Threshing and redrying processing parameter recommendation method and system based on multiple models and storage medium

Abstract

The invention relates to the field of tobacco industry intelligent manufacturing, in particular to a threshing and redrying processing parameter intelligent recommendation method and system based on multi-model machine learning and a storage medium, and the method comprises the steps: obtaining threshing and redrying historical processing data, and carrying out the preprocessing, and obtaining a processing data set; based on the processing data set, adopting multiple machine learning algorithms to train forward prediction models respectively; evaluating the performance of each forward prediction model by adopting a cross validation method, and selecting the forward prediction model with the optimal performance as a processing parameter prediction model; obtaining core machining parameters based on the machining parameter prediction model; and based on the core machining parameters, an optimization algorithm is adopted for parameter fine adjustment. According to the embodiment of the invention, the generalization ability and robustness of the prediction model are remarkably improved through a multi-model integration strategy, and the optimization dimension is greatly reduced through a core parameter screening mechanism based on feature importance.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing in the tobacco industry, specifically to a method, system, and storage medium for intelligent recommendation of leaf re-drying processing parameters based on multi-model machine learning. Background Technology

[0002] Tillage and re-drying is a crucial step in the tobacco industry's production chain, and its quality directly determines the final product's quality and market competitiveness. During this process, dozens of process parameters need to be controlled, including physical parameters such as temperature, humidity, pressure, and flow rate, as well as the operating parameters of various equipment. The settings of these parameters not only affect the physical morphology of the tobacco leaves, such as the percentage of large leaves, medium leaves, and broken leaves, but also influence the retention of the tobacco's chemical composition and sensory quality.

[0003] Traditionally, the setting of parameters for threshing and re-drying tobacco leaves relies primarily on experienced technicians who adjust parameters based on sensory judgment of the raw material characteristics and historical experience. This approach has several problems: First, it is highly dependent on experience, and inconsistent judgment standards among different operators lead to significant fluctuations in processing quality. Second, traditional experience is often insufficient to quickly adapt to tobacco leaves from different origins and of different grades, requiring extensive trial and error. Third, with the modernization and increasing complexity of threshing and re-drying equipment, the parameter space has expanded dramatically, making it difficult for manual experience to cover all possible parameter combinations. Finally, traditional methods lack quantitative performance evaluation, making it impossible to develop replicable and optimizable standardized processes.

[0004] In recent years, with the promotion of Industry 4.0 and intelligent manufacturing concepts, data-driven process parameter optimization methods have gradually attracted attention. Some studies have attempted to apply machine learning techniques to the optimization of leaf re-roasting parameters, but existing technologies still have significant shortcomings: In terms of model selection, existing methods mostly use a single machine learning model, such as neural networks or support vector machines, but different models have varying adaptability to data types and feature distributions. A single model often performs well in certain situations, but its generalization ability is limited when faced with changes in data distribution or noise interference. In terms of feature processing, existing methods lack systematic feature importance analysis, often using all available parameters as optimization variables. This not only increases computational complexity but also easily leads to the optimization process getting trapped in local optima. In fact, different quality indicators correspond to different key process parameters, and blindly optimizing all parameters is neither economical nor efficient. In terms of optimization strategies, traditional methods mostly use a single optimization algorithm, such as genetic algorithms or particle swarm optimization, but the initial solution selection of these methods is often random and lacks domain knowledge guidance. Random initialization easily leads to unstable optimization results, especially in high-dimensional parameter spaces, where the probability of converging to the global optimum is low. In terms of effect verification, existing methods lack effective recommendation result verification mechanisms. Most studies focus only on the convergence of optimization algorithms, neglecting the feasibility and predictive effectiveness of recommendation parameters in actual production. The lack of closed-loop validation makes it difficult to guarantee the reliability of recommendation results. Furthermore, existing technologies also have shortcomings in engineering practicality, such as a lack of batch processing capabilities for multiple datasets, a lack of user-friendly interfaces, and a lack of robust handling of outlier data. Summary of the Invention

[0005] The purpose of this invention is to provide a method, system, and storage medium for intelligent recommendation of processing parameters for leaf re-drying based on multi-model machine learning, in order to solve the technical problems of traditional manual experience methods being highly dependent on them and difficult to standardize, as well as existing data-driven methods having single models and optimization strategies, blind feature processing, and lack of result verification.

