A home textile sleep-aiding power evaluation system based on characteristic function indexes and a construction method thereof

Abstract

This invention relates to the field of textile performance testing and intelligent evaluation technology, specifically providing a home textile sleep-aiding performance evaluation system and construction method based on feature function indicators. The system includes: acquiring five-dimensional feature indicators (tactile, thermal humidity, pressure, interference, and hygiene) through standardized instrument testing, and automatically assigning weights using range normalization and entropy weighting; constructing a physically constrained Bayesian neural network, embedding prior knowledge of materials science and sleep physiology as regularization terms into the loss function, and outputting a sleep-aiding performance score and confidence interval; establishing an adaptive weighted graph convolutional network to achieve knowledge transfer and zero-sample prediction between different home textile products; and fitting the relationship between static indicators and environmental parameters through a dynamic environment adaptive mapping module to output a scenario-based sleep-aiding performance level. This invention integrates instrumental quantitative testing with multi-level neural networks, breaking away from traditional linear regression dependence and achieving rapid, objective, personalized, and scenario-adaptive sleep-aiding performance evaluation.

Description

Technical Field

[0001] This invention relates to the field of textile performance testing and intelligent evaluation technology, and in particular to a home textile sleep-aiding evaluation system and its construction method based on characteristic functional indicators. Background Technology

[0002] As a direct carrier of the sleep environment, home textiles, with their tactile, thermal, pressure, disturbance control, and hygiene characteristics, significantly influence users' sleep onset speed, deep sleep duration, and overall sleep quality. With consumers increasingly valuing sleep health, how to scientifically and objectively evaluate the sleep-aiding properties of home textiles has become a research hotspot in the fields of textile testing and sleep medicine.

[0003] The evaluation of the sleep-aiding performance of home textiles mainly relies on two methods: first, subjective sleep trial questionnaires or scales, where users rate the product's comfort, ease of falling asleep, and post-sleep energy recovery based on their own experiences; second, objective physiological indicators such as sleep latency, sleep stages, and number of awakenings during the sleep trial process are collected using physiological instruments like polysomnography. Some testing institutions also attempt to establish simple linear correlations between instrument test results of a single characteristic of textiles and their sleep-aiding effects to infer the product's sleep-aiding ability.

[0004] The existing technologies mentioned above have significant shortcomings. Subjective evaluations are greatly affected by individual differences, psychological states, and environmental factors, resulting in poor repeatability and long evaluation periods. Simple physiological monitoring is costly and difficult to promote on a large scale. Methods that use linear regression to fit single or a few instrument indicators with sleep-aiding effects ignore the complex physical coupling relationships between these indicators and cannot adapt to the differences in sleep-aiding mechanisms among different product categories or the dynamic changes in the usage environment. Furthermore, they are insufficient for effectively predicting the effectiveness of new products that have not undergone sleep training, leading to insufficient accuracy and universality of the evaluation results. Therefore, a comprehensive evaluation system and method for the sleep-aiding power of home textiles that is automatic, objective, adaptive, and considers multi-physical field coupling is needed. Summary of the Invention

[0005] To address the problems of poor repeatability of subjective sleep trials, high cost of physiological monitoring, neglect of physical coupling between indicators by single or linear regression models, inability to adapt to different product categories and dynamic changes in usage environment, and difficulty in effectively predicting new products in existing methods for evaluating the sleep-aiding performance of home textiles, this invention provides a home textile sleep-aiding performance evaluation system and construction method based on characteristic functional indicators.

[0006] The technical solution adopted by this invention to solve its technical problem is as follows: a home textile sleep-aiding power evaluation system based on feature function indicators. Preferably, the system includes a five-dimensional feature indicator acquisition module, used to acquire five-dimensional feature indicators (tactile, thermal humidity, pressure, interference, and hygiene) through standardized instrument testing, and automatically assigning weights using range normalization and entropy weighting; a sleep-aiding power score and confidence interval output module, used to construct a physically constrained Bayesian neural network, embedding material science and sleep physiology priors as regularization terms into the loss function, and outputting the sleep-aiding power score and confidence interval; a knowledge transfer and zero-shot prediction module, used to establish an adaptive weighted graph convolutional network to achieve knowledge transfer and zero-shot prediction between different home textile products; and a scenario-based sleep-aiding power level output module, used to fit the relationship between static indicators and environmental parameters through a dynamic environment adaptive mapping module, and output a scenario-based sleep-aiding power level.

