High-performance material visual screening system based on machine learning and application method thereof

Abstract

The invention provides a high-performance material visual screening system based on machine learning and an application method thereof. The system comprises a feature extraction module, a machine learning model training module, a prediction module and a feedback optimization module. The data preprocessing module is used for performing primary processing on the original data, wherein the primary processing comprises removal of invalid data and abnormal values; the feature extraction module is used for converting a time domain signal into a frequency domain signal by adopting Fourier transform, extracting a frequency domain feature, and performing data dimension reduction by adopting principal component analysis (PCA); the machine learning model training module is used for ensuring the generalization ability of the model by adopting a cross validation technology in the training process, and accelerating the convergence process of the model through a gradient descent method; the prediction module is used for predicting the material performance of the high-performance material to obtain a prediction result; and the feedback optimization module is used for automatically optimizing model parameters by adopting Bayesian optimization, analyzing a prediction result of the model and dynamically adjusting the model parameters so as to improve the prediction precision. Through the scheme of the embodiment of the invention, the law can be learned from a large amount of experimental data by a machine learning technology, and the material performance can be quickly and accurately predicted.

Description

Technical Field

[0001] The embodiments of the present invention relate to the application of machine learning to the field of materials science and technology, and more particularly to a high-performance material visualization screening system based on machine learning and an application method thereof. Background Art

[0002] With the continuous advancement of science and technology, high-performance materials are increasingly used in aviation, aerospace, electronics, energy and other fields. These materials usually have excellent physical, chemical or mechanical properties, which are of great significance in promoting technological innovation in related fields. However, the research and development of high-performance materials often requires a lot of experiments and tests, which is long, costly, and difficult to ensure the stability and reliability of material performance.

[0003] In recent years, machine learning technology has received extensive attention in the field of materials science, especially showing great potential in material performance prediction, structural optimization, fault diagnosis, etc. Traditional material performance prediction methods mainly rely on physical models and empirical formulas, which have certain limitations when dealing with complex systems. In contrast, machine learning technology can learn rules from a large amount of experimental data and achieve rapid and accurate prediction of material properties. Summary of the invention

[0004] In view of this, an embodiment of the present invention provides a high-performance material visualization screening system based on machine learning and an application method thereof, which at least partially solves the problems existing in the prior art.

[0005] In a first aspect, an embodiment of the present invention provides a high-performance material visualization screening system based on machine learning, comprising:

[0006] Data preprocessing module, feature extraction module, machine learning model training module, prediction module, feedback optimization module;

[0007] The data preprocessing module is used to: perform preliminary processing on the raw data to ensure the quality and consistency of the data. The preliminary processing includes removing invalid data and outliers. The outliers are detected using the Z-score method, and the threshold is set to 3σ, that is, three times the standard deviation, to eliminate the impact of extreme values ​​on model training;

[0008] The feature extraction module is used to extract key features that help improve the model's prediction ability from the data after data preprocessing. The extraction techniques include Fourier transform and principal component analysis (PCA). Fourier transform is used to convert time domain signals into frequency domain signals, extract frequency domain features, and reveal the characteristics of materials at different frequencies. Principal component analysis (PCA) is used to reduce data dimensionality, retain the most useful features for model prediction, reduce data redundancy, and improve computational efficiency.

[0009] The machine learning model training module is used to: perform model training based on preprocessed data and extracted key features. During the training process, cross-validation technology is used to ensure the generalization ability of the model, and the gradient descent method is used to accelerate the convergence process of the model;

[0010] The prediction module is used to: predict the material properties of high-performance materials to obtain prediction results;

[0011] The feedback optimization module is used to automatically optimize the model parameters using Bayesian optimization, and dynamically adjust the model parameters to improve the prediction accuracy by analyzing the prediction results of the model.

