual physical properties, and obtain the feature data after statistical analysis; the feature data after statistical analysis and the input data set are used as the complete input data set.

[0100] Step 2014: Standardize the complete input data set so that the data in the complete input data set are scaled to the same magnitude, thereby obtaining a processed input data set.

[0101] The specific steps are as follows:

[0102] Perform a preliminary analysis of the input data set to detect whether there are missing values ​​in the input data set, and automatically identify and mark the missing feature data X represented by NaN i [j](j=0,1,...,8) for subsequent processing.

[0103] For the detected missing feature data, the Lagrange interpolation method is used to fill it, and its basic formula is as follows:

[0104]

[0105] In the formula, x is the position of the feature to be interpolated in the data set, P(x) is the value of the interpolation point x, and L k (x) is the Lagrange basis polynomial, y k The horizontal axis x k The ordinate (eigenvalue) of the known data point, N is the number of k.

[0106] Among them, the Lagrange basis polynomial L k (x) is calculated as:

[0107]

[0108] Specifically, for the missing data point X i [j] represents the value of the i-th row and j-th column, that is, the value of a feature (column) of the sample graphite raw material in the input data set in the i-th record. If the data at this position is missing, it is filled by interpolation, by identifying the known data points X before and after it. i-1 [j] and X i+1 [j], determine the basic point of interpolation, the interpolation formula is:

[0109] X i [j] = L i-1 (x)·X i-1 [j]+L i+1 (x)·X i+1 [j].

[0110] Among them, L i-1 (x) and L i+1 (x) is a Lagrange basis polynomial based on known points, and the calculation formula is:

[0111]

[0112] x i-1 and x i+1 is the horizontal coordinate of the adjacent known data point, that is, the previous and next data of the missing data in the input data set, X i-1 [j] and X i+1[j] is the vertical coordinate of these adjacent known data points, that is, the characteristics of the finished graphite and raw materials corresponding to the missing data.

[0113] Automatically calculate and fill missing values, and perform statistical analysis on the distribution characteristics of the interpolated data. At the same time, set reasonable physical constraints and value ranges for each feature to ensure that it meets the expected range. If the interpolation result exceeds the reasonable range, it will be automatically adjusted to a reasonable boundary value.

[0114] In order to ensure that the interpolated data can be used as the input of the machine learning model, the interpolated data will be standardized. The standardization formula is as follows:

[0115]

[0116] Among them, X i is the original data, μ is the mean of the data in this column, and σ is the standard deviation of the data in this column. After standardization, an input data set X with uniform dimension and no missing values ​​will be obtained.

[0117] For the missing values ​​detected, Lagrange interpolation method is used to fill them. i [j] is missing, use the known data points X before and after it i-1 [j] and X i+1 [j] Interpolation is performed. If the interpolated data exceeds the reasonable threshold range (such as D90 exceeds 0-20), it will be automatically adjusted to the boundary value within the range.

[0118] The filled data is standardized. The standardization formula is:

[0119]

[0120] Among them, X is the original data, μ is the mean, σ is the standard deviation, and the data range after standardization is between -5 and +5.

[0121] In another exemplary embodiment of the present application, an effective data enhancement strategy is formulated to address the small sample problem, and a random disturbance noise addition method is selected to prevent overfitting caused by small samples.

[0122] Step 202 specifically includes:

[0123] Step 2021: adding uniform noise of different noise amplitudes to the characteristic data of the sample graphite raw materials and the finished product target data in the processed input data set respectively to obtain the disturbed input data set; adding uniform noise to the original data features to simulate the slight fluctuations of the graphite spheroidization production line equipment parameters through slight disturbances to improve the adaptability of the model.

[0124] Step 2022: Use the combined enhancement method to perform sample expansion on the disturbed input data set to obtain an expanded data set; use the combined enhancement method to increase the number of samples to simulate different operating conditions of graphite properties and equipment parameters to prevent model overfitting.

[0125] Determine the data enhancement strategy: By observing and analyzing the data of the graphite spheroidization process, we choose to generate more valid data through random perturbations and noise addition. This strategy can improve the adaptability of the model to the parameters of the graphite spheroidization production line and avoid overfitting due to insufficient samples. In particular, considering the characteristic differences between the raw material and finished product data in the graphite sample, this application uses different noise amplitudes for the volatility of different characteristics.

