Intelligent response type geotextile pore structure prediction and parameter optimization method

Abstract

The application discloses a kind of intelligent response type geotextile pore structure prediction and parameter optimization method, comprising: S1: obtaining input variable, including environmental load parameter, material structure parameter, service state parameter and engineering working condition parameter;S2: establish pore-humidity-pressure mapping model, and generate mechanism constraint data set;S3: construct multi-output prediction model, input variable is input, dynamic porosity, equivalent pore diameter, permeability coefficient and anti-clogging index are synchronous output;S4: construct multi-objective optimization model, with the environmental load parameter, service state parameter and engineering working condition parameter corresponding to the preset engineering scene as boundary condition, solve the multi-objective optimization model, obtain the optimal material structure parameter combination.The application scientifically and dynamically predicts the key performance indicators of geotextile under complex service environment, and optimizes its material parameters accordingly, to balance drainage, soil conservation, anti-clogging and other multiple or even mutually restrictive engineering requirements.

Description

Technical Field

[0001] This invention belongs to the field of geosynthetic material design and performance prediction technology, specifically a smart responsive geotextile pore structure prediction and parameter optimization method. Background Technology

[0002] Geotextiles are widely used geosynthetic materials in municipal engineering, commonly used in roadbeds, urban tunnel drainage, integrated utility tunnel subbases, slope protection, landfill seepage prevention and drainage, and sponge city facilities. They primarily serve functions such as isolation, filtration, drainage, protection, and reinforcement. Current standards typically use indicators such as unit area mass, thickness, tensile strength, puncture strength, equivalent pore size, and vertical permeability coefficient as the basis for selection and acceptance.

[0003] However, the above indicators are mostly static test results under standard conditions, which are difficult to reflect the continuous impact of humidity changes, soil pressure, traffic load, fine particle migration and construction disturbance on the pore structure and drainage performance of geotextiles, thus limiting the dynamic identification of drainage attenuation, filter instability and siltation risks.

[0004] The filtration and drainage performance of geotextiles depends on the matching relationship between the pore structure and soil particles, hydraulic gradient, and pressure state. Classical reverse filtration theory requires a balance between soil retention, permeability, and anti-clogging properties; that is, excessively large pore sizes can lead to the loss of fine particles, while excessively small pore sizes may cause blockage of hydraulic channels. Related studies have shown a clear coupling relationship between pore size distribution, permeability, long-term clogging, and field constraints. Therefore, the performance evaluation of geotextiles should not be limited to a single static index, but should be comprehensively analyzed in conjunction with the service environment and pore evolution process.

[0005] In recent years, the development of intelligent construction, sensing and monitoring, and material functionalization has driven the transformation of geotextiles from passive materials to functional materials that can sense and respond. Intelligent responsive geotextiles can regulate fiber contact states and pore channels through hydrophilic, temperature- and humidity-sensitive, or expansive components, and can achieve feedback on humidity, pressure, pore pressure, and drainage flow rate through on-site monitoring. However, intelligent response does not necessarily mean that larger pore sizes or more responsive components are always better. For municipal drainage filter layers, excessively large pore sizes may weaken soil retention, excessively strong wet expansion response may reduce drainage capacity, and excessively thick composite structures may increase water flow paths. Therefore, there is an urgent need to establish predictive optimization models that can comprehensively describe the relationship between environmental loads, material parameters, and performance targets.

[0006] Existing research has largely focused on filtration criteria, clogging mechanisms, interfacial mechanics, and durability evaluation. However, research on unified prediction and parameter optimization of porosity, equivalent pore size, permeability coefficient, and anti-clogging capacity for smart responsive geotextiles remains relatively insufficient. Especially in municipal scenarios characterized by high humidity in coastal areas, significant groundwater recharge, and the coupling effect of road loads and drainage / filtration requirements, the dynamic porosity evolution and performance regulation mechanisms of geotextiles still require further quantification. Summary of the Invention

[0007] To address the problems existing in the prior art, the present invention aims to provide a method for predicting the pore structure and optimizing the parameters of intelligent responsive geotextiles. This method scientifically and dynamically predicts the key performance indicators (porosity, pore size, permeability, and anti-clogging properties) of intelligent responsive geotextiles in complex and variable service environments (especially in high-humidity and high-pressure municipal engineering scenarios), and optimizes its material parameters (area density, pore size, thickness, and content of responsive components) accordingly, so as to balance multiple or even mutually restrictive engineering requirements such as drainage, soil conservation, and anti-clogging.

[0008] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0009] A smart responsive geotextile pore structure prediction and parameter optimization method includes the following steps: Step S1: Obtain input variables, including environmental load parameters, material structure parameters, service status parameters, and engineering condition parameters; Step S2: Establish a pore-humidity-pressure mapping model, which is used to calculate dynamic porosity, equivalent pore size, permeability coefficient, and anti-clogging index based on the input variables, and generate a mechanism constraint dataset; Step S3: Construct a multi-output prediction model, and train it using the mechanism constraint dataset to obtain a trained multi-output prediction model; the multi-output prediction model takes the input variables as input and outputs dynamic porosity, equivalent pore size, permeability coefficient, and anti-clogging index simultaneously; Step S4: Construct a multi-objective optimization model, using the trained multi-output prediction model as the objective function calculator, taking the material structure parameters as design variables, and using the environmental load parameters, service status parameters, and engineering condition parameters corresponding to the preset engineering scenario as boundary conditions to solve the multi-objective optimization model and obtain the optimal combination of material structure parameters; Step S5: Output the optimal combination of material structure parameters.

[0010] As a further improvement to the above technical solution:

[0011] The environmental load parameters include: relative humidity. Normal pressure and temperature The material structural parameters include: fiber areal density. Initial micropore size Composite layer thickness and response component mass fraction The service status parameters include: service age. and hydraulic gradient The engineering operating parameters The conditions are represented by discrete variables, including seepage control conditions, high pressure bearing conditions, drainage and storage conditions, and comprehensive conditions. The normal pressure range corresponding to the high pressure bearing conditions is 200 to 500 kPa.

[0012] The step S2, establishing the pore-humidity-pressure mapping model, specifically includes: based on the initial micropore size... Fiber surface density and composite layer thickness Calculate the reference porosity using the following formula. :

[0013]

[0014] In the formula, For constant terms, , and The contributions of initial micropore size, fiber areal density, and composite layer thickness to the reference porosity are characterized, respectively. and These are the initial micropore sizes. The minimum and maximum engineering boundaries, and These are fiber areal densities. The minimum and maximum engineering boundaries, and The thickness of the composite layer is respectively The minimum and maximum engineering boundaries.

[0015] The step S2 of establishing the pore-humidity-pressure mapping model also includes: calculating the humidity response shrinkage term using the following formula. Pressure compaction item Structural support items and age decay coefficient :

[0016]

[0017]

[0018]

[0019]

[0020] in, In response to the component mass fraction, The relative humidity is in decimal form. For reference temperature, and These are the normalized values ​​for fiber areal density and composite layer thickness, respectively. As a pressure normalization benchmark, For service age, Set the maximum age for the model. The humidity linear response coefficient is... The second-order humidity response coefficient is... This is a temperature correction factor. As the baseline compaction coefficient, This is a correction factor for the fiber areal density's sensitivity to compaction. The humidity softening effect coefficient, , These are coefficients used to characterize the supporting effect of the composite layer thickness and the compacting effect of the fiber skeleton, respectively. This is a coefficient used to control the degree of long-term damage.

[0021] The step S2 of establishing the pore-humidity-pressure mapping model also includes: calculating the dynamic porosity using the following formula. :

[0022]

[0023] in, This is the porosity disturbance term. and These are the minimum and maximum values ​​of the dynamic porosity boundary, respectively. The upper and lower bound cutoff function is defined as follows:

[0024] .

[0025] The step S2 of establishing the pore-humidity-pressure mapping model also includes: calculating the equivalent pore size using the following formula. and permeability coefficient :

[0026]

[0027]

[0028] in, In response to the reduction factor of effective channels after component swelling due to moisture, For aperture disturbance term, and Equivalent aperture The minimum and maximum physical boundaries; The seepage scale coefficient, This is the reduction factor for permeability due to pressure compaction. This is the path correction factor for the composite layer thickness.

