Process parameter verification method, electronic device, and storage medium

Abstract

This disclosure provides a process parameter verification method, electronic device, and storage medium, relating to the field of biomass composite material processing technology. The process parameter verification method includes: acquiring a sample dataset; determining a prediction model trained based on the sample dataset; performing a global attribution analysis on the prediction model based on Shapley additive interpretation to determine the SHAP values ​​of multiple process parameters for each target macroscopic performance index; determining the feature importance ranking, positive and negative influence directions, and critical thresholds of the multiple process parameters based on their respective SHAP values ​​for the target macroscopic performance index; selecting key process parameters from the multiple process parameters based on their feature importance ranking, positive and negative influence directions, and critical thresholds; acquiring microscopic physical structure characterization data of biomass composite material samples related to the key process parameters; and verifying the influence mechanism of the key process parameters on the target macroscopic performance index based on the microscopic physical structure characterization data.

Description

Technical Field

[0001] This disclosure relates to the field of biomass composite material processing technology, specifically to a process parameter verification method, electronic equipment, and storage medium. Background Technology

[0002] In the manufacturing process of biomass composite materials (such as bamboo-based fiber composites and wood-based reconstituted materials), their macroscopic performance indicators, such as mechanical properties and water resistance, are influenced by a combination of various process parameters, exhibiting typical multivariable, nonlinear, and strongly interactive characteristics. To ensure the stability and optimization of product performance, it is necessary to accurately understand the impact mechanism of each process parameter on macroscopic performance, and then select the process parameters that play a key role in the target performance to achieve targeted process control. However, purely data-driven methods only reach the level of macroscopic statistical regression and lack reverse verification methods based on microscopic physical structures, resulting in insufficient physical credibility and limited predictive reliability when making process decisions. Summary of the Invention

[0003] In view of this, embodiments of the present disclosure provide a process parameter verification method, an electronic device, and a storage medium.

[0004] In a first aspect, one embodiment of this disclosure provides a method for verifying process parameters, comprising: acquiring a sample dataset, the sample dataset including multiple process parameters and target macroscopic performance indicators of biomass composite material samples; determining a prediction model trained based on the sample dataset, the prediction model being used to predict the target macroscopic performance indicators based on the input process parameters; performing a global attribution analysis on the prediction model based on Shapley Additive Explanations (SHAP) to determine the SHAP values ​​of each of the multiple process parameters for the target macroscopic performance indicators; determining the feature importance ranking, positive and negative influence directions, and critical thresholds of the multiple process parameters based on the SHAP values ​​of each of the multiple process parameters for the target macroscopic performance indicators; selecting key process parameters from the multiple process parameters according to the feature importance ranking, positive and negative influence directions, and critical thresholds; acquiring microscopic physical structure characterization data of biomass composite material samples related to the key process parameters, and verifying the influence mechanism of the key process parameters on the target macroscopic performance indicators based on the microscopic physical structure characterization data, wherein the influence mechanism characterizes the physical path by which the key process parameters affect the target macroscopic performance indicators through changes in the microscopic physical structure of the material.

[0005] In conjunction with the first aspect, in certain implementations of the first aspect, microscopic physical structure characterization data of biomass composite material samples related to key process parameters are obtained, and based on the microscopic physical structure characterization data, the influence mechanism of key process parameters on target macroscopic performance indicators is verified. This includes: obtaining microscopic physical structure characterization data of biomass composite material samples corresponding to the values ​​of key process parameters within a preset range on both sides of the critical threshold, based on the critical threshold of the key process parameters, as test microscopic physical structure characterization data; comparing whether the changing trend of the test microscopic physical structure characterization data with the value of the key process parameters is consistent with the changing trend predicted by the positive and negative influence directions of the key process parameters; if consistent, it is determined that the influence mechanism of key process parameters on target macroscopic performance indicators has been verified.

[0006] In conjunction with the first aspect, in some implementations of the first aspect, multiple process parameters include continuous variables and categorical variables; before determining the prediction model trained based on the sample dataset, the process parameter verification method further includes: performing one-hot encoding transformation on the categorical variables and normalizing the continuous variables to obtain a preprocessed sample dataset; wherein, the prediction model is trained based on the preprocessed sample dataset.

[0007] In conjunction with the first aspect, in some implementations of the first aspect, the prediction model is at least one of Lightweight Gradient Boosting Machine (LightGBM), Random Forest (RF), Categorical Boosting (CatBoost), Support Vector Regression (SVR), Neural Network (NN), or eXtreme Gradient Boosting (XGBoost) with built-in L1 / L2 regularization terms; determining the prediction model trained on the sample dataset includes: optimizing the hyperparameters of the prediction model and evaluating its generalization ability based on cross-validation to obtain the prediction model.

[0008] In conjunction with the first aspect, in some implementations of the first aspect, multiple process parameters include carbonization process parameters; if the biomass composite material is a bamboo-based fiber composite material, then the carbonization process parameters include a carbonization severity factor, which is determined by combining carbonization temperature and carbonization time in a dimension-reducing manner.

[0009] In conjunction with the first aspect, in some implementations of the first aspect, after verifying the influence mechanism of key process parameters on the target macroscopic performance index, the process parameter verification method further includes: if the influence mechanism is verified, then using the prediction model as the fitness function surrogate model, and the target macroscopic performance index as the optimization objective, performing a multi-objective optimization search within the preset process parameter constraint space, and outputting a recommended set of process parameters.

[0010] In conjunction with the first aspect, in some implementations of the first aspect, the multi-objective optimization search adopts the Multi-Objective Bayesian Optimization (MOBO) algorithm or the second-generation Nondominated Sorting Genetic Algorithm II (NSGA-II), and the output recommended process parameter set is the Pareto optimal process solution set.

[0011] In conjunction with the first aspect, in some implementations of the first aspect, the target macroscopic performance indicators include mechanical performance indicators and / or water resistance performance indicators; wherein, the mechanical performance indicators include at least one of the modulus of flexural strength (MOR) and modulus of elasticity (MOE); and the water resistance performance indicators include at least one of the thickness swelling rate (TSR) and water absorption rate (WAR).

[0012] In a second aspect, one embodiment of this disclosure provides an electronic device, including a processor and a memory, wherein computer-executable instructions are stored in the memory, and the processor executes the computer-executable instructions to implement the method of the first aspect.

[0013] Thirdly, one embodiment of this disclosure provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a computer, implement the method of the first aspect.

[0014] In this embodiment, a sample dataset containing process parameters and target macroscopic performance indicators is acquired, and a prediction model trained based on the sample dataset is determined. Key process parameters are identified by ranking the feature importance of each process parameter in the prediction model, determining their positive and negative influence directions, and using critical thresholds. Furthermore, microscopic physical structure characterization data related to these key process parameters is obtained, and the microscopic physical path through which these key process parameters influence the target macroscopic performance indicators is verified. This method combines data-driven decision contribution analysis with microscopic physical structure characterization, providing traceable physical path evidence for the selection of key process parameters. It overcomes the black-box nature of pure machine learning models in interpreting process performance relationships, thereby effectively improving the physical credibility of process parameter selection and the reliability of subsequent process optimization predictions. Attached Figure Description

[0015] The above and other objects, features, and advantages of this disclosure will become more apparent from the more detailed description of the embodiments thereof in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this disclosure and form part of the specification. They are used together with the embodiments of this disclosure to explain the disclosure and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.

