Machine learning based method and system for preparing phosphogypsum based self-healing coating

Abstract

This invention discloses a method and system for preparing phosphogypsum-based self-healing coatings based on machine learning, comprising the following steps: collecting characteristic parameters of phosphogypsum raw materials, functional group ratio parameters, preparation process parameters, and corresponding self-healing performance parameters; constructing a sample dataset; and performing data cleaning and standardization to form a training dataset. The training dataset is then input into a kernel extreme learning machine (KEM) model for training to obtain an initial prediction model. A sparrow search algorithm is then used to optimize the kernel function parameters and penalty factor in the model to obtain an optimized KEM model. The target self-healing performance parameters are further input into the optimized model, outputting the corresponding functional group ratio parameters and preparation process parameters. Finally, based on the parameters, raw material metering, mixing and dispersion, pulping, film formation, and curing treatments are performed to obtain the phosphogypsum-based self-healing coating. This invention achieves targeted design and preparation of coating performance.

Description

Technical Field

[0001] This invention relates to the field of building materials and functional coatings, and in particular to a method and system for preparing phosphogypsum-based self-healing coatings based on machine learning. Background Technology

[0002] As a major industrial solid waste generated during the phosphate chemical industry, the resource utilization of phosphogypsum has become an important research direction in the field of building materials. In recent years, the application of phosphogypsum in coating systems has gradually increased. By introducing film-forming components, functional additives, and self-healing components, functional coatings with certain protective properties and self-healing capabilities can be prepared. However, the formulation and preparation of existing phosphogypsum-based coatings typically rely on accumulated experience or multiple rounds of trial and error to determine the proportions of each component and process parameters, resulting in long development cycles, high costs, and poor reproducibility of results.

[0003] In existing technologies, due to the significant fluctuations in the characteristics of phosphogypsum raw materials—with variations in purity, particle size, and impurity content between different batches—and the complex nonlinear coupling between the functional component ratios and preparation process parameters, coating performance, especially self-healing properties, is difficult to accurately predict and control using simple empirical models. Existing methods typically rely on large amounts of experimental data to establish empirical relationships; however, with limited sample sizes, model accuracy is difficult to guarantee, leading to significant uncertainty in the performance optimization process.

[0004] Traditional models often rely heavily on manual experience to select parameters during parameter setting, lacking an effective global optimization mechanism. This makes them prone to getting trapped in local optima, resulting in insufficient model generalization ability and difficulty in achieving stable transfer of formulations and processes under different raw material conditions. Existing technologies lack methods for unified modeling and collaborative optimization of raw material characteristics, formulation parameters, and process conditions, making it impossible to achieve targeted design guided by self-healing performance.

[0005] Therefore, how to provide a method and system for preparing phosphogypsum-based self-healing coatings based on machine learning is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose a method and system for preparing phosphogypsum-based self-healing coatings based on machine learning. This invention introduces a kernel limit learning machine model and a sparrow search algorithm to model and optimize the relationship between the characteristic parameters of phosphogypsum raw materials, the proportion parameters of functional components, the preparation process parameters, and the self-healing performance parameters. This enables the targeted design of coating formulations and preparation processes, and has the advantages of high modeling accuracy, strong parameter optimization capability, and high controllability of formulation design.

[0007] The method and system for preparing phosphogypsum-based self-healing coatings based on machine learning according to embodiments of the present invention include the following steps: Collect raw material characteristic parameters, functional group allocation ratio parameters, preparation process parameters, and self-healing performance parameters corresponding to the raw material characteristic parameters, functional group allocation ratio parameters, and preparation process parameters of phosphogypsum to construct a sample dataset; The sample dataset is cleaned and standardized to generate a training dataset. The training dataset is input into an improved kernel limit learning model with pre-set initial kernel function parameters and initial penalty factors. The nonlinear mapping relationship between the raw material characteristic parameters, functional group ratio parameters, preparation process parameters and self-healing performance parameters of phosphogypsum is trained to generate an initial prediction model. The initial prediction model is input into the sparrow search algorithm, and the parameters are optimized by taking the initial kernel function parameters and the initial penalty factor as the optimization objects to obtain the optimized kernel function parameters and the optimized penalty factor. The initial prediction model is then updated based on the optimized kernel function parameters and the optimized penalty factor to obtain the optimized kernel extreme learning machine model. Input the target self-healing performance parameters into the optimized kernel limit learning machine model, and output the target functional group allocation ratio parameters and target fabrication process parameters corresponding to the target self-healing performance parameters; Based on the target functional group ratio parameters and target preparation process parameters, raw material metering, mixing and dispersion, pulping, film formation and curing treatment are carried out to generate phosphogypsum-based self-healing coating.

[0008] Optionally, the phosphogypsum raw material characteristic parameters, functional component ratio parameters, preparation process parameters, and self-healing performance parameters in the sample dataset specifically include: The characteristic parameters of phosphogypsum raw materials collected include phosphogypsum purity, calcium sulfate dihydrate content, particle size distribution, specific surface area, moisture content, soluble phosphorus content, and soluble fluorine content. The parameters for collecting functional group proportions include the mass fraction of phosphogypsum, the mass fraction of film-forming components, the mass fraction of self-healing components, the mass fraction of dispersing components, and the mass fraction of auxiliary components, wherein the self-healing components include microcapsule-type repair agents or inorganic self-healing components; The preparation process parameters collected include stirring speed, stirring time, dispersion sequence, reaction temperature, reaction time, curing time, and curing temperature; The self-healing performance parameters include crack closure rate, impermeability, coating adhesion, water resistance, and durability. The raw material characteristics, functional component ratio parameters, and preparation process parameters of phosphogypsum are used as input features, and the self-healing performance parameters are used as output features. Correspondence is established in the sample dataset.

[0009] Optionally, the process of cleaning and standardizing the sample dataset includes: The integrity of the raw material characteristic parameters, functional component ratio parameters, preparation process parameters and self-healing performance parameters of phosphogypsum in the sample dataset was checked to identify missing data items. The identified missing data items are processed by removing samples with missing key parameters and filling in samples with missing non-key parameters using statistical values ​​of the corresponding parameters from samples of the same type, thus obtaining complete sample data. Consistency verification was performed on the raw material characteristic parameters, functional component ratio parameters, preparation process parameters, and self-healing performance parameters of phosphogypsum in the complete sample data. Samples with mismatched parameters, inconsistent recorded results, or duplicate entries were removed to obtain consistent sample data. Outlier identification was performed on the characteristic parameters of phosphogypsum raw materials, functional component ratio parameters, preparation process parameters, and self-healing performance parameters in the consistent sample data. Outlier samples that exceeded the preset value range were removed to obtain cleaned sample data. The dimensionality of the raw material characteristic parameters, functional component ratio parameters, preparation process parameters and self-healing performance parameters of phosphogypsum in the cleaned sample data was uniformized, and then standardized based on the dimensionality uniformization to obtain standardized sample data. A training dataset is generated based on the correspondence between the characteristic parameters of phosphogypsum raw materials, the proportion parameters of functional components, the preparation process parameters, and the self-healing performance parameters in the standardized sample data.