[0029] The step S2, which establishes the pore-humidity-pressure mapping model, also includes: calculating the drainage capacity index using the following formula. and anti-fouling index :

[0030]

[0031]

[0032] In the formula, For hydraulic gradient; This is the initial aperture normalized value. This represents the maximum engineering boundary value of the normal pressure. ~ Assigning weights to each influencing factor. This is the disturbance term of the anti-clogging index.

[0033] The multi-output prediction model in step S3 is an extreme random tree model:

[0034]

[0035]

[0036] In the formula, To integrate prediction functions, The number of random regression trees, For the first Random regression trees, For the first The random partitioning parameters, sample perturbation, and feature perturbation corresponding to each random regression tree. The feature vector is composed of the input variables.

[0037] The objective function of the multi-objective optimization model in step S4 for:

[0038]

[0039] in, As a normalized benchmark for permeability coefficient, For the target porosity, To allow for porosity deviation scale, To create feasible penalties, , , and The weights are respectively for permeability, anti-clogging ability, pore stability, and manufacturing feasibility.

[0040] In step S4, the differential evolution algorithm is used to solve the multi-objective optimization model.

[0041] The beneficial effects of this invention are:

[0042] (1) By incorporating humidity response, pressure compaction, thickness support, age decay, and hydraulic gradient effects into a unified variable system, a pore-humidity-pressure mapping model was established, which can explain pore shrinkage, pore size decay, and material structure support effects under high humidity and high pressure conditions. Compared with the current specifications that only provide selection criteria for static test indicators under standard conditions, this invention can reflect the continuous impact of humidity changes, soil pressure, traffic load, fine particle migration, and construction disturbance on the pore structure and drainage performance of geotextiles, thereby enabling dynamic identification and prediction of drainage decay, filter instability, and siltation risks.

[0043] (2) This invention constructs a multi-output ensemble learning prediction model, realizing the simultaneous prediction of dynamic porosity, equivalent pore size, permeability coefficient, and anti-clogging index. This model can effectively identify the pore evolution and hydraulic performance changes of intelligent responsive geotextiles, and through substitution importance analysis, it can identify the main controlling factors such as initial micropore size, response component mass fraction, humidity, normal pressure, and fiber areal density, transforming the black-box prediction results into engineering judgments that can be used for material design. For construction monitoring scenarios, this model can be used to determine whether the pore structure has entered a high-risk attenuation zone; for material design scenarios, it can be used to quickly screen candidate parameter combinations.

[0044] (3) A multi-objective optimization method is proposed that takes into account permeability, anti-clogging ability, target porosity, and manufacturing feasibility. This method achieves automatic multi-objective optimization, balances mutually restrictive engineering requirements, and solves the design problem in the prior art that "intelligent response does not necessarily mean that the larger the pore size or the more response components, the better"—excessive pore size may weaken soil retention, excessively strong wet expansion response may reduce drainage capacity, and excessively thick composite structure may increase water flow path. The optimization function of this invention rewards better drainage capacity and anti-clogging performance, while penalizing schemes that deviate from the target porosity and manufacturing center value. Under the working conditions of the Shenzhen underpass drainage and storage project, the optimal parameter combination obtained by optimization (fiber surface density 448.41 g / m², initial micropore size 186.80 μm, composite layer thickness 2.94 mm, response component mass fraction 0.146) corresponds to a dynamic porosity of 0.3703 and a permeability coefficient of 0. With a flow rate of m / s and an anti-clogging index of 0.5527, it reflects a reasonable trade-off between drainage capacity and pore stability.

[0045] (4) Humidity-pressure response analysis shows that high humidity and high pressure areas are sensitive areas for pore decay. For coastal high humidity municipal engineering projects, humidity, normal pressure, drainage flow and pore pressure changes should be included in the monitoring system, and material parameter verification and construction process control should be carried out in conjunction with prediction models.

[0046] (5) This invention forms a complete technical chain of mechanism mapping, data learning, and target optimization. Although the case data is formed by expanding the engineering scenario boundary and pore-humidity-pressure mapping model, it can be further calibrated by combining indoor permeability test, gradient ratio test, compression creep test and field sensor data. Furthermore, factors such as particle size distribution, sediment concentration, cyclic load and chemical corrosion can be introduced to construct an online update model for field applications. Attached Figure Description

[0047] Figure 1 This is a distribution map of the target response variables calculated by the pore-humidity-pressure mapping model according to an embodiment of the present invention, wherein (a) is a dynamic porosity distribution map, (b) is an equivalent pore size distribution map, (c) is a permeability coefficient distribution map, and (d) is an anti-clogging index distribution map.

[0048] Figure 2 This is a heatmap showing the correlation of engineering variables according to an embodiment of the present invention.

[0049] Figure 3 This is a scatter plot of the pore response mechanism according to an embodiment of the present invention, wherein (a) is a schematic diagram of the distribution of dynamic porosity under the coupling effect of humidity and mass fraction of response components, and (b) is a schematic diagram of the distribution of equivalent pore size under the coupling effect of fiber areal density and normal pressure.

[0050] Figure 4 The following is a comparison chart of simulated measured values ​​and model predicted values ​​in an embodiment of the present invention, wherein (a) is a comparison chart of simulated measured values ​​and model predicted values ​​of dynamic porosity, (b) is a comparison chart of simulated measured values ​​and model predicted values ​​of equivalent pore size, (c) is a comparison chart of simulated measured values ​​and model predicted values ​​of permeability coefficient, and (d) is a comparison chart of simulated measured values ​​and model predicted values ​​of anti-clogging index.

[0051] Figure 5 The following are residual diagnostic results of a model according to an embodiment of the present invention: (a) is the residual distribution of dynamic porosity, (b) is the residual distribution of equivalent pore size, (c) is the residual distribution of permeability coefficient, and (d) is the residual distribution of anti-clogging index.

[0052] Figure 6 This is a ranking of the importance of variables in a multi-output prediction model according to an embodiment of the present invention.

[0053] Figure 7The following is a one-dimensional response curve of humidity and pressure according to an embodiment of the present invention, wherein (a) is a response curve of dynamic porosity and anti-clogging index to humidity, and (b) is a response curve of dynamic porosity and equivalent pore size to normal pressure.

[0054] Figure 8 This is a dynamic porosity response surface under the coupling effect of humidity and pressure in one embodiment of the present invention.

[0055] Figure 9 This is a multi-objective optimization distribution diagram of candidate material schemes according to an embodiment of the present invention. Detailed Implementation

[0056] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0057] A smart responsive geotextile pore structure prediction and parameter optimization method includes the following steps:

[0058] Step S1: Construct the variable system and engineering boundary.

[0059] The model input variables consist of four categories: environmental load parameters, material and structural parameters, service status parameters, and engineering condition parameters.

[0060] Environmental load parameters include relative humidity Normal pressure and temperature .in, It reflects the degree to which soil moisture content, rainfall infiltration, and groundwater recharge stimulate the response components. It reflects the compaction effect of soil pressure, construction compaction, and traffic loads on the fiber network. This is used to correct changes in the hygroscopic swelling of the responsive component and the flexibility of the polymer material. The responsive component refers to a functional material component incorporated or compounded into the geotextile fiber network that can undergo hygroscopic swelling, hydrophilicity regulation, or pore channel control with changes in humidity or temperature, including but not limited to hydrophilic modified fibers, water-absorbing and swelling polymer particles, temperature and humidity sensitive polymer components, or combinations thereof.

[0061] Material structural parameters include fiber areal density Initial micropore size Composite layer thickness and response component mass fraction .in, The larger the fiber network, the denser it is, which usually enhances soil retention, but may reduce initial porosity and permeability. The larger the size, the better the drainage capacity, but the greater the risk of particle penetration; The larger the structure, the better its structural support and durability, but the longer the vertical flow path. This indicates the proportion of the responding component to the total mass of the geotextile composite material. The larger the value, the stronger the humidity response capability. However, in high humidity environments, the hygroscopic expansion of the response component may compress the effective pore channels, thereby reducing the drainage capacity.

[0062] Service status parameters include service age and hydraulic gradient . Used to describe the degradation of material properties caused by ultraviolet radiation, oxidation, hydrochemical environments, and repeated wet-dry cycles; It is used to reflect the seepage driving force and is an important boundary for calculating the drainage capacity index.