[0012] Optionally, in a specific embodiment,

[0013] The machine learning model training module uses the following models for training: Random Forest, Support Vector Machine (SVM), Gradient Boosting, and Multilayer Perceptron (MLP) models, where the Multilayer Perceptron (MLP) contains at least 3 hidden layers.

[0014] Optionally, in a specific embodiment,

[0015] When the machine learning model training module is a multi-layer perceptron MLP model, the activation function of the multi-layer perceptron MLP model adopts the ReLU function, the loss function adopts the mean square error MSE, and batch normalization technology is used in the model training process to accelerate convergence.

[0016] Optionally, in a specific embodiment,

[0017] In the feedback optimization module, the initial exploration times of Bayesian optimization were 10, the subsequent iteration times were 50, and the objective function adopted the negative log-likelihood function to gradually find the global optimal solution and optimize the model performance.

[0018] In a second aspect, an application method is provided in an embodiment of the present invention, and the method is applicable to the high-performance material visual screening system based on machine learning described in claim 1, wherein the visual screening system includes a data preprocessing module, a feature extraction module, a machine learning model training module, a prediction module, and a feedback optimization module;

[0019] The application method comprises:

[0020] S1: Collect relevant data of the high-performance material, including the number of graphene layers, pore size, material area, pressure change, and stress-strain relationship;

[0021] S2: Use the data processing module and feature extraction module to preprocess and extract features from the collected relevant data;

[0022] S3: Train the machine learning model using the machine learning model training module, where the model training process uses cross-validation technology and the Bayesian algorithm is used for hyperparameter optimization; for example, the validation set ratio is 20%;

[0023] S4: using the prediction module to predict the performance of the new high-performance material to obtain prediction results, wherein the prediction results include key performance indicators of the material, such as water flux and salt rejection rate;

[0024] S5: Utilizing a feedback optimization module to optimize the machine learning model according to the prediction results, including adjusting model parameters and hyperparameters to improve prediction accuracy.

[0025] The embodiment of the present invention provides a high-performance material visualization screening system based on machine learning, including: a data preprocessing module, a feature extraction module, a machine learning model training module, a prediction module, and a feedback optimization module; the data preprocessing module is used to: perform preliminary processing on the original data to ensure the quality and consistency of the data, and the preliminary processing includes removing invalid data and outliers to eliminate the influence of extreme values ​​on model training; the feature extraction module is used to: extract key features that help improve the prediction ability of the model from the data after the data preprocessing is completed, and the extraction techniques used include: Fourier transform and principal component analysis PCA, Fourier transform is used to convert time domain signals into frequency domain signals, extract frequency domain signals, and extract frequency domain signals. Domain features reveal the characteristics of materials at different frequencies; principal component analysis PCA is used for data dimensionality reduction, retaining the most useful features for model prediction, reducing data redundancy, and improving computing efficiency; the machine learning model training module is used to: be responsible for model training based on preprocessed data and extracted key features. During the training process, cross-validation technology is used to ensure the generalization ability of the model, and the convergence process of the model is accelerated by the gradient descent method; the prediction module is used to: predict the material properties of high-performance materials to obtain prediction results; the feedback optimization module is used to: automatically optimize the model parameters using Bayesian optimization, and dynamically adjust the model parameters to improve the prediction accuracy by analyzing the prediction results of the model. Through the solution of the embodiment of the present invention, it is possible to solve the problem that machine learning technology can learn laws from a large amount of experimental data and realize rapid and accurate prediction of material properties. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In the accompanying drawings, unless otherwise specified, the same reference numerals throughout the multiple drawings represent the same or similar parts or elements. These drawings are not necessarily drawn to scale. It should be understood that these drawings only depict some embodiments disclosed in the present invention and should not be regarded as limiting the scope of the present invention.