[0016] Figure 1 The diagram shown is a flowchart illustrating a process parameter verification method provided in an embodiment of this disclosure.

[0017] Figure 2 The diagram shown is a structural schematic of a process parameter verification device provided in an embodiment of this disclosure.

[0018] Figure 3 The diagram shown is a structural schematic of an electronic device provided in an embodiment of this disclosure. Detailed Implementation

[0019] The technical solutions of the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this disclosure, and not all embodiments. Based on the embodiments of this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.

[0020] Biomass composite materials, such as bamboo-based fiber composites and wood-based reconstituted materials, have shown broad application prospects in construction, landscaping, and transportation due to their excellent mechanical properties and environmentally friendly characteristics. However, the performance of these materials is affected by the coupling of various process parameters, exhibiting typical multivariate, nonlinear, and strongly interactive characteristics. Traditional process optimization methods, such as single-factor experiments, empirical trial-and-error methods, and artificial neural networks, have accumulated some understanding in performance fitting and parameter selection.

[0021] However, existing process optimization methods still have significant limitations in engineering applications. Traditional methods rely on trial and error or single-factor experiments, which makes it difficult to reveal the complex interaction mechanisms between multiple parameters. Furthermore, these methods are time-consuming and costly, failing to meet the needs of rapid research and development and precise control.

[0022] In recent years, data-driven machine learning methods have provided new insights into process optimization. However, existing research often employs an end-to-end black-box modeling approach focusing on process-performance. While these models possess some fitting ability, they generally lack physical interpretability and fail to clearly define the interaction mechanisms of various process parameters. This makes it difficult for the models to guide precise adjustments to actual production processes, and they lack the ability to recommend process parameters in reverse based on target performance. Furthermore, existing methods often neglect the regulatory role of microstructure (such as pore structure and resin crosslinking degree) on macroscopic performance, resulting in insufficient reliability of model predictions.

[0023] In summary, existing methods for optimizing biomass composite materials processes are insufficient in terms of interpretability and reverse design capabilities, and a closed-loop technical solution capable of supporting the precise manufacturing and customized development of high-performance materials has not yet been formed.

[0024] To address the aforementioned technical problems, this disclosure provides a method for verifying process parameters of biomass composite materials, using bamboo-based fiber composites and wood-based reconstituted materials as examples. The model construction method is applicable to various biomass composite materials and is not limited to bamboo-based fiber composites and wood-based reconstituted materials.

[0025] The following is combined with Figure 1 Please provide a detailed explanation. Figure 1 The diagram shown is a flowchart illustrating a process parameter verification method provided in an embodiment of this disclosure. Figure 1 As shown, the process parameter verification method includes the following steps.

[0026] Step S110: Obtain the sample dataset.

[0027] For example, a sample dataset refers to a collection of data used to train a prediction model. In some embodiments, the sample dataset includes multiple process parameters and target macroscopic performance indicators of biomass composite material samples.

[0028] For example, a biomass composite material sample refers to a composite material sample formed by the recombination of natural biomass through physical or chemical means. In some embodiments, biomass composite material samples under different process conditions can be collected from pilot-scale tests or actual production lines of at least one enterprise. The biomass composite material sample can be a bamboo-based fiber composite material or a wood-based reconstituted material. Other materials can also be used for the biomass composite material sample, without limitation.

[0029] For example, process parameters refer to controllable process conditions during the preparation of biomass composite materials. For instance, when the biomass composite material is a bamboo-based fiber composite, process parameters may include bamboo species, hot-pressing density, resin content, hot-pressing temperature, and hot-pressing time. As another example, when the biomass composite material is a wood-based reconstituted material, process parameters may include tree species, veneer thickness, resin impregnation content, moisture content before installation, reconstituted wood density, and thermosetting insulation temperature. In some embodiments, process parameters can be manually recorded during the preparation of the biomass composite material.

[0030] In actual process parameter acquisition, sample datasets typically contain variables of different types, with varying dimensions and numerical distributions. Directly using these datasets for model training can lead to biased feature weights. To address this issue, this disclosure provides an optional embodiment that performs uniform preprocessing on the raw data to eliminate the interference of data type heterogeneity and dimensional differences on model training. The specific implementation is as follows.

[0031] In some embodiments, the multiple process parameters include continuous variables and categorical variables.

[0032] For example, continuous variables refer to process parameters that can take any value within the real number range, and their values ​​are ordered and have metrical significance. For instance, continuous variables may include hot-pressing temperature, hot-pressing time, resin impregnation amount, density, etc. The magnitude of continuous variables directly reflects the intensity or degree of process conditions, and their distribution range may vary considerably.

[0033] For example, categorical variables refer to process parameters that take values ​​in a finite number of discrete categories and do not have numerical significance for comparison. For example, categorical variables may include bamboo species, tree species, hot pressing and cold pressing process types, etc.

[0034] In some embodiments, before determining the prediction model trained on the sample dataset (i.e., step S120), the process parameter verification method further includes: performing one-hot encoding transformation on categorical variables and normalizing continuous variables to obtain a preprocessed sample dataset.

[0035] For example, one-hot encoding transformation refers to an encoding method that converts categorical variables into numerical vectors. For a categorical variable with H different categories, one-hot encoding expands it into an H-dimensional binary vector, where each dimension corresponds to a category, with the dimension corresponding to that category having a value of 1, and the other dimensions having values ​​of 0. One-hot encoding transformation avoids incorrectly assigning size order to unordered categories.

[0036] For example, normalization is a linear transformation method that scales the numerical distribution of continuous variables to a uniform numerical range. Typically, min-max normalization is used, linearly mapping the original data to the [0,1] interval according to the minimum and maximum values. Normalization eliminates the impact of differences in units and orders of magnitude on model training.

[0037] In some embodiments, the preprocessing operation further includes outlier removal before performing one-hot encoding transformation on categorical variables and normalization on continuous variables.

[0038] For example, outlier removal refers to identifying and removing data points that significantly deviate from the distribution of the rest of the data from the sample dataset, in order to eliminate the interference caused by measurement errors or process anomalies on model training.

[0039] In this embodiment, the prediction model is trained based on the preprocessed sample dataset.

[0040] Some embodiments in this specification effectively address the data heterogeneity problem caused by the coexistence of continuous and categorical variables in biomass composite material process parameters by performing one-hot encoding transformation on categorical variables and normalizing continuous variables in the sample dataset. This preprocessing operation eliminates interference from differences in dimensionality and inconsistent numerical distribution ranges between different variables, enabling subsequent prediction models to fairly learn the mapping relationship between each process parameter and the target macroscopic performance index within a unified feature space, avoiding feature weight bias caused by differences in data type or dimensionality. Simultaneously, one-hot encoding transformation preserves the class independence of categorical variables, avoiding the introduction of spurious order relationships, while normalization accelerates the convergence speed of the model training process. These preprocessing operations collectively improve the usability of the sample dataset and the stability of model training.