[0126] Using the noise adding method, for the feature X of the input data set j ,j=0,1,...,8, at its original value X j On the basis of , noise sampled from uniform distribution is added. Since the fluctuation of particle size related features is large, while the fluctuation of finished product features is small, the method of adding noise is different, which is implemented as follows:

[0127] Addition of noise related to raw material particle size: particle size characteristics usually show large fluctuations, so the feature X related to particle size distribution in the raw material data is added. j ,j=0,1,2,3, the noise amplitude is set to 10% of the standard deviation. Noise value∈ j From the uniform distribution U(-α·σ j ,α·σ j ), where σ j For feature X j The standard deviation of , α=0.1 is the noise amplitude factor.

[0128] X j ′=X j + j,j ~U(-0.1·σ j ,0.1·σ j ).

[0129] Among them, X j is the original granularity data feature, X j ′ is the particle size feature after adding noise.

[0130] Product feature noise addition: Product feature fluctuations are small, usually manifested as small deviations. Therefore, for the product feature X j ,j=4,5,6,7,8, the noise amplitude is set to 1% of the standard deviation. Noise value∈ j It also comes from the uniform distribution U(-α·σ j ,α·σ j ), but the noise amplitude factor α is set to 0.01.

[0131] X j ′=X j + j,j ~U(-0.01·σ j ,0.01·σ j ).

[0132] Among them, X j is the original product data feature, X j ′ is the feature of the finished product after adding noise.

[0133] Verify the suitability of the noise: To ensure the suitability of the noise, this application uses the following verification method.

[0134] Comparison of training error and validation error: After adding noise, the model's training error may increase slightly, but the validation error should decrease significantly, proving that noise enhances the generalization ability of the model. By comparing the training and validation errors, we can ensure that the noise amplitude is not too large, resulting in excessive perturbation of the model.

[0135] Cross-validation: Cross-validate the model under different noise amplitudes and select the noise amplitude that optimizes the model performance. Through cross-validation, you can compare the impact of noise on model performance and ensure that the appropriate noise size is selected.

[0136] By comparing cross-validation and validation errors, we can ensure that the noise amplitude is within an appropriate range. Figure 2 As shown, Figure 2 Part (a) shows the particle size D 10 Kernel density distribution map, Figure 2 Part (b) shows the particle size D 50 Kernel density distribution map, Figure 2 Part (c) shows the particle size D 90 Kernel density distribution map, Figure 2 Part (d) shows the distribution of the density of the tapped nuclei. Figure 2 Part (e) shows the distribution diagram of the relative density of the kernel. Figure 2 Part (f) shows the distribution of kernel density on a 100-mesh sieve. Figure 2 Part (g) shows the distribution diagram of kernel density below 100 mesh and above 200 mesh. Figure 2 Part (h) shows the kernel density distribution diagram of 200 mesh and 325 mesh. Figure 2 Part (i) shows the kernel density distribution map under 325 meshes. Figure 2 It can be seen that the distribution curves of the original data (i.e., the data in the input dataset) and the enhanced data (i.e., the data in the expanded dataset) are different in various feature dimensions (including: particle size distribution (D 10 , D 50 , D90 ), tap density, specific surface area, 100 mesh sieve, 100 mesh, 200 mesh, 200 mesh, 325 mesh and 325 mesh) show similar shapes and trends, indicating that data enhancement does not destroy the overall characteristic distribution; the peak position of the kernel density distribution curve remains basically consistent, indicating that the data after adding noise is still centered on the original distribution, and the addition of noise is mainly to introduce moderate randomness rather than change the main characteristic distribution of the data.

[0137] In another exemplary embodiment of the present application, step 300 specifically includes:

[0138] According to the model training mode, the expanded data set and the output data set are used to train the XGBoost model, and the trained XGBoost model is determined as the initial model for parameter prediction.

[0139] Among them, model training modes include: by device mode, by module mode and by machine mode.

[0140] The per-equipment mode is used to train each set of spherical production line equipment as a separate model.

[0141] The module mode is used to train each module of the spherical production line equipment classified by functional modules as a separate model.

[0142] The machine-based mode is used to train each type of spherical production line equipment classified by machine type as a separate model.

[0143] The user inputs the characteristic data of the target graphite raw materials and the target data of the finished products; according to the technical requirements of spherical graphite, the appropriate model training method is selected, including three training modes: by equipment, by module and by machine. The user can choose the appropriate mode according to specific needs to ensure that the model matches the actual production line application scenario.

[0144] During the model training phase, the goal of the XGBoost algorithm is to minimize the prediction error and regularization term between samples. The objective function of the XGBoost algorithm is defined as follows:

[0145]

[0146] in, is the value of the objective function at the tth iteration; n represents the total number of samples i; y i Represents the true feature value of the i-th sample; represents the model prediction value at the t-1th iteration; f t (x i ) represents the model learned in the tth iteration for the i-th sample x iThe predicted value; l() represents the loss function, which measures the true value y i and predicted values The difference between Ω(f t ) is the regularization term; C is the constant term.