[0027] Figure 1 A schematic diagram of the structure of a high-performance material visual screening system based on machine learning;

[0028] Figure 2 The figure is a schematic diagram of an application flow of the application method. DETAILED DESCRIPTION

[0029] In order to make the purpose, technical scheme and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0030] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention more clearly understood, the embodiments of the present invention are further described in detail below in combination with the embodiments and drawings. The schematic implementation methods of the embodiments of the present invention and their descriptions are only used to explain the embodiments of the present invention and are not intended to limit the embodiments of the present invention.

[0031] In modern materials science, machine learning technology can be used to identify valuable patterns from massive experimental data, which has far-reaching significance for accelerating the development of new materials. This article will describe in detail a high-performance material visualization screening system based on machine learning, which can effectively solve the key problem of how machine learning technology can quickly and accurately predict material properties from a large amount of experimental data.

[0032] Embodiment 1

[0033] like Figure 1 As shown, the high-performance material visualization screening system 10 includes: a data preprocessing module 101, a feature extraction module 102, a machine learning model training module 103, a prediction module 104 and a feedback optimization module 105.

[0034] The data preprocessing module 101 is used to perform preliminary processing on the original data, and the preliminary processing includes removing invalid data and abnormal values.

[0035] The raw data is preliminarily processed to ensure the quality and consistency of the data. The preliminary processing includes removing invalid data and outliers. The outliers are detected using the Z-score method, and the threshold is set to 3σ, which is three times the standard deviation, to eliminate the impact of extreme values ​​on model training.

[0036] The task of this stage is to prepare the experimental data and ensure its high quality and consistency for further analysis. In order to achieve this, in actual operation, it is necessary to exclude obviously erroneous or non-representative data, that is, to remove invalid data and outliers. The Z-score (standardized score) method used can effectively mark and remove these abnormal measurement data. For example, if there are very few resistivity values ​​in the data of a certain experimental batch that deviate significantly from their sample mean, and their Z-score exceeds the threshold value of plus or minus 3 standard deviations (that is, three times the standard deviation), then such values ​​are considered abnormal and will be systematically eliminated.

[0037] The feature extraction module 102 is used to: convert the time domain signal into the frequency domain signal by Fourier transform, extract the frequency domain feature, and perform data dimension reduction by principal component analysis PCA.

[0038] Extract key features that help improve the model's prediction ability from the data after data preprocessing. The extraction techniques include: Fourier transform and principal component analysis (PCA). Fourier transform is used to convert time domain signals into frequency domain signals, extract frequency domain features, and reveal the characteristics of materials at different frequencies; principal component analysis (PCA) is used to reduce data dimensionality, retain the most useful features for model prediction, reduce data redundancy, and improve computational efficiency. The preprocessed data needs to be further processed before it can be used for training models. At this time, Fourier transform is used to capture the performance characteristics of materials in a specific spectral range to help us better understand their behavioral characteristics; at the same time, principal component analysis (PCA) technology is used to reduce dimensionality and focus on the most valuable information set. In order to improve the model's sensitivity and discrimination in capturing target features, we use a combination of multiple feature extraction techniques, including Fourier transform to capture the spectral characteristics of material properties in time-varying data, and principal component analysis (PCA) to reduce data dimensionality and focus on extracting those information features that are most critical to predictive performance, ensuring that efficient computing performance is maintained while reducing input variable redundancy.

[0039] The machine learning model training module 103 is used to: perform model training based on preprocessed data and extracted key features. During the training process, cross-validation technology is used to ensure the generalization ability of the model, and the gradient descent method is used to accelerate the convergence process of the model.

[0040] This module receives preprocessed and selected feature data for learning and fitting. In this example, we assume that the goal is to find a set of optimal structural parameters for producing high-strength and lightweight nanocarbon fiber reinforced materials. In order to make the final result more reliable and applicable to other similar situations, the K-fold cross-validation strategy should be used (k is usually a positive integer between 5 and 10), that is, the test set is selected in turn within each model iteration training cycle to verify the robustness and wide applicability of the trained parameter set, and the speed and stability indicators of convergence to the minimum error result set are accelerated by the gradient descent optimization strategy. In the machine learning model training module, the cross-validation method is used to evaluate and iterate the model, which can effectively improve the accuracy and robustness of the model on future unknown test sets. At the same time, by applying optimization strategies such as gradient descent, the learning process can be accelerated and the model can converge to the ideal weight state as soon as possible.

