Material thermal transition characteristic prediction method combining classification algorithm and machine learning

By combining classification algorithms and machine learning, and utilizing the composition data of ash slag, a thermal transformation characteristic curve of ash slag was generated, which solved the problems of poor repeatability and long cycle in the existing technology and realized efficient thermal transformation characteristic measurement.

CN118981632AActive Publication Date: 2024-11-19THE UNIV OF NOTTINGHAM NINGBO CHINA

Patent Information

Application Number
CN202411012746.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-26
Publication Date
2024-11-19
Estimated Expiration
2044-07-26

AI Technical Summary

Technical Problem

In existing technologies, the methods for measuring the thermal transformation characteristics of ash slag have poor repeatability, making it difficult to accurately predict specific melting characteristic temperatures, and the experimental cycle is long.

Method used

By combining classification algorithms and machine learning, thermal transformation characteristic curves are generated through ashing processing, component detection, ash type analysis, and appropriate machine learning algorithm analysis.

Benefits of technology

It shortens the time required to determine the thermal transformation properties of materials, improves the measurement efficiency, and generates accurate thermal transformation characteristic curves.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118981632A_ABST
    Figure CN118981632A_ABST
Patent Text Reader

Abstract

The invention provides a material thermal transition characteristic prediction method in combination with a classification algorithm and machine learning. The method comprises the following steps: S1, obtaining a test sample, and performing ashing treatment on the test sample to obtain ash; s2, performing component detection on the ash residues to obtain component data; s3, analyzing the component data by using a classification algorithm to obtain ash classification types corresponding to the ash residues; and S4, based on the ash content type corresponding to the ash, selecting a corresponding machine learning algorithm to analyze the component data to obtain a corresponding thermal transformation characteristic curve. The method has the beneficial effects that the time required for measuring the thermal transition characteristic of the material can be shortened, and the efficiency is improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of material property prediction, and in particular to a material thermal transition property prediction method combining classification algorithm and machine learning. Background Art

[0002] The thermal transition properties of materials are one of the properties that have received much attention. The thermal transition properties of materials are of great significance in the fields of materials science, energy and engineering, especially when evaluating the behavior of materials under high temperature conditions. For example, silicon materials are widely used in electronic and optoelectronic devices, and their morphology and properties at different temperatures directly affect the performance and stability of the devices. Similarly, ash is an inorganic substance remaining after the combustion of materials, and its high temperature resistance has an important impact on the stable operation of boilers. At present, the widely used method for measuring the thermal transition properties of ash is the traditional ash melting test, which involves heating the ash sample to above 1500 to 1600°C in an oxidizing or reducing atmosphere and observing the morphological changes of the conical ash sample to determine the four melting characteristic temperatures of each sample: deformation temperature, softening temperature, hemispherical temperature and flow temperature. However, this traditional method has poor repeatability and it is difficult to accurately predict specific melting characteristic temperatures. Usually, a large amount of material thermal transition property data is required for thermal transition process design, which is difficult to collect through experiments.

[0003] Taking ash as an example, the traditional method is to determine the characteristic temperature of the ash sample by conducting an ash fusion test (Ash Fusion Test), including deformation temperature, softening temperature, hemispherical temperature and flow temperature. These methods rely on manual observation and subjective judgment, resulting in large errors, poor repeatability, and unable to determine the melting characteristics of the ash sample at other temperatures. The melting characteristic curve of the ash obtained by the image analysis method not only has good repeatability, but also can obtain the melting characteristics of the ash sample at other temperatures. However, when processing a large number of samples, the cumbersome operation steps will lead to a longer experimental cycle. Summary of the invention

[0004] The technical problem to be solved by the present invention is to shorten the time required for determining the thermal transition characteristics of materials and improve efficiency. In order to overcome the defects of the above-mentioned prior art (or related technology), the present invention provides a material thermal transition characteristics prediction method combining classification algorithm and machine learning.

[0005] The present invention provides a material thermal transition characteristic prediction method combining classification algorithm and machine learning, comprising:

[0006] Step S1, obtaining a test sample and performing an ashing process on the test sample to obtain ash residue;

[0007] Step S2, performing component detection on the ash to obtain component data;

[0008] Step S3, analyzing the component data using a classification algorithm to obtain the ash type corresponding to the ash;

[0009] Step S4: based on the ash type corresponding to the ash, select a corresponding machine learning algorithm to analyze the component data to obtain a corresponding thermal transition characteristic curve.