[0041] In biomass composite materials, there is a strong interactive coupling relationship between carbonization process parameters (such as carbonization temperature and carbonization time). If these parameters are directly input into the model as two independent variables, it will lead to feature dimensionality redundancy and an increased risk of overfitting. To address the above issues, this disclosure provides an optional embodiment that can perform dimensionality reduction characterization on the carbonization process parameters of bamboo-based fiber composite materials to simplify the feature space and retain key physical information. The specific implementation is as follows.

[0042] In some embodiments, the plurality of process parameters include carbonization process parameters.

[0043] For example, carbonization process parameters refer to the process conditions involving carbonization in the preparation of biomass composite materials.

[0044] In some embodiments, if the biomass composite material is a bamboo-based fiber composite material, the carbonization process parameters include a carbonization severity factor. It should be noted that the carbonization process parameters and the subsequent hot-pressing process parameters such as hot-pressing temperature and hot-pressing time belong to different process stages, and both are independently input into the model.

[0045] For example, the carbonization severity factor is a single continuous variable used to comprehensively characterize the intensity of carbonization treatment. The carbonization severity factor reflects the overall impact of the carbonization process on the properties of bamboo-based fiber composites; a higher carbonization severity factor value indicates a deeper degree of carbonization treatment. In some embodiments, the carbonization severity factor can be determined by combining carbonization temperature and carbonization time in a dimension-reduced manner.

[0046] For example, the carbonization temperature is the ambient temperature of the carbonization process, and the carbonization time is the duration for which the material is held at the carbonization temperature.

[0047] For example, dimensionality reduction merging refers to a process that maps two or more original feature variables with strong interaction relationships into a single new feature variable through a specific mathematical transformation formula. In some embodiments, the carbonization severity factor can be determined by fusing carbonization temperature and carbonization time using a carbonization severity factor calculation formula.

[0048] For example, the formula for calculating the carbonization severity factor can be formula (1): Where S is the carbonization severity factor, t is the carbonization time, T is the carbonization temperature, and k is an empirical constant, exemplarily 14.75. Without a carbonization process, the carbonization severity factor is 0.

[0049] In some embodiments of this specification, the strong interactive coupling and non-independent dimensionality between carbonization temperature and carbonization time in bamboo-based fiber composites are addressed by introducing a carbonization severity factor to reduce the dimensionality of these two factors and merge them into a single continuous variable. This approach effectively reduces the dimensionality of the input feature space, eliminates collinearity interference between carbonization temperature and time, and avoids the overfitting risk caused by directly inputting two-dimensional variables into the model. Simultaneously, the carbonization severity factor can comprehensively characterize the overall intensity of the carbonization process as a single parameter, preserving the key physical information of the temperature-time coupling effect, enabling subsequent machine learning models to more accurately capture the nonlinear monotonic influence of carbonization degree on target performance. Furthermore, the single variable after dimensionality reduction and merging improves the model's compatibility with different process routes, enhancing its generalization ability and engineering practicality.

[0050] For example, the target macroscopic performance index refers to an index used to characterize the service performance of biomass composite materials. In some embodiments, the target macroscopic performance index can be obtained by macroscopic performance testing according to the test methods specified in the relevant national or industry standards.

[0051] In practical engineering applications, to meet diverse performance evaluation needs, this disclosure provides an optional embodiment that can specifically limit the target macroscopic performance index to mechanical performance index and / or water resistance performance index, and specify the corresponding specific test parameters, as follows.

[0052] In some embodiments, the target macroscopic performance indicators include mechanical performance indicators and / or water resistance performance indicators.

[0053] For example, mechanical performance indicators refer to quantitative indicators of a material's resistance to deformation and failure under stress. In some embodiments, mechanical performance indicators include at least one of Modulus of Rupture (MOR) and Modulus of Elasticity (MOE). MOR refers to the maximum bending stress that a material can withstand in a bending test until failure. The higher the MOR, the stronger the material's bending load-bearing capacity. MOE refers to the ratio of bending stress to corresponding strain during the elastic deformation stage of a material. The higher the MOE, the greater the rigidity of the material, and the less likely it is to deform under stress.

[0054] For example, water resistance performance indicators refer to quantitative indicators of a material's ability to resist dimensional changes and moisture absorption under humid environments or water immersion conditions. In some embodiments, water resistance performance indicators include at least one of Thickness Swelling Rate (TSR) and Water Absorption Rate (WAR). TSR refers to the percentage increase in thickness of a material after absorbing water under specified conditions relative to its initial thickness. The lower the TSR, the better the dimensional stability and the superior water resistance of the material. WAR refers to the percentage increase in mass of a material after absorbing water under specified conditions relative to its initial mass. The lower the WAR, the higher the density of the material and the weaker its water permeability.

[0055] For example, MOR and MOE can be determined by the three-point bending test according to GB / T 17657-2022 "Test Methods for Physical and Chemical Properties of Wood-based Panels and Decorative Wood-based Panels"; TSR and WAR can be determined by the water immersion test method in the same standard.

[0056] In some embodiments of this specification, by explicitly defining the target macroscopic performance indicators as mechanical performance indicators and / or water resistance performance indicators, and further specifying the mechanical performance indicators as MOR and / or MOE, and the water resistance performance indicators as TSR and / or WAR, the performance evaluation dimensions applicable to the process parameter verification method are effectively defined. Each type of performance indicator provides multiple selectable testing parameters, allowing users to flexibly choose according to actual application scenarios and testing conditions, thus enhancing the adaptability and operability of the technical solution.

[0057] Step S120: Determine the prediction model trained based on the sample dataset.

[0058] For example, a predictive model is a model that can establish a nonlinear mapping relationship between process parameters and target macroscopic performance indicators. A predictive model can be used to predict target macroscopic performance indicators based on input process parameters.

[0059] In some embodiments, the predictive model can be a machine learning model. For example, a deep neural network (DNN) model, a convolutional neural network (CNN) model, a recurrent neural network (RNN) model, or any combination thereof.

[0060] In some embodiments, the input to the prediction model may include multiple process parameters, and the output may include a target macroscopic performance index.

[0061] In some embodiments, the prediction model can be trained using multiple first training samples with a first label. Multiple first training samples with the first label can be input into the initial prediction model. A loss function is constructed using the first labels and the results of the initial prediction model. The parameters of the initial prediction model are then iteratively updated based on the loss function. Model training is complete when the loss function of the initial prediction model satisfies a preset condition, resulting in a trained prediction model. The preset condition may be loss function convergence, the number of iterations reaching a threshold, etc.

[0062] In some embodiments, the first training sample can be a process parameter in the sample dataset, and the first label can be a target macroscopic performance indicator in the sample dataset.

[0063] Some embodiments in this specification use predictive models to process process parameters. By leveraging the self-learning capabilities of machine learning models, patterns can be found in a large amount of biomass composite material production data, and the correlation between process parameters and target macroscopic performance indicators can be obtained, thereby improving the accuracy and efficiency of determining the mapping relationship between process parameters and target macroscopic performance indicators.

[0064] In practical applications, ordinary machine learning models often exhibit insufficient robustness and poor generalization ability when faced with noise interference and the risk of overfitting to small samples, which are common in industrial data. To overcome these problems, this disclosure provides an optional embodiment that can employ a specific prediction model with regularization mechanisms and ensemble learning advantages, and improve model performance through a systematic hyperparameter optimization strategy, as specifically implemented below.