[0010] Optionally, the generation of the initial prediction model includes: The raw material characteristic parameters, functional group ratio parameters, and preparation process parameters of phosphogypsum in the training dataset are combined according to the sample correspondence to generate input data corresponding to each sample, and the self-healing performance parameters corresponding to each sample are determined as output data. An improved kernel extreme learning machine model is constructed by pre-setting initial kernel function parameters and initial penalty factors for input data and output data. The improved kernel extreme learning machine model uses input data as model input and output data as model output. The input data is processed by kernel mapping based on the initial kernel function parameters to generate the kernel mapping result corresponding to the input data, and the training error between the kernel mapping result and the output data is constrained based on the initial penalty factor. Based on the kernel mapping results, output data, and training error constraints, the improved kernel extreme learning machine model is trained to obtain the nonlinear mapping relationship between the input data and the output data. Based on the nonlinear mapping relationship between input and output data, an initial prediction model is generated to characterize the correspondence between the characteristic parameters of phosphogypsum raw materials, the proportion parameters of functional components, the preparation process parameters, and the self-healing performance parameters.

[0011] Optionally, the improved kernel extreme learning machine model includes: The raw material characteristic parameters, functional group proportioning parameters, and preparation process parameters of phosphogypsum in the training dataset are divided according to parameter type to obtain raw material feature subsets, proportioning feature subsets, and process feature subsets; Subspace mapping is performed on the raw material feature subset, the proportion feature subset, and the process feature subset respectively to obtain the raw material sub-feature mapping result, the proportion sub-feature mapping result, and the process sub-feature mapping result; The raw material sub-feature mapping results, proportion sub-feature mapping results, and process sub-feature mapping results are fused to generate fused input features. The fused input features and self-healing performance parameters are then input into a kernel limit learning model with pre-set initial kernel function parameters and initial penalty factors. The fused input features are used as the model input, and the self-healing performance parameters are used as the model output. The fused input features are processed by kernel mapping based on the initial kernel function parameters to obtain the kernel mapping result corresponding to the fused input features. The training error between the kernel mapping result and the self-repair performance parameters is constrained based on the initial penalty factor. The kernel extreme learning machine model is trained based on the kernel mapping results, self-repair performance parameters, and training error constraints to obtain the initial nonlinear mapping relationship between the fused input features and the self-repair performance parameters, and the initial prediction results are output based on the initial nonlinear mapping relationship. A residual dataset is generated based on the difference between the initial prediction results and the self-healing performance parameters. The fused input features and the corresponding input kernel extreme learning machine model of the residual dataset are then subjected to residual learning to obtain the residual mapping relationship between the fused input features and the residual dataset. The initial nonlinear mapping relationship is updated based on the residual mapping relationship to obtain the corrected nonlinear mapping relationship, and an improved kernel extreme learning machine model is generated based on the corrected nonlinear mapping relationship.

[0012] Optionally, the generation of the optimized kernel extreme learning machine model includes: The initial kernel function parameters and initial penalty factor in the initial prediction model are extracted as parameters to be optimized, and a fitness evaluation standard is established based on the error between the prediction results of the initial prediction model on the training dataset and the self-repair performance parameters. Based on the value range of the parameter to be optimized, multiple parameter combinations are generated, and each parameter combination is used as the position parameter of an individual sparrow to construct a sparrow population; The position parameters of each sparrow are input into the initial prediction model, and the position parameters of each sparrow replace the initial kernel function parameters and the initial penalty factor in the initial prediction model, respectively, to obtain the model prediction results corresponding to each sparrow. The fitness value corresponding to each sparrow is calculated based on the error between the model prediction results corresponding to each sparrow and the self-repair performance parameters. A global search is performed based on the location parameters of the discoverer to obtain the updated location parameters of the discoverer. The location parameters of the followers are then updated based on the updated location parameters of the discoverer to obtain the updated location parameters of the followers. Finally, the location parameters of the sparrows are adjusted based on the response of the vigilant individuals to the changes in fitness values ​​to obtain the updated location parameters of the sparrows. The updated sparrow individual position parameters are re-input into the initial prediction model, the fitness value corresponding to the updated sparrow individual is recalculated, and the sparrow individual sorting, sparrow individual division and sparrow individual position parameter update are repeated according to the recalculated fitness value until the preset iteration termination condition is reached. The positional parameters corresponding to the sparrow individual with the best fitness value when the preset iteration termination condition is reached are extracted and determined as the optimized kernel function parameters and optimized penalty factor. The optimized kernel function parameters and optimized penalty factor are then used to replace the initial kernel function parameters and initial penalty factor in the initial prediction model to update the initial prediction model, resulting in the optimized kernel extreme learning machine model.

[0013] Optionally, the process of outputting the target functional group allocation ratio parameter and the target preparation process parameter corresponding to the target self-healing performance parameter includes: Obtain the target self-healing performance parameters, and organize the target self-healing performance parameters according to the output structure of the self-healing performance parameters in the optimized kernel extreme learning machine model to generate target performance data; Candidate functional group allocation ratio parameters and candidate preparation process parameters are generated based on the value range of functional group allocation ratio parameters and preparation process parameters in the training dataset. The candidate functional group allocation ratio parameters and candidate preparation process parameters are then combined according to the input structure of the optimized kernel extreme learning machine model to generate candidate input data. The candidate input data is input into the optimized kernel extreme learning machine model to obtain the predictive self-repair performance parameters corresponding to the candidate input data. The predicted self-healing performance parameters are compared with the target performance data, the performance deviation between the predicted self-healing performance parameters and the target self-healing performance parameters is calculated, and the candidate input data is filtered according to the performance deviation to obtain the candidate input data that meets the preset conditions. Extract the corresponding functional group allocation ratio parameters and preparation process parameters from the candidate input data that meet the preset conditions, and generate the target functional group allocation ratio parameters and target preparation process parameters corresponding to the target self-healing performance parameters.

