A dynamic property-driven method and system for compiling genes for remodeling sand.

Abstract

This invention presents a method and system for compiling dynamic characteristics-driven sand remodeling genes. The method calculates the small-strain shear modulus of indoor and outdoor sand using sand density, and then corrects it using an overlying stress correction formula to obtain a normalized small-strain shear modulus. Indoor and outdoor test results are integrated to obtain an indoor-in-situ fusion dataset, which is used to train an artificial neural network. An intelligent optimization algorithm is then used to find the optimal variable state parameters for indoor remodeled sand. The absolute value of the difference between the predicted value of the artificial neural network and the corresponding normalized small-strain shear modulus outdoors is used as the optimization objective function. Iterative optimization establishes an indoor-in-situ intelligent remodeling model, generating the optimal variable state parameters and the corresponding sand genes. This method solves the problem of dynamic characteristic deviations caused by the reliance on static parameters in traditional sand remodeling techniques and achieves high-precision equivalence of indoor and outdoor soil dynamic responses.

Description

Technical Field

[0001] This invention relates to the field of soil dynamics sand remodeling technology, specifically a method and system for compiling sand remodeling genes driven by dynamic characteristics. Background Technology

[0002] In the field of seismic resistance and vibration control in geotechnical engineering, the core mission of soil dynamics is to accurately reproduce the stiffness response characteristics of soil under dynamic loads (such as seismic loads and traffic loads). With the continuous development of soil dynamics, the research problem has gradually shifted from macroscopic failure phenomena to microscopic dynamic response control. For sandy soil foundations, the elastic stiffness within a small strain range—the small-strain shear modulus—directly controls the wave propagation speed and liquefaction resistance, and has become a key dynamic indicator in modern seismic design. Furthermore, for different types of sand, within the elastic deformation range (under small strain conditions), if the normalized small-strain shear modulus of two types of sand is the same after correction, then their shear deformation resistance (stiffness) must be the same.

[0003] However, for a long time, the development of soil dynamics testing methods has lagged far behind theoretical needs. Current soil dynamics tests, such as centrifuge tests, shaking table model tests, and cyclic triaxial tests, all require remolding of sand. However, traditional sand remolding techniques still use static characteristic parameters, with relative density as the core, as the control standard. When remolded sand is tested, its liquefaction assessment capability and elastic shear deformation resistance differ from those of undisturbed natural sand outdoors. For example, centrifuge tests do not match the normalized small-strain shear modulus and permeability characteristics of undisturbed natural sand outdoors. The pore structure of indoor remolded sand differs from that of outdoor undisturbed natural sand (such as particle contact stiffness and pore connectivity), resulting in a faster rate of pore water pressure rise under dynamic loads than outdoor undisturbed natural sand. This leads to an earlier assessment of liquefaction occurrence and an overestimation of soil liquefaction risk. In shaking table model tests, the normalized small-strain shear modulus of indoor remolded sand deviates significantly from that of outdoor undisturbed natural sand, causing the propagation speed, phase, and energy dissipation of seismic waves in the model to be inconsistent with the actual site conditions. If the normalized small-strain shear modulus of the cyclic triaxial test differs from that of the outdoor undisturbed natural sand, differences in particle contact stiffness and stress transfer paths will occur. This leads to a deviation between the dilatation / contraction characteristics of the indoor remolded sand under cyclic loading and the outdoor undisturbed natural sand, resulting in misjudgment of the soil liquefaction time point. Therefore, traditional sand remolding techniques are difficult to match the high-precision requirements of modern seismic analysis for soil dynamic response. Thus, it is necessary to consider a similar model of outdoor undisturbed natural sand and indoor remolded sand driven by the normalized small-strain shear modulus to reproduce the in-situ dynamic characteristics in indoor tests.

[0004] Meanwhile, with the development of science and technology, artificial neural networks have been gradually applied to the field of soil dynamics. However, traditional sand remolding technology still relies on manual trial and error and experience to adjust static characteristic parameters such as relative density. The experimental cycle is long, the cost is high, and it is difficult to cope with the variable conditions of complex soils. The lack of integration with artificial neural networks has led to differences in the physical and mechanical properties of outdoor undisturbed natural sand and indoor remolded sand.

[0005] The present invention relates to a dynamic characteristic-driven gene compilation process for remodeling sand, which combines normalized small strain shear modulus index control with artificial neural network to obtain various physical and mechanical property parameters of remodeled sand required for indoor remodeling dynamic experiments. Summary of the Invention

[0006] To address the shortcomings of existing technologies, the technical problem this invention aims to solve is: to provide a dynamic characteristic-driven sand remodeling gene compilation method and system, to construct an indoor-in-situ intelligent remodeling model driven by the normalized small strain shear modulus of sand, and to remodel the variable state parameters of sand using the indoor-in-situ intelligent remodeling model, thereby solving the problem of dynamic characteristic deviation caused by the reliance on static parameters—relative density—in traditional sand remodeling technology, and achieving high-precision equivalence of indoor and outdoor soil dynamic responses.

[0007] The technical solution adopted by the present invention to solve the aforementioned technical problem is as follows:

[0008] In a first aspect, the present invention provides a method for compiling genes for sand remodeling driven by dynamic characteristics, characterized in that the compilation method includes the following:

[0009] The small-strain shear modulus of remolded sand was obtained through indoor experiments under different vertical effective stresses, water contents, and relative densities while keeping the particle size, particle size distribution, maximum void ratio, and minimum void ratio constant. The small-strain shear modulus of the indoor remolded sand was then corrected to obtain the normalized small-strain shear modulus. The normalized small-strain shear modulus was then correlated with the particle size, particle size distribution, maximum void ratio, minimum void ratio, and corresponding vertical effective stress, water content, and relative density of the standard sand used to form an indoor dataset.

[0010] The vertical effective stress and small strain shear modulus of undisturbed natural sand at different burial depths were obtained through outdoor experiments. Simultaneously, samples of undisturbed natural sand were taken, and the particle size, particle size distribution, maximum void ratio, minimum void ratio, water content, and relative density of the undisturbed natural sand were obtained in the laboratory. The small strain shear modulus of the undisturbed natural sand was corrected to obtain the normalized small strain shear modulus. The normalized small strain shear modulus of the outdoor sand was then correlated with particle size, particle size distribution, maximum void ratio, minimum void ratio, vertical effective stress, water content, and relative density to form an outdoor undisturbed dataset.

[0011] The indoor dataset and the outdoor untouched dataset are merged to form an indoor-in-situ fused dataset;

[0012] Using particle size, particle size distribution, maximum void ratio, minimum void ratio, vertical effective stress, moisture content, and relative density as input features, and normalized small strain shear modulus as the target output, an artificial neural network is constructed. The artificial neural network is then trained using an indoor-in-situ fusion dataset to obtain a normalized small strain shear modulus prediction model.

[0013] An indoor-in-situ intelligent remodeling model based on the normalized small strain shear modulus prediction model and intelligent optimization algorithm was constructed. The indoor-in-situ intelligent remodeling model takes the outdoor normalized small strain shear modulus as input to obtain the optimal variable state parameters of the indoor remodeled sand. The optimization objective function of the indoor-in-situ intelligent remodeling model is the absolute value of the difference between the predicted value of the normalized small strain shear modulus prediction model and the corresponding input outdoor normalized small strain shear modulus.

[0014] The variable state parameter is at least one of the following: vertical effective stress, water content, and relative density.

[0015] Secondly, the present invention provides a dynamic property-driven sand remodeling gene compilation system, the system comprising:

[0016] Indoor bending element device to acquire shear wave velocity data of standard sand under different working conditions;

[0017] Electro-hydraulic servo universal testing machine is used to simulate vertical effective stress;

[0018] An outdoor bending element device is used to acquire shear wave velocity data of outdoor undisturbed natural sand under different working conditions.

[0019] A function signal generator is used to apply a sinusoidal waveform with a set voltage and frequency to the transmitting bending element sensor;

[0020] A function signal amplifier is used to receive and amplify the electrical signal from the bending element sensor;

[0021] An oscilloscope is used to input waveform parameters and simultaneously display the output and received waveforms.

