Method for predicting heat insulation performance of multi-layer heat insulation material based on random forest model

Abstract

The invention discloses a method for predicting the heat insulation performance of a multi-layer heat insulation material based on a random forest model, and aims to solve the problem that an accurate model among material structure parameters, working condition parameters and heat insulation performance is difficult to establish in a traditional method. The method comprises the following steps: firstly, obtaining structure parameters, working condition parameters and corresponding heat insulation performance data of a variable-density multi-layer heat insulation material sample to form a data set; aiming at the characteristics of the variable-density multi-layer heat-insulating material, constructing enhanced characteristics through characteristic engineering; and inputting the feature vector as input and the heat insulation performance as a prediction target into a random forest model for training, optimizing key parameters of the model in training, verifying the performance of the trained model by using an evaluation index, and predicting the heat insulation performance of the new material by using the model according to structural parameters and working condition parameters of the new material. According to the method, a structure, working condition and performance relation model can be efficiently and accurately established by utilizing the high-dimensional and nonlinear processing capability of the random forest, and a brand-new material design and performance evaluation tool is provided.

Description

Technical Field

[0001] This invention belongs to the field of materials science and technology, especially the application of machine learning in materials science, and specifically relates to a method for predicting the thermal insulation performance of variable density multilayer thermal insulation materials using a random forest model. Background Technology

[0002] Variable-density multilayer insulation materials are high-performance insulation materials constructed by stacking insulation layers of varying densities. Their insulation performance is strongly nonlinearly coupled with numerous structural parameters, including the total number of layers, the density, thickness, and material of each layer, as well as the configuration of the reflector. Traditional performance evaluation methods primarily rely on experimental measurements, which are time-consuming and labor-intensive. While numerical simulations based on physical principles can provide a reference, they are complex to model, computationally expensive, and challenged in accurately describing the coupling of multiple physical fields.

[0003] Because such complex relationships are difficult to describe with simple mathematical formulas, and especially because traditional analytical methods cannot effectively handle the complex interactions inherent in variable-density multilayer insulation materials, it is difficult to establish accurate predictive models. Therefore, there is an urgent need for a new method that can directly learn from data and establish high-precision mapping relationships to quickly and accurately predict material properties. Summary of the Invention

[0004] The purpose of this invention is to provide a method for predicting the thermal insulation performance of multilayer thermal insulation materials based on a random forest model. This method can effectively capture the complex nonlinear relationships between material structural parameters, operating parameters, and thermal insulation performance, achieving rapid and accurate performance prediction.

[0005] The technical solution of this invention utilizes the random forest model, a data-driven tool, to fit the complex mapping from structural parameters and operating parameters to adiabatic performance parameters. This method includes the following steps:

[0006] S1. Data Preparation: Obtain the structural parameters, operating parameters, and corresponding insulation performance parameters of multiple known variable-density multilayer thermal insulation material samples to form a dataset; the structural parameters include the total number of reflective layers, the total number of spacer layers, reflective layer materials, spacer layer materials, insulation material thickness, density near the cold boundary layer, density of the intermediate layer, and density near the hot boundary layer; the operating parameters include the system vacuum degree, cold boundary temperature, and hot boundary temperature; the insulation performance parameters are heat flux density values.

[0007] S2. Preprocess the dataset, including missing value and outlier detection, and encode the reflective layer material and spacer layer material in the dataset. Divide the dataset into training set and test set proportionally, with the training set accounting for 70-80% and the remainder for the test set.

[0008] S3. Feature and Label Definition: The structural parameters and operating parameters of the preprocessed dataset are defined as the original features, and enhanced features are constructed through feature engineering to target the characteristics of variable density multilayer insulation materials; the enhanced features include basic features and advanced features; the insulation performance parameters are defined as the prediction labels.

[0009] The basic features include average layer density, temperature difference, and layer ratio; the advanced features include temperature gradient, layer density variation coefficient, polynomial feature, ratio feature, and statistical feature.

[0010] The average layer density is calculated by averaging the densities near the cold boundary layer, the intermediate layer, and the hot boundary layer; the temperature difference is the difference between the hot boundary temperature and the cold boundary temperature; the layer ratio is the ratio of the total number of reflective layers to the total number of layers; the temperature gradient is calculated by the ratio of the temperature difference to the thickness of the insulation material; the layer density variation coefficient is calculated by the ratio of the standard deviation of the layer density to the average layer density; the polynomial characteristic includes the square of the temperature difference or the square of the thickness; the ratio characteristic is the screen spacing ratio; and the statistical characteristic includes the maximum, minimum, or range of the layer density.