[0147] In each round of iteration, the objective function is optimized using Taylor second-order expansion and converted into an approximate form to simplify the calculation. The expanded objective function is as follows:

[0148]

[0149] in, is the loss function for the model in the last iteration The first derivative of the predicted value, that is, the gradient; It is the second-order derivative of the loss function with respect to the model’s prediction value in the previous iteration, which represents the local curvature of the loss function.

[0150] Rewrite the objective function as follows, then it can be written as follows:

[0151]

[0152] XGBoost is a gradient boosting tree. The process of training the model is to find the best split of the tree. Therefore, in order to further optimize the structure of the model, it can better adapt to the feature distribution of the input data set X and the output data set Y.

[0153] make I j = {q(x i )=j}, where T is the number of leaf nodes in the tree; j is a leaf node, a sample x is mapped to a corresponding leaf node, and w j is the weight of the jth leaf node; λ is the weight regularization parameter; γ is the complexity parameter that penalizes the number of leaf nodes; G j is the sum of the first-order gradients on leaf node j, H j is the sum of the second-order gradients on leaf node j; I j is defined as the set of sample subscripts on each leaf node j. Substituting these definitions into the objective function and combining the regularization term, the objective function is:

[0154]

[0155] In order to minimize the objective function, the weight w of each leaf node needs to be j For each w j Find the partial derivative and set it to zero, and we get Substituting and simplifying, we get the final objective function:

[0156]

[0157] The form of this objective function directly reflects the two major elements of model optimization. The first is the prediction error term: Measures the contribution of each leaf node to the objective function. The larger the gradient, the smaller the error. The second is the complexity term: γT is used to control the number of leaf nodes to avoid overfitting.

[0158] In the process of building a tree, it is necessary to evaluate whether the gain of each split is large enough. The formula for the split gain is:

[0159]

[0160] Among them, G L ,G R is the sum of the gradients of the left and right child nodes, H L ,H R It is the second-order gradient sum of the left and right child nodes. The sum of the left and right child node gradients is the gradient sum of the parent node. If the split gain Gain>0, it means that the model is not optimal, and the split is performed. When the gain Gain≤0, or the number of samples of the leaf node is less than the set threshold, or the maximum tree depth is reached, the split is stopped.

[0161] In order to objectively evaluate the model performance, this application provides the following three methods as evaluation indicators: mean square error (MSE), square residual coefficient (R 2 ) and root mean square error (RMSE), the formulas are as follows:

[0162]

[0163] in, is the predicted value; i is the true value; is i The mean of .

[0164] In order to achieve accurate parameter recommendations for production lines under different raw material characteristics, this application adopts a visual interface design based on PyQt6. Users input the characteristic data of target graphite raw materials and target data of finished products through the PyQt6 visual interface, automatically remove data that does not conform to the rules, and transmit the input data to the prediction model in real time for multi-mode prediction and result export. Figure 3 The system can predict parameters of the entire process and provide accurate processing targets for the production line.

[0165] In addition to predicting the processing parameters of the spheroidization production line through the spheroidization production line model, the PyQt6 visualization interface also has the function of exporting results. After the parameter prediction is completed, it supports exporting the prediction results to Excel file format. The exported Excel file includes detailed production line parameter predictions (such as frequency settings, feed rates, etc.). This export function supports subsequent data analysis, quality traceability, and comparative analysis of historical data, providing reliable data support for production line parameter optimization for subsequent analysis and preservation by users. By analyzing the data in the exported file, users can more intuitively understand the effect of production line parameter adjustments, helping production line managers make scientific production decisions, thereby achieving efficient and precise control of graphite spheroidization production lines.

[0166] This interface is designed to provide a convenient input method for production line operators and ensure the accuracy of input data. Through this interface, users can easily input and adjust raw material and finished product data, directly connect to the prediction model, and ensure fast and accurate data transmission. This function greatly improves the adaptability of the production line to different production batches, especially when the production line faces different batches of raw materials or customer personalized needs, it can efficiently adapt to changing process requirements and maintain the stability of finished product quality.

[0167] According to the specific characteristics of the graphite spheroidization production line equipment parameters, this application further designs three training-prediction modes, providing flexible and diverse options for different production needs, thereby achieving the best configuration of production line equipment parameters. These three modes are designed to cope with different application scenarios of the production line. Users can choose the most suitable mode according to actual production needs to ensure accurate matching and optimization of production line parameters. The details are as follows:

[0168] By device mode: In this mode, the system trains the four key stages of each set of equipment (main engine, internal division, external division, and fan) as separate models, generating a total of 14 independent models, each of which is specifically designed to predict the specific parameters of a device. This mode is suitable for extremely precise parameter optimization of a single device.