[0022] The intelligent reshaping data management platform includes a data processing module, a database, a display module, and an indoor-in-situ intelligent reshaping model;

[0023] The data processing module acquires the normalized small strain shear modulus of different types of standard sand under different moisture contents, relative densities, and vertical effective stresses. It also acquires the corresponding particle size, particle size distribution, maximum void ratio, minimum void ratio, vertical effective stress, moisture content, relative density, and normalized small strain shear modulus of outdoor undisturbed natural sand at different measurement locations and depths. The module then organizes the data to obtain an indoor-in-situ fusion dataset.

[0024] The indoor-in-situ intelligent reshaping model takes the normalized small strain shear modulus of the outdoor soil as input and initializes and generates candidate combinations of variable state parameters within the constraints of variable state parameters. It calls an artificial neural network to predict the corresponding normalized small strain shear modulus of the candidate combinations of variable state parameters and uses an intelligent optimization algorithm to obtain the variable state parameters of the corresponding indoor reshaping sand.

[0025] The database stores variable state parameters and inherent physical parameters of standard sand and outdoor undisturbed natural sand under different working conditions, as well as normalized small strain shear modulus, and supports engineering applications to call them. The database communicates with the data processing module, the indoor-in-situ intelligent reshaping model and the display module through standardized interfaces. The data processed by the data processing module is stored in the database. The indoor-in-situ intelligent reshaping model can call the data in the database to dynamically update and iteratively optimize the indoor-in-situ intelligent reshaping model. The display module supports the visualization of parameters.

[0026] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0027] 1. Existing indoor remolded sand technology uses relative density as the core static characteristic parameter as the control index, relying on manual trial and error adjustments, making it difficult to correlate with the dynamic mechanical response characteristics of the soil. This invention innovatively introduces the normalized small-strain shear modulus as an equivalent benchmark. By unifying the testing standards through indoor and outdoor bending element devices, it ensures the consistency of the normalized small-strain shear modulus between outdoor undisturbed natural sand and indoor remolded sand, thereby reflecting the equivalence of the soil's liquefaction resistance and elastic deformation capacity. This overcomes the problem of the disconnect between traditional static and dynamic characteristics, significantly improving the response fidelity of indoor remolded sand under complex dynamic loads.

[0028] 2. This invention unifies the measurement method of indoor and outdoor shear wave velocity—bending element test, and drives the establishment of an indoor-in-situ intelligent remodeling model through artificial neural network and intelligent optimization algorithm. It establishes a dynamic mapping between the normalized small strain shear modulus and variable state parameters in the outdoor environment, forming a closed loop from in-situ testing to laboratory remodeling.

[0029] 3. Establish a unified and standardized process for obtaining normalized small strain shear modulus both indoors and outdoors, covering the entire process from sample preparation and parameter measurement to data processing. Construct an indoor-in-situ fusion dataset, and combine artificial neural networks and intelligent optimization algorithms for iterative optimization to form a closed-loop system of "measurement-modeling-optimization-verification". This will significantly shorten the test cycle, reduce costs, and improve the accuracy and adaptability of remodeling variable state parameters under complex soil conditions.

[0030] 4. The present invention makes targeted improvements to the indoor and outdoor bending element device, which improves the stability, durability and measurement accuracy of the device, and provides hardware guarantee for the reliability of test data.

[0031] 5. Traditional sand remodeling lacks systematic data management, with test parameters stored in a scattered manner, making it difficult to iteratively optimize. This invention, based on a dynamic characteristic-driven sand remodeling gene compilation system, integrates outdoor testing, laboratory remodeling, and an intelligent remodeling data management platform. It achieves full automation and intelligence throughout the entire process from field testing, data acquisition, model training to parameter optimization. It possesses full data lifecycle management and powerful analysis and processing capabilities, enabling rapid response to different engineering needs and providing efficient and intelligent solutions for soil dynamics testing and engineering applications.

[0032] 6. In this invention, the shear wave velocity of both indoor and outdoor sand is measured using a bending element device. The small-strain shear modulus of the indoor and outdoor sand is obtained from the sand density. After correction using the overlying stress correction formula, the normalized small-strain shear modulus is obtained. The indoor and outdoor test results are integrated to obtain an indoor-in-situ fusion dataset, which is used to train an artificial neural network. Then, an intelligent optimization algorithm is used to find the optimal variable state parameters of the indoor remolded sand. The absolute value of the difference between the predicted value of the artificial neural network and the corresponding normalized small-strain shear modulus of the outdoor sand is used as the optimization objective function. An indoor-in-situ intelligent remolding model is established through iterative optimization. The trained indoor-in-situ intelligent remolding model is written into the intelligent remolding data management platform to monitor and display the optimal variable state parameters and corresponding sand genes generated by the indoor-in-situ intelligent remolding model in real time. Attached Figure Description

[0033] Figure 1 This is a flowchart illustrating one embodiment of the dynamic characteristics-driven sand remodeling gene compilation method of the present invention.

[0034] Figure 2 This is a schematic diagram of the structure of an indoor bending element device according to an embodiment of the present invention.

[0035] Figure 3 This is a schematic diagram of the side plate of an indoor bending element device according to an embodiment of the present invention.

[0036] Figure 4This is a schematic diagram of the structure of the protective sleeve body after placing the bending element sensor according to an embodiment of the present invention.

[0037] Figure 5 This is a schematic diagram of the structure of an outdoor bending element device according to an embodiment of the present invention.

[0038] Figure 6 This is a schematic diagram of the structure of the dynamic characteristics-driven sand remodeling gene compilation system of the present invention.

[0039] Figure 7 This is a schematic diagram of the normalized small strain shear modulus prediction model in this invention.

[0040] Figure 8 This is a schematic diagram of the modeling process for the indoor-in-situ intelligent reshaping model in this invention. Detailed Implementation

[0041] The present invention will be further explained below with reference to the embodiments and accompanying drawings, but this is not intended to limit the scope of protection of this application.

[0042] The present invention provides a method for compiling sand remodeling genes driven by dynamic characteristics, comprising the following:

[0043] The small-strain shear modulus of remolded sand was obtained through indoor experiments under different vertical effective stresses, water contents, and relative densities, with standard sand having constant particle size, particle size distribution, maximum void ratio, and minimum void ratio. The small-strain shear modulus of the indoor remolded sand was corrected using the overlying stress correction formula to obtain the normalized small-strain shear modulus. The normalized small-strain shear modulus was then correlated with the particle size, particle size distribution, maximum void ratio, minimum void ratio, vertical effective stress, water content, and relative density of the standard sand used to form an indoor dataset.

[0044] The indoor tests employed indoor bending element tests, using soil dynamics standard sands such as Fujian standard sand, Toyoura sand from Japan, or Ottawa sand from the United States. With fixed particle size, particle size distribution, maximum void ratio, and minimum void ratio, the shear wave velocity was measured using an indoor bending element device by varying the effective vertical stress, water content, and relative density. Simultaneously, the density data of the standard sand is obtained by combining the sand cone method or the ring cutter method. Through standardized physical models Accurate calculation of small strain shear modulus of remolded sand in the laboratory .

[0045] Using the overburden stress correction formula The small-strain shear modulus of the remolded sand in the laboratory was corrected to obtain the normalized small-strain shear modulus. ,in This represents the effective vertical stress.

[0046] The vertical effective stress and small strain shear modulus of undisturbed natural sand at different burial depths were obtained through outdoor experiments. Samples of the undisturbed natural sand were taken, and the particle size, particle size distribution, maximum void ratio, minimum void ratio, water content, and relative density of the undisturbed natural sand were obtained in the laboratory. The small strain shear modulus of the undisturbed natural sand was corrected using the overlying stress correction formula to obtain the normalized small strain shear modulus of the outdoor sand. The normalized small strain shear modulus of the outdoor sand was then correlated with particle size, particle size distribution, maximum void ratio, minimum void ratio, vertical effective stress, water content, and relative density to form an outdoor undisturbed dataset.

[0047] The outdoor test employed an outdoor bending element test, which used an outdoor bending element device to measure the shear wave velocity of sand at different burial depths h along the soil depth direction. Simultaneously, layered density tests were conducted using the sand cone method and the ring cutter method to obtain the density of sand at corresponding depths. ; through formula Obtain the vertical effective stress of outdoor undisturbed natural sand. ,in For effective severity; through a standardized physical model Accurate calculation of small strain shear modulus of outdoor undisturbed natural sand ;

[0048] Using the overburden stress correction formula The small-strain shear modulus of the outdoor undisturbed natural sand is corrected to obtain the normalized small-strain shear modulus of the outdoor soil.