[0014] Optionally, the process of metering, mixing and dispersing, slurry preparation, film formation, and curing of raw materials based on the target functional group proportioning parameters and target preparation process parameters includes: Based on the mass fraction of each component in the target functional group distribution ratio parameter, the phosphogypsum raw material, film-forming component, self-healing component, dispersing component and auxiliary component are quantitatively weighed to obtain raw materials of each component consistent with the target functional group distribution ratio parameter; The weighed dispersion components were added to a solvent for pre-dispersion treatment. Then, according to the dispersion order in the target preparation process parameters, the phosphogypsum raw material, film-forming component, self-healing component and auxiliary component were added to the pre-dispersion treatment in sequence. The mixture was then mixed and dispersed under the stirring speed and stirring time conditions corresponding to the target preparation process parameters to obtain a uniform mixed system. In the homogeneous mixing system, slurry preparation is carried out according to the reaction temperature and reaction time corresponding to the target preparation process parameters, so that each component forms a stable and dispersed phosphogypsum-based slurry in the system; The phosphogypsum-based slurry is applied to the substrate surface according to the preset coating method, and film-forming treatment is carried out according to the film-forming conditions corresponding to the target preparation process parameters to obtain the initial coating. The initial coating is cured according to the curing time and temperature corresponding to the target preparation process parameters to stabilize the coating structure and complete the curing process, thus obtaining a phosphogypsum-based self-healing coating.

[0015] Optionally, the machine learning-based phosphogypsum-based self-healing coating preparation system includes the following modules: The data acquisition module is used to collect the characteristic parameters of phosphogypsum raw materials, the functional group ratio parameters, the preparation process parameters, and the self-healing performance parameters corresponding to the characteristic parameters of phosphogypsum raw materials, the functional group ratio parameters, and the preparation process parameters, and to construct a sample dataset. The data processing module is used to clean and standardize the sample dataset to generate a training dataset. The initial prediction model generation module is used to input the training dataset into an improved kernel limit learning model with pre-set initial kernel function parameters and initial penalty factors, and to train the nonlinear mapping relationship between the phosphogypsum raw material characteristic parameters, functional group allocation parameters, preparation process parameters and self-healing performance parameters to generate the initial prediction model. The model optimization module is used to extract the initial kernel function parameters and initial penalty factor from the initial prediction model as parameters to be optimized, and to perform parameter optimization based on the sparrow search algorithm to obtain the optimized kernel function parameters and optimized penalty factor. The optimized kernel function parameters and optimized penalty factor are then used to update the initial prediction model to generate the optimized kernel extreme learning machine model. The parameter output module is used to input the target self-healing performance parameters into the optimized kernel limit learning machine model and output the target functional group allocation ratio parameters and target preparation process parameters corresponding to the target self-healing performance parameters. The coating preparation module is used to meter raw materials, mix and disperse them, slurry them, form films and cure them according to the target functional group ratio parameters and the target preparation process parameters, to generate phosphogypsum-based self-healing coatings.

[0016] The beneficial effects of this invention are: This invention introduces a kernel limit learning machine model and a sparrow search algorithm to uniformly model and globally optimize the complex nonlinear relationships between phosphogypsum raw material characteristic parameters, functional group ratio parameters, preparation process parameters, and self-healing performance parameters. This achieves high-precision performance prediction and stable modeling under small sample conditions, solving the problems of low accuracy and insufficient sample utilization in existing technologies that rely on experience-based modeling, and significantly improving the model's generalization ability and prediction reliability.

[0017] This invention constructs an improved kernel extreme learning machine model based on grouped feature mapping and residual feedback mechanism, and combines the sparrow search algorithm to adaptively optimize the kernel function parameters and penalty factors, thereby achieving automatic optimization of model parameters and error compensation. This avoids the problems of parameter dependence and easy getting trapped in local optima in traditional methods, and improves model stability and optimization efficiency.

[0018] This invention establishes a multidimensional coupled model of phosphogypsum raw material characteristics, functional component ratios, and preparation processes, and performs reverse engineering guided by target self-healing performance parameters. This enables directional output of coating formulation and process parameters, transforming formulation design from experience-based trial and error to target-driven intelligent design. This significantly shortens the R&D cycle, reduces experimental costs, and improves the reliability and added value of phosphogypsum solid waste resources in high-performance self-healing coatings. Attached Figure Description

[0019] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 The flowchart shows the preparation method and system of phosphogypsum-based self-healing coating based on machine learning proposed in this invention. Figure 2This is a schematic diagram of the improved kernel extreme learning machine model proposed in this invention. Detailed Implementation

[0020] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0021] refer to Figures 1-2 A method and system for preparing phosphogypsum-based self-healing coatings based on machine learning, comprising the following steps: Collect raw material characteristic parameters, functional group allocation ratio parameters, preparation process parameters, and self-healing performance parameters corresponding to the raw material characteristic parameters, functional group allocation ratio parameters, and preparation process parameters of phosphogypsum to construct a sample dataset; The sample dataset is cleaned and standardized to generate a training dataset. The training dataset is input into an improved kernel limit learning model with pre-set initial kernel function parameters and initial penalty factors. The nonlinear mapping relationship between the raw material characteristic parameters, functional group ratio parameters, preparation process parameters and self-healing performance parameters of phosphogypsum is trained to generate an initial prediction model. The initial prediction model is input into the sparrow search algorithm, and the parameters are optimized by taking the initial kernel function parameters and the initial penalty factor as the optimization objects to obtain the optimized kernel function parameters and the optimized penalty factor. The initial prediction model is then updated based on the optimized kernel function parameters and the optimized penalty factor to obtain the optimized kernel extreme learning machine model. Input the target self-healing performance parameters into the optimized kernel limit learning machine model, and output the target functional group allocation ratio parameters and target fabrication process parameters corresponding to the target self-healing performance parameters; Based on the target functional group ratio parameters and target preparation process parameters, raw material metering, mixing and dispersion, pulping, film formation and curing treatment are carried out to generate phosphogypsum-based self-healing coating.

[0022] In this embodiment, the phosphogypsum raw material characteristic parameters, functional component ratio parameters, preparation process parameters, and self-healing performance parameters in the sample dataset specifically include: The characteristic parameters of phosphogypsum raw materials collected include phosphogypsum purity, calcium sulfate dihydrate content, particle size distribution, specific surface area, moisture content, soluble phosphorus content, and soluble fluorine content. The parameters for collecting functional group proportions include the mass fraction of phosphogypsum, the mass fraction of film-forming components, the mass fraction of self-healing components, the mass fraction of dispersing components, and the mass fraction of auxiliary components, wherein the self-healing components include microcapsule-type repair agents or inorganic self-healing components; The preparation process parameters collected include stirring speed, stirring time, dispersion sequence, reaction temperature, reaction time, curing time, and curing temperature; The self-healing performance parameters include crack closure rate, impermeability, coating adhesion, water resistance, and durability. The raw material characteristics, functional component ratio parameters, and preparation process parameters of phosphogypsum are used as input features, and the self-healing performance parameters are used as output features. Correspondence is established in the sample dataset.