[0063] Engineering operating parameters Discrete variables are used to represent the conditions, including seepage control, high-pressure bearing, drainage and storage, and comprehensive conditions. The purpose of setting engineering condition variables is to allow the same material parameter to generate different evaluation weights under different engineering objectives. For example, seepage control focuses more on pore convergence and soil retention, high-pressure bearing focuses more on compaction stability, and drainage and storage focuses more on permeability and anti-siltation balance.

[0064] It should be noted that in this invention, high humidity refers to relative humidity. ≥70% environmental conditions, high pressure refers to normal pressure Engineering load conditions of ≥200 kPa. The above numerical range is determined based on the following: when the relative humidity exceeds 70%, the hygroscopic expansion effect of the response component is significantly enhanced; when the normal pressure exceeds 200 kPa, the fiber network compaction effect is significantly aggravated. Both of these are important threshold conditions for the significant technical effects of this invention.

[0065] Output variables include dynamic porosity Equivalent aperture Permeability coefficient Drainage capacity index and anti-fouling index . It is a dimensionless representation of the overall openness of the pore structure; An equivalent scale that reflects water flow and particle passage; Reflects vertical seepage capacity; Used to Hydraulic gradient and composite layer thickness Overall, this contributes to the drainage effect of the project; Used to describe a material's ability to maintain channel stability under the combined effects of particle migration, pore shrinkage, and long-term service. Because can be , and Further calculations show that the prediction model mainly uses... , , and As a multi-output objective, i.e., the target response variable. The target response variable refers to the four output indicators that the multi-output prediction model aims to predict—dynamic porosity. Equivalent aperture Permeability coefficient and anti-fouling index During the model training and validation phase, the values ​​calculated by the pore-humidity-pressure mapping model in step S2 are used as simulated measured values ​​(label data).

[0066] The calibration and physical meaning of the model coefficients are explained below.

[0067] The empirical coefficients involved in this invention (including those hereinafter) ~ , ~ , ~ , , , , , , , , ~ (etc.) can all be determined through indoor experimental fitting or by referring to literature and engineering experience to determine initial values. The physical meaning, dimensions, value range and calibration method of each coefficient are as follows:

[0068] , , , : The constant term (dimensionless, ranging from 0.30 to 0.50) represents the intrinsic porosity of the geotextile under reference conditions. (Value range 0.15~0.40) is the contribution coefficient characterizing the initial micropore size to the reference porosity; (Value range 0.08~0.25) is the contribution coefficient characterizing the fiber areal density to the reference porosity; (Values ​​range from 0.03 to 0.12) represent the contribution coefficients of the composite layer thickness to the reference porosity. These coefficients are obtained by regression analysis of the material's physical parameters at the factory, using least squares fitting with at least 30 sets of reference porosity test data covering the engineering boundary.

[0069] , , : (Values ​​range from 0.3 to 0.9) represent the humidity linear response coefficient. (Value range 0.1~0.5) is the humidity quadratic response coefficient (characterizing the nonlinear enhancement effect in high humidity areas), and the two together describe the concentration effect and humidity nonlinear characteristics of the hygroscopic expansion of the response component; (Values ​​range from 0.001 to 0.010, dimensionless 1 / ℃) are temperature correction coefficients, reflecting the promoting effect of increased flexibility near the glass transition temperature of polymer materials on hygroscopic expansion. These coefficients were calibrated through free expansion tests under different humidity conditions.

[0070] , , : (Values ​​ranging from 0.02 to 0.15) represent the baseline compaction coefficient. (Values ​​range from 0.01 to 0.10) is a correction factor for the fiber areal density's sensitivity to compaction. (Value range 0.2~0.7) is the humidity softening effect coefficient (reflecting the softening effect of humidity on fiber modulus). The above coefficient is calibrated through compression and rebound tests.

[0071] , : (Values ​​range from 0.02 to 0.20) represent the composite layer thickness support coefficient. (Values ​​range from 0.02 to 0.15) represents the fiber skeleton density support coefficient. Both describe the structural resistance of the composite layer thickness and fiber areal density to pore shrinkage, respectively.

[0072] : (Dimensionless, value range 0.05~0.20) is the long-term damage control coefficient, which controls the aging and decay rate of materials and is calibrated by accelerated aging tests (UV aging, thermo-oxidative aging, wet-dry cycle).

[0073] , , , : (Dimensionless, value range 0.05~0.35) is the reduction factor of the effective channel after the response component expands due to moisture; (The dimensions are related to the permeability coefficient; the example value is...) value range ~ () is the seepage scale coefficient, which is related to the Kozeny-Carman seepage theory and reflects the intrinsic permeability characteristics of the material; (Dimensionless, value range 0.05~0.35) is the reduction factor of pressure compaction on permeability; (Dimensionless, value range 0.02~0.20) is the composite layer thickness path correction coefficient.

[0074] ~ : (Values ​​ranging from 0.20 to 0.60) are constant terms. (Values ​​ranging from 0.30 to 0.80) represent the dynamic porosity weight. (Values ​​range from 0.10 to 0.40) represent the adjustment weights for the response components. (Values ​​ranging from 0.02 to 0.20) represent the initial aperture weight. (Values ​​ranging from 0.05 to 0.25) represent the fiber areal density penalty weight. (Values ​​ranging from 0.01 to 0.10) represent the pressure penalty weight. (Values ​​range from 0.01 to 0.10) represent the age-related decay weights. These weights were determined based on the Analytic Hierarchy Process (AHP) combined with expert experience.

[0075] and : (The example value is 0.0008, and the unit is related to the areal density.) This is the fiber areal density deviation penalty coefficient. (Example value is 0.09, unit is related to thickness) This is the penalty coefficient for deviation of composite layer thickness. Both are determined based on the production process capability index.

[0076] Disturbance term , , : , Dimensionless, with a value range of 0.015 to 0.030; , Its dimension is μm, and its value ranges from 3 to 8 μm; , Dimensionless, ranging from 0.015 to 0.040. Among them, The mean is variance is The sign of the normal distribution is given. For each calculation, a value is randomly selected according to the corresponding normal distribution and substituted into the formula. The above disturbance parameters can also be determined based on the results of repeated indoor tests or statistical results of field monitoring errors.

[0077] Among the coefficients mentioned above, the example values ​​given in Table 1 are exemplary values ​​in a specific embodiment, used to illustrate the feasible implementation of the model of the present invention. In practical applications, adjustments can be made according to specific material types and engineering scenarios. The adjustment method is as follows: obtain at least 30 sets of test data covering the engineering working condition boundary, with the goal of minimizing the mean square error between the predicted value and the measured value, and re-optimize the coefficients using a genetic algorithm, gradient descent method, or nonlinear least squares method. The specific process of optimization iteration is as follows: ① Initialize the coefficient population; ② Substitute the coefficients into formulas (1) to (10) to calculate the output predicted value of each sample; ③ Calculate the mean square error between the predicted value and the measured value as the objective function; ④ Update the coefficients through selection, crossover, and mutation operations of the genetic algorithm (or backpropagation of the gradient descent method, iterative update of the nonlinear least squares method); ⑤ Repeat steps ② to ④ until the objective function value converges (the change is less than 10%). (or reach the maximum number of iterations). The above process yields optimal coefficient values ​​suitable for specific materials and engineering scenarios. The coefficients given in Table 1 are merely example initial values ​​under typical nonwoven geotextile conditions and do not constitute a limitation on the scope of protection of this invention.

[0078] Step S2: Establish a pore-humidity-pressure mapping model and generate a mechanism constraint dataset.

[0079] This process follows a progressive relationship from initial structure to environmental disturbances and then to hydraulic performance, and includes the following steps:

[0080] Step S21: Calculate the baseline porosity.