[0041] The prediction module 104 is used to predict the material properties of the high-performance material to obtain a prediction result.

[0042] Based on the trained machine learning model, we can calculate and predict the relevant indicators of material performance for the samples to be tested, and obtain the corresponding quantitative prediction results; at the same time, combining all the above steps, the entire screening system can effectively realize the visual analysis and selection of high-performance materials.

[0043] The feedback optimization module 105 is used to automatically optimize the model parameters by using Bayesian optimization, and dynamically adjust the model parameters by analyzing the prediction results of the model.

[0044] Here, the Bayesian optimization algorithm is used to automatically search for hyperparameter configuration solutions under better combination settings, and the parameters are adjusted in combination with the prediction error feedback to ensure a long-term and continuous trend of prediction quality improvement. In this way, an efficient closed-loop mechanism is achieved in the entire screening process from input data preprocessing to the final material prediction effect display, which significantly improves the efficiency and accuracy of intelligent recognition driven by massive experimental evidence, thereby achieving the goal of solving prediction and analysis problems in the field of materials science through machine learning.

[0045] The screening system uses automated data collection to achieve the tedious preliminary data preparation work in materials science research, and uses advanced mathematical tools and technologies to dig out deep material behavior information to drive the intelligent auxiliary development process, significantly shortening the new material research and development cycle and greatly reducing the demand for manpower investment. In short, in practical applications, with the help of such innovative technical solutions, not only can the experimental costs be greatly reduced, the speed of new material products being launched can be shortened, but also more possibilities can be brought to the future science and technology industry.

[0046] Optionally, in one embodiment, the machine learning model training module adopts the following models for training: Random Forest, Support Vector Machine SVM, Gradient Boosting and Multilayer Perceptron MLP model, wherein the Multilayer Perceptron MLP includes at least 3 hidden layers.

[0047] In this process, model training is a core link, which uses historical data to adjust model parameters to improve the accuracy and generalization ability of predictions. The following is an explanation and step-by-step description of several learning algorithms mentioned, and will also be explained through a specific example.

[0048] Random Forest: First, a set of independent decision tree models are established, and then the majority voting mechanism is used to obtain the results. Support Vector Machine SVM: The goal of SVM is to build a classification function that finds the optimal dividing hyperplane during training to achieve the maximum separation of each category. Gradient Boosting (GB) uses the principle of combining weak models into a powerful model, and optimizes the performance of the entire algorithm by repeatedly training multiple decision trees with greater weights for incorrectly predicted samples in the data set. Multilayer Perceptron (MLP), as a typical artificial neural network (ANN), has an input layer, an output layer, and a multi-layer fully connected neural node (Hidden Layers) between them, which is used for complex task processing. In our example, more than three layers of Hidden Layers are set to capture nonlinear relationships, so that deeper features can be captured in fields such as stock market prediction analysis. We first initialize all weights in these layers; then adjust the connection strength through backpropagation, and repeat the forward pass and weight iteration to reach the minimum loss value or meet the stop condition. In practice, feature engineering, hyperparameter optimization and other work steps may be required to improve the overall effect.

[0049] Further optionally, in one embodiment, when the machine learning model training module is a multi-layer perceptron MLP model, the activation function of the multi-layer perceptron MLP model adopts the ReLU function, the loss function adopts the mean square error MSE, and batch normalization technology is used in the model training process to accelerate convergence. For such a model design and training configuration, the meaning of each component must be understood first.