[0007] On the other hand, a method for constructing a home textile sleep-aiding power evaluation based on feature function indicators is characterized by the following steps: S1: Collecting feature function indicators of home textiles in five dimensions—tactile perception, thermal and humidity perception, pressure perception, interference control, and mental health—through standardized instrument testing. The collected indicators are then normalized by range, and the weights of each indicator are automatically calculated using the entropy weight method to form a sample database; S2: Constructing a physically constrained Bayesian neural network, utilizing the weighted feature function indicators in the sample database and corresponding real-person sleep-aiding power scores, and combining prior knowledge of materials science and sleep physiology to set regularization penalty terms, training a basic sleep-aiding power prediction model; S3: Constructing a product indicator scenario-based heterogeneous graph, using a graph convolutional network for cross-product knowledge transfer training to obtain a graph neural network model capable of handling new product categories; S4: Constructing a conditional variational autoencoder, using static product indicators and environmental parameters as input, and sleep-aiding power scores under different environments as output, training a dynamic environment adaptive model; S5: Integrating the basic sleep-aiding power prediction model, the graph neural network model, and the dynamic environment adaptive model to form a multi-level evaluation system.

[0008] The beneficial effects of this invention are as follows: It automatically assigns weights to indicators through range normalization and entropy weighting, eliminating manual intervention; it embeds prior knowledge of materials science and sleep physiology into a physically constrained Bayesian neural network, ensuring the physical rationality of the model; it achieves cross-category knowledge transfer through an adaptive weighted graph convolutional network, supporting zero-shot prediction; and it fits the relationship between static product indicators and environmental parameters through a dynamic environment adaptive mapping module, outputting a scenario-based sleep-aiding power level. This invention integrates the above multi-level models, achieving rapid, objective, personalized, and adaptive sleep-aiding performance evaluation. Compared to traditional linear regression or trial-and-error methods, it significantly improves evaluation efficiency, cost control, and prediction accuracy. Attached Figure Description

[0009] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary, and those skilled in the art can derive other embodiments based on the provided drawings without creative effort.

[0010] Figure 1 This is a schematic diagram of the optimal embodiment of the home textile sleep aid evaluation system and construction method based on characteristic functional indicators of the present invention. Detailed Implementation

[0011] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0012] Example 1 The present invention provides a home textile sleep aid evaluation system based on feature function indicators, which mainly includes four core modules: a five-dimensional feature indicator acquisition module, a sleep aid score and confidence interval output module, a knowledge transfer and zero-sample prediction module, a scenario-based sleep aid level output module, and an integrated evaluation output module.

[0013] The five-dimensional characteristic index acquisition module is used to obtain a set of characteristic functional indicators of the tested home textiles in five dimensions: tactile perception, heat and humidity perception, pressure perception, interference control, and mental health, through standardized instrument testing. This module first measures the textile samples using professional testing equipment according to relevant national, industry, or group standards. For example, the tactile perception dimension includes indicators such as the coefficient of dynamic friction, bending stiffness, and itching sensation, measured using a fabric tactile tester according to the FZ / T01171 standard; the heat and humidity perception dimension includes indicators such as thermal resistance, moisture resistance, moisture permeability index, coolness upon contact, and air permeability, measured using the evaporative hot plate method according to the GB / T11048 standard and the coolness upon contact tester according to the GB / T35263 standard; the pressure perception dimension includes indicators such as compression ratio, resilience, indentation hardness, and bulkiness, measured using a compression elasticity meter according to the GB / T24442.1 standard; the interference control dimension includes indicators such as frictional noise sound pressure level, light-blocking rate, sound insulation coefficient, and antistatic half-life, measured using the silent chamber method and a light-blocking rate tester. This module also uses range normalization and entropy weighting to automatically assign weights to each indicator, completely eliminating manual intervention. The range normalization process is as follows: For positive indicators, those with higher values ​​are more conducive to sleep, such as moisture permeability index, compressibility, and coolness upon contact. The formula is: ; For negative indicators, those with lower values ​​are more conducive to sleep, such as the coefficient of kinetic friction, thermal resistance, and frictional noise. The formula used is: ; in, Indicates the first The product sample in the first The raw measurement values ​​for each indicator have clear physical dimensions, such as thermal resistance in square Kelvin per watt, coefficient of friction in dimensionless form, and compression ratio as a percentage value. and These are the maximum and minimum values ​​of the indicator in all samples, respectively. This is the normalized dimensionless value, with a value range between 0 and 1.