[0049] Meanwhile, samples of unspoiled natural sand were taken outdoors and tested in the laboratory for particle size, particle size distribution, maximum void ratio, minimum void ratio, moisture content, and relative density.

[0050] The indoor dataset and the outdoor untouched dataset are merged to form an indoor-in-situ fused dataset;

[0051] Using particle size, particle size distribution, maximum void ratio, minimum void ratio, vertical effective stress, moisture content, and relative density as input features, and normalized small-strain shear modulus as the target output, an artificial neural network is constructed. This artificial neural network is then trained using an indoor-in-situ fusion dataset to obtain a normalized small-strain shear modulus prediction model (see [link to relevant documentation]). Figure 7 );

[0052] An indoor-in-situ intelligent remodeling model (hereinafter referred to as the indoor-in-situ intelligent remodeling model) based on the normalized small strain shear modulus prediction model and intelligent optimization algorithm is constructed. The indoor-in-situ intelligent remodeling model takes the outdoor normalized small strain shear modulus as input to obtain the optimal variable state parameters of the indoor remodeled sand. The optimization objective function of the indoor-in-situ intelligent remodeling model is the absolute value of the difference between the predicted value of the normalized small strain shear modulus prediction model and the corresponding input outdoor normalized small strain shear modulus.

[0053] The variable state parameter is at least one of vertical effective stress, moisture content, and relative density.

[0054] The present invention provides an indoor-in-situ intelligent remodeling model, which uses the normalized small strain shear modulus of outdoor undisturbed natural sand as a benchmark. It uses an intelligent optimization algorithm to call the normalized small strain shear modulus prediction model to invert the laboratory remodeling parameters, which are the variable state parameters.

[0055] In practical applications, the particle size, particle size distribution, maximum void ratio, and minimum void ratio of outdoor undisturbed natural sand and standard sand are all preset inherent physical parameters in the indoor-in-situ intelligent reshaping model. The candidate combinations of variable state parameters generated in each iteration of the intelligent optimization algorithm are input together with the inherent physical parameters into the normalized small strain shear modulus prediction model. The normalized small strain shear modulus prediction model outputs the corresponding predicted value of normalized small strain shear modulus. The intelligent optimization algorithm takes the input outdoor normalized small strain shear modulus as the target value, and iteratively optimizes by calculating the deviation between the predicted value and the target value (the optimization objective function is the absolute value of the deviation) until the convergence condition is met, and outputs the optimal variable state parameters.

[0056] In the indoor-in-situ intelligent remodeling model, the normalized small-strain shear modulus (target value) of the outdoor soil is input, and an intelligent optimization algorithm is executed to initialize the particle swarm. A series of candidate combinations of variable state parameters are generated through initialization. Then, the normalized small-strain shear modulus prediction model is called to predict the corresponding normalized small-strain shear modulus (predicted value) using the candidate combinations. The fitness is then calculated by optimizing the objective function with the absolute value of the difference between the predicted value and the target value, and the optimal variable state parameters of the indoor remodeled sand corresponding to the current outdoor undisturbed natural sand are obtained.

[0057] In artificial neural networks, intrinsic physical parameters and variable state parameters are used together as input features, with the normalized small-strain shear modulus as the target output. Intrinsic physical parameters include particle size, particle size distribution, maximum void ratio, and minimum void ratio of both indoor standard sand and outdoor undisturbed natural sand. Variable state parameters include vertical effective stress, water content, and relative density. The artificial neural network is trained using an indoor-in-situ fusion dataset containing properties of both standard sand and outdoor undisturbed natural sand. It integrates the complex responses of standard sand and outdoor undisturbed natural sand, learning the normalized small-strain shear modulus mapping patterns of different sands under varying vertical effective stress, water content, and relative density. New experimental data (such as sand from different regions) can be injected into the artificial neural network in real time, continuously optimizing network weights through learning and improving adaptability.

[0058] This invention, after training an artificial neural network, uses an intelligent optimization algorithm for optimization. The normalized small-strain shear modulus of outdoor sand is used as the input to the algorithm, and variable state parameters are used as the output. These are then combined with the inherent physical parameters of standard sand. The combination of these two parameters yields the desired indoor remolded sand. When the type of indoor sand is determined (e.g., Fujian standard sand), particle size, particle size distribution, and maximum / minimum void ratio can be fixed, and only one parameter among vertical effective stress, moisture content, and relative density can be optimized. This approach is suitable for scenarios requiring rapid matching of a single indicator. For complex sites (such as multi-layered sandy soil), the vertical effective stress, moisture content, and relative density can be optimized simultaneously. The optimal combination is searched globally through an intelligent optimization algorithm to ensure the equivalence of the normalized small strain shear modulus. Simultaneously, samples are prepared according to the formula corresponding to these parameters, and indoor bending element tests are conducted. The normalized small strain shear modulus obtained from the indoor bending element tests is used as the verification result and fed back to the indoor-in-situ intelligent reshaping model to evaluate the performance of the indoor-in-situ intelligent reshaping model and continuously iterate and update the indoor-in-situ intelligent reshaping model. The optimized indoor-in-situ intelligent reshaping model can be optimized according to different scenarios and different needs, and its applicability is very high.

[0059] The aforementioned artificial neural network can be at least one of BP neural network, ANN neural network or RNN neural network, and the intelligent optimization algorithm can be at least one of PSO algorithm, GWO algorithm or GA algorithm.

[0060] In this application, the components of the "sand gene" are defined as including inherent physical parameters and variable state parameters. Among them, particle size, particle size distribution, maximum void ratio and minimum void ratio are used as inherent physical parameters, and vertical effective stress, water content and relative density are used as variable state parameters. Gene compilation mainly involves compiling the variable state parameters through intelligent optimization algorithms.

[0061] In this invention, the dynamic characteristic driving factor refers to the calculation driven by the normalized small strain shear modulus. The optimal combination of variable state parameters for indoor remolded sand is found through intelligent optimization algorithms (such as PSO), achieving a directional expression of dynamic characteristics through deep genotyping of sand parameters. The term "genotyping" is directly used to describe the variable state parameter optimization process.

[0062] Specifically, the sandy soil characteristics include particle size D50, uniformity coefficient Cu, and maximum porosity. Minimum porosity Moisture content (w), relative density and vertical effective stress The gene compilation refers to optimization using the PSO algorithm. w The combination of these factors allows the normalized small-strain shear modulus predicted by the artificial neural network to approximate the in-situ target value.

[0063] This invention optimizes by "approaching the in-situ target value with the predicted value" rather than directly optimizing the normalized small strain shear modulus output parameters from the outdoor environment. It uses an intelligent optimization algorithm to deduce the optimal reshaping parameters in reverse. The artificial neural network, acting as a forward surrogate model, essentially learns the "constitutive behavior" of the sand. The objective function of the intelligent optimization algorithm (such as PSO) is to "reduce the gap between the predicted and in-situ target values," which is equivalent to allowing the algorithm to automatically explore feasible solutions in the parameter space that satisfy dynamic stiffness equivalence. The greatest wisdom of this approach lies in its lack of pre-defined parameter combinations; instead, it allows the physical laws themselves to converge to the optimal solution through data-driven methods. Direct parameter fitting cannot achieve this. The error from each experimental verification is fed back into the indoor-in-situ intelligent reshaping model iteration, essentially installing an "adaptive calibrator." "Genetic compilation": This is not a simple parameter replication but a simulation of the "dynamic phenotype" of the sand. The training data for the artificial neural network comes from the indoor-in-situ fusion dataset, which naturally contains the feasible parameter domain. The intelligent optimization algorithm automatically avoids illegal areas during the search, making it more reasonable than direct parameter optimization.

[0064] The function signal generator 001 can generate and precisely adjust various standard waveform signals (such as sine waves, square waves, and triangle waves). By controlling parameters such as frequency, amplitude, and phase, it can generate a stable sine wave signal.

[0065] The function signal amplifier 002 is used to amplify weak electrical signals (such as voltage, current or power) and increase the signal strength for subsequent processing or transmission.

[0066] The oscilloscope 003 is used to capture and visualize electrical signal waveforms and measure parameters such as voltage, frequency, and phase.