[0050] Multilayer Perceptron (MLP) is a type of artificial neural network composed of multiple fully connected layers. Through layer-by-layer propagation calculations, MLP has a strong adaptability to the expression of nonlinear functions, allowing input data to be mapped to multiple intermediate processing layers to extract useful features and make predictions.

[0051] Suppose we want to achieve visual screening of the position and size of an object in an image. We use a multi-layer perceptron with 4 hidden layers to achieve this goal. Each hidden layer is fully connected, and the hidden layer to the output layer is also a fully connected structure. There is no loop or cross-connection between the layers. The overall layout is trapezoidal. The number of input and output nodes can be adjusted appropriately to meet application requirements depending on the problem. This structure helps the model extract complex image features to achieve higher accuracy in target recognition.

[0052] The ReLu function is a commonly used choice as an activation function embedded after each hidden layer to increase the complexity of the neural network and thus improve its generalization performance, so that the multi-layer perceptron can simulate complex nonlinear mapping relationships, thereby solving more complex and difficult classification and regression problems. At the same time, the introduction of ReLu also simplifies the computational cost of parameter updates, making the training process more efficient. Specifically, after the weighted summation of the information from the predecessor node received by each neuron and the corresponding deviation, ReLU activation is used to obtain a nonlinear change result. If the result is positive, the value remains unchanged; if the result is negative, the activation function sets the result to 0. In the above example, through the role of the ReLU activation function, the model is more sensitive to the key elements in the image during the training process of certain specific image data.

[0053] The loss function (mean square error, MSE) is used to guide the speed and direction of the network weights approaching the optimization target. It measures the mean of the square of the difference between the model's predicted output value and the actual label. It is a commonly used loss calculation standard for regression problems, and its mathematical form is simple and clear. In this case, we can express the size of our model error by averaging the square of the distance between the image label coordinates and the predicted coordinates; then use the gradient method to adjust the weights to reduce this gap, and continue to adjust until a more satisfactory result is obtained.

[0054] Batch Normalization (BN) is added to the MLP model to improve the model's expressiveness and training speed. In each batch, a normalized statistical method (such as mean = 0, variance = 1) is performed on the distribution of the data set of each layer of neurons to reduce the displacement of internal covariates, so as to stabilize the model training, accelerate its convergence and improve the final accuracy to a certain extent. For example: after each small batch is trained, the data received by all neurons at the current moment is subtracted from the average value of the features represented by all sample data in this small batch at the neuron nodes in this layer, and then divided by the square root of the variance of these features.

[0055] Through these specific settings, it can be ensured that the multi-layer perceptron training module can further improve its prediction effect on the basis of learning complex feature mapping, and shorten the time required for the model to reach a stable state.

[0056] Optionally, in one embodiment, in the feedback optimization module, the initial exploration times of Bayesian optimization are 10 times, the subsequent iteration times are 50 times, and the objective function adopts a negative log-likelihood function to gradually find the global optimal solution and optimize the model performance.

[0057] Bayesian optimization is a black-box optimization problem, especially in the context of high-cost objective function parameter tuning. In this context, we first need to define the specific steps of the Bayesian optimization process.

[0058] In the initial stage, sampling is performed through 10 purely random searches. The purpose of these samples is mainly to collect information about the range of parameters to be optimized, laying the foundation for more refined adjustments in the next step. For example, when building a machine learning model, the parameters that need to be adjusted include but are not limited to the maximum depth of the decision tree model, the learning rate, etc. The first ten groups of settings may randomly select some parameters, such as the maximum depth can be one of the values ​​from 2 to 15, and the learning rate may be randomly selected in the range of 0.01 to 0.5.