[0014] The five-dimensional feature index acquisition module automatically calculates the weight of each index using the entropy weight method. Calculate the first The first indicator The contribution of each sample is calculated using the following formula: ; in, For the first The first indicator The contribution of each sample is given by the formula, where the denominator is the sum of the normalized values ​​of all samples under that indicator. Calculate the first The information entropy of each indicator is calculated using the following formula: ; in, For the total number of product samples, when When = 0, it is agreed that =0. For the first The information entropy of each indicator ranges from 0 to 1. The larger the information entropy value, the smaller the degree of variation of the indicator and the less information it contains. Calculate the first The entropy weight of each indicator is calculated using the following formula: ; in, The total number of characteristic function indicators, That is, the first Automatic weighting of each indicator to satisfy The five-dimensional feature index acquisition module outputs a weighted comprehensive index vector based on entropy weight, which serves as the input to the subsequent neural network.

[0015] The sleep aid rating and confidence interval output module is used to construct a physically constrained Bayesian neural network. This module takes a weighted feature function index vector as input and outputs a basic sleep aid rating and its uncertainty interval. The neural network is structured as a multilayer perceptron, containing an input layer, two hidden layers, and an output layer. The number of neurons in each layer is set according to the number of indicators. To incorporate the physical coupling relationship between materials science and sleep physiology as a regularization prior, this module adds a physical constraint violation penalty term to the loss function. The physical coupling relationships include: a constraint on the product of thermal resistance and moisture permeability index (the product should be within the physiologically comfortable range; a large product indicates the product is too stuffy and detrimental to sleep, while a small product indicates insufficient warmth); a constraint on the difference between compression ratio and rebound rate (the difference should be below a preset threshold; a large difference leads to unstable product support); and a constraint on the positive correlation between friction noise and friction coefficient (the sound pressure level of friction noise should be positively correlated with the friction coefficient; if the test data violates this physical law, it indicates measurement anomalies or special materials).

[0016] The regularization prior term is a physical constraint violation penalty term added to the loss function: ; in, This is the total loss function value. For the sample size, The sleep-inducing power score predicted by the model. The true sleep aid ability score is obtained by fusing polysomnography and subjective scales. The penalty coefficient is selected based on cross-validation and is typically set to between 0.01 and 0.1. For the first The degree of violation of a physical constraint, such as the constraint of the product of thermal resistance and moisture permeability index. Defined as the deviation of the normalized product value from the boundary of the physiological comfort range. The preset tolerance threshold is automatically determined from real-person sleep trial evaluation data using statistical methods. This module not only outputs a sleep aid rating but also outputs the confidence interval of the rating, i.e., the standard deviation of the predicted value, through Bayesian inference.

[0017] The knowledge transfer and zero-shot prediction module is implemented by establishing an adaptive weighted graph convolutional network. This module includes three sub-units: a heterogeneous graph construction unit, an adaptive weighting unit, and a knowledge transfer unit.

[0018] In the heterogeneous graph construction unit, the system constructs a heterogeneous graph containing various types of nodes. Product nodes include common home textile categories such as sheets, duvet covers, comforters, pillow cores, mattresses, and curtains; indicator nodes include individual indicators under the five dimensions of tactile perception, thermal and moisture perception, pressure perception, interference control, and mental health, such as the coefficient of dynamic friction, thermal resistance, compressibility, friction noise sound pressure level, and antibacterial rate; scenario nodes include usage seasons such as summer and winter, usage regions such as the south and the north, and user groups such as the elderly and children. The edges between these nodes represent their relationships; for example, there is a relationship between the comforter product node and the thermal resistance indicator node, and a relationship between the summer scenario node and the coolness-to-the-touch indicator node.

[0019] In the adaptive weighted unit, the system dynamically calculates the edge weights between different types of nodes through a multi-head attention mechanism. Specifically, for each node pair, its attention coefficient is calculated, which is continuously optimized during training to automatically learn the importance differences between different products, metrics, and scenarios. The multi-head attention mechanism uses multiple independent attention heads to calculate the edge weights, and then concatenates or averages the results of each head to enhance the model's expressive power.

