Artificial intelligence assisted stone plastic floor material performance prediction method

Abstract

The present application belongs to the field of building material performance prediction, and particularly relates to a stone plastic floor material performance prediction method assisted by artificial intelligence. First, data including production filler proportion, thickness, etc. and performance index data are collected, and after processing, key features are screened by using Shapley value calculation introducing feature dependency and interaction, overcoming the limitations of traditional feature engineering. Then, modeling and training are carried out through improved XGBoost algorithm, which is optimized in loss function, incremental training, learning rate adjustment and early stopping mechanism, so that the model can efficiently process data and accurately predict. Through experimental comparison, the error between the predicted value and the true value is small, indicating that this method is effective and reliable, and can provide important reference for stone plastic floor production, quality control, etc., and promote the progress of industry technology and the improvement of product quality.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of building material performance prediction, and particularly relates to an artificial intelligence assisted stone plastic floor material performance prediction method. BACKGROUND

[0002] As a new type of floor decoration material, stone plastic floor is increasingly widely used in the building decoration market due to its environmental protection, wear resistance, moisture resistance, and aesthetic appearance. However, the performance of stone plastic floor is affected by various factors, and in actual production and application, accurate prediction of the performance of stone plastic floor under different environments is of great significance for product quality control, product development, and meeting market demand. Existing stone plastic floor performance testing mainly relies on physical experiments, which can directly obtain material performance data, but has many limitations. On the one hand, physical experiments are time-consuming, labor-intensive, and costly, requiring a large amount of manpower, material resources, and time investment. For example, to test the moisture resistance and stability of stone plastic floor under different humidity and temperature environments, long-term sample testing under different environmental conditions is required, and each experiment can only target specific environmental parameter combinations, making it difficult to cover various actual use scenarios. On the other hand, physical experiments have poor flexibility and are difficult to quickly respond to changes in product design and market demand. When the formula or production process of stone plastic floor needs to be adjusted, traditional experimental methods cannot provide performance prediction data in a timely manner, resulting in a prolonged research and development cycle and slow product iteration speed. SUMMARY

[0003] The present application proposes an artificial intelligence assisted stone plastic floor material performance prediction method to address the technical problems in the background art.

[0004] To achieve the above purpose, the technical solution adopted by the present application is as follows:

[0005] S1, first collect performance data of stone plastic floor material, including production filler ratio, thickness, processing temperature, environmental humidity, and performance indicators of stone plastic floor obtained by experiment;

[0006] S2, calculate the Shapley value by introducing feature dependency and interaction, and select the most influential features for performance prediction;

[0007] S3, model and train the selected features using an improved XGBoost algorithm, and the model adjusts and optimizes itself to continuously improve the accuracy and efficiency of prediction;

[0008] The implementation of the improved XGBoost algorithm in step S3 is as follows:

[0009] S31, first, through the error back propagation in the training process, combined with the contribution of the characteristics, adjust the construction of the loss function: Where y m , are the actual value and the predicted value respectively, M is the total number of samples, ω k is the kth model parameter, K is the number of parameters, λ is the weight coefficient of the regularization term, γ is the weight coefficient of the gradient penalty term, is the gradient of the prediction result to the feature;

[0010] S32, then use the incremental training method, only update the part of the model with large error in each training, set the incremental training threshold, only fine-tune the area where the model performs poorly on the sample, the specific update formula is: Where is the predicted value after incremental training, is the predicted value obtained by previous training;

[0011] S33, then combine the training error and gradient information to adaptively adjust the learning rate: Where η t is the learning rate of the tth round, η0 is the initial learning rate, L(t), L(t-1) are the losses of the tth and t-1th rounds of training respectively, β is the hyperparameter that controls the learning rate adjustment speed;

[0012] S34, finally, introduce the adaptive early stopping mechanism, monitor the performance of the model on the validation set, and automatically stop training when the performance does not improve significantly after multiple rounds of training, the early stopping strategy is as follows: Where δ is the threshold for stopping training, if the loss of the model after multiple rounds of training changes less than the threshold, it means that the model has reached the optimal state, then terminate the training.