[0059] In the subsequent stage, the algorithm will generate a probability model (such as a Gaussian process) based on these ten sets of parameters and results to predict which areas may have better results. This algorithm is based on the results collected in the previous step to calculate the sampling criterion or acquisition function (such as expected improvement, probability improvement, or upper limit trust enhancement, etc.); in the next 50 rounds, this criterion will be used to determine the sampling location of the next group; the 50 subsequent sampling processes are no longer randomly selected but use previous information to guide the next attempt direction; this can effectively help us lock the optimal solution or the location of the solution very close to the global optimal solution more quickly in the search space, thereby improving the efficiency of the parameter adjustment process.

[0060] In this process, the optimization process targets the negative log-likelihood as the objective function. For example, if you are solving a regression problem and use a loss based on the predicted value of a probability distribution, such as the mean square error (MSE) loss, the Bayesian framework will try to maximize the probability of the observed data for the model hypothesis (i.e., the probability of the data occurring is as high as possible). The optimization goal is indirectly achieved by minimizing the negative log-likelihood, that is, by maximizing the likelihood of the data, which is a very common situation in machine learning model training.

[0061] In summary, in a typical case, such as for a complex data set, we have a preliminarily selected set of model parameters but are not sure what the best configuration is. We can use the initial 10 trials to roughly adjust the general direction; then, based on the results of the previous experiments, the parameter configuration of the subsequent 50 iterations can give a more specific parameter combination through the Bayesian optimization algorithm, and finally reach the global optimal solution or a level very close to the optimal; and by using the negative logarithmic probability (log likelihood) loss as the evaluation criterion, we can effectively evaluate the effect of each set of trial results, while also avoiding the local optimal point trap problem often encountered in some model training processes, thereby ultimately helping us find the optimal model configuration and achieve the expected performance within a limited time.

[0062] Embodiment 2

[0063] The application method in the embodiment of the present invention is applicable to a high-performance material visual screening system based on machine learning, and the visual screening system includes a data preprocessing module, a feature extraction module, a machine learning model training module, a prediction module, and a feedback optimization module; the application method:

[0064] The goal of the high-performance material visual screening system is to find those with excellent properties from a large number of material candidates. This system uses multiple modules to form a closed-loop system to achieve this goal.

[0065] like Figure 2 As shown, the application method includes:

[0066] S1: Collect relevant data on high-performance materials, including number of graphene layers, pore size, material area, pressure change, and stress-strain relationship.

[0067] First (S1), relevant experimental data and literature information need to be collected in the system. For example, for a specific new water treatment membrane material such as graphene oxide, the data to be collected may involve the number of layers of the material under different preparation processes, the pore distribution range (usually referred to as pore size), the specific area size of the film, and the stress-strain characteristics of these materials under external pressure or other external forces. For example, a series of samples with different graphene layers and pore size structures are prepared under laboratory conditions.

[0068] S2: Use the data processing module and feature extraction module to preprocess and extract features from the collected relevant data.

[0069] These raw data must go through a preprocessing process (removing invalid records and abnormal data, etc.) to ensure the accuracy of the model, and use the feature extraction module to refine effective characteristics for modeling. For example, for the materials in the aforementioned samples, we may focus on specific characteristics such as the impact of pore size on filtration performance or the impact of changes in the number of material layers on pressure-bearing performance.

[0070] S3: Use the machine learning model training module to train the machine learning model, where the model training process uses cross-validation technology and the Bayesian algorithm is used for hyperparameter optimization.

[0071] After entering (S3), the model training phase begins. Machine learning techniques are used here (for example, advanced parameter adjustment methods such as Bayesian optimization combined with traditional cross-validation methods to ensure that the best model selected during the training process can still perform well on unknown samples). The data that has been cleaned and sorted in the early stage and effectively screened for specificity is used as the training set for model building and hyperparameter optimization. At the same time, a portion of the test data is reserved for monitoring and adjustment to prevent overfitting. During this cross-validation, the validation set ratio can be 20% or other ratios.

[0072] S4: Use the prediction module to predict the performance of new high-performance materials to obtain prediction results, which include key performance indicators of the materials.