[0006] To achieve the above objectives, embodiments of the present invention provide an intelligent recommendation method for leaf re-drying processing parameters based on multi-model machine learning, including: Historical processing data of leaf re-drying was obtained and preprocessed to obtain the processing dataset; Based on the aforementioned processing dataset, a positive prediction model was trained using various machine learning algorithms. The performance of each positive prediction model was evaluated using cross-validation, and the best-performing positive prediction model was selected as the processing parameter prediction model. Based on the aforementioned processing parameter prediction model, the core processing parameters are obtained; Based on the core processing parameters, an optimization algorithm is used to fine-tune the parameters.

[0007] Optionally, based on the processing parameter prediction model, the core processing parameters are obtained as follows: Extract the feature importance index of each processing parameter from the processing parameter prediction model; Based on the aforementioned feature importance index, the top N most important processing parameters are selected as important parameters for each quality index; Obtain the core parameter set based on the important parameters corresponding to each quality indicator.

[0008] Optionally, based on the core processing parameters, fine-tuning the parameters using an optimization algorithm includes: Find K similar cases in historical data based on the characteristics of incoming materials and target quality indicators; Extract recommended initial processing parameters from the K similar cases; The core processing parameters are fine-tuned based on a multi-objective optimization algorithm.

[0009] Optionally, K similar cases can be found in historical data based on incoming material characteristics and target quality indicators, including: Construct a comprehensive feature vector based on incoming material characteristics and target quality indicators; The comprehensive feature vector is then standardized. The similarity is calculated according to formula (1). (1) in, The weighted Euclidean distance is used as the similarity. The global weights of the incoming material features. The global weight of the quality indicator; For the first The weight of each incoming material feature For the first The weights of each target quality indicator, Case 1 The value of the incoming material characteristic. For Case 2 The value of each incoming material characteristic; For Case 1 The value of each target quality indicator For Case 2 The value of each target quality indicator This represents the total number of incoming material characteristics. This represents the total number of quality indicators.

[0010] Optionally, the initial processing parameters extracted from the K similar cases may include: Extract the actual processing parameters from the K similar cases; Calculate the weighted average of each recommended processing parameter across K similar cases, and use this as the initial recommended processing parameter value.

[0011] Optionally, fine-tuning the core processing parameters based on a multi-objective optimization algorithm includes: The multi-objective optimization function is obtained according to formula (2). (2) in, Let be the objective function. For the processing parameter vector, Indicates when input At that time, the first Predicted values ​​for each quality indicator; Indicates the first Target values ​​for each quality indicator; For the first The weight of each quality indicator; Based on the objective optimization function, multiple optimization algorithms are run in parallel, and the globally optimal processing parameters are selected for recommendation.

[0012] Optionally, based on the objective optimization function, the following recommendations are made: Multiple optimization algorithms are run in parallel to select the globally optimal processing parameters: For each algorithm, starting with the recommended initial processing parameters, run it multiple times and obtain the results of each run. From the results of each run, select the processing parameter vector corresponding to the minimum value of the objective optimization function as the optimal solution for the current algorithm; The processing parameter vector corresponding to the minimum value of the objective optimization function is selected from the optimal solutions of various optimization algorithms as the globally optimal processing parameter recommendation.

[0013] Optionally, the recommendation method further includes evaluating the recommendation parameters, including: The recommended parameters are input into the trained processing parameter prediction model to obtain the predicted values ​​of the quality indicators. Based on the predicted values ​​of the quality indicators, calculate the absolute error rate of each quality indicator and conduct a grade evaluation for each individual quality indicator. Based on the predicted values ​​of the quality indicators, the average absolute error rate is calculated and an overall grade assessment is performed.