[0067] Example 1: The dynamic characteristic-driven sand remodeling gene compilation method in this example includes the following:

[0068] 1) Indoor standardized bending element testing

[0069] Specific soil dynamics standard sands were selected, and their inherent physical parameters (particle size, particle size distribution, maximum void ratio, minimum void ratio, etc.) were strictly controlled and recorded. Remolded specimens with a series of relative densities were prepared using a layered compactor, a shaking table, or an air / water sedimentation method based on the target void ratio. For unsaturated specimens, controlled spray wetting was used to precisely prepare specimens with a series of initial moisture contents. Saturated specimens were prepared using standard saturation processes, such as CO2 replacement or back pressure saturation. The prepared specimens were installed in an indoor bending element device, and vertical effective stress was simulated using an electro-hydraulic servo universal testing machine.

[0070] The indoor bending element device is nickel-plated on both the inner and outer surfaces to ensure the rust and corrosion resistance of the iron plate. The indoor bending element device 005 includes a rectangular experimental box formed by four corrosion-resistant side plates and a sealed bottom plate, heavy-duty feet 2 installed at the bottom of the experimental box, and a protective sleeve. Sealing grooves are machined at the connection points of the four side plates, with acid- and alkali-resistant, high-temperature-resistant fluororubber circular sealing strips 1 inside. Adjacent side plates are connected by bolts. When the bolts are tightened, the volume of the fluororubber circular sealing strips 1 exceeds the machined sealing grooves, thus ensuring that the waterproof sandbox will not leak. The four heavy-duty feet 2 are used to bear the weight of the experimental box and the load. The heavy-duty feet 2 are M10 heavy-duty feet, each capable of withstanding 300KG of pressure, and the four heavy-duty feet can fully bear the weight of the experimental box and the load. A threaded hole 7 is provided on the bottom of one side plate of the experimental chamber. The threaded hole is filled with permeable metal stone to ensure that when the sand inside the experimental chamber is under pressure, only water flows out of the sand and does not overflow. A solenoid valve 3 is installed on the outside of the threaded hole 7 to control the drainage conditions during the test.

[0071] The four side plates and the bottom plate are all made of iron plates and are connected to each other by bolts to form a box. After processing, the inner and outer surfaces of the five iron plates are nickel-plated to ensure the iron plates are rust-proof and corrosion-resistant. Pre-drilled holes for installing the protective sleeve are provided on a pair of opposite side plates. The solenoid valve 3 uses AC220V power supply to control the drainage conditions during the test. The protective sleeve includes a protective sleeve cover 5 and a protective sleeve body 6. The protective sleeve cover 5 and the protective sleeve body 6 have pre-drilled encapsulation grooves at their front ends. The protective sleeve cover 5 and the protective sleeve body 6 are aligned and connected by bolts. The bending element sensor 4 is encapsulated with epoxy resin to form a protective layer on its surface, keeping it insulated from the outside. The bending element sensor is fixed in the protective sleeve using bolts. After the protective sleeve cover 5, the protective sleeve body 6, and the bending element sensor 4 are fixedly installed, silicone is applied to the encapsulation groove, and then left to stand until the silicone completely solidifies. At this point, the bending element sensor, except for the exposed part, is completely encapsulated in the protective sleeve, standing still and insulated. The length of the bending element sensor extending from its front end can be controlled by the bolt position, thereby adjusting the accuracy of the measurement data. After the protective sleeve and silicone solidify, the protective sleeve and the bending element sensor are inserted through the pre-drilled hole on the side plate and the protective sleeve is fixed to the side plate with bolts. This constitutes a complete loading test environment, which is completely sealed and will not leak water.

[0072] Function signal generator 001 is connected to the transmitting bending element sensor at one end via a BNC cable and to oscilloscope 003 at the other end. Function signal amplifier 002 is connected to the receiving bending element sensor at one end via a BNC cable and to oscilloscope 003 at the other end. A sinusoidal waveform with a set voltage and frequency is applied to the transmitting bending element sensor by function signal generator 001. The piezoelectric effect forces the transmitting bending element sensor to extend on one side and shorten on the other side, generating bending motion and lateral vibration in the surrounding soil. The shear wave generated in the direction perpendicular to this vibration is transmitted through the soil to the receiving bending element sensor at the other end. After receiving the shear wave, the receiving bending element sensor is forced to generate a lateral swing and converts the mechanical swing into an electrical signal, which is displayed on oscilloscope 003 by function signal amplifier 002.

[0073] After the specimen reaches the target vertical effective stress stability under the electro-hydraulic servo universal testing machine, the integrated intelligent remodeling data management platform 004 is activated to collect shear wave velocity data. Immediately after the bending element sensor test is completed, the density of the specimen is tested, and the small strain shear modulus of the indoor remodeled sand is calculated using a standardized physical model. This is then corrected using the overburden stress correction formula to obtain the normalized small strain shear modulus of the indoor remodeled sand. All parameters of the current specimen in the indoor test (particle size, particle size distribution, maximum void ratio, minimum void ratio, vertical effective stress, moisture content, relative density, and normalized small strain shear modulus) are associated and stored to form a highly structured indoor dataset.

[0074] 2) Outdoor in-situ bending element testing

[0075] The outdoor bending element device includes a horizontal fixing rod 8 and two vertical fixing rods 9. The horizontal fixing rod 8 can adjust the spacing between the two vertical fixing rods 9 and ensure that the vertical fixing rods 9 are in the same plane. Each vertical fixing rod 9 consists of two halves of steel plates, which are connected to the horizontal fixing rod as a whole by bolts. A cone hammer 10 is installed at the bottom of the vertical fixing rod 9 to facilitate insertion into the soil and effectively reduce disturbance to the surrounding soil during the cone insertion process. A screw hole is provided at the lower end of the vertical fixing rod 9 for installing the bending element sensor 4. Flat iron plates 12 are installed above and below the bending element sensor 4 to protect it when the device is inserted into and pulled out of the soil. At the same time, a groove is opened in the vertical fixing rod 9 for the cable of the bending element sensor to pass through. The cable is connected to a function signal generator 001 and a function signal amplifier 002 respectively. A scale 13 is engraved on the outside of the vertical fixing rod to detect the soil penetration depth of the penetrator. The exposed portion of the bending element sensor 4 is encapsulated with epoxy resin, and the gap between it and the vertical fixing rod 9 is sealed with silicone to ensure strict waterproofing and accurate measurement data. Both the transmitting and receiving bending element sensors of the outdoor bending element device need to be inserted into the soil being measured at the same depth.

[0076] After selecting the measurement site, the outdoor bending element device is used to conduct an outdoor test on undisturbed natural sand. A sinusoidal waveform with a pre-set voltage and frequency is applied to the transmitting bending element sensor via a function signal generator 001. The piezoelectric effect forces the transmitting bending element sensor to elongate on one side and shorten on the other, generating bending motion and lateral vibration in the surrounding soil. The shear wave generated in the direction perpendicular to this vibration is transmitted through the soil to the receiving bending element sensor at the other end. Upon receiving the shear wave, the receiving bending element sensor is forced to generate a lateral oscillation, which is converted into an electrical signal and displayed on an oscilloscope 003 via a function signal amplifier 002. The outdoor bending element device is precisely lowered to the predetermined test depth h. A command is sent via the function signal generator 001 to excite the bending element sensor to generate a shear wave and record the waveform signal to obtain the shear wave velocity of the undisturbed natural sand. This velocity is then calculated using the formula... Obtain the vertical effective stress of outdoor undisturbed natural sand. ,in To ensure effective density, a small amount of undisturbed natural sand was collected after the test and sent to the laboratory for precise determination of its density, particle size, particle size distribution, maximum void ratio, minimum void ratio, water content, and relative density. A standardized physical model was used to accurately calculate the small-strain shear modulus of the undisturbed natural sand at that depth, and this was corrected using the overlying stress correction formula to obtain the normalized small-strain shear modulus for the outdoor area. All parameters at the current measurement point (particle size, particle size distribution, maximum void ratio, minimum void ratio, vertical effective stress, water content, relative density, and normalized small-strain shear modulus) were associated and stored to form a highly structured outdoor undisturbed dataset.

[0077] 3) Indoor-in-situ fusion dataset

[0078] Rigorous data cleaning, format standardization, and unit standardization were performed on both indoor and outdoor undisturbed datasets. This ensured that all parameters were clearly defined and measurement methods were explicit. By training the model within the same framework, both the laboratory's "standard" response and the in-situ "real" response, the model gained generalization ability to predict whether selected standard sand could achieve the target in-situ normalized small strain shear modulus under specific parameter combinations.