[0073] When the initial training is completed and relatively stable performance indicators are obtained, the prediction process (S4) can be started to evaluate the performance of new samples: the relevant information of the new type of material that is not directly included in the above process is substituted into the existing optimal version model for simulation calculation to obtain the output value - in the above example, it may refer to the prediction of key performance indicators such as water flux / desalination efficiency. Taking graphene for filtration as an example, based on the preliminary results, researchers can conduct further experimental exploration of the new structure. The key performance values ​​of the material may include but are not limited to water flux and salt rejection rate.

[0074] S5: Use the feedback optimization module to optimize the machine learning model based on the prediction results, including adjusting model parameters and hyperparameters.

[0075] Finally (S5), the system also needs to provide an optimization channel for users to feedback the deviation between the actual test and the simulation analysis, and use this as a basis to iteratively upgrade the parameters and framework of the original settings (this also involves the key link in the model parameter optimization process mentioned above). To ensure that the prediction accuracy remains at a high level over time - in our example, if after repeated experiments, it is found that the results obtained under a specific layer thickness condition deviate greatly from the simulation results, the optimization path provided in this link can be used to adjust the relevant settings accordingly until they finally converge to the expected target range.

[0076] The following describes an example of using a high-performance material visualization screening system based on machine learning to screen graphene-based seawater desalination membranes and carbon nanotube-based seawater desalination membranes.

[0077] Embodiment 3

[0078] High-performance material visual screening system based on machine learning for screening graphene-based desalination membranes

[0079] 1. Implementation of data collection and preprocessing:

[0080] Data on graphene-based desalination membranes were collected from experimental databases and literature. The collected data included the number of graphene layers, pore size, material area, pressure change, stress-strain relationship, etc. The data was cleaned using a data preprocessing module, and the Z-score method was used to remove outliers, with the threshold set to 3σ.

[0081] 2. Implementation of feature extraction module:

[0082] Fourier transform was used to extract the frequency domain features of the graphene film surface characteristics. Then, principal component analysis (PCA) was used to reduce the dimensionality of the multidimensional data and extract the most representative features.

[0083] 3. Implementation of machine learning model training module:

[0084] The extracted features were input into the Random Forest and Support Vector Machine (SVM) models for training. Cross-validation technology was used in the training process, and the validation set ratio was 20%. The model was tuned by gradient descent method and Bayesian optimization method to finally determine the optimal model parameters.

[0085] 4. Implementation of feedback optimization module:

[0086] Parameter optimization: According to the prediction results, the gradient descent method is used to adjust the model parameters. The initial learning rate is set to 0.01 and gradually decays according to the number of iterations.

[0087] Bayesian optimization: Bayesian optimization is used to automatically optimize model parameters. By building a proxy model and continuously updating the posterior distribution of hyperparameters, the global optimal solution is gradually found. The initial number of explorations is set to 10, the number of subsequent iterations is set to 50, and the objective function uses the negative log-likelihood function.

[0088] 5. Implementation of visualization module:

[0089] The prediction results are presented through a visualization module, including the performance comparison of different graphene-based films, helping users to quickly screen out the best graphene-based materials for further development and testing.

[0090] Embodiment 4

[0091] High-performance material visual screening system based on machine learning for screening carbon nanotube-based desalination membranes

[0092] 1. Implementation of data collection and preprocessing:

[0093] Data on nanotube-based desalination membranes were collected from experimental databases and literature. The collected data included the number of graphene layers, pore size, material area, working pressure, stress-strain relationship, etc. The data preprocessing module was used to clean the data, and the Z-score method was used to remove outliers, with the threshold set to 2σ.

[0094] 2. Implementation of feature extraction module:

[0095] Fourier transform was used to extract the frequency domain features of the surface characteristics of nanotube-based desalination membranes. Then, principal component analysis (PCA) was used to reduce the dimensionality of the multidimensional data and extract the most representative features.