[0023] In this embodiment, the process of cleaning and standardizing the sample dataset includes: The integrity of the raw material characteristic parameters, functional component ratio parameters, preparation process parameters and self-healing performance parameters of phosphogypsum in the sample dataset was checked to identify missing data items. By traversing each sample record in the sample dataset, the corresponding values ​​of phosphogypsum raw material characteristic parameters, functional component ratio parameters, preparation process parameters and self-healing performance parameters are checked one by one. Samples with any missing parameter or incomplete records are determined to be incomplete samples. The identified missing data items are processed by removing samples with missing key parameters and filling in samples with missing non-key parameters using statistical values ​​of the corresponding parameters from samples of the same type, thus obtaining complete sample data. Consistency verification was performed on the raw material characteristic parameters, functional component ratio parameters, preparation process parameters, and self-healing performance parameters of phosphogypsum in the complete sample data. Samples with mismatched parameters, inconsistent recorded results, or duplicate entries were removed to obtain consistent sample data. After the missing data processing is completed, the raw material characteristic parameters, functional group ratio parameters, preparation process parameters and corresponding self-healing performance parameters of phosphogypsum are matched and verified according to the sample number to determine that the input parameters and performance parameters in the same sample are consistent with each other. Samples with mismatched parameter combinations, multiple sets of performance results corresponding to the same number or duplicate records are identified and duplicate or conflicting records are deleted, and sample data with unique correspondence are retained. Outlier identification was performed on the characteristic parameters of phosphogypsum raw materials, functional component ratio parameters, preparation process parameters, and self-healing performance parameters in the consistent sample data. Outlier samples that exceeded the preset value range were removed to obtain cleaned sample data. Based on the statistical distribution range of each parameter in the sample dataset, the upper and lower thresholds of the raw material characteristic parameters, functional component ratio parameters, preparation process parameters and self-healing performance parameters of phosphogypsum are calculated respectively. Samples that exceed the range of the upper and lower thresholds are marked as abnormal samples and removed from the sample dataset. The dimensionality of the raw material characteristic parameters, functional component ratio parameters, preparation process parameters and self-healing performance parameters of phosphogypsum in the cleaned sample data was uniformized, and then standardized based on the dimensionality uniformization to obtain standardized sample data. A training dataset is generated based on the correspondence between the characteristic parameters of phosphogypsum raw materials, the proportion parameters of functional components, the preparation process parameters, and the self-healing performance parameters in the standardized sample data.

[0024] In this embodiment, the generation of the initial prediction model includes: The raw material characteristic parameters, functional group ratio parameters, and preparation process parameters of phosphogypsum in the training dataset are combined according to the sample correspondence to generate input data corresponding to each sample, and the self-healing performance parameters corresponding to each sample are determined as output data. An improved kernel extreme learning machine model is constructed by pre-setting initial kernel function parameters and initial penalty factors for input data and output data. The improved kernel extreme learning machine model uses input data as model input and output data as model output. The input data is processed by kernel mapping based on the initial kernel function parameters to generate the kernel mapping result corresponding to the input data, and the training error between the kernel mapping result and the output data is constrained based on the initial penalty factor. Substitute each sample feature vector in the input data into the kernel function determined by the initial kernel function parameters, calculate the kernel function values ​​between samples, and construct a kernel matrix. Use the kernel matrix as the kernel mapping result of the input data. During model training, an initial penalty factor is introduced to constrain the deviation between the model output and the corresponding self-repair performance parameters, so that the training process satisfies the preset constraint relationship between fitting error and model complexity. Based on the kernel mapping results, output data, and training error constraints, the improved kernel extreme learning machine model is trained to obtain the nonlinear mapping relationship between the input data and the output data. Based on the nonlinear mapping relationship between input and output data, an initial prediction model is generated to characterize the correspondence between the characteristic parameters of phosphogypsum raw materials, the proportion parameters of functional components, the preparation process parameters, and the self-healing performance parameters.

[0025] In this embodiment, the improved kernel extreme learning machine model includes: The raw material characteristic parameters, functional group proportioning parameters, and preparation process parameters of phosphogypsum in the training dataset are divided according to parameter type to obtain raw material feature subsets, proportioning feature subsets, and process feature subsets; Subspace mapping is performed on the raw material feature subset, the proportion feature subset, and the process feature subset respectively to obtain the raw material sub-feature mapping result, the proportion sub-feature mapping result, and the process sub-feature mapping result; The raw material feature subset, the proportion feature subset, and the process feature subset are taken as independent inputs. The similarity between samples in each subset is calculated by the corresponding kernel function to generate the kernel mapping result corresponding to each subset, so as to achieve independent mapping of different feature subspaces. The raw material sub-feature mapping results, proportion sub-feature mapping results, and process sub-feature mapping results are fused to generate fused input features. The fused input features and self-healing performance parameters are then input into a kernel limit learning model with pre-set initial kernel function parameters and initial penalty factors. The fused input features are used as the model input, and the self-healing performance parameters are used as the model output. In the kernel extreme learning machine model initialization phase, the kernel function type and its corresponding kernel function parameters are pre-defined as initial kernel function parameters, and a penalty factor is set to constrain the model training error. The initial kernel function parameters and the initial penalty factor are used as the initial configuration parameters of the model. The fused input features are processed by kernel mapping based on the initial kernel function parameters to obtain the kernel mapping result corresponding to the fused input features. The training error between the kernel mapping result and the self-repair performance parameters is constrained based on the initial penalty factor. The kernel function is used to calculate the kernel function values ​​between samples by pairwise inputting the feature vectors of each sample in the fused input features. The kernel matrix is ​​then constructed based on the kernel function values, thereby realizing the mapping of the input features from the original space to the high-dimensional feature space. During model training, the kernel mapping results are fitted with the corresponding self-healing performance parameters, and an initial penalty factor is introduced to constrain the deviation generated during the fitting process, so that the error between the output results and the self-healing performance parameters meets the preset constraint conditions. The kernel extreme learning machine model is trained based on the kernel mapping results, self-repair performance parameters, and training error constraints to obtain the initial nonlinear mapping relationship between the fused input features and the self-repair performance parameters, and the initial prediction results are output based on the initial nonlinear mapping relationship. A kernel matrix is ​​constructed based on the kernel mapping results, and a training equation is formed by combining the self-healing performance parameters. The output weights are solved under the training error constraint to establish a correspondence between the kernel mapping results and the self-healing performance parameters, thus completing the model training. A residual dataset is generated based on the difference between the initial prediction results and the self-healing performance parameters. The fused input features and the corresponding input kernel extreme learning machine model of the residual dataset are then subjected to residual learning to obtain the residual mapping relationship between the fused input features and the residual dataset. The fused input features and corresponding residual data are matched one by one according to the samples and then input into the kernel extreme learning machine model. The residual data is used as the new output target to train the mapping relationship between the fused input features and the residuals, and the residual prediction model is obtained. The initial nonlinear mapping relationship is updated based on the residual mapping relationship to obtain the corrected nonlinear mapping relationship, and an improved kernel extreme learning machine model is generated based on the corrected nonlinear mapping relationship.