[0010] Compared with the prior art, the material thermal transition characteristics prediction method combining classification algorithm and machine learning in this application has the following advantages:

[0011] In the present application, the test sample is ash-treated in step S1; the ash is tested for composition in step S2; the composition data is systematically classified to obtain the ash type in step S3; and the thermal transition characteristic curve of the test sample is generated in step S4. This method can effectively utilize the composition data of the ash to indirectly predict the thermal transition characteristic curve, and combine easy-to-measure physical quantities, classification algorithms and machine learning techniques to achieve indirect determination of the thermal transition characteristic curve of the material, avoiding the cumbersome processing and analysis process required by traditional measurement methods, thereby shortening the time required for determining the thermal transition characteristics of the material and improving efficiency.

[0012] In a possible implementation, in the step S1, the test sample is subjected to an ashing treatment in an environment of 500-600°C.

[0013] In a possible embodiment, in step S2, low-pressure polyethylene is used as the substrate for the inlay boundary, and a press is used at a pressure of 40 to 50 MPa for 30 to 50 seconds to press the ash into a round piece, and then an X-ray fluorescence spectrometer is used to perform a chemical composition analysis on the round piece to obtain the ash type corresponding to the ash.

[0014] In a possible implementation manner, in the step S2, the ash is pressed into the circular tablets having a diameter of 32 to 34 mm.

[0015] In a possible implementation, in step S3, the ash types corresponding to the ash obtained through analysis include high temperature resistant ash, sintered ash, expanded ash and stable melting ash.

[0016] In a possible implementation, step S3 includes:

[0017] Step S31, inputting the component data into the pre-trained random forest model and support vector machine model respectively to obtain corresponding prediction results;

[0018] Step S32, weighting the two prediction results respectively according to the different categories of the ash types;

[0019] Step S33, combining the two weighted prediction results to determine the final prediction category corresponding to the ash as the ash type.

[0020] In a possible implementation, in step S33, weighted voting is performed on the two weighted prediction results, and the prediction result group with a larger weight is selected as the final prediction category.

[0021] In a possible implementation, step S4 includes:

[0022] Step S41, determining whether the component type corresponding to the ash is sintered ash:

[0023] If yes, go to step S42;

[0024] If not, go to step S43;

[0025] Step S42, calling a support vector regression algorithm to analyze the component data to obtain the corresponding thermal transition characteristic curve;

[0026] Step S43, calling the gradient boosting algorithm to analyze the component data to obtain the corresponding thermal transition characteristic curve. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 is a flow chart of the steps of the present invention;

[0028] Figure 2 is a specific flow chart of step S3 of the present invention;

[0029] Figure 3 is a specific flow chart of step S4 of the present invention;

[0030] Figure 4 It is a schematic diagram of the thermal transition characteristic curve prediction and RMSE evaluation results obtained by predicting and evaluating existing data in the present invention. DETAILED DESCRIPTION

[0031] First, those skilled in the art should understand that these implementations are only used to explain the technical principles of the embodiments of the present application, and are not intended to limit the protection scope of the embodiments of the present application. Those skilled in the art can make adjustments to them as needed to adapt to specific application scenarios.

[0032] The present application is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0033] See also Figure 1The embodiment of the present application discloses a method for predicting the thermal transition characteristics of materials by combining a classification algorithm and machine learning, comprising: a first step of ashing a test sample to prepare a test sample; a second step of detecting the composition of the obtained ash; a third step of systematically classifying the ash using a classification algorithm; a fourth step of selecting an appropriate machine learning algorithm for processing according to the category to which the ash belongs; and a fifth step of generating a thermal transition characteristic curve of the test sample through computational analysis. The prediction method can effectively utilize the composition data of the ash to indirectly predict the thermal transition characteristic curve, providing a new analysis tool for related fields.

[0034] Continue to see Figure 1 In the first step, when the test samples are ashed, a batch of fuel samples are first ashed in an environment with a temperature range of 500 to 600°C.

[0035] Continue to see Figure 1 In the second step, when the ash was tested for composition, low-pressure polyethylene was used as the substrate for the inlay boundary, and a press was used to press the ashed sample into a circular piece with a diameter of 32 to 34 mm at a pressure of 40 to 50 MPa for 30 to 50 seconds. Subsequently, the chemical composition of these ashed samples was analyzed using an X-ray fluorescence spectrometer.

[0036] Continue to see Figure 1 ,In the third step, when the ash is systematically classified by the classification algorithm, the categories of the test samples are calculated respectively by the random forest and support vector machine algorithms. By adopting the voting weight method, the ash type to which the test sample belongs is finally determined, including high temperature resistant ash, sintering ash, expansive ash and stable melting ash.