[0020] In the knowledge transfer unit, when a completely new home textile product is input, such as a novel sleep pillow that has never appeared in the training set, the system first adds it as a new product node to the heterogeneous graph and establishes edges between it and the corresponding indicator nodes based on the product's known indicator values. Then, using the weights of the pre-trained graph convolutional network, the state of the new product node is updated through a graph propagation algorithm. After several rounds of graph convolution calculations, the final hidden state of the new product node can be transformed into a predicted value of the product's sleep-aiding effect. This process achieves knowledge transfer from existing product categories to new categories, i.e., zero-shot prediction capability.

[0021] Next, the scenario-based sleep aid level output module uses a dynamic environment adaptive mapping module to fit the relationship between the product's static indicators and environmental parameters. This module includes an environmental response experimental unit, a conditional variational autoencoder unit, and a scenario-based output unit.

[0022] The environmental response experimental unit is responsible for collecting data on changes in characteristic functional indicators of typical products under at least two different sets of environmental conditions. Typical environmental conditions include ambient temperature, ambient humidity, and mattress firmness. For example, under one set of conditions—a temperature of 16 degrees Celsius, relative humidity of 50%, and a medium mattress firmness—and another set of conditions—a temperature of 26 degrees Celsius, relative humidity of 70%, and a soft mattress firmness—the changes in indicators such as thermal resistance, moisture permeability index, and compression ratio of the same comforter product are measured. This experimental data is used to train a conditional variational autoencoder.

[0023] The Conditional Variational Autoencoder (CVA) unit consists of an encoder and a decoder. The encoder takes as input a vector of static product metrics and a vector of environmental parameters, and outputs the mean and variance of the latent variables. Static metrics include inherent properties that do not change with the environment, such as the thermal conductivity of the fabric and the compressive modulus of the core material; environmental parameters include dynamic variables such as temperature, humidity, and mattress firmness. The decoder takes as input latent variables sampled from the latent variable distribution and the environmental parameters, and outputs a reconstructed dynamic sleep-aid score. By minimizing the reconstruction error and the KL divergence of the latent variable distribution, the CVA learns the mapping relationship from the product's static metrics and environmental parameters to the sleep-aid score. The unique feature of this unit is that once trained, for any new product with a given static metric, no multi-environmental testing is required; only the target environmental parameters need to be input to calculate the product's sleep-aid score in that environment.

[0024] The scenario-based output unit receives user-inputted parameters for the expected usage environment. For example, if a consumer wants to know the sleep-aiding effect of a particular mattress in a summer air-conditioned room with high humidity, they input the mattress's static indicators and the parameters for summer, high humidity, and air conditioning into a conditional variational autoencoder. The encoder then outputs the product's sleep-aiding power level and usage suggestions for that scenario. The sleep-aiding power level is divided according to pre-set standards, such as from A to E, corresponding to excellent, good, average, poor, and very poor, respectively.

[0025] The scenario-based sleep aid rating output module is used to fit the relationship between static indicators and environmental parameters through a dynamic environment adaptive mapping module. This module includes an environmental response experiment unit, a conditional variational autoencoder unit, and a scenario-based output unit. The environmental response experiment unit collects data on the changes in characteristic functional indicators of typical products under at least two different combinations of ambient temperature, humidity, and mattress firmness. For example, it tests the changes in thermal resistance, moisture resistance, and compressibility of the same product under conditions of 18℃ temperature, 50% humidity, and a firm mattress, and 26℃ temperature, 70% humidity, and a soft mattress. The conditional variational autoencoder unit consists of an encoder and a decoder. The encoder encodes the product's static indicators and environmental parameters as latent variables, outputting the mean and logarithmic variance of the latent variables during the encoding process. The decoder reconstructs the dynamic sleep aid rating from the latent variables and environmental parameters. During training, this unit uses the data collected from the environmental response experiment and performs backpropagation optimization using reparameterization techniques. The scenario-based output unit is used to receive the expected usage environment parameters input by the user. For example, if the user specifies that the bedroom temperature is 22℃, the humidity is 60%, and the mattress is of medium firmness, the unit calls the trained conditional variational autoencoder and outputs the sleep-aid level of the product under test in this environment and usage suggestions.