[0065] In some embodiments, the prediction model is at least one of Light Gradient Boosting Machine (LightGBM), Random Forest (RF), Categorical Boosting (CatBoost), Support Vector Regression (SVR), Neural Network (NN), or eXtreme Gradient Boosting (XGBoost) with built-in L1 / L2 regularization.

[0066] For example, LightGBM uses a histogram-based decision tree splitting strategy for iterative training, discretizing continuous floating-point features into a fixed number of integer histograms to reduce memory consumption and computational complexity. It also introduces a one-sided gradient sampling algorithm and a mutually exclusive feature binding algorithm to further improve training efficiency. Finally, the predicted values ​​of all decision trees are summed as the final output. LightGBM exhibits high training efficiency and low memory usage when processing large-scale structured tabular data.

[0067] For example, Randomized Randomized Fire (RF) generates multiple training subsets through bootstrapping and randomly selects a subset of features from all features for optimal splitting at the nodes of each decision tree. After each decision tree grows independently, the prediction results of each tree are integrated by averaging to obtain the final output. RF exhibits good anti-overfitting ability and model stability when dealing with high-dimensional feature spaces.

[0068] For example, CatBoost uses an ordered boosting and symmetric tree structure strategy for iterative training. Each new tree is used to correct the prediction error of the preceding model, and a target statistics-based encoding method is used to transform categorical features into numerical features for training. Finally, the prediction values ​​of all trees are summed as the final output. CatBoost has high prediction accuracy and a simplified preprocessing workflow when processing structured tabular data containing many categorical features.

[0069] For example, SVR constructs a regression pipeline with a width of 2ε by introducing an ε-insensitive loss function. It only uses sample points located on or outside the pipeline boundary as support vectors in model construction and uses a kernel function to map the low-dimensional nonlinear problem to a high-dimensional linear space for regression fitting. SVR has strong generalization ability and regression accuracy on small sample datasets.

[0070] For example, a neural network (NN) transmits signals through weighted connections between the input layer, one or more hidden layers, and the output layer. Each neuron in each layer undergoes a nonlinear transformation via an activation function. The model calculates predicted values ​​through forward propagation, calculates the gradient of the loss function with respect to the weights of each layer through backpropagation, and iteratively updates the weights, ultimately converging to the optimal prediction model. NNs possess extremely strong nonlinear fitting capabilities when processing massive amounts of unstructured data.

[0071] For example, XGBoost iteratively generates multiple decision trees, each new tree fitting the residual of the previous tree's prediction, and finally sums the predictions from all trees as the final output. XGBoost exhibits high prediction accuracy and computational efficiency when processing structured tabular data.

[0072] For example, L1 / L2 regularization terms refer to additional penalty terms added to the model's loss function. L1 regularization incorporates the sum of the absolute values ​​of the feature weights into the loss function, which can drive some unimportant feature weights to approach zero, thus achieving the function of feature selection; L2 regularization incorporates the sum of the squares of the feature weights into the loss function, which can limit the feature weights from being too large and suppress the model from overfitting to the noise of the training data.

[0073] In some embodiments, determining a prediction model trained based on sample data includes: optimizing the hyperparameters of the prediction model and evaluating its generalization ability based on cross-validation to obtain the prediction model.

[0074] For example, hyperparameter optimization refers to the process of systematically searching and selecting the best hyperparameters that control the model structure, learning strategy, and regularization strength before model training. For instance, when the prediction model is an XGBoost with built-in L1 / L2 regularization, hyperparameters may include the maximum tree depth, learning rate, number of iterations, and L1 / L2 regularization coefficients. The goal of hyperparameter optimization is to find the optimal combination of hyperparameters that makes the model perform best on the validation set. In some embodiments, grid search or Bayesian optimization can be used to optimize the hyperparameters of the prediction model.

[0075] For example, grid search is a hyperparameter optimization method based on exhaustive traversal. Grid search predefines a set of candidate values ​​for each hyperparameter to be optimized, combines all candidate values ​​of the hyperparameters by Cartesian product to form a complete parameter grid, then traverses each parameter combination in the grid, trains the model and evaluates its performance, and finally selects the parameter combination with the best performance as the optimization result.

[0076] Bayesian optimization is a sequential optimization method based on a probabilistic surrogate model, suitable for optimization scenarios where objective function evaluation is costly. Bayesian optimization approximates the true objective function by constructing a probabilistic surrogate model and intelligently recommends the next most likely parameter combination to improve performance in each iteration using a sampling function, thus approaching the global optimum with as few iterations as possible.

[0077] For example, cross-validation is a method for evaluating the generalization ability of a model. The sample dataset is divided into K subsets of similar size. One subset is selected sequentially as the validation set, and the remaining K-1 subsets are used as the training set for model training and validation. This process is repeated K times, and the average of the validation metrics is taken as the evaluation result of the model performance. Cross-validation can make full use of limited data and reduce the variance of the model evaluation results. In some embodiments, 5-fold cross-validation can be used to evaluate the generalization ability of the prediction model.

[0078] In some embodiments, the prediction model evaluated by hyperparameter optimization and cross-validation can be determined as the final prediction model.

[0079] Some embodiments in this specification employ at least one of LightGBM, RF, CatBoost, SVR, NN, or XGBoost with built-in L1 / L2 regularization as the prediction model, combined with hyperparameter optimization and cross-validation evaluation strategies, significantly improving the model's fitting accuracy and anti-overfitting ability to process data. L1 / L2 regularization constrains the distribution of feature weights during training, effectively suppressing the interference of inherent noise in industrial data on small training sets and reducing the risk of model overfitting. Hyperparameter optimization systematically searches for the optimal hyperparameter combination, ensuring the rationality of the model structure. Cross-validation provides an unbiased evaluation of the model's generalization ability, avoiding performance misjudgments caused by the randomness of data partitioning. These mechanisms collectively endow the prediction model with robustness and reliability in high-dimensional, nonlinear, and noisy process data scenarios, laying an accurate surrogate model foundation for subsequent interpretable analysis and reverse design.

[0080] Step S130: Perform global attribution analysis on the prediction model based on SHapley Additive exPlanations (SHAP) to determine the SHAP values ​​of each of the multiple process parameters for the target macroscopic performance index.

[0081] For example, SHAP is a unified feature attribution framework based on Shapley values ​​in cooperative game theory. Its core idea is to treat the model's predicted output as a "payoff" resulting from the combined effect of all input features, and to fairly distribute this payoff according to the marginal contribution of each feature to the prediction result.

[0082] For example, global attribution analysis refers to assessing the average influence of each process parameter on the target macroscopic performance index in the entire model decision space by statistically aggregating the local feature attribution values ​​(i.e., the SHAP values ​​of each process parameter in each sample) of each sample across the entire sample dataset (such as calculating the absolute mean, ranking, or distribution density).

[0083] For example, the SHAP value refers to the marginal contribution of a given feature to the model output value in a specific prediction. The magnitude of the SHAP value reflects the size of the marginal contribution of the process parameter to the target macroscopic performance index, and the positive or negative sign indicates the direction of the positive or negative influence of the process parameter on the target macroscopic performance index.

[0084] In some embodiments, based on the trained prediction model, SHAP can be applied to each sample in the sample dataset. By traversing the marginal contribution of each process parameter in all possible combinations of feature subsets, the SHAP value of each process parameter in the prediction result of that sample can be calculated.

