Machine learning based natural graphite spherical production line parameter prediction method and device

Abstract

This application relates to the field of deep processing of natural graphite, and provides a method and apparatus for predicting parameters of a natural graphite spherical production line based on machine learning. This addresses the challenges of experience-dependent adjustments, high modeling dimensionality due to multiple equipment connected in series, and limited sample sizes for new product debugging. The method includes: inputting production indicators and raw material characteristics into a pre-built prediction model to obtain stage operating parameters; compressing the input dimensionality by aggregating and feature-encoding historical parameters of the natural graphite spherical production line equipment in stages; integrating meta-learning and fully automated machine learning frameworks to construct a robust prediction model adaptable to varying data sizes from small samples to sufficient data; performing reverse parameter recommendation based on the prediction model, and combining production information from the Manufacturing Execution System (MES) and Enterprise Resource Planning (ERP) systems to compensate for and precisely distribute the recommended parameters in stages. This application can adapt to complex processes, reduce trial-and-error costs, and ensure stable production.

Description

Technical Field

[0001] This application relates to the field of deep processing of natural graphite, specifically a method and apparatus for predicting parameters of a natural graphite spherical production line based on machine learning. Background Technology

[0002] In the lithium-ion battery manufacturing process, the spheroidization of natural graphite is one of the key steps. During spheroidization, the settings of equipment parameters (such as the main unit, internal separator, external separator, and fan frequency) directly affect the particle size distribution of the product (e.g., Dp). 10 D 50 D 90 Key indicators include tap density and other parameters. Currently, natural graphite spherical production lines consist of a continuous processing flow consisting of multiple machines connected in series. The adjustment and optimization of equipment parameters heavily rely on the operator's personal experience and repeated trial and error. This traditional method not only lacks scientific and unified data, leading to large fluctuations and poor consistency in product quality between different batches, but also results in lengthy and costly debugging processes when developing new products or responding to changes in raw materials, consuming large amounts of raw materials, supplies, and electricity.

[0003] While machine learning methods offer new insights for process optimization, their direct application in spherical graphite production presents significant challenges: First, the complete production process involves numerous devices, leading to extremely high-dimensional input parameters and hindering reliable model creation. Second, high-quality training data is limited in industrial production, especially for new products or operating conditions, where small sample sizes are a typical problem, making conventional machine learning methods prone to overfitting or poor performance. Third, the simplistic model structure makes it difficult to accurately depict the complex nonlinear mapping relationships between parameters and indicators in spherical graphite production, resulting in insufficient prediction accuracy and generalization ability. Therefore, a method is urgently needed that can adapt to complex processes and stably and accurately achieve intelligent optimization and recommendation of equipment parameters even under small sample conditions.

[0004] This section is intended to provide background or context for the embodiments of the invention set forth in the claims. The description herein is not an admission that it is prior art simply because it is included in this section. Summary of the Invention

[0005] To address the problems in the existing technology, this application provides a method and apparatus for predicting parameters of a natural graphite spherical production line based on machine learning. This method can adapt to the complex process of graphite spherical production and can still stably and accurately achieve intelligent optimization and recommendation of equipment parameters under small sample conditions.

[0006] To solve the above-mentioned technical problems, this application provides the following technical solution: In a first aspect, this application provides a method for predicting parameters of a natural graphite spherical production line based on machine learning, including: The obtained spherical graphite production indicators and raw material characteristics are input into a pre-built spherical production line parameter prediction model to obtain the stage operation parameters of the spherical production line. Based on the production plan information collected from the Manufacturing Execution System and the equipment macro information collected from the Enterprise Resource Planning System, the stage operation parameters of the spherical production line are adjusted compensatorily and distributed to each piece of equipment in the natural graphite spherical production line in stages.

[0007] Furthermore, the steps of pre-constructing a parameter prediction model for a spherical production line include: The historical parameters of the natural graphite spherical production line equipment are feature-encoded to obtain historical operation code features; The historical product indicators and the historical operation coding features are input into the machine learning model to obtain the parameter prediction model of the spherical production line; wherein, the machine learning model is implemented using a fully automated machine learning framework combined with meta-learning technology.

[0008] Furthermore, the historical parameters of the natural graphite spherical production line equipment include the operating parameters of each machine at each stage; the stages include the crushing stage, the crushing and shaping stage, and the shaping stage; the operating parameters of each machine at each stage include the main machine frequency, internal distribution frequency, external distribution frequency, and fan frequency of each machine at each stage; the feature encoding of the historical parameters of the natural graphite spherical production line equipment to obtain historical operating code features includes: The stage host frequency, stage internal frequency, stage external frequency, and stage fan frequency are determined based on the host frequency, internal frequency, external frequency, and fan frequency of each machine in the stage. The obtained feed parameters, raw material parameters, intermediate product parameters, stage host frequency, intra-stage sub-frequency, extra-stage sub-frequency, and stage fan frequency are combined to obtain a multi-dimensional array; The multidimensional array is feature-encoded to obtain the historical running encoding features.

[0009] Further, determining the stage host frequency, stage internal frequency, stage external frequency, and stage fan frequency corresponding to each stage based on the host frequency, internal frequency, external frequency, and fan frequency of each machine in the stage includes: The arithmetic mean of the host frequencies of each machine in the said phase is determined as the host frequency of the said phase; The arithmetic mean of the internal frequencies of each machine in the said phase is determined as the internal frequency of the said phase; The arithmetic mean of the external frequencies of each machine in the said stage is determined as the external frequency of the said stage; The arithmetic mean of the fan frequencies of each machine in the stated stage is determined as the fan frequency of the stated stage.

[0010] Further, the step of performing feature encoding on the multidimensional array to obtain the historical execution encoding features includes: Determine the high and low level values ​​corresponding to the stage host frequency, intra-stage sub-frequency, extra-stage sub-frequency, and stage fan frequency in the multidimensional array, respectively. Based on the high and low level values ​​corresponding to the stage host frequency, the stage sub-frequency, the stage sub-frequency, and the stage fan frequency, respectively, determine the center value and half-pitch corresponding to the stage host frequency, the stage sub-frequency, the stage sub-frequency, and the stage fan frequency. Based on the center value and half-spacing corresponding to the stage host frequency, the stage sub-frequency, the stage external sub-frequency, and the stage fan frequency, feature encoding is performed on the stage host frequency, the stage sub-frequency, the stage external sub-frequency, and the stage fan frequency to obtain the historical operation encoding features; wherein, the historical operation encoding features include the encoding values ​​corresponding to the stage host frequency, the stage sub-frequency, the stage external sub-frequency, and the stage fan frequency.

[0011] Further, the step of inputting historical product indicators and historical operational coding features into a machine learning model to obtain the parameter prediction model for the spherical production line includes: If the number of historical product indicators and corresponding historical operation coding features does not reach the sample size threshold, a meta-learning training task set is constructed based on historical multi-task data; a meta-learning algorithm is used to perform meta-training on the meta-learning training task set to obtain a shared base learner; the shared base learner is fine-tuned using the historical product indicators and corresponding historical operation coding features; wherein, if the number of historical product indicators and corresponding historical operation coding features never reaches the sample size threshold, the fine-tuned shared base learner is used as the parameter prediction model for the spherical production line. If the number of historical product indicators and corresponding historical operation coding features has reached the sample size threshold, the fully automatic machine learning framework is initialized, and multiple base learners are automatically selected from the preset algorithm space to construct a prediction model candidate set. The base learners in the prediction model candidate set are then used for training. Based on the bagging method, the historical product indicators and historical operation coding features are sampled with replacement to obtain each training data subset. Each training data subset is input into the multiple base learners in the prediction model candidate set for training to obtain multiple candidate models. Based on the prediction outputs of the multiple candidate models, a secondary learner is trained and fused to generate the spherical production line parameter prediction model.

[0012] Further, the prediction outputs based on the multiple candidate models are fused by training a secondary learner to generate the spherical production line parameter prediction model, including: A K-fold cross-validation strategy is employed to obtain the prediction results of each candidate model on each fold validation set among the multiple candidate models, and to calculate their respective performance metrics; the performance metrics include the coefficient of determination and / or root mean square error. Based on the performance metrics, one or more candidate models are selected from the plurality of candidate models for integration, forming a subset of models to be integrated; All prediction results generated by each candidate model in the subset of models to be integrated in K-fold cross-validation are concatenated by sample to form a secondary training feature set, and the corresponding true target value is obtained. A secondary learner is trained using the secondary training feature set and the true target value; The subset of models to be integrated is combined with the secondary learner to form a stacked integrated model, which serves as the parameter prediction model for the spherical production line.