[0026] In this embodiment, the generation of the optimized kernel extreme learning machine model includes: The initial kernel function parameters and initial penalty factor in the initial prediction model are extracted as parameters to be optimized, and a fitness evaluation standard is established based on the error between the prediction results of the initial prediction model on the training dataset and the self-repair performance parameters. Based on the value range of the parameter to be optimized, multiple parameter combinations are generated, and each parameter combination is used as the position parameter of an individual sparrow to construct a sparrow population; Based on the numerical range of the initial kernel function parameters and the initial penalty factor in the initial prediction model, and combined with the distribution of the input data in the training dataset, the upper and lower limits of the initial kernel function parameters and the initial penalty factor are set respectively, and the range of values ​​of the parameters to be optimized is limited by the upper and lower limits. The position parameters of each sparrow are input into the initial prediction model, and the position parameters of each sparrow replace the initial kernel function parameters and the initial penalty factor in the initial prediction model, respectively, to obtain the model prediction results corresponding to each sparrow. The fitness value corresponding to each sparrow is calculated based on the error between the model prediction results corresponding to each sparrow and the self-repair performance parameters. The sparrows are sorted according to their fitness values, and then classified into discoverer sparrows, follower sparrows, and vigilant sparrows based on the sorting results. A global search is performed based on the location parameters of the discoverer to obtain the updated location parameters of the discoverer. The location parameters of the followers are then updated based on the updated location parameters of the discoverer to obtain the updated location parameters of the followers. Finally, the location parameters of the sparrows are adjusted based on the response of the vigilant individuals to the changes in fitness values ​​to obtain the updated location parameters of the sparrows. Based on the current location of the discoverer, the location is updated in a random direction within the range of values ​​of the parameters to be optimized, generating new parameter combinations and enabling the updated location parameters to cover a larger search area, thus achieving parameter search in a global range. Based on the updated discoverer's position parameters, the follower's position parameters are adjusted so that the follower moves toward the discoverer's position, and a new parameter combination is generated during this movement by incorporating random perturbations. Based on the changes in the fitness value of individual vigilant individuals, when the fitness value deteriorates, the position parameters of the relevant sparrow individuals are shifted and adjusted to move them away from the current unfavorable area and redistribute them to other parameter areas. The updated sparrow individual position parameters are re-input into the initial prediction model, the fitness value corresponding to the updated sparrow individual is recalculated, and the sparrow individual sorting, sparrow individual division and sparrow individual position parameter update are repeated according to the recalculated fitness value until the preset iteration termination condition is reached. The iteration is stopped when either the number of iterations reaches the preset maximum number of iterations or the fitness value changes less than a set threshold in multiple consecutive iterations. This is the preset iteration termination condition. The positional parameters corresponding to the sparrow individual with the best fitness value when the preset iteration termination condition is reached are extracted and determined as the optimized kernel function parameters and optimized penalty factor. The optimized kernel function parameters and optimized penalty factor are then used to replace the initial kernel function parameters and initial penalty factor in the initial prediction model to update the initial prediction model, resulting in the optimized kernel extreme learning machine model.

[0027] In this embodiment, the process of outputting the target functional group allocation ratio parameter and the target preparation process parameter corresponding to the target self-healing performance parameter includes: Obtain the target self-healing performance parameters, and organize the target self-healing performance parameters according to the output structure of the self-healing performance parameters in the optimized kernel extreme learning machine model to generate target performance data; Candidate functional group allocation ratio parameters and candidate preparation process parameters are generated based on the value range of functional group allocation ratio parameters and preparation process parameters in the training dataset. The candidate functional group allocation ratio parameters and candidate preparation process parameters are then combined according to the input structure of the optimized kernel extreme learning machine model to generate candidate input data. The candidate input data is input into the optimized kernel extreme learning machine model to obtain the predictive self-repair performance parameters corresponding to the candidate input data. The predicted self-healing performance parameters are compared with the target performance data, the performance deviation between the predicted self-healing performance parameters and the target self-healing performance parameters is calculated, and the candidate input data is filtered according to the performance deviation to obtain the candidate input data that meets the preset conditions. The predicted self-healing performance parameters and the target self-healing performance parameters are calculated item by item according to the corresponding indicators, and the differences of each indicator are weighted and summed to obtain the comprehensive performance deviation; the comprehensive performance deviation is compared with the preset deviation threshold, and candidate input data with a comprehensive performance deviation less than or equal to the preset deviation threshold are selected. Extract the corresponding functional group allocation ratio parameters and preparation process parameters from the candidate input data that meet the preset conditions, and generate the target functional group allocation ratio parameters and target preparation process parameters corresponding to the target self-healing performance parameters.

[0028] In this embodiment, the process of metering, mixing and dispersing, slurry preparation, film formation, and curing based on the target functional group proportioning parameters and target preparation process parameters includes: Based on the mass fraction of each component in the target functional group distribution ratio parameter, the phosphogypsum raw material, film-forming component, self-healing component, dispersing component and auxiliary component are quantitatively weighed to obtain raw materials of each component consistent with the target functional group distribution ratio parameter; Based on the proportion of each component in the target functional group allocation parameters, determine the percentage of phosphogypsum raw material, film-forming component, self-healing component, dispersing component and auxiliary component in the total system mass, and calculate the actual weighed mass of each component based on the total mass. The weighed dispersion components were added to a solvent for pre-dispersion treatment. Then, according to the dispersion order in the target preparation process parameters, the phosphogypsum raw material, film-forming component, self-healing component and auxiliary component were added to the pre-dispersion treatment in sequence. The mixture was then mixed and dispersed under the stirring speed and stirring time conditions corresponding to the target preparation process parameters to obtain a uniform mixed system. In the homogeneous mixing system, slurry preparation is carried out according to the reaction temperature and reaction time corresponding to the target preparation process parameters, so that each component forms a stable and dispersed phosphogypsum-based slurry in the system; The phosphogypsum-based slurry is applied to the substrate surface according to the preset coating method, and film-forming treatment is carried out according to the film-forming conditions corresponding to the target preparation process parameters to obtain the initial coating. Select the coating method according to the coating thickness and construction requirements. Apply the phosphogypsum-based slurry evenly to the substrate surface by scraping, spraying or roller coating, and control the coating thickness and coating speed to ensure uniform coating distribution. The initial coating is cured according to the curing time and temperature corresponding to the target preparation process parameters to stabilize the coating structure and complete the curing process, thus obtaining a phosphogypsum-based self-healing coating.