[0081] First, the baseline porosity of the material under undisturbed humidity and pressure conditions is determined. For nonwoven geotextiles, larger initial micropore sizes allow for more channels for water flow and particles between fibers; higher fiber areal density and composite layer thickness result in a denser fiber skeleton and generally lower pore openness. Therefore, this invention uses normalized initial micropore size, fiber areal density, and composite layer thickness to collectively express the baseline porosity:

[0082] (1)

[0083] In the formula, As the reference porosity, and These are the initial micropore sizes. The minimum and maximum engineering boundaries, and These are fiber areal densities. The minimum and maximum engineering boundaries, and The thickness of the composite layer is respectively The minimum and maximum engineering boundaries are defined. These engineering boundaries can be determined based on the geotextile product specifications, the range of indoor tests, or specific engineering design requirements. In this embodiment, the initial micropore size... The engineering boundary is 45–210 μm, that is , Fiber surface density The engineering boundary is 220–720 g / m², that is , Composite layer thickness The engineering boundary is 1.2–6.5 mm. This formula is not a simple empirical fit, but rather transforms the material's factory structural parameters into the initial state for subsequent dynamic response calculations. Its trend is consistent with the principles of nonwoven geotextile pore size and unit area mass control.

[0084] Formula (1) is based on the semi-empirical relationship between the porosity of nonwoven geotextiles and fiber areal density, initial micropore size, and thickness: fiber areal density The larger the diameter, the higher the fiber mass per unit area, the denser the fiber packing, and the lower the reference porosity. The sign before the item is negative; initial micropore size. The larger the diameter, the more abundant the initial channels between fibers, and the higher the reference porosity. The prefix is ​​positive; composite layer thickness The larger the fiber size, the thicker the fiber layer, resulting in a greater average distance between fibers at the same areal density. However, increased thickness also leads to a more uniform pore distribution within the fiber layer, and overall, porosity tends to decrease with increasing thickness. The item is preceded by a negative sign. The above trend is consistent with the principles for controlling the pore size and unit area quality of nonwoven geotextiles.

[0085] Step S22: Calculate the dynamic influence term.

[0086] After obtaining the baseline porosity, the model further describes the combined effects of humidity, pressure, composite layer thickness support, and age on the pore state. Humidity effects are mainly manifested through the hygroscopic expansion of the response components, often exhibiting nonlinear enhancement in high-humidity regions. Pressure effects are primarily manifested as fiber network compaction, and are influenced by both fiber areal density and humidity softening. Composite layer thickness and fiber areal density provide some structural support, offsetting some pore shrinkage. Age reflects the performance degradation caused by long-term wet-dry cycles, chemical corrosion, and material aging. These four effects are represented as humidity-responsive shrinkage terms. Pressure compaction item Structural support items and age decay coefficient .

[0087] (2)

[0088] (3)

[0089] (4)

[0090] (5)

[0091] In the formula, In response to the component mass fraction, The relative humidity is in decimal form. For reference temperature, and These are the normalized values ​​of fiber areal density and composite layer thickness, respectively. , ,when and When within the corresponding project boundary, and The values ​​of are all in the range of 0 to 1. This is a pressure normalization reference value, used to make... It becomes a dimensionless quantity; it can be taken as a representative value within the range of engineering pressure or the maximum design normal pressure. For service age, Set the maximum age for the model.

[0092] After the above treatment, humidity, pressure, composite layer thickness and age are no longer isolated variables, but together constitute the intermediate state of pore dynamic evolution.

[0093] In formula (2), This expresses the concentration effect of the hygroscopic swelling of the response component. The higher the humidity, the greater the moisture absorption) and the nonlinear effect of humidity (linear expansion is the main phenomenon in low humidity areas, while the swelling of the polymer network intensifies and exhibits secondary enhancement in high humidity areas, i.e., when H>0.7). This is a temperature correction term, reflecting the promoting effect of increased polymer chain segment movement and increased free volume on hygroscopic expansion when the temperature rises.

[0094] In formula (3), It reflects the nonlinear characteristics of fiber network compaction—the logarithmic form describes the mechanical behavior of rapid deformation in the initial stage (low pressure zone) and gradual compaction in the later stage (high pressure zone), which conforms to the classical mechanical model of fiber assembly compression. This reflects the softening effect of humidity on fiber modulus; the higher the humidity, the lower the fiber modulus, and the greater the degree of compaction under the same pressure.

[0095] In formula (4), and All are normalized dimensionless quantities, expressed by a linear combination of the composite layer thickness and fiber areal density to represent the structural resistance to pore shrinkage. The greater the thickness and the higher the areal density, the stronger the fiber skeleton's ability to resist external loads, and the greater the structural support. The larger.

[0096] In formula (5), The logarithmic time scale was used to describe the aging and degradation of materials, reflecting the engineering experience pattern that geotextiles experience rapid performance degradation in the early stage of service (0-180 days) and gradual degradation in the later stage (180-365 days and beyond). Control the overall attenuation level. The larger the value, the more significant the long-term degradation.

[0097] Step S23: Synthesize dynamic porosity.

[0098] By superimposing the aforementioned intermediate states onto the baseline porosity, the dynamic porosity can be obtained. Since the porosity of engineering materials cannot increase or decrease indefinitely, the model employs upper and lower limit cutoff functions to control the calculation results within a reasonable range, and includes a perturbation term to represent un-modeled construction discreteness and material micro-differences.

[0099] (6)

[0100] In the formula, For dynamic porosity, This is the porosity disturbance term. and These are the minimum and maximum values ​​of the dynamic porosity boundary, respectively. This is a cutoff function for upper and lower limits, used to restrict calculated values ​​within a preset physical range. Its definition is:

[0101]

[0102] in, . and The value can be determined based on the type of geotextile material, indoor porosity test results, or the allowable range of the engineering design; in this embodiment, for nonwoven smart responsive geotextiles, the value can be taken as... , .

[0103] Formula (6) uses the reference porosity (Initial material condition) and humidity-responsive shrinkage term (Porosity reduction), pressure compaction item (Reduced porosity), structural support items (Porosity compensation) superimposed, then multiplied by the age-related decay coefficient. (Long-term service reduction), finally apply a random perturbation term. (Simulates construction discreteness and material microstructure differences), and uses the clip function to ensure that the dynamic porosity does not deviate from the material's physical range. ~ This formula embodies the complete pore evolution chain of "initial structure → environmental disturbance → long-term service → construction discreteness".

[0104] Formula (6) is the core of the pore-humidity-pressure mapping model: the humidity response contraction term. and pressure compaction item Reduce the degree of pore opening, structural support items Provides structural compensation, age-related degradation coefficient Reduce the value of long-term service status. The function ensures that the dynamic porosity does not deviate from the material's physical range.

[0105] Step S24: Calculate the equivalent pore size and permeability coefficient.

[0106] Dynamic porosity itself cannot directly represent filtration and drainage performance; it needs to be further transferred to the pore size and seepage scales. This invention assumes that the equivalent pore size adjusts to the square root of the relative change in porosity, while also considering the occupancy of effective water flow channels by the hygroscopic components of the response components; the permeability coefficient is determined by pore connectivity, pore size, pressure reduction, and thickness path.

[0107] (7)

[0108] (8)

[0109] In the formula, For equivalent aperture, For aperture disturbance term, and Equivalent aperture The minimum and maximum physical boundaries; Permeability coefficient, Composite layer thickness path correction factor Used to characterize the effect of increased composite layer thickness on vertical seepage path and permeability.

[0110] In formula (7), This reflects the square root relationship between porosity changes and equivalent pore size—when the dynamic porosity increases relative to the reference porosity, the equivalent pore size increases proportionally to the square root. This relationship is based on the capillary bundle model (pore size is proportional to the square root of porosity). This reflects the pore size reduction caused by the swelling of the responding component occupying the pore channels. The larger the value or the higher the humidity, the more significant the reduction; disturbance term Simulate the discreteness of materials.

[0111] Formula (8) is based on the Kozeny-Carman equation. For geotextiles, porosity Usually much smaller than 1, therefore Approximately 1, retain The term is used to express the amplifying effect of pore connectivity on permeability; The term directly derives from the Kozeny-Carman theory's understanding of the controlling effect of the square of the pore size on the permeability coefficient; the denominator contains... These characters respectively characterize the pore channel compression caused by pressure compaction ( The reduction in permeability due to the increase in composite layer thickness path (item) and (item) (Item). The above treatment is consistent with the understanding of pore connectivity in geotextile permeability tests and filter layer design.

[0112] It should be noted that, and Equivalent aperture The calculated cutoff boundary is used to limit the equivalent pore size, after corrections for humidity response, porosity change, and disturbance, from exceeding a reasonable physical range. In this embodiment, the equivalent pore size... The calculation cutoff boundary is taken , The above values ​​can be adjusted based on the geotextile product specifications, indoor equivalent pore size test results, or engineering reverse filtration requirements.