[0013] Furthermore, the aforementioned machine learning-based parameter prediction method for a natural graphite spherical production line also includes: Real-time acquisition of equipment operating parameters from the equipment's programmable logic controller; The parameter prediction model for the spherical production line is optimized using real-time collected equipment operating parameters.

[0014] Furthermore, the step of compensatoryly adjusting the stage operating parameters of the spherical production line based on production plan information collected from the Manufacturing Execution System and equipment macro-information collected from the Enterprise Resource Planning System, and distributing these adjustments in stages to each piece of equipment in the natural graphite spherical production line, includes: The performance degradation coefficient of each piece of equipment is determined based on the production plan information and the macroscopic information of the equipment. The operating parameters of the spherical production line are adjusted in a compensatory manner based on the performance degradation coefficient. The compensated adjustment of the spherical production line stage operating parameters is distributed to each piece of equipment in the natural graphite spherical production line according to the current production stage.

[0015] Secondly, this application provides a parameter prediction device for a natural graphite spherical production line based on machine learning, comprising: The model prediction unit is used to input the acquired spherical graphite production indicators and raw material characteristics into a pre-built spherical production line parameter prediction model to obtain the stage operation parameters of the spherical production line. The parameter distribution unit is used to distribute the stage operation parameters of the spherical production line to each piece of equipment in the natural graphite spherical production line in stages, based on the production plan information collected from the manufacturing execution system and the equipment macro information collected from the enterprise resource planning system.

[0016] Thirdly, this application provides an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the machine learning-based method for predicting parameters of a natural graphite spherical production line.

[0017] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the machine learning-based method for predicting parameters of a natural graphite spherical production line.

[0018] Fifthly, this application provides a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of the machine learning-based method for predicting parameters of a natural graphite spherical production line.

[0019] To address the problems in existing technologies, this application provides a machine learning-based method and apparatus for predicting parameters in a natural graphite spherical production line. Through an innovative process stage aggregation strategy, it simplifies high-dimensional, continuous equipment parameters into stage features with clear physical meaning, effectively solving the curse of dimensionality problem in modeling multi-machine serial processes. Furthermore, the system deeply integrates an automated machine learning framework and meta-learning technology, enabling end-to-end automatic construction of a high-precision prediction model. It also significantly improves the robustness and generalization ability of the prediction by utilizing stacking integration and repeated bagging strategies, allowing it to quickly adapt and provide reliable predictions even with a small number of new samples. The core advantage of this invention lies in achieving reverse intelligent recommendation from product goals to equipment parameters, transforming the traditional parameter debugging process, which requires extensive physical testing, into efficient digital simulation optimization, reducing trial-and-error costs. Ultimately, through deep integration with Manufacturing Execution Systems (MES) and Enterprise Resource Planning (ERP) systems, the system dynamically transforms static parameter recommendations into personalized settings that sense the real-time health status of equipment. This enables refined and differentiated adjustments to the parameters of various equipment within the production line based on equipment wear and tear. Consequently, while ensuring overall process stability, it effectively compensates for equipment performance degradation, continuously improving product quality consistency and production efficiency. Overall, this invention not only significantly enhances the intelligence level and product quality stability of spherical graphite production but also provides a replicable, efficient, and reliable technical solution for the digital transformation and upgrading of process industries. Attached Figure Description

[0020] 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 some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is an overall logic diagram of the parameter prediction method for a natural graphite spherical production line based on machine learning in the embodiments of this application; Figure 2 This is a schematic diagram of the equipment operating parameters collected in real time based on a programmable logic controller (PLC) in an embodiment of this application. Figure 3 This is a graph showing the online particle size distribution detection results of the laser particle size online detection platform mounted on the production line in this embodiment of the application. Figure 4 This is one of the flowcharts for a machine learning-based parameter prediction method for a natural graphite spherical production line in this application. Figure 5 This is a flowchart illustrating the construction of a parameter prediction model for a spherical production line in an embodiment of this application; Figure 6 This is a flowchart illustrating the historical execution encoding characteristics obtained in this application embodiment; Figure 7 This is a flowchart illustrating the determination of the stage host frequency, the intra-stage sub-frequency, the extra-stage sub-frequency, and the stage fan frequency in the embodiments of this application. Figure 8 This is a detailed flowchart illustrating the process of obtaining historical execution coding features in this application embodiment; Figure 9 This is a flowchart illustrating the construction of a parameter prediction model for a spherical production line in an embodiment of this application; Figure 10 This is a detailed flowchart of the parameter prediction model for the spherical production line obtained in the embodiments of this application; Figure 11 This is the second flowchart of the parameter prediction method for a natural graphite spherical production line based on machine learning in the embodiments of this application; Figure 12 This is a flowchart illustrating the phased issuance of operating parameters for the spherical production line in this embodiment of the application. Figure 13 This is a structural diagram of the parameter prediction device for a natural graphite spherical production line based on machine learning in an embodiment of this application; Figure 14 This is a schematic diagram of the structure of the electronic device in the embodiments of this application. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings. Here, the illustrative embodiments of the present invention and their descriptions are used to explain the present invention, but are not intended to limit the present invention.

[0023] The information collected in the technical solution of this application is information and data authorized by the user or fully authorized by all parties. The collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data all comply with the relevant laws, regulations and standards of the relevant countries and regions, necessary confidentiality measures have been taken, and they do not violate public order and good morals. Corresponding operation portals are provided for users to choose to authorize or refuse.

[0024] Provide users with corresponding operation entry points, allowing them to choose to agree to or reject the automated decision results; if the user chooses to reject, the process will proceed to the expert decision-making process.

[0025] This invention provides a method and apparatus for predicting parameters in a natural graphite spherical production line based on machine learning. (See also...) Figure 1 The system constructs a multi-layered intelligent decision-making system: at the infrastructure layer, it provides elastic computing power through a combination of cloud platform and edge computing, and utilizes 5G and industrial communication protocols to achieve real-time data transmission; the data layer integrates raw material and product databases, historical process databases, real-time production databases, and equipment status databases to form comprehensive data support; the algorithm layer, based on feature encoding and standardized processing, adopts a dual-track modeling strategy combining fully automated machine learning and meta-learning, and constructs a high-precision prediction model by combining stacked integration and bagging methods, and achieves reverse parameter recommendation through multi-objective optimization; the application layer deeply integrates the intelligent decision-making platform with the manufacturing execution system and enterprise resource planning system to achieve coordination of production planning and equipment information; finally, the intelligent operation and decision-making center coordinates the four core functions of intelligent parameter recommendation engine, production process monitoring, closed-loop optimization and learning, and collaborative optimization and visualization, forming a fully intelligent closed loop from order receipt to parameter issuance, from real-time monitoring to model optimization. This method can adapt to the complex process of graphite spherical production and can still stably and accurately achieve intelligent optimization and recommendation of equipment parameters under small sample conditions.

[0026] In one embodiment, see Figure 4 To adapt to the complex process of graphite spherical production and to achieve stable and accurate intelligent optimization and recommendation of equipment parameters even under small sample conditions, this application provides a machine learning-based parameter prediction method for natural graphite spherical production lines, including: S101: Input the obtained spherical graphite production indicators and raw material characteristics into the pre-constructed spherical production line parameter prediction model to obtain the stage operation parameters of the spherical production line; wherein, the spherical production line parameter prediction model is constructed by feature encoding the historical parameters of the natural graphite spherical production line equipment and using a fully automated machine learning framework combined with meta-learning technology. S102: Based on the production plan information collected from the Manufacturing Execution System and the equipment macro information collected from the Enterprise Resource Planning System, the stage operation parameters of the spherical production line are adjusted compensatorily and distributed to each piece of equipment in the natural graphite spherical production line in stages.

[0027] It is understandable that in the process of spherical graphite production, the obtained spherical graphite production indicators (product quality specifications, such as the particle size distribution of the target product (D)) are used to determine the spherical graphite production indicators. 10 D 50 D 90 By inputting the tap density, specific surface area range, and raw material characteristics (referring to the initial characteristics of flake graphite used in processing spherical graphite, such as fineness, moisture content, fixed carbon content, etc.) into a pre-built parameter prediction model for the spherical production line, the stage operation parameters of the spherical production line can be obtained. Based on these stage operation parameters, the equipment parameters corresponding to each machine in the spherical production line (see [reference]) are then determined to achieve the spherical graphite production targets. Figure 2 (As shown).