[0011] The original features and the enhanced features together constitute the feature vector.

[0012] S4. Random Forest Model Construction and Training: Using a training set, with feature vectors as input and predicted labels as output, a random forest model is trained, and key parameters are optimized using hyperparameter optimization methods. The random forest model is an ensemble learning algorithm that autonomously learns decision rules from data, constructs multiple decision trees, and combines them to form a random forest. By constructing multiple decision trees and integrating their predictions, the final prediction is obtained. This mechanism makes it naturally resistant to overfitting and capable of handling high-dimensional features. Key parameters mainly include:

[0013] The number of decision trees in the forest, n_estimators, is used to determine the model's stability, but also to increase computational costs. The maximum depth of the trees, max_depth, controls the complexity of the trees and prevents overfitting. The minimum number of samples required to split internal nodes, min_samples_split, is also used.

[0014] The optimization method is as follows: cross-validation combined with a search algorithm is used to optimize the key hyperparameters of the random forest model. The search algorithm can be grid search, random search, or Bayesian optimization.

[0015] S5. Model Evaluation: Validate the trained random forest model on the reserved test set using evaluation metrics, including multiple metrics such as mean absolute error (MAE), relative mean absolute error (MAE%), root mean square error (RMSE), relative root mean square error (RMSE%), and coefficient of determination (R²), to ensure that the model has satisfactory prediction accuracy and generalization ability.

[0016] S6. Using the validated random forest model, predict the thermal insulation performance of the material under test based on its structural and operating parameters.

[0017] The beneficial effects of this invention are as follows:

[0018] This invention leverages the advantages of random forest models in handling high-dimensional features and complex interactions, enabling in-depth analysis of the intrinsic relationships between structural parameters, operating parameters, and adiabatic performance, resulting in accurate and reliable predictions. It transforms the performance prediction process into a data-driven, real-time prediction; once the model is trained, the prediction process can be completed rapidly, far exceeding traditional experiments. The ensemble learning mechanism of random forests gives it natural tolerance to data noise and incomplete features, resulting in good model stability and practicality in engineering applications. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the overall process of the method of the present invention;

[0020] Figure 2 This is a schematic diagram illustrating the principle of the Random Forest algorithm.

[0021] Figure 3 A scatter plot comparing the model's predicted values ​​with the actual values;

[0022] Figure 4 This is a comparison chart of the actual and predicted values ​​by sample index. Detailed Implementation

[0023] To better understand the above-described objects, features, and advantages of the present invention, the present invention will be further described below in conjunction with the accompanying drawings and embodiments. Many specific details are set forth in the following description to provide a thorough understanding of the present invention; however, the present invention may be practiced in other ways than those described herein, and therefore, the present invention is not limited to the specific embodiments disclosed below.

[0024] like Figure 1 As shown, this embodiment proposes a method for predicting the thermal insulation performance of variable-density multilayer thermal insulation materials based on a random forest model. This method rationally utilizes the structural and operating parameters of multilayer thermal insulation materials for predicting their thermal insulation performance, and includes the following steps:

[0025] 1. Data preparation and collection:

[0026] The experimental dataset used in this embodiment contains several key parameters of variable-density multilayer insulation materials, including:

[0027] Structural parameters: total number of reflective layers, total number of spacer layers, density of each layer (density near the cold boundary layer, density of the intermediate layer, density near the hot boundary layer), total material thickness, material type of reflective layer, material type of spacer layer;

[0028] Operating parameters: vacuum pressure, cold boundary temperature, hot boundary temperature;

[0029] Target variable: heat flux density (W / m²).

[0030] The experimental data were obtained through standard thermal performance testing to ensure the accuracy and reliability of the data.

[0031] 2. Data preprocessing, including the following preprocessing steps:

[0032] 2.1 Data Quality Check: Check data integrity, identify and count missing values, detect outliers, use IQR (interquartile range) method to identify outlier data, analyze data distribution characteristics, and ensure that the data meets the modeling requirements.

[0033] 2.2 Classification variable coding: Classification variables such as reflective layer material and spacer layer material are labeled and coded, and a coding mapping dictionary is established to ensure coding consistency.

[0034] 2.3 Dataset partitioning: Using a random sampling method, the complete dataset is divided into a training set and a test set in an 80:20 ratio. The training set accounts for 80% and is used for model training and parameter optimization, while the test set accounts for 20% and is used for final model performance evaluation.