[0029] In this embodiment, the machine learning-based phosphogypsum-based self-healing coating preparation system includes the following modules: The data acquisition module is used to collect the characteristic parameters of phosphogypsum raw materials, the functional group ratio parameters, the preparation process parameters, and the self-healing performance parameters corresponding to the characteristic parameters of phosphogypsum raw materials, the functional group ratio parameters, and the preparation process parameters, and to construct a sample dataset. The data processing module is used to clean and standardize the sample dataset to generate a training dataset. The initial prediction model generation module is used to input the training dataset into an improved kernel limit learning model with pre-set initial kernel function parameters and initial penalty factors, and to train the nonlinear mapping relationship between the phosphogypsum raw material characteristic parameters, functional group allocation parameters, preparation process parameters and self-healing performance parameters to generate the initial prediction model. The model optimization module is used to extract the initial kernel function parameters and initial penalty factor from the initial prediction model as parameters to be optimized, and to perform parameter optimization based on the sparrow search algorithm to obtain the optimized kernel function parameters and optimized penalty factor. The optimized kernel function parameters and optimized penalty factor are then used to update the initial prediction model to generate the optimized kernel extreme learning machine model. The parameter output module is used to input the target self-healing performance parameters into the optimized kernel limit learning machine model and output the target functional group allocation ratio parameters and target preparation process parameters corresponding to the target self-healing performance parameters. The coating preparation module is used to meter raw materials, mix and disperse them, slurry them, form films and cure them according to the target functional group ratio parameters and the target preparation process parameters, to generate phosphogypsum-based self-healing coatings.

[0030] Example 1: To verify the feasibility of this invention in practice, it was applied to a concentrated area of ​​the phosphate chemical industry, where phosphogypsum, as a byproduct, has been stored for a long time, highlighting the urgent need for its resource utilization. A materials research unit, in developing phosphogypsum-based protective coatings, attempted to improve the coating's durability under microcrack conditions by introducing self-healing components. However, during actual research and development, it was found that due to the complex sources of phosphogypsum raw materials, different batches exhibited significant fluctuations in purity, particle size distribution, and impurity content. Furthermore, a strong coupling relationship existed between the functional component ratio and the preparation process, leading to large fluctuations in coating performance, especially crack-closing ability and impermeability. Traditional methods relying on experience-based formulation and step-by-step testing require numerous repeated experiments, with each round of testing typically lasting 3 to 5 days. Optimizing a single formulation often requires more than 20 experimental combinations, resulting in an overall research and development cycle exceeding 3 months, with poor reproducibility and difficulty in achieving stable control.

[0031] In the aforementioned application scenarios, the machine learning-based self-healing coating preparation method described in this invention was used for research and optimization. First, the phosphogypsum raw material was systematically tested to obtain parameters such as purity, calcium sulfate dihydrate content, particle size distribution, specific surface area, moisture content, soluble phosphorus content, and soluble fluorine content. Simultaneously, the proportions of film-forming components, self-healing components, dispersing components, and additive components were recorded. Combined with process parameters under different stirring speeds, reaction temperatures, and curing conditions, a sample dataset was established. This dataset contains 86 valid samples, with phosphogypsum purity ranging from 82% to 96%, median particle size between 12 μm and 48 μm, and moisture content between 4.2% and 9.7%. Corresponding self-healing performance data includes crack closure rate, impermeability, adhesion, and water resistance, with the crack closure rate ranging from 41% to 92%.

[0032] After cleaning and standardizing the sample dataset, 9 sets of missing and outlier samples were removed, retaining 77 sets of valid training data. In the model construction phase, the raw material characteristic parameters, functional group proportion parameters, and preparation process parameters of phosphogypsum were used as inputs, and the self-healing performance parameters were used as outputs, inputting into an improved kernel extreme learning machine model. This model first divides the input parameters into raw material feature subsets, proportion feature subsets, and process feature subsets, and then performs subspace mapping before fusing them. Finally, a nonlinear relationship is constructed through kernel function mapping. During model training, a residual feedback mechanism is used to compensate for errors in the initial prediction results, enabling the model to further learn complex coupling relationships.

[0033] A sparrow search algorithm was introduced to optimize the kernel function parameters and penalty factor in the model. During parameter optimization, the parameters were set to range from 0.01 to 100 for the kernel function and from 0.1 to 1000 for the penalty factor. An initial population of 30 individuals was generated, and iterative optimization was performed through a collaborative update mechanism involving discoverers, followers, and watchdogs. During the iteration process, the model fitness value gradually decreased from an initial 0.183 to 0.042, eventually converging and stabilizing. The optimized model achieved a mean relative error of 3.7% on the validation data.

[0034] In practical applications, target self-healing performance parameters are set, including a crack closure rate of no less than 88%, an anti-permeability improvement of no less than 35%, and an adhesion greater than 1.6 MPa. These target performance parameters are input into the optimized model. Multiple formulation and process combinations are generated through candidate parameter combination search, and candidate schemes with performance deviations of less than 5% are selected through model prediction. Finally, three optimal parameter combinations are obtained. One formulation consists of 58% phosphogypsum (by mass), 22% film-forming component, 8% self-healing component, 6% dispersing component, and 6% additive component, corresponding to a stirring speed of 950 r / min, a reaction temperature of 52℃, and a curing time of 36 h.

[0035] The coating was prepared according to the above output parameters. A high-speed disperser was used for mixing and dispersion to form a uniform slurry, which was then coated and cured. The resulting coating underwent a self-healing test under artificial crack widths of approximately 120 μm. After 48 hours, the crack closure rate reached 91.3%, an improvement of approximately 18.6 percentage points compared to traditional empirical formulations. The anti-permeability improvement rate reached 39.2%, the adhesion test result was 1.72 MPa, and the water resistance test showed no significant blistering or peeling after 72 hours of immersion.

[0036] Compared to traditional methods, which require approximately 24 sets of experiments to obtain a formula with near-performance under the same R&D objectives using an empirical trial-and-error approach, the method of this invention only requires model training and optimization based on existing data of 77 sets. Candidate solutions are generated and screened during the prediction phase, and actual verification requires only 3 sets of experiments, reducing the number of experiments by approximately 87.5%. The R&D cycle is shortened from approximately 90 days to approximately 18 days, improving efficiency by over 80%. Validation under different batches of phosphogypsum raw materials showed that the model prediction error was controlled within ±5%, demonstrating good stability and generalization ability.