[0028] The term "stages" can be understood as follows: Based on the physical mechanism of natural graphite spheroidization and the functional characteristics of the equipment, the production process of the spheroidizing production line is divided into three key stages: The first stage is the crushing section, mainly involving 80 / 60 series spheroidizing main units, which complete the primary crushing and size reduction of the material; the second stage is the crushing and shaping section, mainly composed of 50 series spheroidizing main units, which have both crushing and spheroidizing functions; the third stage is the shaping section, mainly composed of 30 series spheroidizing main units, which are primarily responsible for the final spheroidization of the material. The term "each piece of equipment" refers to all the equipment involved in the material crushing, shaping, grading, and conveying stages of the natural graphite spheroidizing production line.

[0029] This predictive model is constructed based on historical operational coding features obtained by feature encoding historical parameters of equipment in a natural graphite spherical production line. As an intelligent decision engine, this model can deduce and recommend operating parameters (also known as equipment parameters) for each machine on the complete production line, starting from spherical graphite production indicators and raw material characteristics. Its input is a clearly defined product quality specification target (such as the particle size distribution of the target product including D...). 10 D 50 D 90The parameters (such as tap density, specific surface area range) and the initial characteristics of the raw material flake graphite (such as fineness, moisture content, and fixed carbon content) are used to output a complete parameter setting scheme for the spheroidizing production line, namely, the recommended values ​​for the main frequency, internal distribution frequency, external distribution frequency, and fan frequency of each device from the first feeding spheroidizing device to the last spheroidizing device.

[0030] It should be noted that the key process system of the natural graphite spherical production line consists of core equipment such as an airflow vortex micronizer (referred to as the main unit), an internal classifier (referred to as the internal classifier), an external classifier (referred to as the external classifier), and a Roots blower (referred to as the blower). To achieve the desired spheroidization effect, the equipment parameters listed in the table need to be precisely controlled. By adjusting the set frequency of the motor frequency converter, precise control of the equipment speed and flow rate can be achieved. Specifically, the correlation is as follows:

[0031] The actual physical meanings of the main unit frequency, internal sub-frequency, external sub-frequency, and fan frequency are respectively the rotation speed of the crushing or shaping disc of the natural graphite spheroidizing main unit, the rotation speed of the internal grading impeller of the main unit, and the rotation speed of the external grading impeller; the fan frequency includes the flow rate of the Roots blower.

[0032] The specific implementation steps of the method provided in this application are as follows: (1) Data acquisition: Production planning information, such as output, yield, and energy consumption during the production process, is collected from the Manufacturing Execution System (MES); macro-level equipment information, such as equipment status and maintenance records, is collected from the Enterprise Resource Planning (ERP) system.

[0033] (2) Feature engineering: The production process of the spherical production line is divided into the above three key stages; for each stage, the historical parameters of the natural graphite spherical production line equipment are processed by feature coding, key features are extracted and coded to generate historical operation coding features.

[0034] (3) Model construction: A parameter prediction model for the spherical production line is constructed based on the historical operation coding features. That is, when the sample size is small, a meta-learning track is used for training and optimization. When the sample size is sufficient, a fully automated machine learning track is used, and the method provided in this application can also achieve better training results.

[0035] (4) Parameter prediction: Input the obtained spherical graphite production indicators and raw material characteristics into the prediction model to obtain the stage operation parameters of the spherical production line, such as the recommended values ​​of the main frequency, internal distribution frequency, external distribution frequency and fan frequency corresponding to each stage.

[0036] (5) Parameter distribution: The predicted stage operation parameters are distributed to each piece of equipment in the natural graphite spherical production line in stages to achieve automated control and optimization. For example, the recommended values ​​of the main unit frequency, internal distribution frequency, external distribution frequency and fan frequency corresponding to the first stage are distributed to each machine corresponding to the first stage.

[0037] (5) Real-time monitoring: The production process is monitored in real time through the MES system and ERP system, and parameters are adjusted in a timely manner to ensure production efficiency and product quality.

[0038] (6) Data feedback: Feedback the actual operating parameters in the production process to the prediction model to continuously optimize the model performance and prediction accuracy.

[0039] The method provided in this application achieves seamless integration of production planning and equipment information by combining MES and ERP systems, thereby improving the intelligence level and production efficiency of the spherical graphite production process. Simultaneously, through historical parameter feature engineering and machine learning models, it enables accurate prediction and optimized control of the production process.

[0040] Table 1 shows the daily production report for spherical graphite (MES system). SG represents spherical graphite, and XX represents specific specifications, such as SG17 / SG10, etc.

[0041] Table 1

[0042] As described above, the machine learning-based parameter prediction method for a natural graphite spherical production line provided in this application simplifies high-dimensional, continuous equipment parameters into stage features with clear physical meaning through an innovative process stage aggregation strategy, effectively solving the curse of dimensionality problem in modeling multi-machine serial processes. Based on this, the system deeply integrates an automated machine learning framework and meta-learning technology, enabling end-to-end automatic construction of a high-precision prediction model. Furthermore, it significantly improves the robustness and generalization ability of the prediction by utilizing stacking integration and repeated bagging strategies, allowing it to quickly adapt and provide reliable predictions even when faced with a small number of new samples. The core advantage of this invention lies in achieving reverse intelligent recommendation from product goals to equipment parameters, transforming the traditional parameter debugging process, which requires extensive physical testing, into efficient digital simulation optimization, thus reducing trial-and-error costs. Ultimately, through deep integration with Manufacturing Execution Systems (MES) and Enterprise Resource Planning (ERP) systems, the system dynamically transforms static parameter recommendations into personalized settings that sense the real-time health status of equipment. This enables refined and differentiated adjustments to the parameters of various equipment within the production line based on equipment wear and tear. Consequently, while ensuring overall process stability, it effectively compensates for equipment performance degradation, continuously improving product quality consistency and production efficiency. Overall, this invention not only significantly enhances the intelligence level and product quality stability of spherical graphite production but also provides a replicable, efficient, and reliable technical solution for the digital transformation and upgrading of process industries.

[0043] In one embodiment, see Figure 5 The steps for pre-constructing a parameter prediction model for a spherical production line include: S201: Perform feature encoding on the historical parameters of the natural graphite spherical production line equipment to obtain historical operation code features; for specific steps, please refer to steps S301 to S303. S202: Input the historical product indicators and the historical operation coding features into the machine learning model to obtain the parameter prediction model of the spherical production line; wherein, the machine learning model adopts a fully automated machine learning framework combined with meta-learning technology. For specific steps, please refer to steps S601 to S602.

[0044] In one embodiment, see Figure 6 The historical parameters of the natural graphite spherical production line equipment include the operating parameters of each machine at each stage; the stages include the crushing stage, the crushing and shaping stage, and the shaping stage; the operating parameters of each machine at each stage include the main machine frequency, internal distribution frequency, external distribution frequency, and fan frequency of each machine at each stage; the historical operating parameters of the natural graphite spherical production line equipment are feature-encoded to obtain historical operating code features, including: S301: Determine the stage host frequency, stage internal frequency, stage external frequency, and stage fan frequency corresponding to each machine in the stage based on the host frequency, internal frequency, external frequency, and fan frequency of each machine in the stage. S302: Combine the obtained feed parameters, raw material parameters, intermediate product parameters, stage host frequency, stage sub-frequency, stage sub-frequency, and stage fan frequency to obtain a multi-dimensional array; S303: Perform feature encoding on the multidimensional array to obtain the historical running encoding features.

[0045] It is understandable that steps S301 to S303 are processes for preprocessing training data to build a parameter prediction model for the spherical production line. The data involved are all historical data.

[0046] First, let me explain the process of dividing the process into stages and constructing features.

[0047] As mentioned earlier, based on the physical mechanism of natural graphite spheroidization and the functional characteristics of the equipment, the spheroidization production line process is divided into three key stages. The first stage is the crushing section, mainly involving the 80 / 60 series mainframes, which completes the primary crushing and particle size reduction of the material; the second stage is the crushing and shaping section, mainly composed of the 50 series mainframes, which have both crushing and spheroidizing functions; the third stage is the shaping section, mainly composed of the 30 series mainframes, which is primarily responsible for the final spheroidization treatment of the material.