[0079] 4) An intelligent optimization algorithm is employed, setting the constraint range of the variable state parameters. The normalized small strain shear modulus from the outdoor environment is input, and the optimal combination of variable state parameters is searched. Laboratory personnel conduct indoor bending element tests on the reshaped specimens based on the optimal variable state parameters and corresponding inherent physical parameters provided by the intelligent optimization algorithm. The actual normalized small strain shear modulus of the reshaped specimens is calculated to verify the equivalence of the indoor-in-situ intelligent reshaping model. This verification result is also fed back to the system to evaluate the performance of the indoor-in-situ intelligent reshaping model or to expand the indoor-in-situ fusion dataset as new data points, continuously optimizing the indoor-in-situ intelligent reshaping model and forming a closed loop. The constraint range is determined by the physical boundaries and physical states of the variable state parameters. The constraint range refers to the reshaping parameters (vertical effective stress) calculated by the intelligent optimization algorithm. Moisture content (w), relative density During the search and inversion process, hard or soft constraints are applied. These constraints are derived directly from the physical properties of the soil, the limits of laboratory equipment capabilities, and engineering experience, aiming to ensure that the optimization result is not only mathematically optimal but also meets engineering practice requirements.

[0080] Example 2: The method for compiling sand remodeling genes driven by dynamic characteristics in this example (see...) Figure 1 The process is as follows:

[0081] Step 1: Collect indoor dataset

[0082] For indoor testing, standard sands such as Fujian standard sand, Toyoura sand from Japan, or Ottawa sand from the United States can be used. This example uses Fujian standard sand produced by Xiamen Aisio Standard Sand Co., Ltd. for illustration.

[0083] 1.1 Sand Sample Pretreatment

[0084] According to the "Standard for Geotechnical Testing Methods", the particle size (D50=0.55mm), particle size distribution (Cu=1.54), and maximum void ratio of Fujian standard sand were determined. =0.83) and minimum void ratio ( (=0.48) remained unchanged. The Fujian standard sand was dried in an oven at 105℃ for 24 hours to completely remove moisture and obtain dry sand.

[0085] 1.2 Moisture Content Control

[0086] Preparation of wet Fujian standard sand: Assume that the mass of dried Fujian standard sand is weighed. For a weight of 1000g, the moisture content w is selected as 2%, 4%, 6%, 8%, 10%, 12%, 14%, 16%, 18%, and 20%, and calculated using the formula... Calculate target moisture content Water was sprayed onto the surface of the sand sample in layers using a precision sprayer. Each layer was manually stirred for 3 minutes until the target moisture content was completely sprayed into the Fujian standard sand, thus obtaining wet Fujian standard sand with the corresponding moisture content. The wet Fujian standard sand was then placed in a sealed bag and left to stand in a constant temperature environment (20±1℃) for ≥24 hours to allow the moisture to penetrate evenly.

[0087] Preparation of saturated Fujian standard sand: After the dried Fujian standard sand is loaded into a mold, a vacuum is applied for 30 minutes, degassed water is injected, and a back pressure of ≥200 kPa is applied until the B value is ≥0.95. The B value refers to the pore water pressure coefficient, which is the ratio of the increase in pore water pressure to the increase in confining pressure.

[0088] 1.3 Relative Density Control

[0089] The relative density of Fujian standard sand was selected. The porosity values ​​are 10%, 20%, 30%, 40%, 50%, 60%, 70%, 80%, and 90%, respectively, calculated using the porosity formula. Calculate the void ratio of Fujian standard sand at different relative densities. ,in , These are the maximum void ratio and the minimum void ratio, respectively.

[0090] The bottom area of ​​the experimental chamber of the indoor bending element device was measured in the laboratory as follows: The height is The volume of the experimental chamber can be obtained. for: Through formula The solid volume of Fujian standard sand was calculated. Through formula Obtain the dry sand quality of Fujian standard sand ,

[0091] in, The specific gravity of soil particles, Fujian standard sand. 2.65 g / cm 3 ; The density of water is taken as 1 g / cm³. 3 .

[0092] Wet Fujian standard sand with different relative densities and corresponding moisture contents was prepared according to the moisture content control method: First, the dry sand mass was obtained through the above process according to the set relative density. Then, the target moisture content was calculated based on the target moisture content and the dry sand mass. Water with the target moisture content was sprayed onto the dry sand to obtain wet Fujian standard sand with the set relative density and moisture content. The wet Fujian standard sand was then divided into 3-5 layers and placed in an indoor bending element device. Each layer was compacted with a 2.5 kg compaction hammer at a drop height of 30 cm for 10-15 times (energy approximately 600 kJ / m). 3 To achieve the target relative density.

[0093] 1.4 Vertical Effective Stress Control

[0094] The indoor bending element device containing prepared wet Fujian standard sand was placed under an electro-hydraulic servo universal testing machine for vertical effective stress control. Before pressurization, a pressure block was placed on the wet Fujian standard sand to ensure that the universal testing machine could apply pressure evenly. The electro-hydraulic servo universal testing machine was controlled to apply stable pressures of 100 kPa, 200 kPa, 300 kPa, 400 kPa, 500 kPa, 600 kPa, 700 kPa, and 800 kPa respectively, and kept constant during the test. After each stress loading, the deformation rate was monitored to be ≤0.005 mm / min, which was considered as consolidation stability.

[0095] 1.5 Measurement of Shear Wave Velocity and Density

[0096] The two ends of the function signal generator 001 are connected to the transmitting bending element sensor and the oscilloscope 003 respectively. The function signal amplifier 002 is electrically connected to the receiving bending element sensor. The data collected by the receiving bending element sensor is connected to the oscilloscope 003 through the function signal amplifier 002.

[0097] Turn on the function signal generator 001, function signal amplifier 002, and oscilloscope 003. Input waveform parameters into oscilloscope 003. Oscilloscope 003 will simultaneously display the output wave and the received wave. Read the time difference between the first peaks of the two waveforms. .

[0098] Use a ruler to measure the horizontal distance between the transmitting and receiving bending element sensors of the indoor bending element device. Through formula The shear wave velocity of the standard sand was calculated. ;

[0099] After the indoor test was completed, the density of the sample was measured using the ring cutter method. Through formula Obtain the corresponding small strain shear modulus of the room .

[0100] Using formula The small-strain shear modulus of the remolded sand in the laboratory was corrected to obtain the normalized small-strain shear modulus. .

[0101] Different types of standard sand with varying moisture content, relative density, and vertical effective stress, along with their corresponding normalized small-strain shear moduli in the laboratory, were collected and mapped one-to-one using the intelligent reshaping data management platform 004. The collected data underwent preliminary processing to check its rationality and consistency. All processed data was then stored in a standardized format to form an indoor dataset.

[0102] Step 2: Collect outdoor untouched dataset

[0103] 2.1 Calculation of Vertical Effective Stress

[0104] Align the outdoor bending element with the measuring point, and use a cone hammer to press it into the soil to the target depth at a uniform speed of 0.5 cm / s. Record the penetration depth h using the external scale of the outdoor bending element, and calculate the vertical effective stress layer by layer. .

[0105] For soil layers above the groundwater level, the formula is used. Obtaining vertical effective stress ,

[0106] Where i is the number of the target soil layer below the soil; The natural unit weight of the j-th soil layer can be determined from regional geological data; Let be the thickness of the j-th layer of soil below ground, and The top surface of the corresponding soil layer is higher than the groundwater level.

[0107] For soil layers below the groundwater level, the formula is used. Where k is the soil layer number where the groundwater level is located. The natural unit weight of the j-th soil layer below the groundwater level. The natural effective unit weight of the j-th soil layer below the groundwater level. ; The specific weight of water is generally taken as 9.8 kN / m³. 3 ; Let be the thickness of the j-th layer of soil.