[0014] On the other hand, the present invention also includes an intelligent recommendation system for processing parameters of leaf re-drying based on multi-model machine learning, the system including a processor configured to perform any of the methods described above.

[0015] In another aspect, the present invention also includes a computer-readable storage medium storing instructions that, when executed by a processor, implement any of the methods described above.

[0016] The beneficial effects of this invention are: This invention significantly improves the generalization ability and robustness of the prediction model through a multi-model ensemble strategy, enabling it to adapt to different types of tobacco raw materials and varying processing conditions, resulting in more accurate recommendations. Secondly, regarding optimization efficiency, the core parameter selection mechanism based on feature importance drastically reduces the optimization dimensionality, shortening computation time by more than 50% while maintaining optimization effectiveness, greatly improving the operability of practical applications. In terms of recommendation stability, the K-nearest neighbor initial recommendation strategy fully utilizes the experience of historical successful cases, avoiding the instability caused by random initialization, thus significantly improving the consistency and reproducibility of the recommendation results. Furthermore, in terms of quality assurance, the positive model validation mechanism provides quantitative effect prediction and error analysis for each recommendation result, enabling production managers to make scientific decisions based on expected effects and risk assessments, avoiding production risks that may arise from blindly adopting recommended parameters.

[0017] Other features and advantages of the embodiments of the present invention will be described in detail in the following detailed description section. Attached Figure Description

[0018] The accompanying drawings are provided to further illustrate embodiments of the present invention and form part of the specification. They are used together with the following detailed description to explain the embodiments of the present invention, but do not constitute a limitation thereof. In the drawings: Figure 1 A flowchart of a method for intelligent recommendation of leaf re-drying processing parameters based on multi-model machine learning according to an embodiment of the present invention; Figure 2 A flowchart of a method for obtaining core processing parameters from a processing parameter prediction model according to an embodiment of the present invention; Figure 3 This is a schematic diagram illustrating the contribution of each feature in a LightGBM model according to an embodiment of the present invention. Figure 4 A flowchart illustrating a method for fine-tuning parameters using an optimization algorithm according to an embodiment of the present invention; Figure 5 A schematic diagram showing the MAPE values ​​predicted for quality indicators based on the initial recommended processing parameters according to an embodiment of the present invention. Figure 6 This is a schematic diagram showing the comparison results of different algorithms according to one embodiment of the present invention. Detailed Implementation

[0019] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit the scope of the present invention.

[0020] It should be noted that the acquisition, transmission, storage, use, and processing of data in the technical solution of this application all comply with relevant laws and regulations. In the embodiments of this application, certain existing industry solutions such as software, components, and models may be mentioned. These should be considered exemplary, intended only to illustrate the feasibility of implementing the technical solution of this application, and do not imply that the applicant has already used or necessarily used such solutions.

[0021] like Figure 1 The diagram shows a flowchart of a method for intelligent recommendation of leaf re-drying processing parameters based on multi-model machine learning, according to an embodiment of the present invention. Figure 1 In this context, the recommendation method may include the following steps: In step S10, historical processing data of leaf re-drying is obtained and preprocessed to obtain a processing dataset; In step S11, based on the processing dataset, a positive prediction model is trained using various machine learning algorithms. In step S12, the performance of each positive prediction model is evaluated using cross-validation, and the positive prediction model with the best performance is selected as the processing parameter prediction model. In step S13, core processing parameters are obtained based on the processing parameter prediction model; In step S14, based on the core processing parameters, an optimization algorithm is used to fine-tune the parameters.