[0026] The integrated evaluation output module, an optional but preferred additional module, receives the characteristic functional indicators of the home textile product under test, sequentially calls the four modules mentioned above, and finally outputs the product's sleep-aiding power level, the corresponding confidence interval, and the sleep-aiding power variation curve under different usage environments. This module is also used to generate product design improvement suggestions based on the sensitivity analysis results of each characteristic functional indicator and sleep-aiding power when the product's sleep-aiding power level is lower than a preset threshold. The sensitivity analysis uses partial derivatives based on entropy weighting, and the calculation formula is as follows: ; in, For the first The sensitivity of each indicator to the sleep aid rating. The entropy weight of this indicator. For the prediction model output, the normalized first... The partial derivatives of each indicator are calculated using automatic differentiation technology, which records the gradient of the output with respect to the input during the backpropagation of the neural network. If the sensitivity of a certain indicator is positive and the value is large, it means that increasing the value of that indicator can significantly improve its sleep-inducing effect. Based on this, product design improvement suggestions are generated. For example, if the sensitivity of the compression ratio is the highest and positive, it is recommended to increase the thickness of the filling or use a high-resilience material.

[0027] Example 2 A method for constructing an evaluation system for the sleep-aiding power of home textiles based on characteristic functional indicators includes: S1: Collect characteristic functional indicators of home textiles in five dimensions: tactile perception, heat and humidity perception, pressure perception, disturbance control, and mental health through standardized instrument testing. Normalize the range of the collected indicators and automatically calculate the weight of each indicator using the entropy weight method to form a sample database.

[0028] S2: Construct a physically constrained Bayesian neural network. Utilize weighted feature function indicators from the sample database and corresponding real-person sleep aid ratings. Combine prior knowledge of materials science and sleep physiology to set regularization penalty terms, and train to obtain a basic sleep aid prediction model. In this step, the physical constraint regularization penalty terms should include at least: the product of thermal resistance and moisture permeability should be within the physiological comfort range; the difference between compression ratio and rebound rate should be lower than a preset threshold; and the friction noise sound pressure level should be positively correlated with the friction coefficient. The physiological comfort range and preset threshold are automatically determined from real-person sleep evaluation data using statistical methods. For example, the distribution of the product of thermal resistance and moisture permeability in a large number of real-person sleep samples can be collected, and the 95% confidence interval in the middle of the distribution can be taken as the physiological comfort range. The network training uses the Adam optimizer with a learning rate of 0.001 and 200 training epochs. After each epoch, the loss is evaluated on the validation set. Training stops when the validation loss no longer decreases for 10 consecutive epochs.

[0029] S3: Construct a heterogeneous graph combining product metrics and scenarios, i.e., a directed heterogeneous graph combining product nodes, metric nodes, and scenario nodes. Use a graph convolutional network for cross-product knowledge transfer training to obtain a graph neural network model capable of handling new product categories. In this step, the structure of the heterogeneous graph is the same as the aforementioned knowledge transfer module. The preferred number of graph convolutional layers is two, each using a 128-dimensional hidden representation, with ReLU as the activation function. Semi-supervised learning is used during training, with the sleep-aid rating of some product nodes as labels, and the label mask for the remaining nodes set to 0. Label information is propagated through graph convolution. After model training is complete, zero-shot prediction can be performed on any newly added product node.

[0030] S4: Construct a conditional variational autoencoder (FUE) with product static indicators and environmental parameters as input, and sleep-aiding power scores under different environments as output, to train a dynamic environment adaptive model. In this step, the product static indicators are normalized five-dimensional feature indicators, and the environmental parameters are numerical representations of temperature, humidity, and mattress firmness. The latent variable dimension of the FUE is set to 16, and both the encoder and decoder are three-layer fully connected networks with 64 neurons in the intermediate layers. The loss function consists of two parts: reconstruction loss and KL divergence loss. The reconstruction loss uses mean squared error.

[0031] S5: Integrate the basic sleep aid prediction model, the graph neural network model, and the dynamic environment adaptive model to form a multi-level evaluation system. The integration method is as follows: first, the basic sleep aid prediction model provides an initial score; then, the graph neural network model refines the score based on product type and scenario nodes; finally, the dynamic environment adaptive model outputs a scenario-specific level. The three models can be used sequentially or in a weighted fusion manner, with the weights of each model dynamically adjusted based on performance on the validation set.