[0048] To address the issue of excessively high model input dimensionality (factors) due to the large number of devices, this application employs an innovative method of stage parameter aggregation to significantly reduce model complexity and dimensionality. Specifically, for all machines within each process stage, their operating parameters are no longer treated as independent variables. Instead, the arithmetic mean of similar parameters within that stage is calculated, thereby aggregating parameters from multiple machines (sets) into stage-level features. For example, for a crushing section containing four machines, the host frequency values ​​of all machines in that section are averaged to obtain a new aggregated feature named "Crushing Section Host Frequency." Similarly, four core operating features can be extracted from each stage: host frequency, internal distribution frequency, external distribution frequency, and fan frequency. The "feed frequency," which exists only in the first machine of the crushing section, is retained as an independent feature, "Feed Frequency."

[0049] In other words, assuming a spherical production line has N machines, the original model needs to consider four operating parameters for each machine, totaling 4N input features. After three-stage aggregation, regardless of the actual number of machines N, the input features are compressed to 13-14. That is, the characteristic number is: Three stages × four operating characteristics (main unit, internal separation, external separation, fan) + raw material (intermediate product) characteristics + one feeding parameter (required for the crushing stage) This process fundamentally solves the "curse of dimensionality" problem, making subsequent machine learning modeling feasible and stable.

[0050] As can be seen from the above description, the machine learning-based parameter prediction method for natural graphite spherical production lines provided in this application can perform feature encoding on the historical parameters of the natural graphite spherical production line equipment to obtain historical operation code features.

[0051] In one embodiment, see Figure 7 The step of determining the stage host frequency, stage internal frequency, stage external frequency, and stage fan frequency corresponding to each machine in the stage, based on the host frequency, internal frequency, external frequency, and fan frequency of each machine in the stage, includes: S401: Determine the arithmetic mean of the host frequencies of each machine in the stage as the host frequency of the stage; S402: Determine the arithmetic mean of the internal frequencies of each machine in the stage as the internal frequency of the stage; S403: Determine the arithmetic mean of the external frequencies of each machine in the said stage as the external frequency of the said stage; S404: Determine the arithmetic mean of the fan frequencies of each machine in the stage as the fan frequency of the stage.

[0052] In one embodiment, see Figure 8 The step of performing feature encoding on the multidimensional array to obtain the historical execution encoding features includes: S501: Determine the high-level and low-level values ​​corresponding to the stage host frequency, intra-stage sub-frequency, extra-stage sub-frequency, and stage fan frequency in the multi-dimensional array, respectively. S502: Determine the center value and half-pitch corresponding to the stage host frequency, the stage sub-frequency, the stage sub-frequency and the stage fan frequency respectively based on the high level value and low level value corresponding to the stage host frequency, the stage sub-frequency, the stage sub-frequency and the stage fan frequency. S503: Based on the center value and half-spacing corresponding to the stage host frequency, the stage sub-frequency, the stage sub-frequency, and the stage fan frequency, feature encoding is performed on the stage host frequency, the stage sub-frequency, the stage sub-frequency, and the stage fan frequency to obtain the historical operation encoding features; wherein, the historical operation encoding features include the encoding values ​​corresponding to the stage host frequency, the stage sub-frequency, the stage sub-frequency, and the stage fan frequency.

[0053] Understandably, the following will continue with the aforementioned embodiments to explain the factor coding and standardization process.

[0054] To eliminate dimensional differences between parameters, achieve fair data comparison, and improve the stability and accuracy of numerical calculations, the aggregated parameters undergo coding standardization. First, based on historical data distribution and process knowledge, appropriate high and low level values ​​are determined for each parameter. A coding system of -1 and +1 is used to convert the natural factors of the operating equipment parameters into coded factors. That is, the low level is set to -1, and the high level is set to +1. For example, based on historical experience, the common operating range of the "crushing section host frequency" may be between 40Hz and 50Hz, so the low level can be set to 40 and the high level to 50. This process makes the coefficients more intuitive and also enhances the training stability and convergence speed of the machine learning model. Simultaneously, the coded parameters facilitate the setting of reasonable constraint boundaries during optimization, avoiding invalid solutions that exceed the equipment's capabilities. The conversion formula between natural factors and coded factors is as follows: ; The center value and half-spacing value are respectively: ; .

[0055] This encoding process has multiple benefits: First, it brings all features to the same scale, improving the training speed and convergence stability of the machine learning model; second, in subsequent optimization modules, the encoded parameter space (usually around [-1,1]) makes it easier to define search boundaries and constraints, effectively avoiding the recommendation of invalid parameters that are out of touch with the actual capabilities of the device; third, the magnitude of the encoded value can directly indicate the degree of deviation of the parameter from the normal operating range, enhancing the interpretability of the model.

[0056] As can be seen from the above description, the machine learning-based parameter prediction method for natural graphite spherical production lines provided in this application can perform feature encoding on the historical parameters of the natural graphite spherical production line equipment to obtain historical operation code features.

[0057] In one embodiment, see Figure 9 The step of inputting historical product indicators and historical operational coding features into a machine learning model to obtain the parameter prediction model for the spherical production line includes: S601: If the number of historical product indicators and corresponding historical operation coding features does not reach the sample size threshold, a meta-learning training task set is constructed based on historical multi-task data; a meta-learning algorithm is used to perform meta-training on the meta-learning training task set to obtain a shared base learner; the shared base learner is fine-tuned using the historical product indicators and corresponding historical operation coding features; wherein, if the number of historical product indicators and corresponding historical operation coding features never reaches the sample size threshold, the fine-tuned shared base learner is used as the parameter prediction model for the spherical production line. Historical multi-task data can be understood as extracting multiple independent machine learning tasks from accumulated historical production data. Each task simulates a complete historical production debugging scenario, including: (1) Task support set: A small amount of sample data corresponding to the historical batch or product specification (i.e., a subset of the "historical operation coding features" and the corresponding "historical product indicators"), used to simulate the "rapid adaptation" stage in meta-learning.

[0058] (2) Task query set: The remaining sample data in the same task, used to evaluate the model performance after rapid adaptation on the task.

[0059] S602: If the number of historical product indicators and corresponding historical operation coding features has reached the sample size threshold, the fully automatic machine learning framework is initialized, and multiple base learners are automatically selected from the preset algorithm space to construct a prediction model candidate set. The base learners in the prediction model candidate set are then used for training. Based on the bagging method, the historical product indicators and historical operation coding features are sampled with replacement to obtain each training data subset. Each training data subset is input into the multiple base learners in the prediction model candidate set for training to obtain multiple candidate models. Based on the prediction output of the multiple candidate models, a secondary learner is trained and fused to generate the spherical production line parameter prediction model.

[0060] It is understandable that steps S601 to S602 are the process of constructing a spherical production line parameter prediction model using a dual-track system combining fully automated machine learning and meta-learning. The above process includes: (1) inputting historical product indicators and historical operation coding features into the fully automated machine learning track; (2) when facing the situation that the historical product indicators are insufficient (i.e., only a small sample) during the debugging stage of new spherical graphite products, using meta-learning technology to quickly adapt to the small sample and obtain the spherical production line parameter prediction model.

[0061] Specifically, a training task set for meta-learning is constructed based on historical production data (historical product indicators and historical operational coding features). That is, multiple meta-training tasks are extracted from historical production data, each simulating a debugging scenario for a historical product batch or specification, and including the corresponding historical operational coding features and historical product indicator data. Then, a meta-learning algorithm independent of the model to be trained (the spherical production line parameter prediction model) is used for meta-training on the (meta-learning) training task set. This process aims to enable the model to master the intrinsic patterns of quickly learning new tasks from a small number of samples, ultimately obtaining a shared base learning model with good generalization initialization parameters. Next, the candidate set of prediction models for the fully automated machine learning framework is initialized. The shared base learning model obtained from meta-training is used as a pre-trained model and included in the initial candidate set of prediction models along with other types of base learners (such as linear models, decision trees, etc.) automatically selected by the fully automated machine learning framework from a preset algorithm space. Then, the model is quickly adapted to the limited sample data of the new product: when facing the debugging task of a new spherical graphite product and only having a small sample data, the shared base learning model provided by meta-learning in the candidate set is quickly fine-tuned using this data to adapt it to the new task. If the sample size is sufficient, all historical data can be directly applied to train the model for other types of base learners in the candidate set. Next, multiple training data subsets are generated using the bagging method. Multiple subsets are constructed from the available training data (including historical data and small sample data of the new task) through sampling with replacement. Then, the prediction candidate models are trained in parallel: each subset is input into other types of base learners in the prediction model candidate set for parallel training, resulting in candidate prediction results for multiple candidate models. Finally, the prediction candidate models are fused based on the stacked ensemble method to generate the final prediction model: all candidate prediction results obtained in the previous step are used as new feature vectors and input into a secondary learner for high-order fusion, thereby obtaining an integrated, higher-performance spherical production line parameter prediction model.