[0113] The cubic term is used to express the amplification effect of pore connectivity on permeability. The squared term is used to express the control effect of pore size on water permeability; the pressure term and composite layer thickness term in the denominator are used to characterize the reduction effect of pressure compaction and composite layer thickness path increase on permeability, respectively. This treatment is consistent with the understanding of pore connectivity and pore size control in geotextile permeability tests and filter layer design.

[0114] Step S25: Calculate drainage capacity and anti-clogging index.

[0115] In engineering evaluation, the permeability coefficient needs to be converted into drainage capacity by considering the hydraulic gradient and drainage path; siltation resistance requires simultaneous consideration of pore openness, response component regulation capacity, initial pore size, fiber density, pressure, and long-term aging. Based on this approach, the drainage capacity index... and anti-fouling index Defined as:

[0116] (9)

[0117] (10)

[0118] In the formula, For hydraulic gradient; Here is the initial aperture normalized value, where, ; For the anti-fouling index disturbance term, This represents the maximum engineering boundary value of the normal pressure. Anti-clogging index. The value ranges from 0 to 1. A larger value indicates a stronger ability of the geotextile to maintain effective drainage channels under the influence of particle migration, pore shrinkage, and long-term service, i.e., stronger resistance to clogging; a smaller value indicates that the pore channels are more prone to blockage or drainage attenuation. Therefore, This indicates the lowest resistance to clogging or the highest risk of clogging. This indicates the highest resistance to clogging or the lowest risk of clogging.

[0119] In formula (9), the amplification factor of 1000 is a dimension adjustment factor, the purpose of which is to make... The values ​​fall within the order of magnitude convenient for engineering readings. Because... The dimension is m / s. Dimensionless The dimension of is m, therefore Its dimension is 1 / s, and its value is usually 1 / s. ~ The magnitude, multiplied by 1000, is converted to ~ The magnitude is used for comparison in engineering projects. This magnification factor only serves to scale the values ​​and does not affect the relative ranking of drainage capacity among different schemes or the optimization decision.

[0120] Formula (10) is constructed based on the principle of coordinating soil retention, permeability, and anti-clogging properties in classical reverse filtration theory. The physical meanings of each term in the formula are as follows: The constant term reflects the material's intrinsic resistance to clogging; The item reflects the positive contribution of the degree of pore openness to resist clogging (the more open the pores, the less likely the particles are to clog). The item reflects the contribution of the response component's humidity regulation ability to combat clogging (appropriate response components can dynamically regulate pores). The larger the initial aperture, the less likely it is to become clogged; This item reflects the negative effect of reduced pore channels when the fiber areal density is too high; The item reflects the compression effect of pressure compaction on the channel; This reflects the decline in the body's ability to resist clogging over long-term aging; This represents a random perturbation term. The formula reflects an engineering trade-off: moderately open pores, appropriate response components, and larger initial pore sizes are beneficial for reducing the risk of clogging, but excessive fiber areal density, high normal pressure, and long-term aging can weaken the stability of pore channels.

[0121] Formulas (6), (7) and (10) , and These are used to characterize the material discreteness, construction disturbance, and test error that are not explicitly modeled in the calculation of porosity, equivalent pore size, and anti-clogging index, respectively. In this embodiment, all three types of disturbance terms can be set as normal random disturbances with a mean of 0, i.e. , , .in, A value of 0.015 to 0.030 is acceptable. 3–8 μm is acceptable. The value can be taken as 0.015 to 0.040. Each time the value is substituted into the formula for calculation, a specific value is randomly selected according to the corresponding normal distribution as the perturbation value for that calculation. To avoid the calculation results exceeding the physical range due to random perturbation, the dynamic porosity, equivalent pore size, and anti-clogging index after perturbation are respectively determined by... The function is limited to the corresponding boundary range. The above disturbance parameters can also be determined based on the results of repeated indoor tests or the statistical results of field monitoring errors.

[0122] The anti-clogging index integrates pore openness, response component regulation capability, initial pore size, fiber density, pressure, and long-term aging, reflecting an engineering trade-off: moderately open pores, appropriate response components, and larger initial pore size are beneficial to reducing the risk of clogging, but excessive fiber areal density, high normal pressure, and long-term aging will weaken the stability of pore channels.

[0123] Thus, the model has completed the continuous transfer from the initial material structure to environmental disturbances, and then to dynamic porosity, equivalent pore size, permeability and anti-clogging performance.

[0124] Step S26: Generate the mechanism constraint dataset. The mechanism constraint dataset refers to the sample data used to train the subsequent multi-output prediction model, which is calculated by the pore-humidity-pressure mapping model according to the physical mechanism.

[0125] Based on the above mapping relationship, samples are generated within the preset engineering boundaries. Specifically, relative humidity... Normal pressure ,temperature Fiber surface density Initial micropore size Composite layer thickness Mass fraction of response components Service age and hydraulic gradient For continuous variables, candidate samples are generated by Latin hypercube sampling or uniform random sampling within their respective engineering boundaries; for engineering operating parameters, discrete random sampling or assignment according to the proportion of engineering parts are used to form samples of seepage control operating conditions, high pressure bearing conditions, drainage and storage operating conditions and comprehensive operating conditions.

[0126] The sample size can be determined based on the variable dimensionality, model complexity, and engineering calculation accuracy requirements. In this embodiment, there are 9 continuous input variables and 1 discrete engineering condition variable, with a sample size of 1800 sets to cover the material response characteristics under conditions of low humidity in the dry season, high humidity in the rainy season, groundwater recharge, road structure layer compaction, and different drainage driving forces. Alternatively, the initial sample size can be determined according to the principle of no less than 100-200 equivalent sampling points for each continuous variable, and the sample size can be further increased based on the convergence of prediction errors.

[0127] For each set of input variables obtained from sampling, substitute them into formulas (1) to (10) in sequence to calculate the dynamic porosity. Equivalent aperture Permeability coefficient Drainage capacity index and anti-fouling index The input variables and corresponding output results are then paired to form sample pairs. For intermediate results that exceed engineering boundaries or physical limits, a different approach is adopted. The function performs truncation; samples with negative permeability coefficients, invalid pore sizes, missing values, or that do not meet engineering logic are discarded or resampled. The final mechanistic constraint dataset is obtained as follows:

[0128]

[0129] in, For the first The input variable vector of each sample, For the corresponding output variable vector, This represents the number of valid samples.

[0130] Step S3: Build and train a multi-output prediction model.

[0131] The pore-humidity-pressure mapping model provides a basic path for data generation and mechanistic constraints. However, in practical engineering, the coupling between parameters does not necessarily conform entirely to explicit functional forms. For example, the sensitivity of pressure compaction under high humidity conditions may change with the fiber density, and the adjustment of pore size by response components may also be affected by both the initial micropore size and the composite layer thickness. Therefore, in the prediction stage, this invention does not directly use a single analytical expression to replace all engineering responses, but instead uses input variables to form a feature vector. The dynamic porosity, equivalent pore size, permeability coefficient, and anti-clogging index are combined into an output vector. Numerical variables are standardized, and engineering condition parameters are encoded using one-hot encoding to avoid unreasonable impacts on the model due to differences in dimensions and the encoding order of discrete variables. Eigenvectors and output vector Represented as:

[0132] (11)

[0133] (12)

[0134] In the formula, These are engineering operating parameters, which, after being thermally encoded, are used to characterize seepage control conditions, high-pressure bearing conditions, drainage and storage conditions, and comprehensive conditions.

[0135] This embodiment employs an extreme random tree model as a multi-output prediction model. The multi-output prediction model described in this invention refers to a regression model capable of simultaneously predicting multiple target variables within the same model. The extreme random tree model is a specific implementation of this multi-output prediction model, composed of multiple random regression trees, with the final output obtained by averaging the prediction results of these trees. This model is not a replacement for the mechanism formula in step S2, but rather a continuation: the mechanism formula ensures that the direction of sample change has engineering significance, while the extreme random tree model is used to learn the nonlinear response surface under the combined effects of multiple variables. Compared to a single decision tree, the extreme random tree model can reduce variance through averaging multiple random regression trees; compared to linear regression, it can express the interaction of humidity thresholds, nonlinear pressure compaction, and material parameters; compared to deep neural networks, it has lower parameter tuning requirements for medium-scale engineering data and can analyze variable contributions through permutation importance analysis. The extreme random tree model is suitable for integration with geotechnical engineering reliability analysis and intelligent sensing frameworks to identify controlling factors from multi-source variables.