[0169] Module mode: This mode divides the production line into four modules (main unit, internal division, external division, and fan) according to the functional modules. Each module contains 14 devices, which are modeled as a whole, and a total of four models are trained. This mode is suitable for the situation where the equipment parameters of the entire module are consistent. It can achieve unified parameter settings for all devices in the module, which is suitable for production lines that require high consistency and standardization.

[0170] By machine mode: In the by machine type mode, the system trains 60, 50, and 30 series devices as a class of machines, each of which contains 4, 6, and 4 devices, respectively, generating 3 main models, and each model contains 4 stages (that is, a total of 12 models). This mode is suitable for situations where the parameters of different types of devices vary greatly.

[0171] The design of multiple training-prediction modes significantly improves the flexibility of the system, enabling the graphite spheroidization production line to quickly make adaptive adjustments when facing different raw material characteristics and production requirements. Users can flexibly choose different modes to adapt to diverse production needs from fine single equipment control to modular unified control.

[0172] In another exemplary embodiment of the present application, the grid parameter search can only make a preliminary estimate of the model parameters and provide a range. With the increase of the range of XGBoost algorithm parameter values, the search time of this type of optimization method that needs to list parameter combinations will grow geometrically, and the efficiency is not high. In order to efficiently optimize the hyperparameters of the XGBoost model, to obtain the optimal hyperparameter combination and improve the model prediction accuracy, the present application uses an improved BAS (BeetleAntennae Search, Longicorn beard search) algorithm to optimize the parameters of the XGBoost model.

[0173] Step 402 specifically includes:

[0174] The search range for parameter settings in the preliminary optimization model is based on parameter prediction.

[0175] Specifically, the search space is set to limit the range of the given 6 parameters, the parameters obtained by grid search are set as initial values, and the parameter combination is set in the limited interval to form a closed interval for search. The optimization parameters include:

[0176] n_estimators (number of trees): controls the complexity of the XGBoost model, with a range of ±150% of the initial value.

[0177] max_depth (maximum tree depth): controls the complexity of model fitting to adapt to the nonlinear processing relationship of the production line. The range is set to ±30%-50% of the initial value.

[0178] learning_rate: Adjust the step size of model training to ensure that the model gradually converges within a smaller error range, and the range is set to ±50% of the initial value.

[0179] alpha (L1 regularization term): used to prevent model overfitting, the range is set to ±100% of the initial value.

[0180] lambda (L2 regularization term): used to control model complexity, the range is set to ±100% of the initial value.

[0181] subsample (random sampling ratio): Improve the model's adaptability to fluctuations in production line equipment parameters, with the range set to ±50% of the initial value.

[0182] The parameters in the parameter prediction preliminary optimization model are taken as the current hyperparameter values ​​of the centroid of the longicorn beetle.

[0183] Specifically, a basic BAS algorithm is established, and the initial hyperparameter combination position and search direction are set. In the graphite spheroidization production line, the initial hyperparameter combination is first set:

[0184] P[n_estimatorsn,max_depth,learning_rate,alpha,lambd,subsample].

[0185] This combination includes the six key hyperparameters of the model. Based on this initial combination, two new hyperparameter combinations P are defined. + and P - , representing slightly different hyperparameter combinations, which are used as the current hyperparameter values ​​for the left and right whiskers of the longhorn beetle, respectively, to explore the impact of different combinations on model performance, and set the initial step size s (1) is 10.

[0186] Determine the hyperparameter values ​​of the centroid of the longhorn beetle at the current number of iterations; wherein the parameters in the parameter prediction preliminary optimization model are used as the hyperparameter values ​​of the centroid of the longhorn beetle at the initial iteration.

[0187] The hyperparameter values ​​of the left and right whisker directions of the beetle at the current number of iterations are determined according to the hyperparameter value of the beetle's centroid and the search range at the current number of iterations; the calculation formula of the hyperparameter value is:

[0188] P + (t) =P (t) +δ·D

[0189] P - (t) =P (t) -δ·D.

[0190] Among them, P + (t) is the hyperparameter value of the left whisker direction of the longicorn under the current number of iterations; P - (t) is the hyperparameter value of the right whisker direction of the longicorn under the current number of iterations; P (t)is the hyperparameter value of the centroid of the longhorn beetle under the current number of iterations; t is the current number of iterations; δ is the adjustment amplitude, which controls the range of change of the hyperparameters each time and is obtained according to the search range; D is the random direction vector, which ensures random exploration in different dimensions to fully cover the possible hyperparameter combination space; this step provides a preliminary exploration direction for each hyperparameter dimension, which is convenient for finding the optimal combination in the model hyperparameter space.