[0062] The ensemble of multiple models often outperforms the prediction of a single model, typically significantly reducing the variance of the final prediction. Based on this theory, this application abandons the traditional practice of manually specifying machine learning model types. Instead, it employs a fully automated machine learning (AutoML) framework to automatically search for the optimal model end-to-end within a predefined "algorithm space." This algorithm space comprehensively covers various machine learning paradigms: linear models, decision trees, random forests, gradient boosting trees, support vector machines, and other algorithm models. The system uses the particle size distribution index (D) of spherical graphite. 10 D 50 D 90Regression models were independently constructed for key quality indicators such as tap density, specific surface area, and raw materials used in the spheroidization process. Each model was constructed based on previously obtained historical operational coding characteristics and historical product indicators.

[0063] The model construction follows an end-to-end automated process. It employs stacking as the top-level fusion strategy, combined with bagging as one of the base learners. Stacking uses the outputs of multiple base learners as input to secondary learners. Specifically, in the first stage, diverse base algorithms are used to predict on the same training set, with each model outputting a scalar prediction value, forming a feature vector of length n. In the second stage, these feature vectors are used as input to train a secondary learner for final prediction, thereby reducing prediction variance and improving generalization ability through the complementarity between models. Simultaneously, bagging is used internally within the model, generating multiple sub-training sets through bootstrapping and training multiple base learners in parallel. For regression models, results are aggregated by averaging, effectively reducing model variance and preventing overfitting.

[0064] To ensure the model's rapid adaptability to new operating conditions, the system employs meta-learning techniques that do not rely on specific model architectures. Instead, it constructs a cross-task knowledge transfer framework that works in conjunction with a fully automated machine learning framework. Through meta-training on multiple batches and specifications of historical tasks, the system acquires the ability to "learn to learn," thereby learning the intrinsic patterns that allow it to quickly adapt to new tasks from a small sample size. When faced with new raw material batches or product specifications, the system requires only a small amount of experimental data. The meta-learning algorithm can quickly match similar historical scenarios, providing intelligent initial parameter initialization for the prediction model, enabling the system to maintain good performance even with small sample sizes.

[0065] For training data, a scientific training-validation-test set partitioning strategy is employed. Based on the data distribution characteristics of each quality indicator, the training, validation, and test sets are randomly divided in a 70%:15%:15% ratio to ensure that each set has a representative distribution within the target value range. The training set is used for learning model parameters, the validation set is used for hyperparameter tuning and model selection, and the test set is used to finally evaluate the model's generalization ability on unseen data. For data with obvious time series characteristics, a time-series partitioning method is used to ensure that the training set is earlier than the validation and test sets, simulating time series prediction scenarios in actual production.

[0066] As can be seen from the above description, the machine learning-based parameter prediction method for natural graphite spherical production lines provided in this application can pre-build a parameter prediction model for spherical production lines.

[0067] In one embodiment, see Figure 10The prediction outputs based on the multiple candidate models are fused by training a secondary learner to generate the spherical production line parameter prediction model, including: S701: Using a K-fold cross-validation strategy, the prediction results of each candidate model in the plurality of candidate models are obtained on each fold validation set, and their respective performance indicators are calculated; the performance indicators include the coefficient of determination and / or root mean square error; S702: Based on the performance metrics, select one or more candidate models from the plurality of candidate models for integration to form a subset of models to be integrated; S703: All prediction results generated by each candidate model in the subset of models to be integrated in K-fold cross-validation are concatenated by sample to form a secondary training feature set, and the corresponding true target value is obtained. S704: Using the secondary training feature set and the true target value, train a secondary learner; S705: Combine the subset of models to be integrated (as the first-layer model) with the secondary learner (as the second-layer model) to form a stacked integrated model, which serves as the parameter prediction model for the spherical production line.

[0068] Understandably, in constructing a parameter prediction model for a spherical production line, the first step is to comprehensively evaluate multiple candidate models using a K-fold cross-validation strategy. This process divides the dataset into K equal-sized subsets, using one subset as the validation set and the remaining K-1 subsets as the training set in turn. Each candidate model is trained and validated multiple times to obtain prediction results for each model on different validation sets. In each round of validation, corresponding performance metrics are calculated, primarily including the coefficient of determination (COP). R² ) and root mean square error ( RMSEThese metrics reflect the model's prediction accuracy and goodness of fit from different perspectives. This cross-validation method effectively avoids evaluation bias caused by the randomness of data partitioning, resulting in more stable and reliable model performance evaluation results. Based on these performance metrics, the next step is to select the best-performing models from all candidate models to form a subset of models to be integrated. The selection criteria are usually based on the performance metrics, choosing models that perform stably and have good prediction results in cross-validation to ensure that the subsequent integration process can achieve a synergistic effect and improve the overall prediction capability. After selecting the subset of models to be integrated, the prediction results generated by these models during K-fold cross-validation need to be integrated. Specifically, all prediction results obtained by each model in K-fold validation are concatenated according to the sample order to form a higher-dimensional secondary training feature set. Simultaneously, the corresponding true target values ​​of these samples are collected to construct a complete training dataset. This secondary feature set contains the prediction information of each basic model, providing rich feature information for the subsequent training of secondary learners. Subsequently, this constructed secondary training feature set and corresponding true target values ​​are used to train a secondary learner, often referred to as a meta-learner or second-layer model. Its role is to learn how to optimally combine and weight the predictions of the various base models in the first layer. Finally, the subset of models to be integrated in the first layer is organically combined with the secondary learner in the second layer to construct a complete stacked ensemble model. This model can fully utilize the advantages of multiple base models, optimizing and integrating the predictions of the base models through the meta-learner, thereby achieving more accurate and reliable predictions of the parameters of the spherical production line.

[0069] It should be noted that in the ensemble stage of constructing the parameter prediction model for the spherical production line, the candidate prediction results are first used as feature vectors input to the secondary learner. The secondary learner learns how to combine the outputs of the base learners by analyzing these features. Subsequently, the coefficient of determination (COP) can be used... R² The base learner's fit is evaluated using a metric that reflects the model's ability to explain data variation; a value closer to 1 indicates a better fit. Based on the candidate predictions and the fit, the secondary learner is trained to generate an optimized secondary learner. Then, the root mean square error (RMSE) is used to evaluate the fit. RMSE Mean absolute percentage error ( MAPE The meta-learner was further analyzed using the coefficient of determination to evaluate its prediction accuracy and stability. RMSE Measuring the deviation between predicted and actual values, MAPE Reflecting the magnitude of relative error, combined with R²The model's performance can be comprehensively evaluated. Finally, based on the analysis results, the base learners are weighted and fused, with higher weights assigned to the better-performing models, thus generating the final spherical production line parameter prediction model. This process ensures that the model possesses both high accuracy and robustness, effectively addressing the complex prediction needs of production line parameters.

[0070] Furthermore, the model evaluation adopts a multi-dimensional comprehensive evaluation system. Key technical indicators include: root mean square error (RMSE). RMSE Mean absolute percentage error ( MAPE ) and coefficient of determination ( R² ).

[0071] Among them, root mean square error ( RMSE The mean absolute percentage error (MAE) reflects the absolute magnitude of the prediction error and has strong interpretability; MAPE The coefficient of determination (COP) can measure the level of relative error and is not affected by the unit of data. R² This has two functions. First, it serves to improve the training of the secondary learner. R² The evaluation process involves assessing the fit of different base learners on the training (or validation) set, followed by evaluating the goodness of fit within the ensemble model to account for data variation. R² As a dimensionless index, it is suitable for comparing data with three different dimensions—particle size distribution, tap density, and specific surface area—during the prediction process. The formulas are as follows:

[0072] The above formula SS R and SS E These represent the regression sum of squares and the error sum of squares, respectively, and their expressions are:

[0073] In the above formula It is the actual value. It is a predicted value. Dependent variable y of n Arithmetic mean of the observations n This represents the number of samples.

[0074] As can be seen from the above description, the machine learning-based method for predicting parameters of a natural graphite spherical production line provided in this application can use a stacked ensemble method to input the candidate prediction results as feature vectors into a secondary learner, and fuse multiple candidate models to obtain the spherical production line parameter prediction model.

[0075] In one embodiment, see Figure 2 , Figure 11 The aforementioned machine learning-based parameter prediction method for natural graphite spherical production lines further includes: S801: Real-time acquisition of equipment operating parameters from the equipment's programmable logic controller; S802: Optimize the parameter prediction model of the spherical production line using real-time collected equipment operating parameters.