[0136] The extreme random tree model can be represented as:

[0137] (13)

[0138] In the formula, To integrate prediction functions, The number of random regression trees, For the first Random regression trees, The parameters for random partitioning, sample perturbation, and feature perturbation are defined for the tree. The multi-output model stores the mean vectors of the four target variables at each leaf node, thus preserving the correlation between different performance metrics. In other words, the model does not predict in isolation. , , and Instead, these responses are estimated synchronously under the same material-environment combination, which is more consistent with the engineering fact that pore structure and hydraulic performance change synergistically.

[0139] In this embodiment, the hyperparameter of the extreme random tree model is set as: the number of random regression trees. The node splitting criterion adopts the squared error criterion, the maximum tree depth is not limited, the minimum leaf node sample size is 2, the minimum node splitting sample size is 2, and the number of features participating in the random selection for each split is the square root of the total number of features. The random seed is set to 2026. The above hyperparameters can also be adjusted according to the training sample size and cross-validation results.

[0140] The dataset generated in step S2 is randomly divided into a training set and a test set in an 8:2 ratio for model training and validation. The training set is used for model parameter learning, and the test set is used for evaluating the model's generalization ability. Alternatively, cross-validation methods (such as 5-fold cross-validation) can be used to fine-tune the model's hyperparameters based on the total sample size. After model training, the following evaluation metrics are used: coefficient of determination. Mean absolute error Root mean square error :

[0141]

[0142]

[0143]

[0144] In the formula, For the first Simulated measured values ​​for each sample, For the first The model prediction value for each sample. The test set simulates the mean of the measured values. This represents the number of samples in the test set. The closer the value is to 1, the stronger the model's explanatory power. Reflects the mean absolute deviation; It is more sensitive to larger errors. For engineering applications, if and The predicted trends are reliable, and even if local errors exist, they can be used for construction monitoring and early warning, as well as material selection.

[0145] In this embodiment, the simulated measured values ​​refer to the output variable values ​​calculated by the pore-humidity-pressure mapping model (formulas (1)~(10)) in step S2, which are used as label data for training and validating the multi-output prediction model. In subsequent applications, this label data can be replaced with measured data obtained from indoor permeability tests, gradient ratio tests, or on-site monitoring to further calibrate the model.

[0146] Providing only the prediction accuracy is insufficient to support materials design; it is also necessary to answer which variables dominate the model's judgments, i.e., to explain the basis of the model's judgments. Therefore, this invention employs substitution importance analysis:

[0147] (17)

[0148] In the formula, For the first The importance of each variable The average coefficient of determination for the original test set. To disrupt the first The average coefficient of determination is obtained after considering all variables. If the model performance significantly decreases after a certain variable is shuffled, it indicates that the variable contributes more to the multi-objective prediction. This method avoids bias caused by simply relying on the number of splits within the tree model and helps to transform black-box prediction results into engineering judgments that can be used for materials design.

[0149] Step S4: Construct a multi-objective optimization model and solve it.

[0150] The prediction model addresses how performance changes under given material and environmental conditions. Engineering design further requires determining optimal material parameter values ​​under specific engineering conditions. Therefore, this invention embeds the trained prediction model into the optimization process, using fiber areal density, initial micropore size, composite layer thickness, and response component mass fraction as design variables. ,Right now =[ , , , ];Will , , , , and As boundary conditions given by the engineering scenario, the optimization objective is not to maximize a single permeability coefficient, but to achieve a balance between drainage capacity, anti-fouling capability, target porosity, and manufacturing feasibility. If only increasing... or reduce Although it can improve However, this may weaken soil conservation; if only the soil conservation capacity is increased... While this can enhance the humidity response, it may excessively shrink porosity in high-humidity environments. Therefore, the objective function is set according to the principle of coordinating soil conservation, permeability, and anti-siltation, and a comprehensive optimization objective is established. :

[0151] (18)

[0152] In the formula, As a normalized benchmark for permeability coefficient, For the target porosity, To allow for porosity deviation scale, To create feasible penalties, , , and The weights are respectively for permeability, anti-clogging ability, pore stability, and manufacturing feasibility. ( / ,2) is used to limit the infinite contribution of the permeability coefficient to the objective function when it is too high, so that the optimization result is not biased towards schemes with excessively large pore size. With this setting, the optimization function (i.e., formula (18)) rewards better drainage capacity and anti-clogging performance, while penalizing schemes that deviate from the target porosity and manufacturing center value.

[0153] The feasibility penalty for manufacturing is expressed as follows:

[0154] (19)

[0155] In the formula, and The recommended manufacturing center values ​​represent the fiber areal density center value and the composite layer thickness center value, which are relatively easy to control and have good engineering economy during material production and on-site laying. and To determine the manufacturing deviation penalty coefficient, a recommended manufacturing center value can be determined based on commonly used geotextile product specifications, production experience, indoor test results, or engineering design requirements. In this embodiment, for nonwoven smart responsive geotextiles used in municipal drainage backfiltration, a value of [value missing] can be used. , When the fiber areal density or composite layer thickness of the candidate solution deviates significantly from the above-mentioned center value, This increases the overall optimization target value of the scheme, thereby reducing it.

[0156] This item is used to express constraints related to material production, laying thickness control, and engineering economics. The boundary value of the optimization variables is set to 260–680 g / m². 2 55~190 m, 1.6–5.8 mm, and 0.08–0.32 correspond to the engineering feasible ranges for fiber areal density, initial micropore size, composite layer thickness, and mass fraction of the response component, respectively. Because As indirectly given by the prediction model, the objective function may exhibit non-convex and non-smooth characteristics. This invention employs a differential evolution algorithm for global search, whose mutation, crossover, and selection processes are suitable for handling such implicit objective functions. In this embodiment, the key parameters of the differential evolution algorithm are set as follows: population size... ,in To optimize the variable dimensions, in this embodiment... Maximum number of iterations: 300; Mutation factor Take values ​​between 0.5 and 0.8; crossover probability Take 0.7; convergence tolerance is taken as... The optimization variable boundaries are fiber areal density of 260–680. Initial micropore size 55–190 Composite layer thickness: 1.6–5.8 mm The response component mass fraction is 0.08–0.32. When the optimal objective function value changes less than the convergence tolerance for several consecutive iterations, or when the maximum number of iterations is reached, the iteration stops and the optimal material structure parameter combination is output. The above algorithm parameters can be adjusted according to the required computational accuracy and sample size.

[0157] Step S5: Output the optimal combination of material parameters.

[0158] The optimized combination of material structure parameters is output to guide the material design, test program development, and construction monitoring and control of intelligent responsive geotextiles.

[0159] Through the above steps, the model forms a complete chain of mechanism mapping, data learning, and goal optimization.

[0160] It should be noted that the empirical coefficients involved in this invention (including...) ~ , ~ , ~ , , , , , , , , ~ (etc.) can all be determined through indoor experimental fitting or by referring to literature and engineering experience to obtain initial values ​​(setting initial values ​​based on existing literature or engineering experience). For indoor experimental fitting, for specific types of smart responsive geotextiles, permeability tests, compression creep tests, and gradient ratio tests are carried out under different humidity and pressure conditions according to the "Test Procedure for Geosynthetics" to obtain multiple sets of results. The optimal coefficients were calculated using the nonlinear least squares method based on the measured data. The initial values ​​based on engineering experience refer to the initial values ​​set based on engineering experience. When experimental data is lacking, the example initial values ​​given in Table 1 can be used as a reference. These initial values ​​are derived from the engineering experience of typical nonwoven geotextiles (fiber surface density 300~500g / m², initial pore size 100~150 μm).

[0161] The model coefficients given in Table 1 are exemplary initial values ​​in a specific embodiment, used to illustrate the feasible implementation of the model of the present invention, and do not limit the range of values ​​of the present invention in other material types or engineering scenarios.

[0162] Table 1. Example values ​​for model coefficients.