[0191] Calculate the fitness function value of the hyperparameter value (i.e., hyperparameter combination) of the left and right whiskers of the longhorn beetle under the current number of iterations, and determine the hyperparameter value of the direction with the smallest fitness function value under the current number of iterations; the fitness function is a mean square error function.

[0192] By calculating the fitness of different hyperparameter combinations and determining the optimization direction, the mean square error (MSE) is set as the fitness function f(P), whose formula has been given above. Calculate P + and P - The fitness f(P + ) and f(P - ), by minimizing f(P), the sphericalized production line parameters predicted by the XGBoost model can be made more consistent with the actual production line data.

[0193] Dynamically adjust the step size, gradually reduce the step size as the number of iterations increases, and gradually reduce it to 1, so that the later optimization is more refined to achieve fine optimization. In order to ensure that the hyperparameter combination can reach the optimal configuration in the later adjustment stage, this application introduces a dynamic adjustment strategy for the step size, so that the step size s (t) It gradually decreases with the number of iterations.

[0194] Dynamically decay the step size at the current iteration number according to the current iteration number; the calculation formula of the step size is:

[0195]

[0196] Among them, s (t) is the step size at the current iteration number t, which is used to control the amplitude of each hyperparameter combination adjustment, s max is the maximum setting step size, s min is the minimum setting step size, s max and min Controls the ratio of large-scale exploration in the initial stage and fine-tuning in the later stage. T is the maximum number of iterations set.

[0197] Update the XGBoost hyperparameter combination. According to the optimization direction, update the hyperparameter combination of the XGBoost model so that it gradually approaches the optimal setting. The update formula is as follows:

[0198] P (t+1) =P (t) +s(t) ·sign(f(P + )-f(P - ))·D.

[0199] Among them, P (t+1) is the hyperparameter value of the centroid of the longhorn beetle after the current number of iterations; sign(f(P + )-f(P - )) controls the updated direction to ensure that the direction of each update step is consistent with the current optimal hyperparameter combination. This is a local exploration update formula, which is suitable for optimization in the early stage. When the model has not yet found a clear optimization direction, it is mainly used for exploration in a local range.

[0200] In order to accelerate the convergence to the best combination, this application introduces a global optimal guidance mechanism, which guides the next round of updates through the currently found optimal hyperparameter combination.

[0201] According to the hyperparameter value of the direction where the fitness function value is the smallest at the current number of iterations and the step size at the current number of iterations, the hyperparameter value of the centroid of the longhorn beetle at the current number of iterations is updated; the update formula of the hyperparameter value of the centroid of the longhorn beetle is:

[0202] P (t+) =P (t) +r·(P * -P (t) )+s (t) ·sign(f(P + (t )-f(P - t ))·D.

[0203] Among them, P (t+1) is the hyperparameter value of the centroid of the longhorn beetle after the current number of iterations; P * is the currently found global optimal hyperparameter value, which is the optimal solution of a historical record; r is a random number between 0 and 1, sign(f(P + )-f(P - )) is the updated direction. This is a formula combining global guidance and local exploration. Here, a global guidance term r·(P * -P (t) ), which is to use the currently found global optimal solution P * , guide the current solution P (t) Approaching the global optimum to balance exploration and utilization, so that the hyperparameter update can quickly approach the current optimal combination while preventing it from falling into the local optimum; this guidance mechanism ensures that in each round of hyperparameter optimization, the hyperparameter combination of the XGBoost model gradually moves towards the current optimal setting, thereby accelerating convergence and improving the algorithm's search efficiency.

[0204] When the termination condition is reached, the hyperparameter value of the longhorn beetle's centroid after the current iteration number is updated is used as the final optimization parameter of the preliminary optimization model for parameter prediction to obtain the parameter prediction model; otherwise, the hyperparameter value of the longhorn beetle's centroid after the current iteration number is updated is used as the hyperparameter value of the longhorn beetle's centroid at the next iteration number. After updating the iteration number, the step of determining the hyperparameter values ​​of the longhorn beetle's left whisker direction and the longhorn beetle's right whisker direction at the current iteration number is returned based on the hyperparameter value of the longhorn beetle's centroid at the current iteration number and the search range.

[0205] Among them, the termination conditions include: the current number of iterations reaches the set maximum number of iterations, the change range between the fitness function value of the hyperparameter value of the left whisker direction of the longhorn beetle at the current number of iterations and the fitness function value of the hyperparameter value of the center of mass of the longhorn beetle after the current number of iterations is less than the set threshold, or the change range between the fitness function value of the hyperparameter value of the right whisker direction of the longhorn beetle at the current number of iterations and the fitness function value of the hyperparameter value of the center of mass of the longhorn beetle after the current number of iterations is less than the set threshold.