[0037] As can be seen from this embodiment, the present invention effectively solves the problems of reliance on experience, difficulty in precise control of performance, and low optimization efficiency in the development of phosphogypsum-based self-healing coatings. It realizes multi-dimensional collaborative modeling and targeted optimization from raw material characteristics and formulation design to process control, which not only improves the controllability of self-healing performance, but also significantly reduces R&D costs and enhances the high-value utilization level of industrial solid waste resources.

[0038] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for preparing a machine learning based phosphogypsum based self-healing coating, characterized by, include: Collect raw material characteristic parameters, functional group allocation ratio parameters, preparation process parameters, and self-healing performance parameters corresponding to the raw material characteristic parameters, functional group allocation ratio parameters, and preparation process parameters of phosphogypsum to construct a sample dataset; The sample dataset is cleaned and standardized to generate a training dataset. The training dataset is input into an improved kernel limit learning model with pre-set initial kernel function parameters and initial penalty factors. The nonlinear mapping relationship between the raw material characteristic parameters, functional group ratio parameters, preparation process parameters and self-healing performance parameters of phosphogypsum is trained to generate an initial prediction model. The initial prediction model is input into the sparrow search algorithm, and the parameters are optimized by taking the initial kernel function parameters and the initial penalty factor as the optimization objects to obtain the optimized kernel function parameters and the optimized penalty factor. The initial prediction model is then updated based on the optimized kernel function parameters and the optimized penalty factor to obtain the optimized kernel extreme learning machine model. Input the target self-healing performance parameters into the optimized kernel limit learning machine model, and output the target functional group allocation ratio parameters and target fabrication process parameters corresponding to the target self-healing performance parameters; Based on the target functional group ratio parameters and target preparation process parameters, raw material metering, mixing and dispersion, pulping, film formation and curing treatment are carried out to generate phosphogypsum-based self-healing coating.

2. The machine learning based method of preparing a phosphogypsum based self-healing coating according to claim 1, wherein, The phosphogypsum raw material characteristic parameters, functional component ratio parameters, preparation process parameters, and self-healing performance parameters in the sample dataset specifically include: The characteristic parameters of phosphogypsum raw materials collected include phosphogypsum purity, calcium sulfate dihydrate content, particle size distribution, specific surface area, moisture content, soluble phosphorus content, and soluble fluorine content. The parameters for collecting functional group proportions include the mass fraction of phosphogypsum, the mass fraction of film-forming components, the mass fraction of self-healing components, the mass fraction of dispersing components, and the mass fraction of auxiliary components; The preparation process parameters collected include stirring speed, stirring time, dispersion sequence, reaction temperature, reaction time, curing time, and curing temperature; The self-healing performance parameters include crack closure rate, impermeability, coating adhesion, water resistance, and durability. The raw material characteristics, functional component ratio parameters, and preparation process parameters of phosphogypsum are used as input features, and the self-healing performance parameters are used as output features. Correspondence is established in the sample dataset.

3. The method for preparing phosphogypsum-based self-healing coatings based on machine learning according to claim 1, characterized in that, The process of cleaning and standardizing the sample dataset includes: The integrity of the raw material characteristic parameters, functional component ratio parameters, preparation process parameters and self-healing performance parameters of phosphogypsum in the sample dataset was checked to identify missing data items. The identified missing data items are processed by removing samples with missing key parameters and filling in samples with missing non-key parameters using statistical values ​​of the corresponding parameters from samples of the same type, thus obtaining complete sample data. Consistency verification was performed on the raw material characteristic parameters, functional component ratio parameters, preparation process parameters, and self-healing performance parameters of phosphogypsum in the complete sample data. Samples with mismatched parameters, inconsistent recorded results, or duplicate entries were removed to obtain consistent sample data. Outlier identification was performed on the characteristic parameters of phosphogypsum raw materials, functional component ratio parameters, preparation process parameters, and self-healing performance parameters in the consistent sample data. Outlier samples that exceeded the preset value range were removed to obtain cleaned sample data. The dimensionality of the raw material characteristic parameters, functional component ratio parameters, preparation process parameters and self-healing performance parameters of phosphogypsum in the cleaned sample data was uniformized, and then standardized based on the dimensionality uniformization to obtain standardized sample data. A training dataset is generated based on the correspondence between the characteristic parameters of phosphogypsum raw materials, the proportion parameters of functional components, the preparation process parameters, and the self-healing performance parameters in the standardized sample data.

4. The method for preparing phosphogypsum-based self-healing coatings based on machine learning according to claim 1, characterized in that, The generation of the initial prediction model includes: The raw material characteristic parameters, functional group ratio parameters, and preparation process parameters of phosphogypsum in the training dataset are combined according to the sample correspondence to generate input data corresponding to each sample, and the self-healing performance parameters corresponding to each sample are determined as output data. An improved kernel extreme learning machine model is constructed by pre-setting initial kernel function parameters and initial penalty factors for input data and output data. The improved kernel extreme learning machine model uses input data as model input and output data as model output. The input data is processed by kernel mapping based on the initial kernel function parameters to generate the kernel mapping result corresponding to the input data, and the training error between the kernel mapping result and the output data is constrained based on the initial penalty factor. Based on the kernel mapping results, output data, and training error constraints, the improved kernel extreme learning machine model is trained to obtain the nonlinear mapping relationship between the input data and the output data. Based on the nonlinear mapping relationship between input and output data, an initial prediction model is generated to characterize the correspondence between the characteristic parameters of phosphogypsum raw materials, the proportion parameters of functional components, the preparation process parameters, and the self-healing performance parameters.

5. The method for preparing phosphogypsum-based self-healing coatings based on machine learning according to claim 4, characterized in that, The improved kernel extreme learning machine model includes: The raw material characteristic parameters, functional group proportioning parameters, and preparation process parameters of phosphogypsum in the training dataset are divided according to parameter type to obtain raw material feature subsets, proportioning feature subsets, and process feature subsets; Subspace mapping is performed on the raw material feature subset, the proportion feature subset, and the process feature subset respectively to obtain the raw material sub-feature mapping result, the proportion sub-feature mapping result, and the process sub-feature mapping result; The raw material sub-feature mapping results, proportion sub-feature mapping results, and process sub-feature mapping results are fused to generate fused input features. The fused input features and self-healing performance parameters are then input into a kernel limit learning model with pre-set initial kernel function parameters and initial penalty factors. The fused input features are used as the model input, and the self-healing performance parameters are used as the model output. The fused input features are processed by kernel mapping based on the initial kernel function parameters to obtain the kernel mapping result corresponding to the fused input features. The training error between the kernel mapping result and the self-repair performance parameters is constrained based on the initial penalty factor. The kernel extreme learning machine model is trained based on the kernel mapping results, self-repair performance parameters, and training error constraints to obtain the initial nonlinear mapping relationship between the fused input features and the self-repair performance parameters, and the initial prediction results are output based on the initial nonlinear mapping relationship. A residual dataset is generated based on the difference between the initial prediction results and the self-healing performance parameters. The fused input features and the corresponding input kernel extreme learning machine model of the residual dataset are then subjected to residual learning to obtain the residual mapping relationship between the fused input features and the residual dataset. The initial nonlinear mapping relationship is updated based on the residual mapping relationship to obtain the corrected nonlinear mapping relationship, and an improved kernel extreme learning machine model is generated based on the corrected nonlinear mapping relationship.