[0013] As preferred, the performance indicators of the stone plastic floor in step S1 include wear resistance, moisture resistance and hardness.

[0014] As preferred, the implementation steps of the data introduction feature dependence and interaction Shapley value calculation in step S2 are:

[0015] S21, first, through the weighted Shapley value calculation, consider the interaction effect between features, adjust the Shapley value calculation to the weighted marginal contribution, the weighting formula is: Where, φ i (v) is the Shapley value of feature i, subset S is a combination of all features in feature set N after removing feature i, subset It is the feature combination without feature i, v(S) represents the model prediction value corresponding to subset S, is the measure of the contribution of each feature in the Shapley value, v(S∪{i}) is the output value of the model after adding feature i to subset S, and ω(S,i) is the correlation weight between feature i and subset S.

[0016] S22. In addition to the contribution of individual features, consider the contribution of feature interactions to the overall performance. Use the Shapley interaction index to quantify the interactions between features and optimize the feature selection process. The interaction formula is: Among them I ij Shapley values ​​represent the interaction between features i and j, M is the number of Monte Carlo samplings, and Δ ij (S m ) are features i and j in subset S m Marginal contribution when the two forces work together, Δ i (S m ), Δ j (S m ) are features in subset S m The marginal contribution;

[0017] S23. Finally, by calculating the dependency Shapley value, features that have an impact on the target variable greater than the set threshold of 0.4 and have a strong correlation with other features greater than the threshold of 0.3 are selected. Based on the degree of mutual influence between features, a set of features that can provide important information while avoiding redundancy is selected.

[0018] Compared with existing technologies, the advantages and positive effects of this invention lie in its use of Shapley value calculation to screen key features by introducing feature dependencies and interactions, thus overcoming the limitations of traditional feature engineering. Furthermore, by improving the XGBoost algorithm for modeling and training, with optimizations in loss function, incremental training, learning rate adjustment, and early stopping mechanism, the model can efficiently process data and make accurate predictions, thereby promoting technological progress and quality improvement in the SPC flooring industry. Detailed Implementation

[0019] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described below with reference to embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0020] Numerous specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways than those described herein, and therefore the invention is not limited to the specific embodiments disclosed in the following specification.

[0021] In the embodiment, the performance test of the existing stone plastic floor depends on physical experiments, which can directly obtain data but has great limitations. The physical experiments are time-consuming, labor-intensive and high-cost. For example, to test the moisture resistance stability in different environments, the samples need to be tested in a specific environment for a long time, which is difficult to cover all scenarios. Moreover, the flexibility is poor. When the product design or process is adjusted, the prediction data cannot be provided in time, which leads to a long research and development cycle and slow iteration, affecting the development of the industry. In order to accurately predict the performance of the stone plastic floor in different environments and solve the problem of time-consuming and poor flexibility of the traditional physical experiment method, an artificial intelligence assisted stone plastic floor material performance prediction method is proposed, which includes data collection, feature selection and improved XGBoost algorithm modeling and training, and has the advantages of high efficiency, accuracy and adaptability to various demand changes.

[0022] Firstly, the data related to the stone plastic floor material is comprehensively collected, including the production filler ratio, thickness, processing temperature, environmental humidity and other production and use environment factors, and the performance indexes of the stone plastic floor determined by experiments, including wear resistance, moisture resistance and hardness. The data is derived from actual production records, environmental monitoring equipment and standard experiment tests. The collected data is arranged and preprocessed to ensure the accuracy and integrity of the data, and abnormal values and error data are removed.