[0206] Update the current hyperparameter value of the centroid of the longicorn according to the current hyperparameter value of the direction where the fitness function value is the smallest and the current step size;

[0207] To ensure that optimization is stopped in time after the best hyperparameter combination is obtained, this application sets the termination conditions of the BAS algorithm, including:

[0208] Reach the maximum number of iterations T (e.g., T = 100): ensure that hyperparameter optimization is completed within a reasonable computational cost.

[0209] The fitness converges to a preset threshold (such as 0.001): When the change in fitness is less than the set threshold, the optimization is stopped to ensure that the model parameters have reached the optimal value.

[0210] This termination condition can stop unnecessary iterations after finding the best hyperparameter combination of the XGBoost model, ensuring the optimization efficiency while guaranteeing the optimal prediction effect of the parameter prediction model.

[0211] This application uses an improved BAS algorithm to optimize the XGBoost model parameters, involving the number of iterations, step decay coefficient, and fitness function to accelerate the convergence process, ensure that the model achieves optimal performance on a small sample data set, and achieve accurate optimization of model parameters. Including:

[0212] Improvement of the position update mechanism: Through the typical optimal solution guidance mechanism, the position of the longhorn beetle is updated to make it approach the global optimal solution more quickly.

[0213] Dynamic adjustment of search step size: A linear decreasing strategy based on the current number of iterations is used to adjust the step size to increase the convergence speed of the algorithm.

[0214] like Figure 4 As shown, the overall design steps of the parameter prediction method applied to the graphite spheroidization production line of the present application mainly include the following aspects:

[0215] In terms of sample graphite raw material data acquisition, the finite element simulation data containing defects extracted from the specified file are divided into a test set and a training set.

[0216] In terms of XGBoost prediction model establishment, the hyperparameters of the XGBoost model are initialized through the training set, and the hyperparameters of the XGBoost model are optimized using grid search, optimized BAS algorithm and cross-validation method (5-fold cross-validation in the figure). RMSE is used as the fitness function to determine whether it is the optimal hyperparameter. If so, the optimal model is obtained and tested with the test set; if not, the optimization continues.

[0217] In terms of the prediction of spheroidization production line parameters, actual samples (i.e., target graphite raw material data) are exported and input into the prediction model to obtain the prediction results.

[0218] This application uses a small sample learning method based on a machine learning model to provide a preliminary parameter solution for the technical indicators of spherical graphite, so as to improve the accuracy of parameter prediction of the graphite spheroidization production line under small sample conditions, achieve parameter optimization of the graphite spheroidization production line under small sample conditions, and thus improve the efficiency and quality of graphite processing. Through the prediction method and system of this application, it is possible to provide different preliminary parameter solutions for the technical indicators of spherical graphite, which is particularly suitable for application scenarios with small sample data.

[0219] Based on the same inventive concept, the embodiment of the present application also provides a parameter prediction system applied to a graphite spheroidization production line for implementing the parameter prediction method applied to a graphite spheroidization production line involved above. The implementation scheme for solving the problem provided by the system is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more embodiments of the parameter prediction system applied to a graphite spheroidization production line provided below can refer to the limitations of the parameter prediction method applied to a graphite spheroidization production line above, and will not be repeated here.

[0220] In an exemplary embodiment, a parameter prediction system for a graphite spheroidization production line is provided, comprising:

[0221] The test data acquisition module is used to acquire characteristic data of the target graphite raw material and finished product target data of the target graphite raw material; the finished product target data includes particle size distribution data, tap density data and specific surface area during the spheroidization process of the graphite raw material.

[0222] The model determination module is used to determine the parameter prediction model.

[0223] The processing parameter prediction module is used to input the characteristic data of the target graphite raw material and the finished product target data of the target graphite raw material into the parameter prediction model to determine the spheroidization production line parameters of the target graphite raw material; the spheroidization production line parameters include the feeding frequency of the spheroidization production line equipment during the spheroidization process of the graphite raw material.

[0224] The model determination module specifically includes:

[0225] The sample data set acquisition unit is used to acquire an input data set and an output data set; the input data set includes: characteristic data of the sample graphite raw material and finished product target data; the output data set includes: spheroidization production line parameters of the sample graphite raw material.

[0226] The sample data expansion unit is used to expand the input data set by adopting a small sample data expansion and enhancement method to obtain an expanded data set.

[0227] The model training unit is used to train the XGBoost model using the augmented data set and the output data set, and determine the trained XGBoost model as the initial model for parameter prediction.