[0076] Understandably, during the system integration and parameter decomposition and deployment phases, it is necessary to ensure the accurate and reliable application of optimized recommended parameters in real industrial environments. To this end, this application's embodiment constructs a cloud-edge collaborative architecture centered on an intelligent operation and decision-making center, deeply integrating the Manufacturing Execution System (MES) and Enterprise Resource Planning (ERP). This center, acting as the "intelligent brain" of the entire spherical production line, transforms static algorithm recommendations into a dynamically perceptive, real-time decision-making, and precisely executed intelligent control closed loop.

[0077] The core of the system lies in the operation mechanism of the intelligent operation and decision-making center as a unified decision-making hub. It achieves deep bidirectional integration with the manufacturing execution system and enterprise resource planning system through standard interfaces, forming a closed-loop information flow of "production intention - equipment status - optimization decision".

[0078] The data acquisition layer directly obtains real-time operating parameters (equipment operating frequency, current, etc.) from the device's programmable logic controller (PLC) via Industrial Internet of Things (IIoT) protocols, forming an instantaneous perception of the production process. This is combined with an online particle size detection system (see...). Figure 3 (As shown) Real-time feedback of granularity and other physical parameters allows for fine-tuning of the equipment parameters recommended by the prediction model. This actual production data can also be used to optimize the parameter prediction model for spherical production lines.

[0079] In one embodiment, see Figure 12 The step of distributing the stage operation parameters of the spherical production line to each piece of equipment in the natural graphite spherical production line in stages, based on production plan information collected from the Manufacturing Execution System and equipment macro-information collected from the Enterprise Resource Planning System, includes: S901: Determine the performance degradation coefficient of each piece of equipment based on the production plan information and the equipment macro information; S902: Adjust the stage operating parameters of the spherical production line according to the performance degradation coefficient; S903: The compensated adjustment of the spherical production line stage operation parameters is distributed to each piece of equipment in the natural graphite spherical production line according to the current production stage.

[0080] Understandably, the system is deeply integrated with MES and ERP systems through standard interfaces.

[0081] The integration of the system with MES and ERP forms a closed-loop information flow from "production planning - equipment status - optimization decision-making". The MES system provides information such as production orders, target product specifications, and raw material batch numbers, and can upload manual or automatic machine adjustment parameters to this system, forming a continuously growing adjustment experience library. More importantly, the MES system provides key performance indicators such as actual output, pass rate, and energy consumption, as shown in Table 1. The ERP system contributes macro-level status information on equipment maintenance and purchase order information for equipment upgrades, such as historical maintenance records, replacement cycles and current cumulative wear time of key wear parts (such as crushing gears and shaping modules), and equipment upgrade information.

[0082] Specifically, the system establishes an intelligent decomposition mechanism from stage-aggregated parameters to specific equipment parameters. In model application, the recommended parameters output by the predictive model are the average parameters of each process stage (such as the average frequency of the main unit in the crushing section). These stage-average parameters need to be differentiated according to the actual status of each piece of equipment within the same stage. Through the integrated MES and ERP systems, the system can obtain data such as the cumulative working time, recent maintenance records, and historical replacement cycles of key wear parts (such as crushing gears and shaping modules) of each piece of equipment. While decomposing the average parameters, directly distributing them evenly to each piece of equipment might result in severely worn equipment failing to achieve the expected processing effect, while new equipment might over-process. Therefore, the average parameters need to be adjusted according to the performance degradation coefficient of each piece of equipment; that is, the "performance degradation coefficient" of each piece of equipment needs to be calculated. The performance degradation coefficient reflects the degree of decrease in processing capacity caused by wear; a higher value indicates more severe wear and lower processing efficiency.

[0083] The approach is to provide positive compensation (i.e., increase the parameter values ​​to make up for the decrease in processing efficiency) for equipment with severe performance degradation, and negative compensation (i.e., decrease the parameter values ​​to avoid over-grinding, etc.) for equipment with minor performance degradation. However, the magnitude of the compensation should be controlled within a reasonable range, and the overall processing effect of the entire stage should be consistent with the expected effect of the stage's average parameters.

[0084] Based on the performance degradation coefficient, the system uses a weighted allocation algorithm to decompose the stage average parameter into specific device parameters. For multiple devices within the same stage, the actual recommended parameter calculation formula for each device is as follows: Equipment parameters = stage average parameters × (1 + compensation coefficient × (equipment performance degradation coefficient - stage average degradation coefficient)).

[0085] The stage average attenuation coefficient refers to the average value of the performance attenuation coefficients of all devices within the same stage.

[0086] The compensation coefficient is pre-calibrated based on the equipment type and process characteristics to ensure that the compensated parameters are within the equipment's safe range and effectively offset the processing efficiency loss caused by equipment wear. For example, for equipment with severe wear, the system will appropriately increase its frequency parameters to compensate for the decrease in processing capacity; for newly replaced gear rings or well-maintained equipment, its parameters will be maintained or appropriately reduced to optimize energy consumption while avoiding over-grinding.

[0087] As can be seen from the above description, the machine learning-based parameter prediction method for natural graphite spherical production lines provided in this application can distribute the stage operation parameters of the spherical production line to each piece of equipment in the natural graphite spherical production line in stages, based on the production plan information collected from the manufacturing execution system and the equipment macro information collected from the enterprise resource planning system.

[0088] Based on the same inventive concept, this application also provides a machine learning-based parameter prediction device for a natural graphite spherical production line, which can be used to implement the method described in the above embodiments, as shown in the following embodiments. Since the principle of the machine learning-based parameter prediction device for a natural graphite spherical production line is similar to that of the machine learning-based parameter prediction method, the implementation of the machine learning-based parameter prediction device for a natural graphite spherical production line can refer to the implementation of the software performance benchmark determination method, and will not be repeated. As used below, the terms "unit" or "module" can refer to a combination of software and / or hardware that performs a predetermined function. Although the system described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0089] In one embodiment, see Figure 13 To adapt to the complex process of graphite spherical production and achieve stable and accurate intelligent optimization and recommendation of equipment parameters even under small sample conditions, this application provides a parameter prediction device for a natural graphite spherical production line based on machine learning, comprising: The model prediction unit 1301 is used to input the acquired spherical graphite production indicators and raw material characteristics into a pre-constructed spherical production line parameter prediction model to obtain the stage operation parameters of the spherical production line; wherein, the spherical production line parameter prediction model is constructed based on the historical operation coding features obtained by feature encoding the historical parameters of the natural graphite spherical production line equipment; The parameter distribution unit 1302 is used to distribute the stage operation parameters of the spherical production line to each piece of equipment in the natural graphite spherical production line in stages, based on the production plan information collected from the manufacturing execution system and the equipment macro information collected from the enterprise resource planning system.

[0090] From a hardware perspective, in order to adapt to the complex process of graphite spherical production and to stably and accurately achieve intelligent optimization and recommendation of equipment parameters even under small sample conditions, this application provides an embodiment of an electronic device for implementing all or part of the aforementioned machine learning-based natural graphite spherical production line parameter prediction method. The electronic device specifically includes the following components: The system comprises a processor, a memory, a communications interface, and a bus; wherein the processor, memory, and communications interface communicate with each other via the bus; the communications interface is used to transmit information between the machine learning-based natural graphite spherical production line parameter prediction device and core business systems, user terminals, and related databases and other related equipment; the logic controller can be a desktop computer, tablet computer, or mobile terminal, etc., and this embodiment is not limited to these. In this embodiment, the logic controller can be implemented with reference to the embodiments of the machine learning-based natural graphite spherical production line parameter prediction method and the machine learning-based natural graphite spherical production line parameter prediction device in the embodiments, the contents of which are incorporated herein, and repeated details will not be described again.

[0091] It is understood that the user terminal may include smartphones, tablet computers, network set-top boxes, portable computers, desktop computers, personal digital assistants (PDAs), in-vehicle devices, smart wearable devices, etc. Among these, the smart wearable devices may include smart glasses, smartwatches, smart bracelets, etc.

[0092] In practical applications, the parameter prediction method for natural graphite spherical production lines based on machine learning can be partially executed on the electronic device side as described above, or all operations can be completed in the client device. The choice can be made based on the processing power of the client device and the limitations of the user's usage scenario. This application does not impose any limitations on this. If all operations are completed in the client device, the client device may further include a processor.