[0023] Considering that the existing feature engineering method often analyzes features separately and ignores the interaction, it is difficult to handle the complex relationship of the stone plastic floor with multiple factors, and it is easy to select redundant features to cause model overfitting. The hapley value calculation of the present invention comprehensively considers the dependence and interaction between features, quantifies the synergistic effect through a weighted formula and the number of interactions, and selects key and strongly related features based on this. Firstly, the weighted Shapley value calculation considers the interaction effect between features. The calculation of the Shapley value is adjusted to the marginal contribution with weight, and the weighted formula is: wherein φ i (v) is the Shapley value of feature i, the subset S is a combination of all features after feature i is removed from the feature set N, and the subset is the feature combination without feature i, v(S) represents the model prediction value corresponding to the subset S, which is a measure of the contribution of each feature in the Shapley value, v(S∪{i}) is the output value of the model after feature i is added to the subset S, and ω(S,i) is the correlation weight between feature i and the subset S; In addition to the contribution of a single feature, the contribution of the interaction between features to the overall performance is considered, the interaction between features is quantified by the Shapley interaction index, and the feature selection process is optimized. The interaction formula is: wherein I ij is the Shapley value of the interaction between features i and j, M is the number of Monte Carlo samples, Δ ij (S m ) is the Shapley value of the interaction between feature i and feature j in the subset S mMarginal contribution when the two forces work together, Δ i (S m ), Δ j (S m ) are features in subset S m The marginal contribution of the feature is calculated; finally, by calculating the dependency Shapley value, features that have an impact on the target variable greater than a set threshold of 0.4 and a correlation strength with other features greater than a threshold of 0.3 are selected. Based on the degree of mutual influence between features, a set of features that can provide important information while avoiding redundancy is selected. This method can deeply mine effective information when dealing with complex data, build a more accurate model for predicting the performance of SPC flooring, effectively overcome the limitations of traditional methods, and promote technological progress and product quality improvement in the SPC flooring industry.

[0024] Next, considering the lack of flexibility in setting the loss function in existing technologies, low incremental training efficiency, unintelligent learning rate adjustment, and the frequent absence of an effective early stopping mechanism, this invention's improved algorithm precisely adjusts the loss function through error backpropagation, better adapting to data complexity and reducing prediction errors; incremental training provides targeted updates, saving resources and enabling rapid optimization; adaptive learning rate adjustment balances training speed and accuracy; and an adaptive early stopping mechanism prevents overfitting and provides a reliable model in a timely manner. Firstly, through error backpropagation during training, combined with the contribution of features, the construction of the loss function is adjusted: Where y m , These represent the actual value and the predicted value, respectively, where M is the total number of samples, and ω is the actual value and the predicted value. k Let λ be the k-th model parameter, K be the number of parameters, λ be the weight coefficient of the regularization term, and γ be the weight coefficient of the gradient penalty term. To predict the gradient of the result with respect to the features, an incremental training method is then used. In each training iteration, only the parts of the model with large errors are updated. An incremental training threshold is set, and only the regions where the model performs poorly on the samples are fine-tuned. The specific update formula is as follows: in These are the predicted values ​​after incremental training. It is the predicted value obtained from previous training; then, combining the training error and gradient information, the learning rate is adaptively adjusted: Where η t Let η be the learning rate in round t, η0 be the initial learning rate, L(t) and L(t-1) be the losses in rounds t and t-1, respectively, and β be a hyperparameter controlling the speed of learning rate adjustment with a value of 0.15. Finally, an adaptive early stopping mechanism is introduced. By monitoring the model's performance on the validation set, training is automatically stopped when the performance does not significantly improve after multiple rounds of training. The early stopping strategy is as follows: Wherein δ is the threshold value of stopping training is 0.5%, if the loss change of the model after multiple rounds of training is less than the threshold value, it means that the model has reached the optimal state, then terminate the training. This technical means in stone plastic floor performance prediction, can more efficient processing data, fast response changes, provide strong support for product research and development and production, promote the progress and development of industry technology.

[0025] Finally, the experimental data and the predicted experimental data are compared, as shown in Table 1. In Table 1, the comparison of stone plastic floor performance data is shown, including two groups of experimental data, respectively, the first group and the second group, each group covers hardness, wear resistance, moisture resistance three performance indicators, and each indicator has true value and predicted value comparison. In general, the error between the predicted value and the true value of each performance indicator is relatively small, indicating that the prediction method used has certain effectiveness and reliability in stone plastic floor performance prediction. These data provide important reference for stone plastic floor production, quality control and performance evaluation, which helps to understand product performance, optimize production process and product quality.