[0108] 2.2 Shear wave velocity measurement

[0109] Connect the two ends of the function signal generator 001 to the transmitting bending element sensor of the outdoor bending element device and the oscilloscope 003, respectively. Connect the receiving bending element sensor of the outdoor bending element device to the oscilloscope 003 through the function signal amplifier 002. Turn on the function signal generator 001, the function signal amplifier 002, and the oscilloscope 003, input the waveform parameters, and the oscilloscope 003 will simultaneously display the output wave and the received wave. Read the time difference between the first peaks of the two waveforms. Measure the horizontal distance between the two bending element sensors of the outdoor bending element device using a ruler. Through formula The shear wave velocity of the outdoor undisturbed natural sand was calculated. .

[0110] 2.3 Outdoor undisturbed natural sand sampling and laboratory parameter determination

[0111] At the same depth, within a horizontal distance ≤30 cm from the outdoor bending element device, a thin-walled soil sampler was used to press the sample in at a uniform speed of 1 m / min to the measuring point. The sample was immediately sealed with wax at both ends after extraction. After sampling, the density of the sample was determined in the laboratory using the ring sampler method. Through formula The small strain shear modulus of the corresponding outdoor undisturbed natural sand was obtained. Using the formula The small-strain shear modulus of undisturbed natural sand is corrected to obtain the normalized small-strain shear modulus. .

[0112] The moisture content of the soil at the testing points was obtained by the standard drying method; particle size D50, particle size distribution C u and the maximum and minimum void ratio ( , The results were obtained by standard sieving, laser spectroscopy, and vibration table spectroscopy, respectively; and obtained through formulas. ( The specific gravity of soil particles. The density of water, The void ratio of the undisturbed natural sand (where the dry sand density can be measured in a laboratory) is obtained. ; through formula The relative density of the outdoor undisturbed natural sand was obtained. .

[0113] By inserting outdoor bending element devices at different positions and depths, multiple sets of outdoor undisturbed data can be measured. This data is collected through the Intelligent Reshaping Data Management Platform 004, and the particle size, particle size distribution, maximum void ratio, minimum void ratio, vertical effective stress, moisture content, relative density, and normalized small strain shear modulus are mapped one-to-one. The collected outdoor undisturbed data is then preliminarily processed to check its rationality and consistency. All processed data is stored in a unified format to form an outdoor undisturbed data set.

[0114] Step 3: Merge the indoor dataset with the outdoor untouched dataset to form an indoor-in-situ fused dataset.

[0115] The data from the indoor and outdoor untouched datasets are preprocessed using the Intelligent Reshaping Data Management Platform 004. Rigorous data cleaning, format standardization, and unit standardization are performed on both datasets to ensure clear parameter definitions and explicit measurement methods. The indoor and outdoor untouched datasets are then merged to obtain the indoor-in-situ fused dataset.

[0116] Step 4: Establish an indoor-in-situ intelligent remodeling model driven by the normalized small-strain shear modulus of sand.

[0117] The indoor-in-situ intelligent reshaping model consists of a normalized small-strain shear modulus prediction model and an intelligent optimization algorithm. The construction process is as follows: First, an artificial neural network is established. Then, the intelligent optimization algorithm performs back-optimization to obtain the optimal variable state parameters for indoor reshaping of sand. The indoor-in-situ intelligent reshaping model can take an outdoor normalized small-strain shear modulus as input. The intelligent optimization algorithm generates a series of candidate combinations of variable state parameters (water content, relative density, and vertical effective stress). Then, the normalized small-strain shear modulus prediction model predicts the corresponding normalized small-strain shear modulus based on this series of candidate combinations. The intelligent optimization algorithm is then called again to calculate the fitness (where fitness is the sum of the optimization objective function and the penalty constraint term), and the historical optimal is updated. It is then determined whether the termination criterion (i.e., convergence condition) is met. If the convergence condition is met, the optimal reshaping parameters are obtained. The optimization objective function is the absolute value of the difference between the predicted value of the normalized small-strain shear modulus prediction model and the corresponding outdoor normalized small-strain shear modulus.

[0118] In this embodiment, a BP neural network can be used as the artificial neural network, and the PSO algorithm can be used as the intelligent optimization algorithm. The optimization of three variable state parameters—moisture content, relative density, and vertical effective stress—is explained below.

[0119] 1) Constructing artificial neural networks

[0120] All feature values ​​(particle size, particle size distribution, maximum void ratio, minimum void ratio, vertical effective stress, moisture content, relative density) and target value (normalized small strain shear modulus) in the indoor-in-situ fusion dataset are normalized to obtain the normalized indoor-in-situ fusion dataset. This dataset is then mapped to the [0, 1] interval to eliminate the influence of dimensions. The normalization formula is as follows: ,in: Eigenvalues The normalized value To and The maximum value among similar eigenvalues To and The minimum value among similar characteristic values.

[0121] The normalized indoor-in-situ fusion dataset was divided into training and testing sets. The training set accounted for 75% and the testing set accounted for 25%.

[0122] The basic architecture of a BP neural network: The input layer of the BP neural network has seven nodes, representing particle size, particle size distribution, maximum void ratio, minimum void ratio, vertical effective stress, water content, and relative density. The output layer of the BP neural network has one node, namely the normalized small strain shear modulus. There is one hidden layer, and the number of nodes in the hidden layer is determined by an empirical formula. In the formula, Let m be the number of hidden layer nodes, m be the number of input layer nodes, and n be the number of output layer nodes. The constant is between 1 and 10. According to the formula, the number of hidden layer nodes is calculated to be between 5 and 14. In this embodiment, it is initially selected as 10, thus obtaining the basic architecture of the BP neural network.

[0123] Training the BP neural network: First, initial parameters are set by assigning random numbers to the weight matrices W and V, setting the sample pattern counter and training count counter to 1, setting the error E to 0, and setting the learning rate η to a random number between 0 and 1. The training set is then input into the BP neural network for training to obtain the normalized small strain shear modulus prediction value. The output error of each layer of neurons is calculated using the error function and backpropagated. Then, the weights and thresholds of each layer of neurons are adjusted according to the error function.

[0124] Once the error function is determined, the weights and thresholds of the output and hidden layers can be changed using the gradient descent method to set the error limit. ,when < If training is stopped at a certain point, it indicates that the training error has reached the required level and the training effect is good; if... ≥ Then continue training until... < So far, the trained BP neural network is obtained, which is the normalized small strain shear modulus prediction model.

[0125] 2) PSO algorithm

[0126] The optimization objective function of the PSO algorithm

[0127]

[0128] in, The predicted value is from the BP neural network (unit: kPa). Let be the normalized small-strain shear modulus (unit: kPa) for outdoor conditions. The objective function represents the difference in dynamic stiffness between indoor remolded sand and outdoor undisturbed natural sand. The optimization direction is to find... Minimizing the dynamic stiffness allows the algorithm to automatically explore feasible solution sets in the parameter space that satisfy dynamic stiffness equivalence, achieving high-precision reproduction of soil dynamic response.

[0129] Choose the decision variable vector:

[0130]

[0131] Wherein, vector x represents the three core control parameters of indoor remolded sand, and its component w represents the moisture content, with a constraint range of [0, 20%], which is set according to the indoor test and covers the unsaturated to saturated state; The relative density is constrained within the range of [10%, 90%], corresponding to a loose to a dense state; The effective vertical stress is constrained within a range of [100, 800] kPa, covering the range of common foundation stresses. The constraints are based on the physical limits of laboratory equipment and the feasible domain of soil physical state. This is the vector transpose symbol, representing a column vector (actually a row vector).

[0132] Perform particle swarm initialization:

[0133]

[0134] This expression represents the initial position of the i-th particle in the Particle Swarm Optimization (PSO) algorithm. The first component represents the initial moisture content. This is the lower limit of moisture content (e.g., 0%). The feasible range for moisture content (e.g., 20% - 0% = 20%). Use random numbers to make the initial moisture content within [ Uniformly distributed within. The first component represents the upper limit of moisture content (e.g., 20%); the second component represents the initial relative density. This is the lower limit of relative density (e.g., 10%). The relative density range (e.g., 90% - 10% = 80%). The numbers are random, ensuring that the initial density is randomly distributed within the feasible region. The upper limit of relative density (e.g., 90%); the third component represents the initialization of vertical effective stress. This is the lower limit of stress (e.g., 100 kPa). The stress range (e.g., 800 kPa - 100 kPa = 700 kPa). Use random numbers to generate compliant initial stress values.

[0135] By using random linear mapping, a set of feasible and diverse candidate combinations of initial variable state parameters are generated for each particle, establishing a search starting point for global optimization.