[0093] The aforementioned client device may have a communication module (i.e., a communication unit) that can communicate with a remote server to achieve data transmission. The server may include a server on the task scheduling center side; in other implementation scenarios, it may also include a server on an intermediate platform, such as a server on a third-party server platform that has a communication link with the task scheduling center server. The server may include a single computer device, a server cluster consisting of multiple servers, or a distributed server structure.

[0094] Figure 14 This is a schematic block diagram illustrating the system configuration of the electronic device 9600 according to an embodiment of this application. Figure 14 As shown, the electronic device 9600 may include a central processing unit 9100 and a memory 9140; the memory 9140 is coupled to the central processing unit 9100. It is worth noting that... Figure 14 This is an example; other types of structures can also be used to supplement or replace this structure to achieve telecommunications functions or other functions.

[0095] In one embodiment, the parameter prediction method for a natural graphite spherical production line based on machine learning can be integrated into a central processing unit 9100. The central processing unit 9100 can be configured to perform the following control: S101: Input the obtained spherical graphite production indicators and raw material characteristics into the pre-constructed spherical production line parameter prediction model to obtain the stage operation parameters of the spherical production line; wherein, the spherical production line parameter prediction model is constructed based on the historical operation coding features obtained by feature encoding the historical parameters of the natural graphite spherical production line equipment; S102: Based on the production plan information collected from the Manufacturing Execution System and the equipment macro information collected from the Enterprise Resource Planning System, the stage operation parameters of the spherical production line are adjusted compensatorily and distributed to each piece of equipment in the natural graphite spherical production line in stages.

[0096] As described above, the machine learning-based parameter prediction method for a natural graphite spherical production line provided in this application simplifies high-dimensional, continuous equipment parameters into stage features with clear physical meaning through an innovative process stage aggregation strategy, effectively solving the curse of dimensionality problem in modeling multi-machine serial processes. Based on this, the system deeply integrates an automated machine learning framework and meta-learning technology, enabling end-to-end automatic construction of a high-precision prediction model. Furthermore, it significantly improves the robustness and generalization ability of the prediction by utilizing stacking integration and repeated bagging strategies, allowing it to quickly adapt and provide reliable predictions even when faced with a small number of new samples. The core advantage of this invention lies in achieving reverse intelligent recommendation from product goals to equipment parameters, transforming the traditional parameter debugging process, which requires extensive physical testing, into efficient digital simulation optimization, thus reducing trial-and-error costs. Ultimately, through deep integration with Manufacturing Execution Systems (MES) and Enterprise Resource Planning (ERP) systems, the system dynamically transforms static parameter recommendations into personalized settings that sense the real-time health status of equipment. This enables refined and differentiated adjustments to the parameters of various equipment within the production line based on equipment wear and tear. Consequently, while ensuring overall process stability, it effectively compensates for equipment performance degradation, continuously improving product quality consistency and production efficiency. Overall, this invention not only significantly enhances the intelligence level and product quality stability of spherical graphite production but also provides a replicable, efficient, and reliable technical solution for the digital transformation and upgrading of process industries.

[0097] In another embodiment, the machine learning-based parameter prediction device for a natural graphite spherical production line can be configured separately from the central processing unit 9100. For example, the data composite transmission device for the machine learning-based parameter prediction device for a natural graphite spherical production line can be configured as a chip connected to the central processing unit 9100, and the function of the machine learning-based parameter prediction method for a natural graphite spherical production line can be realized through the control of the central processing unit.

[0098] like Figure 14 As shown, the electronic device 9600 may further include: a communication module 9110, an input unit 9120, an audio processor 9130, a display 9160, and a power supply 9170. It is worth noting that the electronic device 9600 does not necessarily need to include these components. Figure 14 All components shown; in addition, the electronic device 9600 may also include Figure 14 For components not shown, please refer to existing technologies.

[0099] like Figure 14 As shown, the central processing unit 9100, sometimes also referred to as a controller or operating control, may include a microprocessor or other processor device and / or logic device, which receives inputs and controls the operation of various components of the electronic device 9600.

[0100] The memory 9140 may be, for example, one or more of a cache, flash memory, hard drive, removable media, volatile memory, non-volatile memory, or other suitable devices. It may store the aforementioned failure-related information, and also store a program for executing that information. The central processing unit 9100 may execute the program stored in the memory 9140 to perform information storage or processing, etc.

[0101] Input unit 9120 provides input to central processing unit 9100. Input unit 9120 may be, for example, a keypad or touch input device. Power supply 9170 provides power to electronic device 9600. Display 9160 displays images and text. Display may be, for example, an LCD display, but is not limited thereto.

[0102] The memory 9140 can be a solid-state memory, such as a read-only memory (ROM), random access memory (RAM), a SIM card, etc. It can also be a memory that retains information even when power is off, can be selectively erased, and contains more data; examples of this type of memory are sometimes referred to as EPROMs. The memory 9140 can also be some other type of device. The memory 9140 includes a buffer memory 9141 (sometimes referred to as a buffer). The memory 9140 may include an application / function storage unit 9142 for storing application programs and function programs or processes for executing the operation of the electronic device 9600 via the central processing unit 9100.

[0103] The memory 9140 may also include a data storage unit 9143 for storing data, such as contacts, digital data, pictures, sounds, and / or any other data used by the electronic device. The driver storage unit 9144 of the memory 9140 may include various drivers for the electronic device's communication functions and / or for performing other functions of the electronic device (such as messaging applications, address book applications, etc.).

[0104] The communication module 9110 is a transmitter / receiver that sends and receives signals via the antenna 9111. The communication module (transmitter / receiver) 9110 is coupled to the central processing unit 9100 to provide input signals and receive output signals, which is the same as in a conventional mobile communication terminal.

[0105] Based on different communication technologies, multiple communication modules 9110 can be configured in the same electronic device, such as cellular network modules, Bluetooth modules, and / or wireless LAN modules. The communication module (transmitter / receiver) 9110 is also coupled to a speaker 9131 and a microphone 9132 via an audio processor 9130 to provide audio output via the speaker 9131 and receive audio input from the microphone 9132, thereby realizing typical telecommunications functions. The audio processor 9130 may include any suitable buffer, decoder, amplifier, etc. Additionally, the audio processor 9130 is also coupled to a central processing unit 9100, enabling on-device recording via the microphone 9132 and on-device playback of stored sound via the speaker 9131.

[0106] Embodiments of this application also provide a computer-readable storage medium capable of implementing all steps of the machine learning-based parameter prediction method for a natural graphite spherical production line, where the execution subject is a server or client, as described in the above embodiments. The computer-readable storage medium stores a computer program that, when executed by a processor, implements all steps of the machine learning-based parameter prediction method for a natural graphite spherical production line, where the execution subject is a server or client, as described in the above embodiments. For example, when the processor executes the computer program, it implements the following steps: S101: Input the obtained spherical graphite production indicators and raw material characteristics into the pre-constructed spherical production line parameter prediction model to obtain the stage operation parameters of the spherical production line; wherein, the spherical production line parameter prediction model is constructed based on the historical operation coding features obtained by feature encoding the historical parameters of the natural graphite spherical production line equipment; S102: Based on the production plan information collected from the Manufacturing Execution System and the equipment macro information collected from the Enterprise Resource Planning System, the stage operation parameters of the spherical production line are adjusted compensatorily and distributed to each piece of equipment in the natural graphite spherical production line in stages.

[0107] As described above, the machine learning-based parameter prediction method for a natural graphite spherical production line provided in this application simplifies high-dimensional, continuous equipment parameters into stage features with clear physical meaning through an innovative process stage aggregation strategy, effectively solving the curse of dimensionality problem in modeling multi-machine serial processes. Based on this, the system deeply integrates an automated machine learning framework and meta-learning technology, enabling end-to-end automatic construction of a high-precision prediction model. Furthermore, it significantly improves the robustness and generalization ability of the prediction by utilizing stacking integration and repeated bagging strategies, allowing it to quickly adapt and provide reliable predictions even when faced with a small number of new samples. The core advantage of this invention lies in achieving reverse intelligent recommendation from product goals to equipment parameters, transforming the traditional parameter debugging process, which requires extensive physical testing, into efficient digital simulation optimization, thus reducing trial-and-error costs. Ultimately, through deep integration with Manufacturing Execution Systems (MES) and Enterprise Resource Planning (ERP) systems, the system dynamically transforms static parameter recommendations into personalized settings that sense the real-time health status of equipment. This enables refined and differentiated adjustments to the parameters of various equipment within the production line based on equipment wear and tear. Consequently, while ensuring overall process stability, it effectively compensates for equipment performance degradation, continuously improving product quality consistency and production efficiency. Overall, this invention not only significantly enhances the intelligence level and product quality stability of spherical graphite production but also provides a replicable, efficient, and reliable technical solution for the digital transformation and upgrading of process industries.