[0022] In such Figure 1In the intelligent recommendation method for leaf re-drying processing parameters based on multi-model machine learning, step S10 is used to obtain the processing dataset from the historical leaf re-drying database. In this embodiment, the historical leaf re-drying processing data includes incoming material characteristic data, processing parameters, and quality indicators (final quality inspection results). Incoming material characteristic data includes the original tobacco-producing region, flue-cured tobacco variety, and digital code of the tobacco leaves. Processing parameters cover over a hundred process parameters, such as steam pressure, temperature setting, feeding speed, roller frequency, and airflow control of the vacuum rehumidifier. Quality indicators include eight key physical indicators: large leaf rate, medium leaf rate, large and medium leaf rate, small leaf rate, fragment rate, breakage rate, coarse stem content rate, and stem content in the leaf. In this example, to ensure the scientific nature of the model training, a data partitioning strategy based on product codes is adopted. All product codes are randomly divided into training and testing sets in an 8:2 ratio to ensure that data from all batches of the same product do not appear simultaneously in the training and testing sets, thus avoiding data leakage. This classification method better simulates real-world application scenarios, namely, using historical data of known products to predict the processing parameters of new products or batches.

[0023] Step S11 is used to train a positive prediction model based on the processed dataset using various machine learning algorithms. In this embodiment, the machine learning algorithms used in step S11 can be of various forms known to those skilled in the art. In one example of the present invention, step S11 can use the LightGBM model, the ElasticNet model, and the CatBoost model. Further, the training methods for these three models are as follows: LightGBM model training: Gradient boosting decision tree algorithm is used, with the objective function set to regression, the evaluation metric to root mean square error, the number of leaf nodes to be 31, the learning rate to be 0.03, the feature sampling ratio to be 0.9, the sample sampling ratio to be 0.8, and the sampling frequency to be 4. An early stopping strategy is used to prevent overfitting, with 100 early stopping rounds. ElasticNet model training: Elastic network regression algorithm is used, with regularization strength to be 1.0, ElasticNet mixture parameters to be 0.5, the maximum number of iterations to be 1000, and the feature selection strategy to be cyclic coordinate descent. CatBoost model training: Symmetric decision tree and ordered boosting strategy are used, with 1200 iterations, a learning rate to be 0.03, a tree depth to be 6, an L2 regularization parameter to be 3, and a random seed to be 42.

[0024] Step S12 is used to evaluate the performance of each forward prediction model using cross-validation and select the best-performing forward prediction model as the processing parameter prediction model. Specifically, in this example, this can involve calculating the root mean square error (RMSE), mean absolute error (MAE), and coefficient of determination (R²) for each quality indicator, and selecting the model with the best overall performance as the base model for subsequent recommendations.

[0025] Step S13 is used to obtain core processing parameters based on the processing parameter prediction model. In this embodiment, the specific method for obtaining the core processing parameters in step S13 can be of various forms known to those skilled in the art. In one example of the present invention, step S13 may include, for example... Figure 2 The steps shown are described. Figure 2 In this context, step S13 may include: In step S20, the feature importance index of each processing parameter is extracted from the processing parameter prediction model; In step S21, based on the feature importance index, the top N most important processing parameters are selected as important parameters for each quality index; In step S22, the core parameter set is obtained based on the important parameters corresponding to each quality indicator.

[0026] In such Figure 2 In the method shown, step S20 is used to extract feature importance indices. Feature importance analysis is the foundation for core parameter selection. For different types of models, corresponding importance calculation methods are used. Specifically, in this example, gain importance is used for the LightGBM model to reflect the contribution of each feature to the model, such as... Figure 3 As shown. The gain importance calculation formula is the sum of the reductions in the loss function caused by each feature splitting across all trees. The ElasticNet model uses the absolute value of the regression coefficients as the feature importance index. Because ElasticNet has feature selection capabilities, features with larger absolute coefficient values ​​have a more significant impact on the target variable. The CatBoost model uses a built-in feature importance calculation method, comprehensively evaluating based on the frequency and effectiveness of feature usage during the splitting process.

[0027] Step S21 extracts the feature importance ranking of the corresponding model for each quality indicator, selecting the top N most important processing parameters. Then, step S22 calculates the union of the important parameters corresponding to all quality indicators, forming the core parameter set. This method ensures that the optimization process focuses on both the key parameters of individual quality indicators and the need for collaborative optimization of multiple indicators. In this example, the value of N can be adjusted between 5 and 20 according to the actual situation.