[0096] 3. Implementation of machine learning model training module:

[0097] The extracted features were input into the Random Forest and Support Vector Machine (SVM) models for training. Cross-validation technology was used in the training process, and the validation set ratio was 30%. The model was tuned by gradient descent method and Bayesian optimization method to finally determine the optimal model parameters.

[0098] 4. Implementation of feedback optimization module:

[0099] Parameter optimization: According to the prediction results, the gradient descent method was used to adjust the model parameters. The initial learning rate was set to 0.02 and gradually decayed according to the number of iterations.

[0100] Bayesian optimization: Bayesian optimization is used to automatically optimize model parameters. By building a proxy model and continuously updating the posterior distribution of hyperparameters, the global optimal solution is gradually found. The initial number of explorations is set to 20, the number of subsequent iterations is set to 100, and the objective function uses the negative log-likelihood function.

[0101] 5. Result visualization and decision making

[0102] Through the visualization module, users can intuitively view the performance prediction results of different carbon nanotube-based membranes, compare the advantages and disadvantages of various design schemes, and select the optimal scheme for further experimental verification.

[0103] In general, the high-performance material visual screening system based on machine learning and its application method is a well-structured and technologically innovative product. It uses a series of advanced data science technologies to deal with complex material analysis problems, which can greatly improve the speed and accuracy of new material exploration in R&D work and effectively promote the development of new material applications in the industrial field.

[0104] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A high-performance material visual screening system based on machine learning, characterized in that: include: Data preprocessing module, feature extraction module, machine learning model training module, prediction module, feedback optimization module; The data preprocessing module is used to: perform preliminary processing on the original data, including removing invalid data and outliers; The feature extraction module is used to: convert the time domain signal into the frequency domain signal using Fourier transform, extract the frequency domain features, and use principal component analysis PCA to reduce the data dimension; The machine learning model training module is used to: During the training process, cross-validation technology is used to ensure the generalization ability of the model, and the gradient descent method is used to accelerate the convergence process of the model; The prediction module is used to: predict the material properties of high-performance materials to obtain prediction results; The feedback optimization module is used to automatically optimize the model parameters using Bayesian optimization and dynamically adjust the model parameters by analyzing the prediction results of the model.

2. The visual screening system according to claim 1, characterized in that: The machine learning model training module uses the following models for training: Random Forest, Support Vector Machine (SVM), Gradient Boosting, and Multilayer Perceptron (MLP) models, where the Multilayer Perceptron (MLP) contains at least 3 hidden layers.

3. The visual screening system according to claim 2, characterized in that: When the machine learning model training module is a multi-layer perceptron MLP model, the activation function of the multi-layer perceptron MLP model adopts the ReLU function, the loss function adopts the mean square error MSE, and batch normalization technology is used in the model training process to accelerate convergence.

4. The visual screening system according to claim 1, characterized in that: In the feedback optimization module, the initial exploration times of Bayesian optimization were 10, the subsequent iteration times were 50, and the objective function adopted the negative log-likelihood function to gradually find the global optimal solution and optimize the model performance.

5. An application method, characterized in that: The method is applicable to the high-performance material visual screening system based on machine learning described in claim 1, which includes a data preprocessing module, a feature extraction module, a machine learning model training module, a prediction module, and a feedback optimization module; The application method comprises: S1: Collect relevant data of the high-performance material, including the number of graphene layers, pore size, material area, pressure change, and stress-strain relationship; S2: Use the data processing module and feature extraction module to preprocess and extract features from the collected relevant data; S3: Use the machine learning model training module to train the machine learning model, where the model training process uses cross-validation technology and the Bayesian algorithm is used for hyperparameter optimization; S4: using the prediction module to predict the performance of the new high-performance material to obtain a prediction result, wherein the prediction result includes a key performance index of the material; S5: Utilize the feedback optimization module to optimize the machine learning model according to the prediction results, including adjusting model parameters and hyperparameters.

Images