[0163]

[0164] The above coefficients are merely an exemplary value for this invention, and can be adjusted according to the specific material type and engineering scenario in practical applications. The adjustment method is as follows: obtain at least 30 sets of test data covering the engineering working condition boundary, and optimize the coefficients again using a genetic algorithm, gradient descent method, or nonlinear least squares method with the goal of minimizing the mean square error between the predicted and measured values.

[0165] The above method will be further explained through a specific application example below.

[0166] Taking a municipal road underpass and slope drainage renovation project in a coastal area of ​​Shenzhen as a case study, this paper verifies the prediction and optimization process of the established model. This project is located in a coastal urban area with high rainfall, high humidity, and significant groundwater recharge. The drainage filter layer needs to simultaneously meet the requirements of drainage on the outer sidewalls and bottom slab of the underpass, drainage at the slope toe, and fine particle interception. Based on the project boundaries, 1800 case samples were constructed, with humidity ranging from 35% to 98%, normal pressure from 20 to 420 kPa, and temperature from 5 to 38℃. Material parameters included fiber areal density of 220–720 g / m², initial micropore size of 45–210 μm, composite layer thickness of 1.2–6.5 mm, and response component mass fraction of 0.05–0.35; service life of 0–365 days; and hydraulic gradient of 0.2–2.6. These ranges cover the material response characteristics under conditions of low humidity in the dry season, high humidity in the rainy season, groundwater recharge, road structure layer compaction, and different drainage driving forces.

[0167] Based on the combination of humidity, pressure, and hydraulic gradients, the engineering conditions are divided into seepage control conditions, high-pressure bearing conditions, drainage and storage conditions, and comprehensive conditions. Specifically, the drainage and storage conditions and seepage control conditions are primarily considered for the underpass's bottom slab and outer sidewalls; the high-pressure bearing conditions are considered for the pavement structure layers and backfill compaction zones; drainage filtration and anti-clogging are considered for the slope toes and the area around blind drains; and the comprehensive condition corresponds to the general service condition. This classification helps to compare the impact of different engineering parts and design objectives on material parameter selection within a unified variable system.

[0168] The calculation process includes three steps: First, case samples are generated based on the boundaries of the Shenzhen engineering scenario, and boundary checks are performed on porosity, equivalent pore size, permeability coefficient, and anti-clogging index to remove outliers that do not conform to physical meaning; Second, the samples are divided into training and testing sets, a multi-output prediction model is established, and the model's ability to identify pore evolution laws is verified through evaluation indicators, residual analysis, and variable importance tests; Finally, for the working conditions of drainage and water storage projects, the differential evolution algorithm is used to search for material parameters, and the results are compared with random candidate schemes to verify the applicability of the optimization results.

[0169] To analyze data distribution and variable coupling relationships, this embodiment plots heatmaps of the distribution of target response variables and the correlation between engineering variables, as shown below. Figure 1 and Figure 2 . Figure 1 This indicates that the dynamic porosity and anti-clogging index are approximately unimodal, the equivalent pore size covers the range from low pore size to high pore size, and the permeability coefficient is right-skewed, which is consistent with the characteristics of nonlinear amplification of seepage index. Figure 2 The results showed that the initial micropore size was strongly positively correlated with the equivalent pore size, permeability coefficient, and drainage capacity index, while the fiber surface density, normal pressure, and humidity were negatively correlated with the degree of pore opening to varying degrees.

[0170] To visually demonstrate the influence of humidity, pressure, and material structural parameters on the pore response, a scatter plot of the pore response mechanism was created. (See attached image) Figure 3 Among them, (a) reflects the coupling effect of humidity and response components on dynamic porosity, and (b) reflects the discrete characteristics of the equivalent pore size changing with fiber surface density under pressure compaction conditions, which can be used as the graphical basis for interpreting subsequent model prediction results.

[0171] Table 2 Evaluation Indicators for Multi-Output Prediction Models

[0172]

[0173] As shown in Table 2, the equivalent aperture test set A value of 0.9756 indicates a clear relationship between the initial pore size, porosity, and response components on the equivalent pore size, demonstrating that the model can effectively recover the pore size evolution pattern. (Dynamic porosity test set) A value of 0.8224 indicates that the model can identify the dominant trends of humidity expansion, pressure compaction, and structural support. (Permeability coefficient test set) It is 0.8048, because Simultaneously affected Summation of cubic terms The squared term can amplify local errors nonlinearly, but the overall prediction still has engineering reference value. Anti-clogging index test set. The value was 0.6936, lower than other indicators, because... Integrating multiple factors such as pore structure, response components, pressure, and age, its physical meaning is closer to the comprehensive evaluation index, and it exhibits stronger random perturbation and nonlinear coupling. The model coefficients used in this case are shown in Table 1.

[0174] See the comparison between simulated measured values ​​and predicted values. Figure 4 The residual diagnosis results are shown in Figure 5 .from Figure 4 It can be seen that the equivalent pore size point cloud is basically distributed along the ideal prediction line. The dynamic porosity and permeability coefficient fit well in the low and medium value range, while the dispersion increases in the high permeability coefficient range. Figure 5 The results show that the residuals of each indicator are generally distributed around zero, without any obvious unidirectional shift, indicating that the model does not systematically overestimate or underestimate. For construction monitoring scenarios, this model can be used to determine whether the pore structure has entered a high-risk decay zone; for material design scenarios, it can be used to quickly screen candidate parameter combinations.

[0175] See the ranking of variable importance. Figure 6 It should be noted that, Figure 6 The middle vertical axis "average" Loss (i.e.) The value of may be greater than 1. When a key variable is completely shuffled, the model's predictive ability may become extremely poor, and the permuted coefficient of determination may also be significantly reduced. Negative values ​​may occur, leading to >1. This reflects the extreme importance of this variable to the model's predictions, which is a normal phenomenon in permutation importance analysis, rather than a physical contradiction. Figure 6 The results show that the initial micropore size contributes the most to the multi-objective prediction, followed by the mass fraction of the response component, humidity, normal pressure, and fiber areal density. This is consistent with the model mechanism. Directly determines the initial hydraulic channel dimensions. and Together they determine the intensity of the moisture expansion response. Controlling the degree of compaction, The thickness, temperature, age, and hydraulic gradient of the composite layer have relatively low importance. This does not mean that these variables are meaningless, but rather that within the current sample range, their effects are more reflected through auxiliary terms.

[0176] After verifying the model's accuracy, the influence of typical environmental variables on the pore response was further analyzed. The one-dimensional response curves for humidity and pressure are shown below. Figure 7 The dynamic porosity response surface under humidity-pressure coupled conditions is shown in [reference needed]. Figure 8 .

[0177] It should be noted that, Figure 7 (b) and Figure 8 This reflects the physical laws under different conditions: Figure 7 (b) is a univariate response curve, plotted under the condition of humidity. The humidity is fixed at 50% (medium humidity conditions), and all other parameters are taken as baseline values, except for the normal pressure. Changes; under these conditions, due to the low humidity (50%), the hygroscopic expansion effect of the response component is not significant, and the compaction effect of normal pressure on porosity is mitigated by the structural support term. The partial offsetting effect results in a relatively flat curve. Figure 8 This is a humidity-pressure coupled response surface, with humidity as the horizontal axis. (0.40~0.98), with the vertical axis representing normal pressure. (40~400kPa), color indicates dynamic porosity When observed along the vertical axis from the bottom (40 kPa) upwards (400 kPa), the actual reflection is the difference in pressure effect under different humidity levels—under high humidity conditions (the right area in the figure), the responding components absorb moisture and expand, and the fiber modulus decreases. At this time, applying high pressure amplifies the compaction effect dramatically, leading to a significant decrease in porosity (the color changes from dark blue to light yellow). The two figures are under different conditions, but there is no contradiction.

[0178] Depend on Figure 7 It can be seen that, with other parameters remaining constant, increased humidity leads to a decrease in dynamic porosity, while the anti-clogging index changes relatively slowly. This indicates that the responsive component occupies part of the pore space after absorbing moisture, but at the same time improves local particle migration and pore regulation capabilities. The pressure response curve shows that with increasing normal pressure, the equivalent pore size continues to decrease, and the dynamic porosity also undergoes compaction decay. Figure 8 Further analysis reveals that high humidity and high pressure areas are sensitive areas where porosity rapidly declines, and these areas should be given special attention, particularly the drainage layers behind tunnel linings and the drainage filter layers at the toe of slopes.