[0037] Continue to see Figure 1 In the fourth step, according to the category of the ash, an appropriate machine learning algorithm is selected for processing. After the category of the test sample is clarified, if the test sample is sintered ash, the support vector regression algorithm is called for further calculation; if the test sample is not sintered ash, the gradient boosting algorithm is used for calculation.

[0038] Continue to see Figure 1 The composition data obtained in the second step include silicate, calcium, magnesium, potassium, sodium, phosphorus and sulfur.

[0039] Continue to see Figure 1, Support Vector Machine (SVM) is an effective supervised learning algorithm, mainly used for classification and regression tasks. Their operation method is to identify the decision boundary (or hyperplane) with the largest margin between data points, and optimize the support vectors around the boundary to enhance the generalization ability of the model. SVM can also manage nonlinear data through kernel tricks, allowing them to perform admirably on complex data sets; Random Forest (RF) is an ensemble learning algorithm that improves prediction accuracy by building multiple decision trees and aggregating their predictions. It can effectively handle classification and regression tasks, shows strong resistance to outliers and overfitting, and is applicable to large-scale data sets. In addition, it provides an assessment of variable importance, making it a powerful tool for solving various machine learning challenges. Tools; Gradient Boosting Machine (GBM) is an ensemble learning algorithm that gradually adds weak prediction models, usually decision trees, optimizing the residual of the previous model at each step to reduce model error. GBM excels at handling the challenges of classification and regression, especially complex and nonlinear data sets. Through careful hyperparameter tuning, high-precision prediction performance can be achieved; Support Vector Regression (SVR) is a regression algorithm based on the principles of support vector machines, which aims to identify a hyperplane that minimizes the distance between data points while ensuring that as many data points as possible remain within a predetermined distance from the hyperplane (called the ε-loniny zone). By optimizing the error margin and regularization term, SVR provides an effective method for predicting continuous variables, especially when the data exhibits nonlinear characteristics.

[0040] See also Figure 2 Step S3 includes step S31, step S32 and step S33, wherein step S31 is: inputting the component data into the pre-trained random forest model and support vector machine model respectively to obtain the corresponding prediction results; step S32 is: weighting the two prediction results respectively according to different categories of ash types; step S33 is: combining the two weighted prediction results to determine the final prediction category corresponding to the ash as the ash type.

[0041] See also Figure 3 Step S4 includes step S41, step S42 and step S43, wherein step S41 is: determine whether the component type corresponding to the ash is sintered ash: if so, turn to step S42; if not, turn to step S43; step S42 is: call the support vector regression algorithm to analyze the component data to obtain the corresponding thermal transition characteristic curve; step S43 is: call the gradient boosting algorithm to analyze the component data to obtain the corresponding thermal transition characteristic curve.

[0042] Continue to see Figure 1 , the specific test sample category calculation process in this application is:

[0043] 1. Data preprocessing

[0044] First, load data from the specified Excel file, including composition training data, composition verification data, curve training data, and curve verification data, which are used to train and test the model respectively;

[0045] 2. Random Forest Classification Process

[0046] 2.1 Construct multiple decision trees: The random forest algorithm randomly extracts multiple subsets from the training data and constructs multiple decision trees based on each subset. The construction process of each decision tree includes selecting features and dividing nodes, with the aim of making the samples of the same category as large as possible.

[0047] 2.2 Training decision trees: Each decision tree is trained on its corresponding subset, and the nodes are divided by selecting the best features, so that the samples in each child node are purer;

[0048] 2.3 Decision tree independent prediction: Each trained decision tree independently predicts the test sample and outputs a category label;

[0049] 2.4 Majority voting mechanism: The prediction results of all decision trees are voted, and the category predicted by the majority of decision trees is used as the final prediction category;

[0050] 3. Support vector machine classification process

[0051] 3.1 Constructing the optimal hyperplane: The support vector machine algorithm will find an optimal hyperplane in the feature space, which can maximize the interval between samples of different categories;

[0052] 3.2 Training model: Through the optimization algorithm, the support vector machine finds the optimal hyperplane, which can effectively separate samples of different categories in the feature space. For nonlinearly separable data, the kernel function can be used to map the data to a high-dimensional space and find the optimal linear hyperplane in the high-dimensional space.

[0053] 3.3 Support vector: During the training process, only a small number of key samples determine the specific position and direction of the hyperplane. The support vector is the sample point closest to the decision boundary.