[0228] The model parameter optimization unit is used to optimize and adjust the parameters in the parameter prediction initial model by using the grid search method and the longicorn beetle whisker search algorithm, and determine the optimized and adjusted parameter prediction initial model as the parameter prediction model.

[0229] The system of this application adopts a modular design, modularizing data reading, sample enhancement, model training, hyperparameter optimization and parameter prediction functions, and is flexible and scalable to meet the diverse needs of different graphite properties and equipment requirements.

[0230] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Figure 5As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. Among them, the processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store parameter prediction data applied to the graphite spheroidization production line. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a parameter prediction method applied to the graphite spheroidization production line is implemented.

[0231] Those skilled in the art will understand that Figure 5 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components. In an exemplary embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.

[0232] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0233] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0234] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0235] The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. The non-relational database may include a distributed database based on blockchain, etc., but is not limited thereto. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but is not limited thereto.

[0236] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0237] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A parameter prediction method applied to a graphite spheroidization production line, characterized in that: The parameter prediction method applied to the graphite spheroidization production line includes: Acquire characteristic data of target graphite raw material and target finished product data of target graphite raw material; the target finished product data includes particle size distribution data, tap density data and specific surface area of ​​graphite raw material during spheroidization; Inputting characteristic data of the target graphite raw material and target finished product data of the target graphite raw material into a parameter prediction model to determine the spheroidization production line parameters of the target graphite raw material; the spheroidization production line parameters include the feeding frequency of the spheroidization production line equipment during the spheroidization process of the graphite raw material; The method for establishing the parameter prediction model specifically includes: Acquire an input data set and an output data set; the input data set includes: characteristic data of sample graphite raw materials and finished product target data; the output data set includes: spheroidization production line parameters of sample graphite raw materials; The input data set is expanded using a small sample data expansion and enhancement method to obtain an expanded data set; The expanded data set and the output data set are used to train an XGBoost model, and the trained XGBoost model is determined as an initial model for parameter prediction; The parameters in the parameter prediction initial model are optimized and adjusted by using a grid search method and a longicorn beetle whisker search algorithm, and the optimized and adjusted parameter prediction initial model is determined as the parameter prediction model.

2. The parameter prediction method for a graphite spheroidization production line according to claim 1, characterized in that: The input data set is expanded by using a small sample data expansion and enhancement method to obtain an expanded data set, specifically including: Performing missing value processing, data cleaning and standardization on the input data set to obtain a processed input data set; The processed input data set is expanded using a small sample data expansion and enhancement method to obtain an expanded data set.

3. The parameter prediction method for a graphite spheroidization production line according to claim 1, characterized in that: The parameters in the parameter prediction initial model are optimized and adjusted by using a grid search method and a longicorn beetle whisker search algorithm, and the optimized and adjusted parameter prediction initial model is determined as the parameter prediction model, specifically including: The parameters in the parameter prediction initial model are preliminarily optimized by using a grid search method to obtain the parameter prediction preliminary optimization model; The parameters in the parameter prediction preliminary optimization model are finally optimized by using a longicorn beard search algorithm to obtain the parameter prediction model.

4. The parameter prediction method for a graphite spheroidization production line according to claim 2, characterized in that: The input data set is subjected to missing value processing, data cleaning and standardization processing to obtain a processed input data set, specifically including: Detecting and marking the characteristic data of the sample graphite raw material that is missing in the input data set due to the lack of corresponding finished product target sample data; and treating the missing characteristic data of the sample graphite raw material as the missing characteristic data; Filling the missing feature data using Lagrange interpolation method to obtain filled feature data; Performing statistical analysis on the filled feature data to ensure that the filled feature data conforms to the actual physical properties, and obtaining the feature data after statistical analysis; and using the feature data after statistical analysis and the input data set as a complete input data set; The complete input data set is standardized so that the data in the complete input data set are scaled to the same magnitude, thereby obtaining a processed input data set.

5. The parameter prediction method for a graphite spheroidization production line according to claim 2, characterized in that: The processed input data set is expanded using a small sample data expansion and enhancement method to obtain an expanded data set, which specifically includes: adding uniform noises of different noise amplitudes to the characteristic data of the sample graphite raw material and the finished product target data in the processed input data set respectively, to obtain the disturbed input data set; The combined enhancement method is used to perform sample expansion on the perturbed input data set to obtain an expanded data set.