[0028] The optimization boundaries of the core parameters are determined based on statistical analysis of historical data. For each core parameter, its 1% and 99th percentiles in historical data are calculated as optimization boundaries, which avoids recommending parameters beyond a reasonable range while preserving sufficient optimization space.

[0029] Step S14 is used to fine-tune the parameters based on the core processing parameters using an optimization algorithm. In this embodiment, the specific method for fine-tuning the parameters using the optimization algorithm in step S14 can be of various forms known to those skilled in the art. In one example of the present invention, step S14 may include, for example... Figure 4 The steps shown are described. Figure 4 In this context, step S14 may include: In step S30, K similar cases are found in historical data based on the characteristics of incoming materials and target quality indicators; In step S31, initial processing parameter recommendations are extracted from K similar cases; In step S32, the core processing parameters are fine-tuned based on a multi-objective optimization algorithm.

[0030] In such Figure 4 In the method shown, step S30 is used to find K similar cases. In this example, it can be done by first constructing a comprehensive feature vector, then standardizing the material features and target quality indicators and merging them. Standardization uses the Z-score method to eliminate the influence of different feature dimensions. Weighted Euclidean distance is used as the similarity. Specifically, it can be calculated using formula (1), for example. (1) in, The weighted Euclidean distance is used as the similarity. The global weights of the incoming material features. The global weight of the quality indicator; For the first The weight of each incoming material feature For the first The weights of each target quality indicator, Case 1 The value of the incoming material characteristic. For Case 2 The value of each incoming material characteristic; For Case 1 The value of each target quality indicator For Case 2 The value of each target quality indicator This represents the total number of incoming material characteristics. This represents the total number of quality indicators. In this example, and It is usually set to 0.5 to balance the influence of the two types of features; and It can be adjusted based on expert experience or data analysis results.

[0031] Step S31 is used to extract recommended initial processing parameters from K similar cases. In this example, the specific method for extracting recommended initial processing parameters in step S31 may include: In step S40, the actual processing parameters are extracted from K similar cases; In step S41, the weighted average of each recommended processing parameter in K similar cases is calculated as the initial recommended processing parameter value.

[0032] Step S40 involves finding K most similar cases (K is typically set to 3-20) in historical data and extracting the corresponding processing parameters as candidate initial recommendations. Step S41 involves calculating the final recommended values ​​of the initial processing parameters using a weighted average. Specific weighting methods can include inverse weighting, mean weighting, softmax weighting, etc. The MAPE value of the initial recommended processing parameters for predicting the quality index is shown in [link to relevant documentation]. Figure 5 .

[0033] Step S32 is used to fine-tune the core processing parameters based on a multi-objective optimization algorithm. In this example, the multi-objective optimization function can be defined first according to formula (2): (2) in, Let be the objective function. For the processing parameter vector, Indicates when input At that time, the first Predicted values ​​for each quality indicator; Indicates the first Target values ​​for each quality indicator; For the first The weight of each quality indicator; Based on the objective optimization function, various optimization algorithms are used for parameter fine-tuning. In this example, the specific methods for using various optimization algorithms for parameter fine-tuning may include: In step S50, for each algorithm, the recommended initial processing parameters are used as the starting point, the algorithm is run multiple times, and the results of each run are obtained. In step S51, the processing parameter vector corresponding to the minimum value of the target optimization function is selected from the results of each run as the optimal solution of the current algorithm; In step S52, the processing parameter vector corresponding to the minimum value of the objective optimization function is selected from the optimal solutions of various optimization algorithms as the globally optimal processing parameter recommendation.