[0136] Initialize particle velocities according to the following formula:

[0137]

[0138] in, This represents the initial velocity vector of the i-th particle in the PSO algorithm. The velocity limiting factor (typically 0.2) constrains the velocity range to prevent oscillations and ensure particle motion stability. , , The random number is uniformly distributed in [0,1], which gives the particles a random exploration direction. The random term makes the particle swarm uniformly cover the search direction, avoiding premature convergence.

[0139] Individual historical optimal initialization is performed according to the following formula:

[0140]

[0141] The above formula defines the initialization rule for the individual historical best position of the i-th particle in the PSO algorithm. Let be the initial, historical best position of particle i. Let be the initial position of particle i. This formula defines the default value of the individual optimal solution during PSO initialization, that is, when a particle is first generated, its initial position is the best position it can find at that time. This is the starting point of the PSO algorithm's "memorizing its own historical best" mechanism.

[0142] Then, the normalized small-strain shear modulus prediction model is invoked, and the fitness is calculated according to the following formula:

[0143]

[0144] Let be the fitness function value of the i-th particle. Let be the normalized small strain shear modulus predicted by the BP neural network for the i-th particle. This indicates a penalty constraint. This is the penalty coefficient (typically 1000), used to amplify the cost of violations and force parameters to stay within a safe range. The k-th constraint function (out of 6) represents six violation conditions: sample saturation liquefaction, undefined moisture content, excessive compaction energy, loose and collapsed sand sample, overload of the universal testing machine, and inability to simulate shallow soil stress state. Let be the variable state parameter vector of the i-th particle. To constrain violations of the activation function.

[0145] Perform velocity update during particle state update:

[0146]

[0147] This formula is the velocity update equation in the PSO algorithm. Let i be the velocity vector of particle i in the t-th iteration. For the inertia of particle motion, Let i be the velocity vector of particle i in the (t-1)th iteration. Trust level based on individual experience For the level of trust in group knowledge, and For random disturbance factors, For the cognitive term (the weights for moving towards the particle's historical best position), This represents the social term (the weights used to move towards the group's globally optimal historical position). Let be the position of particle i at the previous moment. This represents the individual's historical best position (the best solution for particle i to date). This is the globally historical best position (the best solution for the entire group to date).

[0148] Perform position update during particle state update:

[0149]

[0150] in, Let be the position of particle i in the t-th iteration. Let i be the position of particle i in the (t-1)th iteration. Let be the velocity vector of particle i in the t-th iteration.

[0151] The individual historical optimal update and global historical optimal update mechanisms in the PSO algorithm are as follows:

[0152]

[0153]

[0154] In the first formula Let i be the individual historical best position of particle i (the best solution found so far). Let i be the position of particle i in iteration t; Let t be the fitness value of particle i in the tth iteration (the smaller the value, the better). The fitness value is the historical best position. The first formula states that if the fitness of the new position is lower (e.g., 5.2 kPa < 8.7 kPa), then the current position is used. Best historical coverage ;

[0155] In the second formula This is the globally historical best position (the best solution for the entire group to date). To select the parameter that minimizes the function value; Let i be the individual historical best position of particle i (the best solution found so far), i = 1, 2, ..., N, where N is the total number of particles.

[0156] The first formula refers to the individual historical optimal update, where particles "learn on their own" and accumulate local experience; the second formula refers to the global historical optimal update, where the group "shares knowledge" to avoid local optima.

[0157] After updating to the historical best, determine if the termination criterion is met. If not, return to particle swarm optimization and call the normalized small strain shear modulus prediction model; if the criterion is met, output the optimal solution.

[0158]

[0159] in, These are the optimal variable state parameters; For the optimal moisture content, For optimal relative density, The optimal vertical effective stress; This is the vector transpose symbol, representing a column vector (actually a row vector).

[0160] The termination criterion is that optimization will terminate if any of the following conditions are met:

[0161]

[0162] Where t is the current iteration number, To preset the maximum number of iterations, The normalized small-strain shear modulus predicted by the globally optimal parameters is the output of the BP neural network. The normalized small-strain shear modulus of the in-situ target outdoors. This is the modulus tolerance threshold, which is 100 kPa in this example. The distance between the particle's position and the global optimum represents the degree of dispersion in the parameter space. Represents the average distance of the particle swarm. The convergence radius threshold represents a percentage of the parameter range; in this embodiment, it is 1% × parameter range.

[0163] The final model obtained after optimization based on artificial neural networks (using BP neural networks as an example) and intelligent optimization algorithms (using PSO algorithms as an example) is called the indoor-in-situ intelligent reshaping model. This model takes a normalized small-strain shear modulus from an outdoor environment as input, and the intelligent optimization algorithm starts working, generating a series of candidate combinations of variable state parameters (water content, relative density, vertical effective stress). Then, the normalized small-strain shear modulus prediction model is called to predict the corresponding normalized small-strain shear modulus for this series of candidate combinations. The intelligent optimization algorithm is then called again to calculate the fitness, and the fitness value is fed back to the intelligent optimization algorithm to evaluate the merits of the current candidate combinations and update the particle positions. When the convergence condition is met after penalty constraints, the optimal combination of variable state parameters is obtained (see...). Figure 8 ).

[0164] Step 5: Connect to the Intelligent Reshaping Data Management Platform

[0165] The trained indoor-in-situ intelligent reshaping model is burned into the intelligent reshaping data management platform 004. The intelligent reshaping data management platform 004 includes a data processing module, a database, a display module, and the indoor-in-situ intelligent reshaping model. The database achieves physical connection and data interaction with the data processing module through a standardized data interface. The normalized small strain shear modulus data of the outdoor bending element device is processed by the data preprocessing module and then imported into the database, triggering the platform to call the indoor-in-situ intelligent reshaping model to obtain the optimal combination of variable state parameters of the indoor reshaping sand.

[0166] Laboratory personnel prepare samples based on the optimal combination of sand genes and conduct indoor bending element tests. The obtained normalized small strain shear modulus is fed back to the platform as a verification result and stored in the database for evaluating model performance. This result can also be used as a new data point to expand the indoor dataset. Furthermore, the database contains soil parameters for outdoor measurement areas. Correlating the current outdoor normalized small strain shear modulus with these parameters creates new outdoor data points, expanding the outdoor undisturbed dataset and enabling dynamic updates to the outdoor undisturbed dataset. This drives continuous model optimization, forming a closed-loop process of "data acquisition - model calculation - experimental verification - data feedback - model optimization." The intelligent reshaping data management platform 004 has an OpenSees interface, enabling seamless integration of "reshaping parameters - numerical simulation."

[0167] Example 3: The dynamic characteristics-driven sand remodeling gene compilation system in this example (see...) Figure 6 ),include:

[0168] Indoor bending element device 005, to acquire shear wave velocity data of standard sand under different working conditions;

[0169] Electro-hydraulic servo universal testing machine is used to simulate vertical effective stress;

[0170] Outdoor bending element device 006, to acquire shear wave velocity data of outdoor undisturbed natural sand under different working conditions;

[0171] Function signal generator 001 is used to apply a sinusoidal waveform with a set voltage and frequency to the transmitting bending element sensor;

[0172] Function signal amplifier 002 is used to receive and amplify the electrical signal from the receiving bending element sensor;

[0173] Oscilloscope 003 is used to input waveform parameters and simultaneously display the output and received waveforms;

[0174] The Intelligent Remodeling Data Management Platform 004 includes a data processing module, a database, a display module, and an indoor-in-situ intelligent remodeling model;

[0175] The indoor bending element device is a one-dimensional consolidated compression bending element device (see...). Figures 2-4 The test chamber consists of a protective cover and a rectangular test chamber formed by four corrosion-resistant side panels and a bottom panel.

[0176] Pre-drilled holes for installing protective sleeves are provided on a pair of opposite side plates, and threaded holes are provided on at least one remaining side plate. The threaded holes are filled with permeable metal stones. A solenoid valve is installed on the threaded holes to control the drainage conditions during the test.