[0035] 3. Feature Engineering Method: Based on the structural parameters and operating parameters of the original experimental data as features, enhanced features are constructed through the following methods:

[0036] (1) Construction of basic features:

[0037] Calculate the average layer density: ;

[0038] Calculate the temperature difference: ;

[0039] Calculate the layer ratio: .

[0040] (2) Advanced feature construction:

[0041] Temperature gradient characteristics: ; ;

[0042] Polynomial characteristics: square of temperature difference, square of thickness; ratio characteristics: ;

[0043] Statistical characteristics: maximum, minimum, and range of layer density.

[0044] 4. Feature selection:

[0045] A dual feature selection strategy is employed. First, feature selection is based on correlation. The absolute correlation coefficient between each feature and the heat flux density value is calculated, and the top k most relevant features are selected according to their correlation strength. Second, feature selection is based on statistical tests. SelectKBest combined with F-regression test is used to select the features with the highest statistical significance to the target variable. The intersection of the two methods is taken as the final feature set. When the number of intersecting features is small, the correlation-based selection result is prioritized.

[0046] 5. Random Forest Model Construction and Optimization:

[0047] like Figure 2 The random forest algorithm is based on decision trees. It randomly selects N new data subsets with replacement from the training set to generate N decision trees, forming a random forest. The final prediction result is the average of the prediction results of all decision trees. In this embodiment, the random forest model is constructed and optimized based on the above principle. The specific process is as follows:

[0048] 5.1 Standardization:

[0049] Feature standardization: The Z-score standardization method is used to standardize numerical features. The standardization method is as follows: ;in, It is the mean of the features in the dataset. It is the standard deviation of the dataset features. These are the feature values ​​of the dataset. These are the standardized values ​​of the dataset features.

[0050] 5.2 Hyperparameter Optimization Space Design:

[0051] Establish a six-dimensional hyperparameter search space, which includes:

[0052] Number of decision trees n_estimators: random integer distribution randint(50, 300);

[0053] Maximum depth max_depth: a set of discrete values ​​[None, 10, 20, 30, 50];

[0054] Minimum number of split samples min_samples_split: random integer distribution randint(2, 11);

[0055] Minimum number of leaf node samples min_samples_leaf: random integer distribution randint(1, 5);

[0056] Maximum number of features max_features: Policy set [auto, sqrt, log2, 0.8, 0.9], where log2 represents the number of features considered in each split as the logarithm of the total number of features to the base 2. For example, with 100 features, log2(100) ≈ 6.64, rounded down to 6;

[0057] Bootstrap: A set of boolean values ​​[True, False].

[0058] 5.3 Execution of the random search optimization algorithm:

[0059] A stochastic search optimization algorithm is used to search the hyperparameter optimization space. For each hyperparameter combination, k-fold cross-validation is used to calculate the average cross-validation score, thereby selecting the combination with the optimal cross-validation score.

[0060] 5.4 Optimal Model Training and Ensemble:

[0061] The final model is trained on the complete training set using the optimal parameters, and the optimal parameter combination is saved for further training. The random forest makes predictions by averaging the prediction results of multiple trees.

[0062] 6. Model Evaluation and Validation:

[0063] The model uses mean absolute error (MAE), relative mean absolute error (MAE%), and coefficient of determination (R²) to evaluate the prediction results: , , ;in The total number of samples, Let i be the true value of the i-th sample. Let be the predicted value for the i-th sample. This is the average of the true values ​​in the sample.

[0064] Mean absolute error and relative mean absolute error measure the difference between predicted and actual values; the closer their values ​​are to 0, the better the model's predictive ability. Correlation coefficient. Used to test how well the regression model fits the true values, with values ​​between 0 and 1. The closer the value is to 1, the better the fit. In this embodiment, the mean absolute error (MAE) of the random forest prediction model is 0.1197, the relative mean absolute error (MAE%) is 7.92%, and the correlation coefficient is... =0.9836, indicating that the model's prediction results are good, and the random forest prediction model can effectively predict the true value.

[0065] Figure 3 This is a scatter plot comparing the model's predicted values ​​with the actual values. The plot visually demonstrates the performance evaluation results of the prediction model built based on the random forest model through the scatter distribution and the fitted line. The horizontal axis represents the actual values ​​of the adiabatic performance parameters, and the vertical axis represents the corresponding predicted values ​​of the model. The data points are all distributed near the ideal fitted line indicated by the red dashed line. This distribution pattern indicates a high linear correlation between the predicted and actual values. Specifically, the concentrated distribution of the test set data points along the fitted line proves that the constructed random forest prediction model has excellent generalization ability and prediction accuracy, and can accurately and stably predict the adiabatic performance of unknown samples. The training set data points also closely follow this trend line, reflecting that the model has learned sufficiently from the training data and that no significant overfitting has occurred.