[0032] In summary, this invention proposes a home textile sleep-aiding performance evaluation system and its construction method based on feature function indicators. This aims to overcome the shortcomings of traditional subjective sleep testing, physiological monitoring, and linear regression models in sleep-aiding performance evaluation, such as poor repeatability, high cost, neglect of multi-physical field coupling, and difficulty in adapting to new products and environmental changes. The system collects feature indicators across five dimensions—touch, heat and humidity, pressure, interference, and hygiene—through standardized instrument testing. Range normalization and entropy weighting are used to automatically assign weights to the indicators, completely eliminating manual intervention. Based on this, a physically constrained Bayesian neural network is constructed, embedding prior knowledge of materials science and sleep physiology as regularization penalty terms into the loss function to ensure that the sleep-aiding performance score output by the model is not only accurate but also possesses physical rationality and a confidence interval. Furthermore, an adaptive weighted graph convolutional network is established, linking products, indicators, and scene nodes through heterogeneous graphs. Multi-head attention mechanisms are used to dynamically optimize edge weights, achieving knowledge transfer and zero-shot prediction between different home textile categories, effectively solving the problem of insufficient training data for new products. Simultaneously, a dynamic environment adaptive mapping module is designed to fit the nonlinear relationship between static product indicators and environmental parameters using a conditional variational autoencoder, outputting scenario-based sleep-aiding power levels and usage suggestions. Finally, by integrating the evaluation output module, the three models are concatenated or weighted and fused to output sleep-aiding power levels, confidence intervals, and performance change curves under multiple environments in a single output. Furthermore, when the score falls below a threshold, design improvement suggestions are generated based on sensitivity analysis. This invention integrates instrumental quantitative testing with multi-level neural networks, achieving rapid, objective, personalized, and scenario-adaptive sleep-aiding performance evaluation. It significantly improves evaluation efficiency, prediction accuracy, and universality, providing scientific and intelligent technical support for the certification and optimized design of sleep-aiding functions in home textiles.

[0033] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. The above descriptions are only preferred embodiments of this application. It should be noted that due to the limitations of written expression, while there are objectively infinite specific structures, those skilled in the art can make several improvements, modifications, or changes without departing from the principles of this application, and can also combine the above technical features in an appropriate manner. These improvements, modifications, changes, or combinations, or the direct application of the inventive concept and technical solution to other situations without modification, should all be considered within the scope of protection of this application.

Claims

1. A home textile sleep-aiding power evaluation system based on characteristic function indicators, characterized in that, include: The five-dimensional feature index acquisition module is used to acquire five-dimensional feature indicators of touch, heat and humidity, pressure, interference and hygiene through standardized instrument testing, and automatically assign weights using range normalization and entropy weighting method. The module for outputting sleep aid power score and confidence interval is used to construct a physically constrained Bayesian neural network. It embeds the prior knowledge of materials science and sleep physiology as regularization terms into the loss function and outputs the sleep aid power score and confidence interval. The knowledge transfer and zero-shot prediction module is used to build an adaptive weighted graph convolutional network to achieve knowledge transfer and zero-shot prediction between different home textile products. The scenario-based sleep aid level output module is used to fit the relationship between static indicators and environmental parameters through the dynamic environment adaptive mapping module, and output the scenario-based sleep aid level.

2. The home textile sleep-aiding evaluation system based on characteristic functional indicators according to claim 1, characterized in that: The range normalization process is as follows: For positive indicators, the formula is as follows: ； For negative indicators, the formula is used: ； in, Indicates the first The product sample in the first The original measured values ​​of each indicator and These are the maximum and minimum values ​​of the indicator in all samples, respectively. The normalized dimensionless value; The positive indicator is one whose larger value is more conducive to sleep, and the negative indicator is one whose smaller value is more conducive to sleep.

3. The home textile sleep-aiding evaluation system based on characteristic functional indicators according to claim 1, characterized in that, The specific method for automatically calculating the weights of each indicator using the entropy weight method is as follows: Calculate the first The first indicator The contribution of each sample is calculated using the following formula: ； Calculate the first The information entropy of each indicator is calculated using the following formula: ； Calculate the first The entropy weight of each indicator is calculated using the following formula: ； in, For the first The first indicator The contribution of each sample For the first Information entropy of each indicator For the first Entropy weight of each indicator, The total number of product samples. The total number of characteristic function indicators, That is, the first Automatic weighting of each indicator to satisfy ; The indicator is obtained by weighting the comprehensive indicator vector based on the entropy weight.

4. The home textile sleep-aiding evaluation system based on characteristic functional indicators according to claim 1, characterized in that, In the physically constrained Bayesian neural network, the physical coupling relationships include: Constraints include the product of thermal resistance and moisture permeability index, the difference between compression ratio and resilience, and the positive correlation between friction noise and friction coefficient. The regularization prior term is a physical constraint violation penalty term added to the loss function: ； in, This is the total loss function value. For the sample size, The sleep-inducing power score predicted by the model. Rate the actual sleep aid effect. For the first The degree of violation of a physical constraint, To preset the tolerance threshold, This is the penalty coefficient.