[0085] Step S140: Based on the SHAP values ​​of multiple process parameters for the target macroscopic performance index, determine the feature importance ranking, positive and negative influence direction, and critical threshold of the multiple process parameters.

[0086] For example, feature importance ranking refers to the relative ranking of the influence of each process parameter on the target macroscopic performance index. Feature importance ranking is used to identify the dominant and secondary factors affecting the target performance.

[0087] In some embodiments, feature importance is typically quantified by calculating the arithmetic mean of the absolute values ​​of the SAP values ​​of each process parameter across all samples. For example, the SAP values ​​of each process parameter output in step S130 are collected from all samples in the full sample set, and the mean absolute value of the SAP for each process parameter is calculated. The process parameters are then sorted in descending order of their mean absolute values ​​of SAP to form a feature importance ranking. The higher the ranking of a process parameter, the stronger its overall control ability over the target macroscopic performance index.

[0088] For example, the positive and negative influence directions refer to the promoting or inhibiting effects of changes in the process parameter values ​​on the target macroscopic performance index. A positive influence direction indicates that increasing the value of the process parameter will lead to an improvement in the target macroscopic performance index, while a negative influence direction indicates that increasing the value of the process parameter will lead to a decrease in the target macroscopic performance index.

[0089] In some embodiments, the direction of positive or negative influence can be determined by analyzing the sign of the SHAP value and its correlation trend with the process parameter values. For example, for each process parameter, the sign distribution characteristics of its SHAP value and its correlation with the original value of the process parameter are analyzed. Specifically, the correlation coefficient between the original value of the process parameter and its SHAP value is calculated. If the correlation coefficient is positive, the process parameter is determined to have a positive influence on the target macroscopic performance index; if the correlation coefficient is negative, it is determined to have a negative influence. Simultaneously, cross-validation is performed by combining the distribution trends of process parameter values ​​and SHAP values ​​in the SHAP summary chart.

[0090] For example, a critical threshold refers to the boundary value at which the effect of a process parameter on a target macroscopic performance index changes significantly. For instance, a critical threshold may include a turning point in the direction of effect and a saturation point in the effect. The turning point in the direction of effect refers to the change in the direction of the positive or negative influence of the process parameter on the target performance before and after the critical threshold; the saturation point in the effect refers to the point where, after exceeding the critical threshold, the SHAP value of the process parameter on the target performance tends to level off or become insignificant. Identifying the critical threshold provides a quantitative basis for defining the optimal range for process parameters.

[0091] In some embodiments, a SHAP dependency graph can be constructed for each process parameter. This involves plotting a scatter plot with the original value of the process parameter on the x-axis and the corresponding SHAP value on the y-axis, and fitting its trend curve. Two types of critical thresholds are identified on the SHAP dependency graph: inflection points in the direction of action and saturation points in the effect. Inflection points in the direction of action can be determined by locating the boundary where the SHAP value transitions from a positive to a negative interval or vice versa. Saturation points in the effect can be determined by identifying the starting point of a plateau region where the growth rate of the SHAP value significantly decreases with increasing process parameters.

[0092] Step S150: Select key process parameters from multiple process parameters based on feature importance ranking, positive and negative influence direction, and critical threshold.

[0093] For example, key process parameters refer to a subset of process parameters that have a significant impact on the target macroscopic performance indicators.

[0094] In some embodiments, process parameters that rank M in importance or whose mean absolute value of SHAP is greater than an importance threshold can be included in the candidate critical process parameter set based on feature importance ranking. M can be determined based on historical experience, and the values ​​of G and M are independent of each other. The candidate critical process parameter set refers to the intermediate solution set in the process of determining critical process parameters.

[0095] In some embodiments, for each process parameter in the candidate critical process parameter set, its positive and negative influence directions are obtained. If there is a clear and stable positive and negative influence direction between the process parameter and the target macroscopic performance index, the parameter is retained in the candidate critical process parameter set; if the influence direction of the process parameter is ambiguous, unstable, or lacks statistical significance, it is removed from the candidate critical process parameter set.

[0096] In some embodiments, for candidate process parameters retained after filtering by positive and negative influence directions, their critical threshold identification results are obtained. Process parameters with clear critical thresholds are selected, marked as primary critical process parameters, and included in the final set of critical process parameters; for process parameters with high feature importance but for which clear critical thresholds have not yet been identified, they are marked as secondary critical process parameters, and whether to include them in the final set of critical process parameters is determined based on actual needs.

[0097] Some embodiments in this specification introduce SHAP (Shape Attribution Analysis) for global attribution analysis of the prediction model, transforming the decision-making logic of the black-box model into quantified SHAP values, significantly improving the transparency and reliability of process parameter selection. Furthermore, determining feature importance ranking, positive and negative influence directions, and critical thresholds based on SHAP values ​​not only identifies the dominant process parameters affecting target performance but also reveals the direction of action and nonlinear boundary of each process parameter, overcoming the shortcomings of traditional methods that only output importance ranking without providing information on the direction of action and thresholds. Based on this, key process parameters are accurately identified from numerous process parameters according to selection rules, effectively reducing the variable dimensionality of subsequent micro-validation and process optimization, and improving the overall execution efficiency and engineering practicality of the method.

[0098] After completing the screening of key process parameters, in order to further explore the possible synergistic enhancement or mutual inhibition relationships among the process parameters, this disclosure also provides an optional embodiment, which can identify the interaction effects among multiple process parameters based on the aforementioned analysis results, thereby more comprehensively revealing the coupling mechanism of process parameters on target performance.

[0099] In some embodiments, the interaction effects between multiple process parameters can also be identified based on feature importance ranking, positive and negative influence direction, and critical threshold.

[0100] For example, for any two or more process parameters, a SHAP dependency graph of each parameter acting alone on the target macroscopic performance index is constructed, as well as a two-dimensional interactive SHAP dependency graph between the parameters. In the two-dimensional interactive SHAP dependency graph, one process parameter is used as the x-axis, and the other process parameter is used as the coloring dimension or facet dimension to observe the distribution law of the SHAP value of the target performance index as a function of the two parameters. If, within different ranges of the value of one parameter, the SHAP value of the other parameter shows significantly different trends, i.e., there is a nonlinear coupling relationship of synergistic enhancement or mutual inhibition between the two parameters, then it is determined that there is an interaction effect between the set of process parameters.

[0101] For example, for different bamboo species, the rate of change of the SHAP value of hot-pressed density in terms of bending strength varies with the increase of density. Some bamboo species can obtain a higher SHAP value at a lower density, while other bamboo species require a higher density to achieve a similar contribution level. This indicates that bamboo species and hot-pressed density jointly affect the pore structure of the final shaped board, and thus synergistically affect water resistance and mechanical properties.

[0102] For example, the two-dimensional interaction SHAP dependency graph of density and resin content shows obvious ridges or valleys. That is, in the low-density region, increasing the resin content has a significant positive contribution to the SHAP of mechanical properties; while in the high-density region, the contribution of further increasing the resin content to mechanical properties tends to saturate or even turn negative, indicating that density and resin content jointly affect the quality of the bonding interface formed by the adhesive inside the board.