6. The method for preparing phosphogypsum-based self-healing coatings based on machine learning according to claim 1, characterized in that, The generation of the optimized kernel extreme learning machine model includes: The initial kernel function parameters and initial penalty factor in the initial prediction model are extracted as parameters to be optimized, and a fitness evaluation standard is established based on the error between the prediction results of the initial prediction model on the training dataset and the self-repair performance parameters. Based on the value range of the parameter to be optimized, multiple parameter combinations are generated, and each parameter combination is used as the position parameter of an individual sparrow to construct a sparrow population; The positional parameters of each sparrow individual are input into the initial prediction model to obtain the model prediction results for each sparrow individual. The fitness value of each sparrow individual is calculated based on the error between the model prediction results for each sparrow individual and the self-healing performance parameters. A global search is performed based on the location parameters of the discoverer to obtain the updated location parameters of the discoverer. The location parameters of the followers are then updated based on the updated location parameters of the discoverer to obtain the updated location parameters of the followers. Finally, the location parameters of the sparrows are adjusted based on the response of the vigilant individuals to the changes in fitness values ​​to obtain the updated location parameters of the sparrows. The updated sparrow individual position parameters are re-input into the initial prediction model, the fitness value corresponding to the updated sparrow individual is recalculated, and the sparrow individual sorting, sparrow individual division and sparrow individual position parameter update are repeated according to the recalculated fitness value until the preset iteration termination condition is reached. The positional parameters corresponding to the sparrow individual with the best fitness value when the preset iteration termination condition is reached are extracted and determined as the optimized kernel function parameters and optimized penalty factor. The optimized kernel function parameters and optimized penalty factor are then used to replace the initial kernel function parameters and initial penalty factor in the initial prediction model to update the initial prediction model, resulting in the optimized kernel extreme learning machine model.

7. The method for preparing phosphogypsum-based self-healing coatings based on machine learning according to claim 1, characterized in that, The process of outputting the target functional group allocation ratio parameter and the target preparation process parameter corresponding to the target self-healing performance parameter includes: Obtain the target self-healing performance parameters, and organize the target self-healing performance parameters according to the output structure of the self-healing performance parameters in the optimized kernel extreme learning machine model to generate target performance data; Candidate functional group allocation ratio parameters and candidate preparation process parameters are generated based on the value range of functional group allocation ratio parameters and preparation process parameters in the training dataset. The candidate functional group allocation ratio parameters and candidate preparation process parameters are then combined according to the input structure of the optimized kernel extreme learning machine model to generate candidate input data. The candidate input data is input into the optimized kernel extreme learning machine model to obtain the predictive self-repair performance parameters corresponding to the candidate input data. The predicted self-healing performance parameters are compared with the target performance data, the performance deviation between the predicted self-healing performance parameters and the target self-healing performance parameters is calculated, and the candidate input data is filtered according to the performance deviation to obtain the candidate input data that meets the preset conditions. Extract the corresponding functional group allocation ratio parameters and preparation process parameters from the candidate input data that meet the preset conditions, and generate the target functional group allocation ratio parameters and target preparation process parameters corresponding to the target self-healing performance parameters.

8. The method for preparing phosphogypsum-based self-healing coatings based on machine learning according to claim 1, characterized in that, The process of raw material metering, mixing and dispersion, pulping, film formation and curing based on the target functional group ratio parameters and target preparation process parameters includes: Based on the mass fraction of each component in the target functional group distribution ratio parameter, the phosphogypsum raw material, film-forming component, self-healing component, dispersing component and auxiliary component are quantitatively weighed to obtain raw materials of each component consistent with the target functional group distribution ratio parameter; The weighed dispersion components were added to a solvent for pre-dispersion treatment. Then, according to the dispersion order in the target preparation process parameters, the phosphogypsum raw material, film-forming component, self-healing component and auxiliary component were added to the pre-dispersion treatment in sequence. The mixture was then mixed and dispersed under the stirring speed and stirring time conditions corresponding to the target preparation process parameters to obtain a uniform mixed system. In the homogeneous mixing system, slurry preparation is carried out according to the reaction temperature and reaction time corresponding to the target preparation process parameters, so that each component forms a stable and dispersed phosphogypsum-based slurry in the system; The phosphogypsum-based slurry is applied to the substrate surface according to the preset coating method, and film-forming treatment is carried out according to the film-forming conditions corresponding to the target preparation process parameters to obtain the initial coating. The initial coating was cured according to the curing time and temperature corresponding to the target preparation process parameters to obtain a phosphogypsum-based self-healing coating.

9. A machine learning-based phosphogypsum-based self-healing coating preparation system, characterized in that, Includes the following modules: The data acquisition module is used to collect the characteristic parameters of phosphogypsum raw materials, the functional group ratio parameters, the preparation process parameters, and the self-healing performance parameters corresponding to the characteristic parameters of phosphogypsum raw materials, the functional group ratio parameters, and the preparation process parameters, and to construct a sample dataset. The data processing module is used to clean and standardize the sample dataset to generate a training dataset. The initial prediction model generation module is used to input the training dataset into an improved kernel limit learning model with pre-set initial kernel function parameters and initial penalty factors, and to train the nonlinear mapping relationship between the phosphogypsum raw material characteristic parameters, functional group allocation parameters, preparation process parameters and self-healing performance parameters to generate the initial prediction model. The model optimization module is used to extract the initial kernel function parameters and initial penalty factor from the initial prediction model as parameters to be optimized, and to perform parameter optimization based on the sparrow search algorithm to obtain the optimized kernel function parameters and optimized penalty factor. The optimized kernel function parameters and optimized penalty factor are then used to update the initial prediction model to generate the optimized kernel extreme learning machine model. The parameter output module is used to input the target self-healing performance parameters into the optimized kernel limit learning machine model and output the target functional group allocation ratio parameters and target preparation process parameters corresponding to the target self-healing performance parameters. The coating preparation module is used to meter raw materials, mix and disperse them, slurry them, form films and cure them according to the target functional group ratio parameters and the target preparation process parameters, to generate phosphogypsum-based self-healing coatings.

Images