[0054] 3.4 Hyperplane prediction: The test sample is predicted by the trained hyperplane, and the position of the sample point on one side or the other side of the hyperplane determines its category;

[0055] 4. Prediction process combining random forest and support vector machine

[0056] 4.1 Independent prediction: The test samples are input into the trained random forest and support vector machine models respectively to obtain two independent prediction results;

[0057] 4.2 Weighted voting mechanism: A weight dictionary is defined in the code to weight the prediction results of random forest and support vector machine according to different categories. For example, for some categories, the prediction weight of random forest is larger; for other categories, the prediction weight of support vector machine is larger;

[0058] 4.3 Final prediction results: According to the weighted voting mechanism, the prediction results of random forest and support vector machine are combined to determine the final prediction category of each sample. Specifically, the code performs weighted voting on the prediction results of the two models and selects the prediction result of the model with a larger weight as the final prediction category;

[0059] Through this method, classification prediction combining random forest and support vector machine is achieved, making the prediction results more accurate and stable.

[0060] Continue to see Figure 1 , the specific process of determining the ash type by voting weight in this application is as follows:

[0061] 1. Calculate the prediction results of each model

[0062] The trained random forest and support vector machine are used to predict the test samples, and the prediction results of the two models are obtained:

[0063] `pred_rf`: prediction results of the random forest model;

[0064] `pred_svm`: prediction results of the support vector machine model;

[0065] 2. Get weight

[0066] Get the weights of various prediction results from the predefined weight dictionary, which is defined as follows: self.weights = {

[0067] 0:{'svm':2,'rf':1},

[0068] 1:{'svm':1,'rf':2},

[0069] 2:{'svm':2,'rf':1},

[0070] 3:{'svm':1,'rf':2}}

[0071] This means that for different categories, the weights of random forest and support vector machine are different. For example, the weight of category 0 is 1 for random forest and 2 for support vector machine.

[0072] 3. Weighted voting determines the final category

[0073] The prediction results of the two models are combined through a weighted voting mechanism. First, a list is initialized to store the final prediction results. Then, the prediction results of each test sample are traversed. For each test sample, its prediction results in random forest and support vector machine are obtained respectively, and weighted voting is performed according to the weight dictionary. Finally, the optimal solution is selected as the result.

[0074] Continue to see Figure 1 , the process of obtaining the thermal transition characteristic curve using the support vector regression (SVR) algorithm in this application is as follows:

[0075] 1. Data preprocessing

[0076] First, the data is loaded from an Excel file, which contains the following four tables: composition training data, composition verification data, curve training data, and curve verification data. For the thermal transition characteristic curve, the curve training data and curve verification data need to be processed for dimensionality reduction;

[0077] 2. Model training

[0078] In the `CurvePredictionSVR` class, initialize the SVR model, filter the data of a specific category, retain only the data of the current category, perform PCA dimensionality reduction on the filtered curve training data to reduce the feature dimension and improve the model training efficiency, and use the reduced-dimensional curve training data and the corresponding component training data to train the SVR model;

[0079] 3. Curve prediction

[0080] The model is verified using the validation data set to obtain the predicted data. Since the predicted data corresponds to the data after PCA dimensionality reduction, the predicted data is inversely transformed by PCA to restore it to the original feature space, that is, the predicted thermal transition characteristic curve data is obtained;

[0081] In order to complete the above process, the following data volumes are required: training data set, which contains multiple samples, each sample has multiple composition features, a category label, and thermal transition characteristic curves at different temperatures; validation data set: similar to the composition training data, but used for model validation;

[0082] 4. Curve meaning and drawing

[0083] The horizontal axis of the thermal transition characteristic curve represents the temperature range, extending from room temperature to 1300°C, and the vertical axis shows the change in the physical properties of the ash cone pressed from ash slag as the temperature changes, which is specifically manifested as the change in relative height. The data obtained through the model is presented in the form of Excel data. Each test sample contains a series of data points, which can be imported into drawing software for visual analysis.

[0084] Continue to see Figure 1 , the process of obtaining the thermal transition characteristic curve using the gradient boosting machine (GBM) algorithm in this application is as follows:

[0085] 1. Data preprocessing

[0086] First, the data is loaded from an Excel file, which contains the following four tables: composition training data, composition verification data, curve training data, and curve verification data. For the thermal transition characteristic curve, the curve data needs to be processed for dimensionality reduction;

[0087] 2. Model training

[0088] In the `CurvePredictionGBM` class, initialize the GBM model, filter the data of a specific category, retain only the data of the current category, perform PCA dimensionality reduction on the filtered curve training data to reduce the feature dimension and improve the model training efficiency, and use the reduced-dimensional curve training data and the corresponding component training data to train the GBM model;

[0089] 3. Curve prediction

[0090] The validation data set is used to validate the model and obtain the predicted data. Since the predicted data corresponds to the data after dimensionality reduction using PCA, the predicted data is inversely transformed by PCA and restored to the original feature space to obtain the predicted thermal transition characteristic curve.