[0108] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, apparatus, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0109] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (devices), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0110] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0111] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0112] Specific embodiments have been used to illustrate the principles and implementation methods of this invention. The descriptions of the embodiments above are only for the purpose of helping to understand the method and core ideas of this invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this invention. Therefore, the content of this specification should not be construed as a limitation of this invention.

Claims

1. A machine learning-based natural graphite spherical production line parameter prediction method, characterized by, include: The obtained spherical graphite production indicators and raw material characteristics are input into a pre-built spherical production line parameter prediction model to obtain the stage operation parameters of the spherical production line. Based on the production plan information collected from the Manufacturing Execution System and the equipment macro information collected from the Enterprise Resource Planning System, the stage operation parameters of the spherical production line are adjusted compensatorily and distributed to each piece of equipment in the natural graphite spherical production line in stages.

2. The machine learning-based natural graphite spheroidization line parameter prediction method according to claim 1, characterized by, The steps for pre-constructing a parameter prediction model for a spherical production line include: The historical parameters of the natural graphite spherical production line equipment are feature-encoded to obtain historical operation code features; The historical product indicators and the historical operation coding features are input into the machine learning model to obtain the parameter prediction model of the spherical production line; wherein, the machine learning model is implemented using a fully automated machine learning framework combined with meta-learning technology.

3. The machine learning-based natural graphite spheroidization line parameter prediction method according to claim 2, characterized by, The historical parameters of the natural graphite spherical production line equipment include the operating parameters of each machine at each stage; the stages include the crushing stage, the crushing and shaping stage, and the shaping stage; the operating parameters of each machine at each stage include the main machine frequency, internal distribution frequency, external distribution frequency, and fan frequency of each machine at each stage; the historical operating parameters of the natural graphite spherical production line equipment are feature-encoded to obtain historical operating code features, including: The stage host frequency, stage internal frequency, stage external frequency, and stage fan frequency are determined based on the host frequency, internal frequency, external frequency, and fan frequency of each machine in the stage. The obtained feed parameters, raw material parameters, intermediate product parameters, stage host frequency, intra-stage sub-frequency, extra-stage sub-frequency, and stage fan frequency are combined to obtain a multi-dimensional array; The multidimensional array is feature-encoded to obtain the historical running encoding features.

4. The machine learning-based natural graphite spheroidization line parameter prediction method according to claim 3, characterized by, The step of determining the stage host frequency, stage internal frequency, stage external frequency, and stage fan frequency corresponding to each machine in the stage, based on the host frequency, internal frequency, external frequency, and fan frequency of each machine in the stage, includes: The arithmetic mean of the host frequencies of each machine in the said phase is determined as the host frequency of the said phase; The arithmetic mean of the internal frequencies of each machine in the said phase is determined as the internal frequency of the said phase; The arithmetic mean of the external frequencies of each machine in the said stage is determined as the external frequency of the said stage; The arithmetic mean of the fan frequencies of each machine in the stated stage is determined as the fan frequency of the stated stage.

5. The machine learning-based natural graphite spheroidization line parameter prediction method according to claim 3, characterized by, The step of performing feature encoding on the multidimensional array to obtain the historical execution encoding features includes: Determine the high and low level values ​​corresponding to the stage host frequency, intra-stage sub-frequency, extra-stage sub-frequency, and stage fan frequency in the multidimensional array, respectively. Based on the high and low level values ​​corresponding to the stage host frequency, the stage sub-frequency, the stage sub-frequency, and the stage fan frequency, respectively, determine the center value and half-pitch corresponding to the stage host frequency, the stage sub-frequency, the stage sub-frequency, and the stage fan frequency. Based on the center value and half-spacing corresponding to the stage host frequency, the stage sub-frequency, the stage external sub-frequency, and the stage fan frequency, feature encoding is performed on the stage host frequency, the stage sub-frequency, the stage external sub-frequency, and the stage fan frequency to obtain the historical operation encoding features; wherein, the historical operation encoding features include the encoding values ​​corresponding to the stage host frequency, the stage sub-frequency, the stage external sub-frequency, and the stage fan frequency. 6.The machine learning-based natural graphite spheroidization line parameter prediction method according to claim 2, wherein The step of inputting historical product indicators and historical operational coding features into a machine learning model to obtain the parameter prediction model for the spherical production line includes: If the number of historical product indicators and corresponding historical operation coding features does not reach the sample size threshold, a meta-learning training task set is constructed based on historical multi-task data; a meta-learning algorithm is used to perform meta-training on the meta-learning training task set to obtain a shared base learner; the shared base learner is fine-tuned using the historical product indicators and corresponding historical operation coding features; wherein, if the number of historical product indicators and corresponding historical operation coding features never reaches the sample size threshold, the fine-tuned shared base learner is used as the parameter prediction model for the spherical production line. If the number of historical product indicators and corresponding historical operation coding features has reached the sample size threshold, the fully automatic machine learning framework is initialized, and multiple base learners are automatically selected from the preset algorithm space to construct a prediction model candidate set. The base learners in the prediction model candidate set are then used for training. Based on the bagging method, the historical product indicators and historical operation coding features are sampled with replacement to obtain each training data subset. Each training data subset is input into the multiple base learners in the prediction model candidate set for training to obtain multiple candidate models. Based on the prediction outputs of the multiple candidate models, a secondary learner is trained and fused to generate the spherical production line parameter prediction model.

7. The machine learning-based natural graphite spheroidization line parameter prediction method according to claim 6, characterized by, The prediction outputs based on the multiple candidate models are fused by training a secondary learner to generate the spherical production line parameter prediction model, including: A K-fold cross-validation strategy is employed to obtain the prediction results of each candidate model on each fold validation set among the multiple candidate models, and to calculate their respective performance metrics; the performance metrics include the coefficient of determination and / or root mean square error. Based on the performance metrics, one or more candidate models are selected from the plurality of candidate models for integration, forming a subset of models to be integrated; All prediction results generated by each candidate model in the subset of models to be integrated in K-fold cross-validation are concatenated by sample to form a secondary training feature set, and the corresponding true target value is obtained. A secondary learner is trained using the secondary training feature set and the true target value; The subset of models to be integrated is combined with the secondary learner to form a stacked integrated model, which serves as the parameter prediction model for the spherical production line. 8.The machine learning based natural graphite spheroidization line parameter prediction method of claim 1, wherein, Also includes: Real-time acquisition of equipment operating parameters from the equipment's programmable logic controller; The parameter prediction model for the spherical production line is optimized by utilizing real-time collected equipment operating parameters. 9.The machine learning based natural graphite spheroidization line parameter prediction method of claim 1, wherein, The process involves compensatory adjustments to the stage operating parameters of the spherical production line based on production plan information collected from the Manufacturing Execution System (MAS) and equipment macro-information collected from the Enterprise Resource Planning (ERP) system, and then distributing these adjustments in stages to each piece of equipment within the natural graphite spherical production line. This includes: The performance degradation coefficient of each piece of equipment is determined based on the production plan information and the macroscopic information of the equipment. The operating parameters of the spherical production line are adjusted in a compensatory manner based on the performance degradation coefficient. The adjusted operating parameters for each stage of the spherical production line are distributed to each piece of equipment in the natural graphite spherical production line according to the current production stage.

10. A machine learning-based natural graphite spherical production line parameter prediction device, characterized by, include: The model prediction unit is used to input the acquired spherical graphite production indicators and raw material characteristics into a pre-built spherical production line parameter prediction model to obtain the stage operation parameters of the spherical production line. The parameter distribution unit is used to distribute the stage operation parameters of the spherical production line to each piece of equipment in the natural graphite spherical production line in stages, based on the production plan information collected from the manufacturing execution system and the equipment macro information collected from the enterprise resource planning system.

11. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the machine learning-based parameter prediction method for natural graphite spherical production lines as described in any one of claims 1 to 9.

12. A computer readable storage medium having stored thereon a computer program, characterized in that, When executed by a processor, the computer program implements the steps of the machine learning-based method for predicting parameters of a natural graphite spherical production line as described in any one of claims 1 to 9.

13. A computer program product comprising computer programs / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the machine learning-based method for predicting parameters of a natural graphite spherical production line as described in any one of claims 1 to 9.

Images