[0103] For example, in the SHAP dependency graph, the SHAP value is negative in the low temperature short time region and positive in the high temperature long time region. There is also a contour ridge extending from the low temperature short time region to the high temperature long time region, which indicates that the hot pressing temperature and hot pressing time jointly affect the degree of crosslinking of thermosetting phenolic resin adhesives, thereby synergistically affecting mechanical properties and water resistance.

[0104] The interaction effects identified in the above manner can be further used as input for the verification of microscopic physical mechanisms in step S160, and physical characterization experiments targeting the interaction mechanisms can be designed accordingly.

[0105] Some embodiments in this specification identify the interaction effects between multiple process parameters based on feature importance ranking, positive and negative influence direction, and critical thresholds. This overcomes the limitations of traditional single-parameter analysis methods that cannot reveal the coupling relationships between parameters. Furthermore, the identified interaction effects can be used to optimize prediction models. For example, by constructing interaction feature terms, the prediction accuracy and interpretability of the model can be further improved, overcoming the shortcomings of existing technologies that treat each process parameter as an independent variable and ignore interaction effects.

[0106] Step S160: Obtain microscopic physical structure characterization data of biomass composite material samples related to key process parameters, and verify the influence mechanism of key process parameters on target macroscopic performance indicators based on the microscopic physical structure characterization data.

[0107] For example, microscopic physical structure characterization data refers to data reflecting the internal microstructure and structural characteristics of biomass composite materials. For instance, microscopic physical structure characterization data may include pore structure, interfacial bonding state, resin distribution uniformity, resin crosslinking degree, fiber orientation, and surface chemical properties, etc., without limitation.

[0108] For example, the influence mechanism characterization describes the physical path by which key process parameters affect the target macroscopic performance indicators through changes in the material's microscopic physical structure. For instance, if the decision contribution analysis shows that the resin impregnation amount has a positive contribution to the MOR and that there is a saturation threshold, then the microscopic physical structure characterization data can be used to verify that the saturation threshold corresponds to the microscopic state with the best resin distribution uniformity and the most complete pore filling.

[0109] In some embodiments, multi-scale physical characterization experiments can be designed to verify the influence mechanism of key process parameters on target macroscopic performance indicators. For example, scanning electron microscopy, nanoindentation testing, and single fiber removal experiments can be used to investigate the influence mechanism of bamboo species on MOR (Modulus of Rupture). As another example, micro-computed tomography (Micro-CT), mercury intrusion porosimetry (MIP), and dynamic mechanical analysis (DMA) can be used to investigate the influence mechanism of density on the Modulus of Rupture (MOR) and Thickness Swelling Rate (TSR). For example, laser scanning confocal microscopy (LSCM), scanning electromicroscopy-energy dispersive spectroscopy (SEM-EDS), Fourier transform infrared spectroscopy (FTIR), X-ray photoelectron spectroscopy (XPS), dynamic vapor sorption (DVS), and water contact angle experiments were used to investigate the influence mechanism of resin content on MOR and TSR. Furthermore, dielectric analysis (DEA), thermogravimetry (TG), and molecular dynamics simulation experiments were used to investigate the influence mechanism of hot-pressing conditions (hot-pressing temperature and hot-pressing time) on TSR.

[0110] Some embodiments in this specification involve acquiring a sample dataset containing process parameters and target macroscopic performance indicators, determining a prediction model trained based on the sample dataset, and identifying key process parameters by ranking the feature importance, positive and negative influence directions, and critical thresholds of each process parameter in the prediction model. This process then acquires microscopic physical structure characterization data related to the key process parameters, verifying the microscopic physical paths through which the key process parameters influence the target macroscopic performance indicators. This method combines data-driven decision contribution analysis with microscopic physical structure characterization, providing traceable physical path evidence for the selection of key process parameters. It overcomes the black-box nature of pure machine learning models in interpreting process performance relationships, thereby effectively improving the physical credibility of process parameter selection and the reliability of subsequent process optimization predictions.

[0111] To further verify the physical authenticity of the impact of each key process parameter on the target macroscopic performance index, this disclosure provides an optional embodiment that can verify the impact mechanism by comparing the microscopic physical structures of samples on both sides of the critical threshold based on the SHAP interpretability analysis results.

[0112] In some embodiments, biomass composite material samples corresponding to the values ​​of key process parameters within a preset range on both sides of a critical threshold can be selected based on the critical threshold of the key process parameters. The selected samples are then subjected to microscopic physical structure characterization to obtain test microscopic physical structure characterization data. The trend of the test microscopic physical structure characterization data changing with the value of the key process parameters is compared to the trend predicted by the positive and negative influence directions of SHAP. If they are consistent, the influence mechanism of the key process parameters on the target macroscopic performance indicators is verified. The preset range can be determined based on historical experience.

[0113] In some embodiments of this specification, samples corresponding to the critical thresholds of key process parameters are selected for microscopic physical structure characterization. The trend of microscopic physical structure characterization data changing with parameters is compared and verified with the positive and negative influence directions of SHAP (Shape-Based Analysis). This allows for the confirmation of the physical authenticity of the machine learning interpretability analysis results using intuitive microscopic experimental data. This verification method combines data-driven feature attribution analysis with material microphysical experiments, providing experimental evidence to support the process-performance mapping relationship revealed by the machine learning model, effectively overcoming the limitations of pure black-box models lacking physical verification. By comparing and verifying samples on both sides of the critical threshold, the microscopic structural critical points where the effect of process parameters on performance changes significantly can be accurately located. This provides a clear structural basis for the precise control of subsequent process parameters, further enhancing the credibility and engineering reliability of the predictive model in process optimization decisions.

[0114] After verifying the microscopic physical mechanism, if the verification results show that the influence mechanism of the key process parameters on the target macroscopic performance index is physically reasonable, then the prediction model has a reliable process-performance mapping foundation and can be further used for reverse recommendation of process parameters. Therefore, this disclosure provides an optional embodiment that can build a reverse design framework based on the verified prediction model, and achieve intelligent reverse derivation from desired performance to specific process parameters through multi-objective optimization search. The specific implementation is as follows.

[0115] In some embodiments, after verifying the influence mechanism of key process parameters on the target macroscopic performance index, the process parameter verification method further includes: if the influence mechanism is verified, using the prediction model as the fitness function surrogate model and the target macroscopic performance index as the optimization objective, performing a multi-objective optimization search within a preset process parameter constraint space, and outputting a recommended set of process parameters.

[0116] For example, a fitness function surrogate model refers to an alternative model used to evaluate the merits of a set of recommended process parameters during the optimization search process. In some embodiments, since directly evaluating the performance indicators corresponding to each set of process parameters through physical experiments is costly, a trained and validated prediction model is used as the fitness function. By inputting the recommended process parameters, the corresponding predicted values ​​of the target macroscopic performance indicators can be quickly output, which serve as the fitness score for that set of recommended process parameters.

[0117] For example, the optimization objective refers to the target macroscopic performance index that needs to be maximized or minimized in a multi-objective optimization process. For instance, in bamboo-based fiber composites or wood-based reconstituted materials, maximizing MOR and minimizing TSR are usually used as dual optimization objectives.