6. The parameter prediction method for a graphite spheroidization production line according to claim 1, characterized in that: The XGBoost model is trained using the expanded data set and the output data set, and the trained XGBoost model is determined as an initial model for parameter prediction, specifically including: According to the model training mode, the XGBoost model is trained using the expanded data set and the output data set, and the trained XGBoost model is determined as the parameter prediction initial model; The model training modes include: by device mode, by module mode and by machine mode; The equipment-based mode is used to train each set of spherical production line equipment as a separate model; The module-based mode is used to train each module of the spherical production line equipment classified by functional modules as a separate model; The machine-based mode is used to train each type of spherical production line equipment classified by machine type as a separate model.

7. The parameter prediction method for a graphite spheroidization production line according to claim 3, characterized in that: The parameters in the parameter prediction preliminary optimization model are finally optimized by using the longicorn beard search algorithm to obtain the parameter prediction model, which specifically includes: Setting a search range based on the parameters in the parameter prediction preliminary optimization model; Determine the hyperparameter value of the centroid of the longhorn beetle at the current number of iterations; wherein the parameters in the parameter prediction preliminary optimization model are used as the hyperparameter value of the centroid of the longhorn beetle at the initial iteration; Determine the hyperparameter values ​​of the left whisker direction and the right whisker direction of the longicorn beetle at the current number of iterations according to the hyperparameter value of the longicorn beetle's centroid at the current number of iterations and the search range; Calculate the fitness function values ​​of the hyperparameter values ​​of the left whisker direction and the right whisker direction of the longhorn beetle under the current number of iterations, and determine the hyperparameter value of the direction where the fitness function value is the smallest under the current number of iterations; Dynamically decay the step size at the current iteration number according to the current iteration number; the calculation formula of the step size is: Among them, s (t) is the step size at the current iteration number t, s max is the maximum setting step size, s min is the minimum set step size, T is the maximum number of iterations set; Update the hyperparameter value of the centroid of the longicorn at the current number of iterations according to the hyperparameter value of the direction where the fitness function value is the smallest at the current number of iterations and the step size at the current number of iterations; When the termination condition is reached, the hyperparameter value of the longhorn beetle mass center updated at the current iteration number is used as the final optimization parameter of the parameter prediction preliminary optimization model to obtain the parameter prediction model; otherwise, the hyperparameter value of the longhorn beetle mass center updated at the current iteration number is used as the hyperparameter value of the longhorn beetle mass center at the next iteration number, and after updating the iteration number, the step of determining the hyperparameter values ​​of the longhorn beetle left whisker direction and the longhorn beetle right whisker direction at the current iteration number according to the hyperparameter value of the longhorn beetle mass center at the current iteration number and the search range is returned; Among them, the termination conditions include: the current number of iterations reaches the set maximum number of iterations, the change range between the fitness function value of the hyperparameter value of the left whisker direction of the longhorn beetle at the current number of iterations and the fitness function value of the hyperparameter value of the center of mass of the longhorn beetle after the current number of iterations is less than the set threshold, or the change range between the fitness function value of the hyperparameter value of the right whisker direction of the longhorn beetle at the current number of iterations and the fitness function value of the hyperparameter value of the center of mass of the longhorn beetle after the current number of iterations is less than the set threshold.

8. A parameter prediction system applied to a graphite spheroidization production line, characterized in that: The parameter prediction system applied to the graphite spheroidization production line includes: The test data acquisition module is used to acquire characteristic data of the target graphite raw material and finished product target data of the target graphite raw material; the finished product target data includes particle size distribution data, tap density data and specific surface area during the spheroidization process of the graphite raw material; A model determination module, used for determining a parameter prediction model; A processing parameter prediction module is used to input characteristic data of the target graphite raw material and target finished product data of the target graphite raw material into a parameter prediction model to determine the spheroidization production line parameters of the target graphite raw material; the spheroidization production line parameters include the feeding frequency of the spheroidization production line equipment during the spheroidization process of the graphite raw material; Wherein, the model determination module specifically includes: A sample data set acquisition unit is used to acquire an input data set and an output data set; the input data set includes: characteristic data of the sample graphite raw material and target data of the finished product; the output data set includes: spheroidization production line parameters of the sample graphite raw material; A sample data expansion unit, used to expand the input data set by using a small sample data expansion and enhancement method to obtain an expanded data set; A model training unit, used to train an XGBoost model using the expanded data set and the output data set, and determine the trained XGBoost model as an initial model for parameter prediction; The model parameter optimization unit is used to optimize and adjust the parameters in the parameter prediction initial model by using a grid search method and a longicorn beetle whisker search algorithm, and determine the optimized and adjusted parameter prediction initial model as the parameter prediction model.

9. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the parameter prediction method for a graphite spheroidization production line 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 computer program is executed by a processor, the parameter prediction method for a graphite spheroidization production line described in any one of claims 1 to 7 is implemented.

Images