[0026] Table 1: Comparison of stone plastic floor performance data

[0027]

[0028] The above is only the preferred embodiment of the present application, not other forms of the present application, any skilled in the art may use the above disclosed technical content to change or modify the equivalent embodiment applied to other fields, but any simple modification, equivalent change and modification of the above embodiment without departing from the technical scheme of the present application, according to the technical essence of the present application, still belongs to the protection scope of the present application.

Claims

1. An artificial intelligence-assisted method for predicting the performance of stone-plastic flooring materials, characterized in that, Includes the following steps: S1. First, collect performance data of stone plastic flooring materials, including production filler ratio, thickness, processing temperature, ambient humidity, and performance indicators of stone plastic flooring obtained from experiments. S2. The Shapley value is calculated for the data, which introduces feature dependencies and interactions, and the feature with the greatest influence on performance prediction is selected. S3. The improved XGBoost algorithm is used to model and train the selected features. The model then self-adjusts and optimizes to continuously improve the accuracy and efficiency of prediction. The implementation of the improved XGBoost algorithm in step S3 consists of the following steps: S31. First, by backpropagating the error during the training process and considering the contribution of the features, the construction of the loss function is adjusted: Where y m , These represent the actual value and the predicted value, respectively, where M is the total number of samples, and ω is the actual value and the predicted value. k Let λ be the k-th model parameter, K be the number of parameters, λ be the weight coefficient of the regularization term, and γ be the weight coefficient of the gradient penalty term. The gradient of the prediction result with respect to the features; S32. Next, an incremental training method is used. During each training iteration, only the parts of the model with large errors are updated. An incremental training threshold is set, and only the regions where the model performs poorly on the samples are fine-tuned. The specific update formula is as follows: in These are the predicted values ​​after incremental training. These are the predicted values ​​obtained from previous training; S33. Then, combining the training error and gradient information, the learning rate is adaptively adjusted: Where η t Let η be the learning rate in round t, η0 be the initial learning rate, L(t) and L(t-1) be the losses in round t and round t-1 respectively, and β be the hyperparameter controlling the speed of learning rate adjustment. S34. Finally, an adaptive early stopping mechanism is introduced. By monitoring the model's performance on the validation set, training is automatically stopped when the performance does not improve significantly after multiple rounds of training. The early stopping strategy is as follows: Where δ is the threshold for stopping training. If the loss of the model after multiple rounds of training is less than this threshold, it means that the model has reached the optimal state, and then training is terminated.

2. The artificial intelligence-assisted method for predicting the performance of stone-plastic flooring materials according to claim 1, characterized in that, The performance indicators of the stone-plastic flooring in step S1 include three performance indicators: wear resistance, moisture resistance, and hardness.

3. The artificial intelligence-assisted method for predicting the performance of stone-plastic flooring materials according to claim 1, characterized in that, The implementation steps for calculating the Shapley value of the data's feature dependencies and interactions in step S2 are as follows: S21. First, the weighted Shapley value is calculated. Considering the interaction effect between features, the calculation of the Shapley value is adjusted to a weighted marginal contribution. The weighting formula is as follows: Where, φ i (v) represents the Shapley value of feature i, and subset S is a combination of all features in feature set N after removing feature i. It is the feature combination without feature i, v(S) represents the model prediction value corresponding to subset S, is the measure of the contribution of each feature in the Shapley value, v(S∪{i}) is the output value of the model after adding feature i to subset S, and ω(S,i) is the correlation weight between feature i and subset S. S22. In addition to the contribution of individual features, consider the contribution of feature interactions to the overall performance. Use the Shapley interaction index to quantify the interactions between features and optimize the feature selection process. The interaction formula is: Among them I ij Shapley values ​​represent the interaction between features i and j, M is the number of Monte Carlo samplings, and Δ ij (S m ) are features i and j in subset S m Marginal contribution when the two forces work together, Δ i (S m ), Δ j (S m ) are features in subset S m The marginal contribution; S23. Finally, by calculating the dependency Shapley value, features that have an impact on the target variable greater than the set threshold of 0.4 and have a strong correlation with other features greater than the threshold of 0.3 are selected. Based on the degree of mutual influence between features, a set of features that can provide important information while avoiding redundancy is selected.

Images