[0118] For example, the preset process parameter constraint space refers to the allowed value boundaries of each process parameter within the physically achievable range. In some embodiments, the preset process parameter constraint space can be pre-set based on factors such as the physical limits of the actual production equipment, process safety boundaries, and cost-effectiveness. For example, the lower limit of the hot-pressing temperature must not be lower than the minimum temperature required for resin curing, and the upper limit must not exceed the thermal degradation temperature of bamboo.

[0119] For example, multi-objective optimization search refers to an optimization algorithm that seeks the best set of trade-off solutions among multiple conflicting optimization objectives.

[0120] For example, a recommended process parameter set refers to at least one combination of process parameters output after multi-objective optimization search. The recommended process parameter set contains specific values ​​for each key process parameter and can be directly used to guide the setting of process parameters in actual production.

[0121] In some embodiments, before performing a multi-objective optimization search, the verification result of the influence mechanism in step S160 is first determined. If the verification result shows that the influence mechanism of the key process parameters on the target macroscopic performance index has physical rationality, that is, it proves that the process-performance mapping relationship captured by the prediction model conforms to the real physical laws, then the multi-objective optimization search is triggered; if the verification fails, the multi-objective optimization search is not performed, and the model is readjusted or experimental data is supplemented.

[0122] In some embodiments of this specification, provided that the influence mechanism of key process parameters on the target macroscopic performance index has been verified, a predictive model is used as a fitness function surrogate model to replace physical experimental evaluation. A multi-objective optimization search is then performed within a preset process parameter constraint space, with the target macroscopic performance index as the optimization objective, ultimately outputting a recommended set of process parameters. This establishes a reverse mapping channel from the target macroscopic performance index to specific process parameters, allowing users to obtain a recommended set of process parameters that meets performance objectives without relying on experience-based trial and error or extensive experiments. This significantly shortens the process development cycle and reduces experimental costs. Furthermore, since the reverse optimization search is triggered only after the influence mechanism has been verified, the physical rationality and engineering reliability of the optimization results are ensured, avoiding the unreliability risks associated with model predictions.

[0123] In the process of using a predictive model as a surrogate model for fitness function in multi-objective optimization search, the choice of optimization algorithm and the expression form of the recommended process parameter set directly determine the computational efficiency and solution quality of reverse engineering. To balance search efficiency and optimality of the solution set, this disclosure provides an optional embodiment that can employ an efficient multi-objective optimization algorithm and output a recommended process parameter set that facilitates engineering decision-making. Its specific implementation is as follows.

[0124] In some embodiments, the multi-objective optimization search employs the Multi-Objective Bayesian Optimization (MOBO) algorithm or the second-generation Nondominated Sorting Genetic Algorithm II (NSGA-II), and the output recommended process parameter set is the Pareto optimal process solution set.

[0125] For example, the MOBO algorithm is a global optimization algorithm based on a probabilistic surrogate model, suitable for optimization scenarios where the objective function evaluation cost is high. The MOBO algorithm approximates the real objective function by constructing a Gaussian process equiprobable surrogate model, and uses a sampling function to balance exploration and exploitation, efficiently approximating the global optimum with fewer iterations.

[0126] For example, the NSGA-II algorithm is a classic multi-objective evolutionary algorithm. The NSGA-II algorithm stratifies individuals in the population through non-dominated sorting, maintains the diversity of the solution set by crowding distance, and retains superior individuals from the parent generation by employing an elitist strategy, thus enabling fast convergence in multi-objective optimization problems.

[0127] For example, the Pareto optimal process solution set refers to the set of solutions in a multi-objective optimization problem that are simultaneously superior to any other solution on all optimization objectives. For any solution in this set, attempting to improve the performance of one objective will inevitably lead to the degradation of the performance of at least one other objective.

[0128] In some embodiments of this specification, the MOBO algorithm or NSGA-II algorithm is used for multi-objective optimization search, and the Pareto optimal process solution set is output as a recommended process parameter set, which effectively improves the computational efficiency and solution set quality of reverse engineering. Each set of process parameters in the Pareto optimal process solution set is a non-dominated solution, corresponding to the best trade-off relationship between different optimization objectives, providing users with diverse engineering options and avoiding the limitation that a single solution cannot meet the needs of multiple objectives.

[0129] In some embodiments, validation samples can be prepared and tested according to the recommended process parameter set, and it can be determined whether the error between the measured values ​​and the model predictions of the validation samples is within an acceptable range. For example, the acceptable range can be set to limit the mean absolute error (MAE) to within 17 MPa for MOR and to limit the mean absolute error (MAE) to within 1.3% for TSR.

[0130] For example, if the error is within an acceptable range, the recommended set of process parameters is deemed valid and the accuracy of the prediction model meets the requirements of engineering applications; if the error exceeds the acceptable range, the verification is deemed unsuccessful, and the process is returned to adjust the prediction model or re-execute the optimization search.

[0131] In some embodiments of this specification, by preparing and testing verification samples according to the recommended process parameter set, and determining whether the error between the measured value and the model prediction value of the verification samples is within an acceptable range, a complete technical path from data acquisition, modeling and prediction, interpretable analysis, microscopic mechanism verification, reverse optimization to experimental closed loop can be formed.

[0132] Figure 2 The diagram shown is a structural schematic of a process parameter verification device provided in an embodiment of this disclosure. Figure 2 As shown, the process parameter verification device 200 provided in this embodiment includes: The acquisition module 210 is used to acquire a sample dataset, which includes multiple process parameters and target macroscopic performance indicators of biomass composite material samples. The first determining module 220 is used to determine the prediction model trained based on the sample dataset. The prediction model is used to predict the target macroscopic performance index based on the input process parameters. The second determination module 230 is used to perform global attribution analysis on the prediction model based on Shapley additive interpretation, and determine the SHAP values ​​of multiple process parameters for the target macroscopic performance index. The third determining module 240 is used to determine the feature importance ranking, positive and negative influence direction and critical threshold of multiple process parameters based on the SHAP values ​​of each of the multiple process parameters on the target macroscopic performance index. The filtering module 250 is used to filter out key process parameters from multiple process parameters based on feature importance, positive and negative influence direction, and critical threshold. The verification module 260 is used to acquire microscopic physical structure characterization data of biomass composite material samples related to key process parameters, and to verify the influence mechanism of key process parameters on target macroscopic performance indicators based on the microscopic physical structure characterization data. The influence mechanism characterizes the physical path through which key process parameters affect target macroscopic performance indicators via changes in the microscopic physical structure of the material.

[0133] For more information on the functions of the process parameter verification device 200, please refer to the process parameter verification method section above, which will not be repeated here.

[0134] Below, for reference Figure 3 This describes an electronic device provided according to embodiments of the present disclosure. Figure 3 The diagram shown is a structural schematic of an electronic device provided in an embodiment of this disclosure.

[0135] like Figure 3 As shown, the electronic device 300 includes one or more processors 310 and memory 320.

[0136] The processor 310 may be a central processing unit (CPU) or other form of processing unit with data processing capabilities and / or instruction execution capabilities, and may control other components in the electronic device 300 to perform desired functions.

[0137] The memory 320 may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may, for example, include random access memory (RAM) and / or cache memory. The non-volatile memory may, for example, include read-only memory (ROM), hard disk, flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage medium, and the processor 310 may execute the program instructions to implement the methods of the various embodiments of this disclosure described above and / or other desired functions.