[0066] Figure 4 This chart compares the true and predicted values ​​by sample index. With the sample index on the horizontal axis, it visually compares the true values ​​of the thermal insulation performance parameters with the predicted values ​​from the random forest model. The predicted values ​​almost perfectly match the true values, demonstrating the model's powerful learning ability. This visualization clearly shows that the model did not overfit the training data, and that the learned patterns are universally applicable, enabling stable and reliable application to predicting the thermal insulation performance of new samples.

[0067] Instance prediction results:

[0068] The trained model can be used to predict the heat leakage performance of newly designed variable-density multilayer insulation materials. Input parameters include:

[0069] Total number of reflective screen layers: 42;

[0070] Total number of interlayers: 126;

[0071] Density near the cold boundary layer: 6.35 layers / cm;

[0072] Intermediate layer density: 12.7 layers / cm;

[0073] Density near the thermal boundary layer: 19.05 layers / cm;

[0074] Vacuum pressure: 0.001 Pa;

[0075] Reflective layer material: Double-layer aluminum-coated thin film DAM;

[0076] Spacer layer material: Non-woven fiber cloth;

[0077] Thickness: 33 mm;

[0078] Cold boundary temperature: 77K;

[0079] Thermal boundary temperature: 323K.

[0080] The output is the predicted heat flux density value of 0.6345 W / m².

[0081] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for predicting the thermal insulation performance of multilayer thermal insulation materials based on a random forest model, characterized in that, Specifically as follows: S1. Obtain the structural parameters, operating parameters, and corresponding insulation performance parameters of multiple known variable-density multilayer insulation material samples to form a dataset; the structural parameters include the total number of reflective layers, the total number of spacer layers, the reflective layer material, the spacer layer material, the insulation material thickness, the density near the cold boundary layer, the density of the intermediate layer, and the density near the hot boundary layer; the operating parameters include the system vacuum degree, the cold boundary temperature, and the hot boundary temperature; the insulation performance parameters are heat flux density values. S2. Preprocess the dataset, including missing value and outlier detection, encode the reflective layer material and spacer layer material in the dataset, and divide the dataset into training set and test set according to the proportion, with the training set accounting for 70-80% and the rest being the test set; S3. Feature and Label Definition: The structural parameters and operating parameters of the preprocessed dataset are defined as original features, and enhanced features are constructed through feature engineering to target the characteristics of variable density multilayer insulation materials; the enhanced features include basic features and advanced features; the insulation performance parameters are defined as prediction labels; the original features and enhanced features together constitute the feature vector; The basic features include average layer density, temperature difference, and layer ratio; the advanced features include temperature gradient, layer density variation coefficient, polynomial feature, ratio feature, and statistical feature. S4. Random Forest Model Construction and Training: Using the training set, with feature vectors as input and predicted labels as output, train the random forest model, and use hyperparameter optimization methods to optimize key parameters; key parameters mainly include: the number of decision trees in the forest, the maximum depth of the trees, and the minimum number of samples required for internal node repartitioning; S5. Model Evaluation: Validate the trained random forest model on the reserved test set using evaluation metrics, including multiple metrics such as mean absolute error, relative mean absolute error, root mean square error, relative root mean square error, and coefficient of determination. S6. Using the validated random forest model, predict the thermal insulation performance of the material under test based on its structural and operating parameters.

2. The method for predicting the thermal insulation performance of multilayer thermal insulation materials based on a random forest model as described in claim 1, characterized in that: Among the basic features, the average layer density is calculated by averaging the densities near the cold boundary layer, the intermediate layer, and the hot boundary layer; the temperature difference is the difference between the hot boundary temperature and the cold boundary temperature; and the layer ratio is the ratio of the total number of reflective layers to the total number of layers. Among the advanced features, the temperature gradient is calculated by the ratio of the temperature difference to the thickness of the insulation material; the layer density variation coefficient is calculated by the ratio of the standard deviation of the layer density to the average layer density; the polynomial feature includes the square of the temperature difference or the square of the thickness; the ratio feature is the screen spacing ratio; and the statistical feature includes the maximum, minimum, or range of the layer density.

3. The method for predicting the thermal insulation performance of multilayer thermal insulation materials based on a random forest model as described in claim 1, characterized in that, The optimization method in step S4 is as follows: cross-validation combined with a search algorithm is used to optimize the key hyperparameters of the random forest model. The search algorithm can be grid search, random search, or Bayesian optimization.

Images