[0034] Steps S50 and S51 are used for intra-algorithm optimization; for each algorithm, the loss function value is selected from the results of its multiple runs. The solution with the smallest value is taken as the optimal solution of the algorithm. Specifically, in this example, the Nelder-Mead simplex algorithm, the COBYLA algorithm, and the Powell direction search algorithm can be used. The Nelder-Mead simplex algorithm is suitable for unconstrained optimization problems, finding the optimal solution by constructing a simplex and performing reflection, expansion, and contraction operations. This algorithm does not require gradient information and has low requirements for the smoothness of the objective function, making it suitable for handling complex machine learning model prediction functions. The COBYLA (Constrained Optimization By Linear Approximations) algorithm is suitable for constrained optimization problems and can handle parameter boundary constraints. This algorithm handles constraints through linear approximation methods, making it suitable for handling physical boundary limitations of processing parameters. The Powell direction search algorithm uses a conjugate direction search strategy and has strong local search capabilities. This algorithm converges quickly when close to the optimal solution, making it suitable for fine-tuning based on good initial values ​​provided by K-nearest neighbors. Each optimization algorithm is run multiple times, and the result with the smallest loss function value is selected as the optimal solution of the algorithm. Then, the results of different algorithms are compared in step S52, such as... Figure 6 As shown, the globally optimal solution is selected as the final recommendation.

[0035] The intelligent recommendation method for processing parameters of leaf re-drying based on multi-model machine learning in this invention can also include evaluating the recommended parameters. Specifically, in this example, the method for evaluating the effectiveness of the recommended parameters can be to input the recommended processing parameters into a trained positive prediction model to predict the corresponding quality indicators. Since the recommended parameters only involve core parameters, non-core parameters need to be set to historical averages or K-nearest neighbor recommended values. Then, error calculation and grade evaluation are performed. Multiple error indicators are calculated, including absolute error, relative error, and absolute error rate. The formula for calculating the absolute error is: The formula for calculating relative error is: The formula for calculating the absolute error rate is: The grade is assessed based on the absolute error rate: Excellent: Absolute error rate <5%, indicating that the recommendation effect is very good and can be applied directly; Good: 5% ≤ absolute error rate <10%, indicating good recommended performance and application is advised; Generally: 10% ≤ absolute error rate < 20% indicates that the recommended effect is average, and small-scale trials are suggested; Needs optimization: An absolute error rate ≥ 20% indicates poor recommendation performance and requires further optimization.

[0036] Calculate the overall evaluation index according to formula (3): (3) The overall grade is determined based on the mean absolute error rate, providing users with an overall evaluation of the recommendation results.

[0037] The system also generates detailed analysis reports, including recommended parameter tables, prediction effect analysis tables, and error statistics charts, providing comprehensive data support for production decisions.

[0038] On the other hand, the present invention also includes an intelligent recommendation system for leaf re-drying processing parameters based on multi-model machine learning, the system including a processor configured to perform any of the methods described in the intelligent recommendation method for leaf re-drying processing parameters based on multi-model machine learning.

[0039] In another aspect, the present invention also includes a computer-readable storage medium storing instructions that, when executed by a processor, implement any of the methods described in the intelligent recommendation method for leaf re-drying processing parameters based on multi-model machine learning.

[0040] The beneficial effects of this invention are: This invention significantly improves the generalization ability and robustness of the prediction model through a multi-model ensemble strategy, enabling it to adapt to different types of tobacco raw materials and varying processing conditions, resulting in more accurate recommendations. Secondly, regarding optimization efficiency, the core parameter selection mechanism based on feature importance drastically reduces the optimization dimensionality, shortening computation time by more than 50% while maintaining optimization effectiveness, greatly improving the operability of practical applications. In terms of recommendation stability, the K-nearest neighbor initial recommendation strategy fully utilizes the experience of historical successful cases, avoiding the instability caused by random initialization, thus significantly improving the consistency and reproducibility of the recommendation results. Furthermore, in terms of quality assurance, the positive model validation mechanism provides quantitative effect prediction and error analysis for each recommendation result, enabling production managers to make scientific decisions based on expected effects and risk assessments, avoiding production risks that may arise from blindly adopting recommended parameters.

[0041] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application 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.

[0042] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. 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... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0043] 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.

[0044] 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.