[0091] Continue to see Figure 1 In this application, the RMSE of the model can also be checked to quantify the gap between the model prediction results and the actual values. A lower RMSE indicates that the model prediction results are more accurate. By calculating the RMSE of different models, the performance of different models can be compared, and the best model can be selected for further tuning. In the last step, the RMSE of each category is calculated and printed out, so that the prediction performance of the model in different categories can be comprehensively evaluated, where the predicted value represents the model prediction value of a certain ash slag, and the actual value represents its actual thermal transformation characteristic curve.

[0092] Continue to see Figure 1 , RMSE is calculated using the following formula:

[0093]

[0094] in,

[0095] n represents the sample size;

[0096] Y 预测 It represents the model prediction value of a certain ash;

[0097] Y 实际 Represents the actual thermal transition characteristic curve.

[0098] See also Figure 4 , predict and evaluate the existing data through the method of this application, Figure 4 The prediction of thermal transition characteristic curves of the four samples and their RMSE evaluation results show that the prediction accuracy is high, and the RMSE value is generally lower than 0.1, indicating that the method of this application can improve the prediction accuracy. Figure 4 The blue lines in the graph represent predicted values, and the black lines represent actual values.

[0099] In the description of the present application, the description with reference to the terms "one embodiment", "some embodiments", "in the present embodiment", "specific example", or "some examples" etc. means that the specific features, mechanisms, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, mechanisms, materials or characteristics described may be combined in a suitable manner in any one or more embodiments or examples. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0100] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be based on the protection scope of the claims.

Claims

1. A method for predicting thermal transition characteristics of materials combining classification algorithm and machine learning, characterized in that: The following steps are involved: Step S1, obtaining a test sample and performing an ashing process on the test sample to obtain ash residue; Step S2, performing component detection on the ash to obtain component data; Step S3, using a classification algorithm to analyze the component data to obtain the ash type corresponding to the ash; Step S4: based on the ash type corresponding to the ash, select a corresponding machine learning algorithm to analyze the component data to obtain a corresponding thermal transition characteristic curve.

2. The material thermal transition characteristics prediction method according to claim 1, characterized in that: In the step S1, the test sample is subjected to an ashing treatment in an environment of 500-600°C.

3. The material thermal transition characteristics prediction method according to claim 1, characterized in that: In the step S2, low-pressure polyethylene is used as the substrate for the inlay boundary, and a press is used at a pressure of 40 to 50 MPa for 30 to 50 seconds to press the ash into a round piece. Subsequently, an X-ray fluorescence spectrometer is used to perform a chemical composition analysis on the round piece to obtain the ash type corresponding to the ash.

4. The material thermal transition characteristic prediction method according to claim 3, characterized in that: In the step S2, the ash is pressed into the circular tablets with a diameter of 32 to 34 mm.

5. The material thermal transition characteristic prediction method according to claim 1, characterized in that: In the step S3, the ash types corresponding to the ash obtained by analysis include high temperature resistant ash, sintered ash, expanded ash and stable melting ash.

6. The material thermal transition characteristic prediction method according to claim 5, characterized in that: The step S3 comprises: Step S31, inputting the component data into the pre-trained random forest model and support vector machine model respectively to obtain corresponding prediction results; Step S32, weighting the two prediction results respectively according to the different categories of the ash types; Step S33, combining the two weighted prediction results to determine the final prediction category corresponding to the ash as the ash type.

7. The material thermal transition characteristic prediction method according to claim 6, characterized in that: In the step S33, weighted voting is performed on the two weighted prediction results, and the prediction result group with a larger weight is selected as the final prediction category.

8. The material thermal transition characteristic prediction method according to claim 5, characterized in that: The step S4 comprises: Step S41, determining whether the component type corresponding to the ash is sintered ash: If yes, go to step S42; If not, go to step S43; Step S42, calling a support vector regression algorithm to analyze the component data to obtain the corresponding thermal transition characteristic curve; Step S43, calling the gradient boosting algorithm to analyze the component data to obtain the corresponding thermal transition characteristic curve.

Citation Information

Patent Citations

  • Ocean buoy life prediction method based on multi-class machine learning method

    CN112288191A

  • Method for identifying combustion ash of different cigarettes based on Fourier infrared spectrum technology and chemometrics algorithm

    CN118067650A

  • Method for Predicting Benchmark Value of Unit Equipment Based on XGBoost Algorithm and System thereof

    US20230213895A1

Cited By

  • Medical glass staged temperature adjustment treatment method and device based on melting characteristics

    CN119841532A