[0138] In one example, the electronic device 300 may also include an input device 330 and an output device 340, which are interconnected via a bus system and / or other forms of connection mechanism (not shown).

[0139] The input device 330 may include, for example, a keyboard, a mouse, etc.

[0140] The output device 340 can output various information to the outside. The output device 340 may include, for example, a display, a speaker, a printer, and a communication network and its connected remote output devices, etc.

[0141] Of course, for the sake of simplicity, Figure 3 Only some of the components of the electronic device 300 relevant to this disclosure are shown, omitting components such as buses and input / output interfaces. In addition, the electronic device 300 may include any other suitable components depending on the specific application.

[0142] In addition to the methods and apparatus described above, embodiments of this disclosure may also be computer program products comprising computer program instructions that, when executed by a processor, cause the processor to perform the steps of the methods described above according to various embodiments of this disclosure.

[0143] The computer program product can be written in any combination of one or more programming languages ​​to perform the operations of the embodiments of this disclosure. The programming languages ​​include object-oriented programming languages ​​such as Java and C++, as well as conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on a user's computing device, partially on a user's computing device, as a standalone software package, partially on a user's computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0144] Furthermore, embodiments of this disclosure may also be computer-readable storage media storing computer-executable instructions that, when executed by a computer, cause the computer to perform the steps in the methods described above according to various embodiments of this disclosure.

[0145] The computer-readable storage medium may be any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of readable storage media (a non-exhaustive list) include: electrical connections having one or more wires, portable disks, hard disks, RAM, ROM, erasable programmable read-only memory (EPROM or flash memory), optical fibers, compact disc read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0146] The basic principles of this disclosure have been described above with reference to specific embodiments. However, it should be noted that the advantages, benefits, and effects mentioned in this disclosure are merely examples and not limitations, and should not be considered as essential features of each embodiment of this disclosure. Furthermore, the specific details disclosed above are for illustrative and facilitative purposes only, and are not limitations. These details do not limit the scope of this disclosure to the necessity of employing the aforementioned specific details for implementation.

[0147] The block diagrams of devices, apparatuses, devices, and systems disclosed herein are merely illustrative examples and are not intended to require or imply that they must be connected, arranged, or configured in the manner shown in the block diagrams. As those skilled in the art will recognize, these devices, apparatuses, devices, and systems can be connected, arranged, and configured in any manner. Words such as “comprising,” “including,” “having,” etc., are open-ended terms meaning “including but not limited to,” and are used interchangeably with them. The terms “or” and “and” as used herein refer to the terms “and / or,” and are used interchangeably with them unless the context clearly indicates otherwise. The term “such as” as used herein refers to the phrase “such as but not limited to,” and is used interchangeably with it.

[0148] It should also be noted that in the apparatus, devices, and methods of this disclosure, the components or steps can be disassembled and / or recombined. These disassemblies and / or recombinations should be considered as equivalent solutions to this disclosure.

[0149] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use this disclosure. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein may be applied to other aspects without departing from the scope of this disclosure. Therefore, this disclosure is not intended to be limited to the aspects shown herein, but rather to be carried out within the widest scope consistent with the principles and novel features disclosed herein.

[0150] The above description has been given for purposes of illustration and description. Furthermore, this description is not intended to limit the embodiments of this disclosure to the forms disclosed herein. Although numerous exemplary aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, alterations, additions, and sub-combinations thereof.

Claims

1. A method for verifying process parameters, characterized in that, include: Obtain a sample dataset, which includes multiple process parameters and target macroscopic performance indicators of biomass composite material samples; A prediction model trained based on the sample dataset is determined, and the prediction model is used to predict the target macroscopic performance index based on the input process parameters; A global attribution analysis is performed on the prediction model based on the Shapley additive interpretation to determine the SHAP value of each of the multiple process parameters for the target macroscopic performance index. Based on the SHAP values ​​of each of the multiple process parameters for the target macroscopic performance index, the feature importance ranking, positive and negative influence direction, and critical threshold of the multiple process parameters are determined. Based on the importance ranking of the features, the direction of positive and negative influence, and the critical threshold, the key process parameters are selected from the plurality of process parameters; Microscopic physical structure characterization data of biomass composite material samples related to the key process parameters are obtained, and the influence mechanism of the key process parameters on the target macroscopic performance index is verified based on the microscopic physical structure characterization data. The influence mechanism characterizes the physical path by which the key process parameters affect the target macroscopic performance index through changes in the microscopic physical structure of the material.

2. The method according to claim 1, characterized in that, The step of acquiring microscopic physical structure characterization data of biomass composite material samples related to the key process parameters, and verifying the influence mechanism of the key process parameters on the target macroscopic performance indicators based on the microscopic physical structure characterization data, includes: Based on the critical threshold of the key process parameters, obtain the microscopic physical structure characterization data of biomass composite material samples corresponding to the values ​​of the key process parameters within a preset range on both sides of the critical threshold, and use them as test microscopic physical structure characterization data. Compare whether the trend of the test microscopic physical structure characterization data with the value of the key process parameter is consistent with the trend predicted by the positive and negative influence directions of the key process parameter. If they match, then the mechanism by which the key process parameters affect the target macroscopic performance index is verified.

3. The method according to claim 1, characterized in that, The multiple process parameters include continuous variables and categorical variables; Before determining the prediction model trained based on the sample dataset, the method further includes: The categorical variables are transformed by one-hot encoding, and the continuous variables are normalized to obtain a preprocessed sample dataset; wherein, the prediction model is trained based on the preprocessed sample dataset.

4. The method according to claim 1, characterized in that, The prediction model is at least one of lightweight gradient boosting machine, random forest, classification feature gradient boosting, support vector regression, neural network, or extreme gradient boosting with built-in L1 / L2 regularization; determining the prediction model trained based on the sample dataset includes: The prediction model is obtained by optimizing its hyperparameters and evaluating its generalization ability based on cross-validation.

5. The method according to claim 1, characterized in that, The multiple process parameters include carbonization process parameters; if the biomass composite material is a bamboo-based fiber composite material, then the carbonization process parameters include a carbonization severity factor, which is determined by combining carbonization temperature and carbonization time in a dimension-reducing manner.

6. The method according to claim 1, characterized in that, After verifying the impact mechanism of the key process parameters on the target macroscopic performance index, the following is also included: If the aforementioned influence mechanism is verified, then the prediction model is used as the fitness function surrogate model, and the target macroscopic performance index is used as the optimization objective. Multi-objective optimization search is performed within the preset process parameter constraint space to output a recommended set of process parameters.

7. The method according to claim 6, characterized in that, The multi-objective optimization search employs a multi-objective Bayesian optimization algorithm or a second-generation non-dominated sorting genetic algorithm, and the output set of recommended process parameters is a Pareto optimal process solution set.

8. The method according to any one of claims 1 to 7, characterized in that, The target macroscopic performance indicators include mechanical performance indicators and / or water resistance performance indicators; wherein... The mechanical performance indicators include at least one of bending strength and elastic modulus; The water resistance performance indicators include at least one of water absorption thickness swelling rate and water absorption rate.

9. An electronic device, characterized in that, It includes a processor and a memory, the memory storing computer-executable instructions, the processor executing the computer-executable instructions to implement the method as described in any one of claims 1 to 8.

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

Images