[0179] Under the operating conditions of a drainage and water storage project with 82% humidity, 180 kPa normal pressure, 22℃ temperature, 120 days of service life, and a hydraulic gradient of 1.4, the optimized fiber areal density was 448.41. Initial micropore size: 186.80 The composite layer thickness is 2.94 mm, and the mass fraction of the response component is 0.146. Substituting these parameters into the prediction model, we obtain a dynamic porosity of 0.3703, an equivalent pore size of 150.51 μm, and a permeability coefficient. m / s, anti-clogging index 0.5527. This scheme does not choose the lowest fiber areal density or the highest responsive component, but rather strikes a trade-off between drainage capacity and pore stability. A larger initial micropore size ensures drainage channels, a moderate fiber areal density prevents excessive pore closure, an approximately 3 mm thickness provides necessary structural support, and an appropriate amount of responsive component enhances humidity regulation and anti-clogging capabilities. The multi-objective distribution of candidate material schemes is shown in [reference needed]. Figure 9 .

[0180] Depend on Figure 9 It is evident that the relationship between permeability coefficient and anti-clogging index is not a simple linear one. Some schemes exhibit high permeability coefficients but low anti-clogging indices, indicating that excessively large pores may weaken particle interception and channel stability; conversely, some schemes have high anti-clogging indices but insufficient permeability coefficients, suggesting that excessive pore shrinkage is detrimental to drainage and water storage. As pressure gradually increases and humidity rises in stages, porosity exhibits a process of first decreasing and then slowly recovering, indicating that compaction, rainfall, and drainage status need to be monitored jointly during construction. This Shenzhen engineering case study demonstrates that the design focus of intelligent responsive geotextiles should be on controllable response rather than strong response. For the drainage and water storage objectives of underpasses, the response components should be controlled within a range that improves anti-clogging without significantly sacrificing permeability; for road backfill compaction and locally high-pressure areas, fiber surface density and composite layer support capacity should be increased; for the outer sidewalls and slope toe filter areas, it is necessary to balance equivalent pore size control with long-term drainage capacity. It is important to emphasize that the case study conclusions reflect the model response patterns under the boundary conditions of a high-humidity coastal engineering scenario in Shenzhen. Their value lies in illustrating the methodological chain and parameter sensitivity, rather than directly providing product specifications that can be used for engineering procurement without experimental verification. These differences demonstrate that a unified model framework can adjust weights and boundaries according to the engineering location and design objectives, and can be further calibrated into a material parameter design method for specific municipal scenarios after obtaining indoor experimental or field monitoring data.

[0181] Finally, it is necessary to state that the above embodiments are only used to further illustrate the technical solution of the present invention in detail, and should not be construed as limiting the scope of protection of the present invention. Any non-essential improvements and adjustments made by those skilled in the art based on the above content of the present invention shall fall within the scope of protection of the present invention.

Claims

1. A smart responsive geotextile pore structure prediction and parameter optimization method, characterized in that, Includes the following steps: Step S1: Obtain input variables, including environmental load parameters, material structure parameters, service status parameters, and engineering condition parameters; Step S2: Establish a pore-humidity-pressure mapping model, which is used to calculate dynamic porosity, equivalent pore size, permeability coefficient and anti-clogging index based on the input variables, and generate a mechanism constraint dataset; Step S3: Construct a multi-output prediction model and train it using the mechanism constraint dataset to obtain the trained multi-output prediction model; the multi-output prediction model takes the input variables as input and takes dynamic porosity, equivalent pore size, permeability coefficient and anti-clogging index as synchronous outputs; Step S4: Construct a multi-objective optimization model, using the trained multi-output prediction model as the objective function calculator, with the material structure parameters as design variables, and with the environmental load parameters, service state parameters, and engineering condition parameters corresponding to the preset engineering scenario as boundary conditions, solve the multi-objective optimization model to obtain the optimal combination of material structure parameters; Step S5: Output the optimal combination of material structure parameters.

2. The optimization method according to claim 1, characterized in that: The environmental load parameters include: relative humidity. Normal pressure and temperature The material structural parameters include: fiber areal density. Initial micropore size Composite layer thickness and response component mass fraction The service status parameters include: service age. and hydraulic gradient The engineering operating parameters The conditions are represented by discrete variables, including seepage control conditions, high pressure bearing conditions, drainage and storage conditions, and comprehensive conditions. The normal pressure range corresponding to the high pressure bearing conditions is 200 to 500 kPa.

3. The optimization method according to claim 2, characterized in that: The step S2, establishing the pore-humidity-pressure mapping model, specifically includes: based on the initial micropore size... Fiber surface density and composite layer thickness Calculate the reference porosity using the following formula. : ； In the formula, For constant terms, , and The contributions of initial micropore size, fiber areal density, and composite layer thickness to the reference porosity are characterized, respectively. and These are the initial micropore sizes. The minimum and maximum engineering boundaries, and These are fiber areal densities. The minimum and maximum engineering boundaries, and The thickness of the composite layer is respectively The minimum and maximum engineering boundaries.

4. The optimization method according to claim 3, characterized in that: The step S2 of establishing the pore-humidity-pressure mapping model also includes: calculating the humidity response shrinkage term using the following formula. Pressure compaction item Structural support items and age decay coefficient : ； ； ； ； in, In response to the component mass fraction, The relative humidity is in decimal form. For reference temperature, and These are the normalized values ​​for fiber areal density and composite layer thickness, respectively. As a pressure normalization benchmark, For service age, Set a maximum age for the model. The humidity linear response coefficient is... The second-order humidity response coefficient is... This is a temperature correction factor. As the baseline compaction coefficient, This is a correction factor for the fiber areal density's sensitivity to compaction. The humidity softening effect coefficient, , These are coefficients used to characterize the supporting effect of the composite layer thickness and the compacting effect of the fiber skeleton, respectively. This is a coefficient used to control the degree of long-term damage.

5. The optimization method according to claim 4, characterized in that: The step S2 of establishing the pore-humidity-pressure mapping model also includes: calculating the dynamic porosity using the following formula. : ； in, This is the porosity disturbance term. and These are the minimum and maximum values ​​of the dynamic porosity boundary, respectively. The upper and lower bound cutoff function is defined as follows: 。 6. The optimization method according to claim 5, characterized in that: The step S2 of establishing the pore-humidity-pressure mapping model also includes: calculating the equivalent pore size using the following formula. and permeability coefficient : ； ； in, In response to the reduction factor of effective channels after component swelling due to moisture, For aperture disturbance term, and Equivalent aperture The minimum and maximum physical boundaries; The seepage scale coefficient, This is the reduction factor for permeability due to pressure compaction. This is the path correction factor for the composite layer thickness.

7. The optimization method according to claim 6, characterized in that: The step S2, which establishes the pore-humidity-pressure mapping model, also includes: calculating the drainage capacity index using the following formula. and anti-fouling index : ； ； In the formula, For hydraulic gradient; This is the initial aperture normalized value. This represents the maximum engineering boundary value of the normal pressure. ~ Assigning weights to each influencing factor. This is the disturbance term for the anti-clogging index.

8. The optimization method according to claim 2, characterized in that: The multi-output prediction model in step S3 is an extreme random tree model: ； ； In the formula, To integrate prediction functions, The number of random regression trees. For the first Random regression trees, For the first The random partitioning parameters, sample perturbation, and feature perturbation corresponding to each random regression tree. The feature vector is composed of the input variables.

9. The optimization method according to claim 7, characterized in that: The objective function of the multi-objective optimization model in step S4 for: ； in, As a normalized benchmark for permeability coefficient, For the target porosity, To allow for porosity deviation scale, To create feasible penalties, , , and The weights are respectively for permeability, anti-clogging ability, pore stability, and manufacturing feasibility.

10. The optimization method according to claim 1, characterized in that: In step S4, the multi-objective optimization model is solved using the differential evolution algorithm. The population size of the differential evolution algorithm is 20 times the dimension of the optimization variables, the maximum number of iterations is 300, the mutation factor is 0.5–0.8, the crossover probability is 0.7, and the convergence tolerance is [missing value]. .

Images