[0045] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0046] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0047] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0048] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0049] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A multi-model machine learning-based intelligent recommendation method for leaf threshing and redrying processing parameters, characterized in that, The recommendation method includes: Historical processing data of leaf re-drying was obtained and preprocessed to obtain the processing dataset; Based on the aforementioned processing dataset, a positive prediction model was trained using various machine learning algorithms. The performance of each positive prediction model was evaluated using cross-validation, and the best-performing positive prediction model was selected as the processing parameter prediction model. Based on the aforementioned processing parameter prediction model, the core processing parameters are obtained; Based on the core processing parameters, an optimization algorithm is used to fine-tune the parameters.

2. The recommendation method of claim 1, wherein, Based on the aforementioned processing parameter prediction model, the core processing parameters are obtained as follows: Extract the feature importance index of each processing parameter from the processing parameter prediction model; Based on the aforementioned feature importance index, the top N most important processing parameters are selected as important parameters for each quality index; Obtain the core parameter set based on the important parameters corresponding to each quality indicator.

3. The recommendation method of claim 1, wherein, Based on the aforementioned core processing parameters, parameter fine-tuning using optimization algorithms includes: Find K similar cases in historical data based on the characteristics of incoming materials and target quality indicators; Extract recommended initial processing parameters from the K similar cases; The core processing parameters are fine-tuned based on a multi-objective optimization algorithm.

4. The recommended method according to claim 3, characterized in that, Based on the characteristics of incoming materials and target quality indicators, find K similar cases in historical data, including: Construct a comprehensive feature vector based on incoming material characteristics and target quality indicators; The comprehensive feature vector is then standardized. The similarity is calculated according to formula (1). ，（1） in, The weighted Euclidean distance is used as the similarity. The global weights of the incoming material features. The global weight of the quality indicator; For the first The weight of each incoming material feature For the first The weights of each target quality indicator, Case 1 The value of the incoming material characteristic. For Case 2 The value of each incoming material characteristic; For Case 1 The value of each target quality indicator For Case 2 The value of each target quality indicator This represents the total number of incoming material characteristics. This represents the total number of quality indicators.

5. The recommended method according to claim 3, characterized in that, The initial processing parameters extracted from the K similar cases are recommended as follows: Extract the actual processing parameters from the K similar cases; Calculate the weighted average of each recommended processing parameter across K similar cases, and use this as the initial recommended processing parameter value.

6. The recommended method according to claim 3, characterized in that, Fine-tuning the core processing parameters based on a multi-objective optimization algorithm includes: The multi-objective optimization function is obtained according to formula (2). ，（2） in, Let be the objective function. For the processing parameter vector, Indicates when input At that time, the first Predicted values ​​for each quality indicator; Indicates the first Target values ​​for each quality indicator; For the first The weight of each quality indicator; Based on the objective optimization function, multiple optimization algorithms are run in parallel, and the globally optimal processing parameters are selected for recommendation.

7. The recommended method according to claim 6, characterized in that, Based on the aforementioned objective optimization function, the following processing parameters are recommended for parallel operation of multiple optimization algorithms to select the globally optimal parameters: For each algorithm, starting with the recommended initial processing parameters, run it multiple times and obtain the results of each run. From the results of each run, select the processing parameter vector corresponding to the minimum value of the objective optimization function as the optimal solution for the current algorithm; The processing parameter vector corresponding to the minimum value of the objective optimization function is selected from the optimal solutions of various optimization algorithms as the globally optimal processing parameter recommendation.

8. The recommended method according to claim 1, characterized in that, The recommendation method further includes evaluating the recommendation parameters, including: The recommended parameters are input into the trained processing parameter prediction model to obtain the predicted values ​​of the quality indicators. Based on the predicted values ​​of the quality indicators, calculate the absolute error rate of each quality indicator and conduct a grade evaluation for each individual quality indicator. Based on the predicted values ​​of the quality indicators, the average absolute error rate is calculated and an overall grade assessment is performed.

9. A smart recommendation system for leaf re-drying processing parameters based on multi-model machine learning, characterized in that, The system includes a processor configured to perform the method as described in any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores instructions that, when executed by a processor, implement the method as described in any one of claims 1 to 8.

Images