5. The home textile sleep-aiding evaluation system based on characteristic functional indicators according to claim 1, characterized in that, The knowledge transfer and zero-shot prediction module includes: The heterogeneous graph construction unit takes bed sheets, duvet covers, duvet cores, pillow cores, mattresses, and curtains as product nodes, takes each individual indicator under the dimensions of tactile perception, heat and humidity perception, pressure perception, interference control, and mental health as indicator nodes, and takes the usage season, usage region, and user group as scenario nodes. An adaptive weighted unit calculates the edge weights between different types of nodes through a multi-head attention mechanism, and the edge weights are dynamically optimized during the training process. The knowledge transfer unit, when a new product is input, utilizes the graph convolution weights between the trained product nodes and indicator nodes to update the state of the new product nodes through graph propagation, and outputs the predicted sleep-aiding power value of the new product.

6. The home textile sleep-aiding evaluation system based on characteristic functional indicators according to claim 1, characterized in that, The scenario-based sleep aid level output module includes: The environmental response testing unit is used to collect data on changes in characteristic functional indicators of typical products under at least two different combinations of ambient temperature, humidity, and mattress firmness. The conditional variational autoencoder unit encodes product static indicators and environmental parameters as latent variables, and its decoder reconstructs the dynamic sleep aid score from the latent variables and environmental parameters. The scenario-based output unit is used to receive the expected usage environment parameters input by the user and output the sleep-aiding power level and suggestions of the product under test in that environment.

7. The home textile sleep-aiding evaluation system based on characteristic functional indicators according to claim 1, characterized in that, It also includes an integrated evaluation output module: The integrated evaluation output module is used to receive the characteristic function indicators of the home textile to be tested, and sequentially call the sleep aid power score and confidence interval output module, the knowledge transfer and zero sample prediction module, and the scenario-based sleep aid power level output module to output the sleep aid power level of the product, the corresponding confidence interval, and the sleep aid power change curve under different usage environments.

8. The home textile sleep-aiding evaluation system based on characteristic functional indicators according to claim 7, characterized in that, The integrated evaluation output module is also used to generate product design improvement suggestions based on the sensitivity analysis results of each feature function index and sleep aid power when the product's sleep aid power level is lower than a preset threshold. The sensitivity analysis is calculated using partial derivatives weighted by entropy weights, and the calculation formula is as follows: ； in, For the first The sensitivity of each indicator to the sleep aid rating. The entropy weight of this indicator. To predict the model output for the normalized first... The partial derivatives of each indicator.

9. A method for constructing an evaluation system for the sleep-aiding power of home textiles based on characteristic functional indicators, characterized in that, include: S1: Collect characteristic functional indicators of home textiles in five dimensions: tactile perception, heat and humidity perception, pressure perception, disturbance control, and mental health through standardized instrument testing. Normalize the range of the collected indicators and automatically calculate the weight of each indicator using the entropy weight method to form a sample database. S2: Construct a physically constrained Bayesian neural network, utilize the weighted feature function indicators in the sample database and the corresponding real-person sleep aid power scores, combine prior knowledge of materials science and sleep physiology to set regularized penalty terms, and train to obtain a basic sleep aid power prediction model. S3: Construct a product indicator scenario combined with a heterogeneous graph, and use a graph convolutional network for cross-product knowledge transfer training to obtain a graph neural network model that can handle new product categories; S4: Construct a conditional variational autoencoder, taking product static indicators and environmental parameters as inputs and sleep aid scores under different environments as outputs, to train a dynamic environment adaptive model. S5: Integrate the basic sleep aid prediction model, graph neural network model, and dynamic environment adaptive model to form a multi-level evaluation system.

10. A method for constructing an evaluation system for the sleep-aiding power of home textiles based on characteristic functional indicators according to claim 9, characterized in that, The physical constraint regularization penalty term includes: The product of thermal resistance and moisture permeability should be within the physiological comfort range, the absolute value of the difference between compression ratio and resilience should be lower than the preset threshold, and the friction noise sound pressure level should be positively correlated with the friction coefficient. The physiological comfort range and preset threshold are automatically determined by statistical methods from real-person sleep trial evaluation data.

Images