[0177] The protective sleeve includes a protective sleeve cover and a protective sleeve body. The protective sleeve cover and the protective sleeve body are fixed together by bolts to form a protective sleeve with a pre-reserved encapsulation groove at the front end and a closed rear end. The encapsulation groove is used to fill the bending element sensor, and the front end of the bending element sensor is exposed outside the encapsulation groove. The protective sleeve is T-shaped in general, including a horizontal part and a vertical part. The vertical projection area of ​​the horizontal part of the T-shape is larger than and completely covers the vertical projection area of ​​the vertical part. The vertical part of the protective sleeve is passed out of the experimental box through the pre-reserved hole. At this time, the horizontal part is attached to the inside of the pre-reserved hole. The horizontal part is fixed to the inner wall of the experimental box by bolts, and the vertical part is exposed outside the experimental box.

[0178] The space between the encapsulation recess and the bent element sensor is sealed with silicone.

[0179] The vertical section is provided with a number of adjustment holes along the height direction to adjust the area of ​​the bending element sensor exposed at the front end of the protective sleeve.

[0180] Two bending element sensors are installed on the two side plates: one is a transmitting bending element sensor, and the other is a receiving bending element sensor.

[0181] The outdoor bending element device (see) Figure 5 The device includes a horizontal fixed rod and two vertical fixed rods with conical hammers symmetrically mounted on the horizontal fixed rod. The vertical fixed rods are engraved with a depth scale. Each of the two vertical fixed rods has a bending element sensor at its lower part. One bending element sensor is a receiving bending element sensor, and the other is a transmitting bending element sensor. Protective iron plates are set on the vertical fixed rods above and below the bending element sensors. The bending element sensors are sealed with silicone after being encapsulated with epoxy resin.

[0182] The function signal generator 001 is connected to the transmitting bending element sensor at one end via a BNC cable and to the oscilloscope 003 at the other end. The function signal amplifier 002 is connected to the receiving bending element sensor at one end via a BNC cable and to the oscilloscope 003 at the other end. Both the transmitted and received sine waves are displayed on the oscilloscope 003.

[0183] The Intelligent Reshaping Data Management Platform 004 is the core hub of the digital ecosystem. It mainly includes a data processing module, a database, a display module, and an indoor-in-situ intelligent reshaping model. The data processing module can acquire different types of standard sand and their corresponding normalized small strain shear modulus under different moisture contents, relative densities, and vertical effective stresses. It can also acquire outdoor undisturbed natural sand at different measurement locations and depths, along with their corresponding particle size, particle size distribution, maximum void ratio, minimum void ratio, vertical effective stress, moisture content, relative density, and normalized small strain shear modulus. The module then processes the data to obtain an indoor-in-situ fused dataset.

[0184] The indoor-in-situ intelligent remodeling model takes the normalized small strain shear modulus of the outdoor soil as input, calls the artificial neural network to predict the corresponding normalized small strain shear modulus, and uses the intelligent optimization algorithm to drive the acquisition of the variable state parameters of the corresponding indoor remodeled sand.

[0185] The database stores the variable state parameters and inherent physical parameters of standard sand and outdoor undisturbed natural sand under different working conditions, as well as the normalized small strain shear modulus, and supports engineering applications to call it. The database is connected to the data processing module, indoor-in-situ intelligent reshaping model and display module through standardized interfaces. The data processed by the data processing module is stored in the database. The indoor-in-situ intelligent reshaping model can call the data in the database to dynamically update the model and iteratively optimize it. The display module supports the visualization of parameters.

[0186] Any aspects not covered in this invention are applicable to existing technologies.

Claims

1. A method for compiling gene remodeling of sandy soil driven by dynamic characteristics, characterized in that, The compilation method Includes the following: The small-strain shear modulus of remolded sand was obtained through indoor experiments under different vertical effective stresses, water contents, and relative densities while keeping the particle size, particle size distribution, maximum void ratio, and minimum void ratio constant. The small-strain shear modulus of the indoor remolded sand was then corrected to obtain the normalized small-strain shear modulus. The normalized small-strain shear modulus was then correlated with the particle size, particle size distribution, maximum void ratio, minimum void ratio, and corresponding vertical effective stress, water content, and relative density of the standard sand used to form an indoor dataset. The vertical effective stress and small strain shear modulus of undisturbed natural sand at different burial depths were obtained through outdoor experiments. Simultaneously, samples of undisturbed natural sand were taken, and the particle size, particle size distribution, maximum void ratio, minimum void ratio, water content, and relative density of the undisturbed natural sand were obtained in the laboratory. The small strain shear modulus of the undisturbed natural sand was corrected to obtain the normalized small strain shear modulus. The normalized small strain shear modulus of the outdoor sand was then correlated with particle size, particle size distribution, maximum void ratio, minimum void ratio, vertical effective stress, water content, and relative density to form an outdoor undisturbed dataset. The indoor dataset and the outdoor untouched dataset are merged to form an indoor-in-situ fused dataset; Using particle size, particle size distribution, maximum void ratio, minimum void ratio, vertical effective stress, moisture content, and relative density as input features, and normalized small strain shear modulus as the target output, an artificial neural network is constructed. The artificial neural network is then trained using an indoor-in-situ fusion dataset to obtain a normalized small strain shear modulus prediction model. An indoor-in-situ intelligent remodeling model based on the normalized small strain shear modulus prediction model and intelligent optimization algorithm was constructed. The indoor-in-situ intelligent remodeling model takes the outdoor normalized small strain shear modulus as input to obtain the optimal variable state parameters of the indoor remodeled sand. The optimization objective function of the indoor-in-situ intelligent remodeling model is the absolute value of the difference between the predicted value of the normalized small strain shear modulus prediction model and the corresponding input outdoor normalized small strain shear modulus. The construction process of the indoor-in-situ intelligent remodeling model is as follows: Input the normalized small strain shear modulus of the outdoor undisturbed natural sand, and denote it as the target value; set the constraint range of each variable state parameter, and generate candidate combinations of variable state parameters through random linear mapping within the constraint range. Then, the normalized small strain shear modulus prediction model is called, and the corresponding normalized small strain shear modulus is predicted using the candidate combination of variable state parameters, which is denoted as the predicted value. Fitness is calculated by summing the absolute value of the difference between the predicted value and the target value with the penalty constraint term; The optimal variable state parameter is obtained by performing an iterative process to minimize the absolute value of the difference between the predicted value and the target value. The penalty constraint term includes six constraint violation activation functions, which correspond to six violation conditions. The six violation conditions are: sample saturation liquefaction, undefined moisture content, excessive compaction energy, loose and collapsed sand sample, overload of universal testing machine, and inability to simulate shallow soil stress state. The variable state parameter is at least one of vertical effective stress, water content, and relative density; The components of the "sand gene" are defined as inherent physical parameters and variable state parameters. Among them, particle size, particle size distribution, maximum void ratio, and minimum void ratio are used as inherent physical parameters, while vertical effective stress, water content, and relative density are used as variable state parameters. Gene compilation is the compilation of variable state parameters through intelligent optimization algorithms.

2. The method for compiling sand remodeling genes driven by dynamic characteristics according to claim 1, characterized in that, The artificial neural network is at least one of a BP neural network, an ANN neural network, or an RNN neural network, and the intelligent optimization algorithm is at least one of a PSO algorithm, a GWO algorithm, or a GA algorithm.

3. A dynamic property-driven gene compilation system for reshaping sandy soil, characterized in that, The system includes: Indoor bending element device to acquire shear wave velocity data of standard sand under different working conditions; Electro-hydraulic servo universal testing machine is used to simulate vertical effective stress; An outdoor bending element device is used to acquire shear wave velocity data of outdoor undisturbed natural sand under different working conditions. A function signal generator is used to apply a sinusoidal waveform with a set voltage and frequency to the transmitting bending element sensor; A function signal amplifier is used to receive and amplify the electrical signal from the bending element sensor; An oscilloscope is used to input waveform parameters and simultaneously display the output and received waveforms. The intelligent reshaping data management platform includes a data processing module, a database, a display module, and an indoor-in-situ intelligent reshaping model; The data processing module acquires the normalized small strain shear modulus of different types of standard sand under different moisture contents, relative densities, and vertical effective stresses. It also acquires the corresponding particle size, particle size distribution, maximum void ratio, minimum void ratio, vertical effective stress, moisture content, relative density, and normalized small strain shear modulus of outdoor undisturbed natural sand at different measurement locations and depths. The module then organizes the data to obtain an indoor-in-situ fusion dataset. The indoor-in-situ intelligent reshaping model takes the normalized small strain shear modulus of the outdoor soil as input and initializes and generates candidate combinations of variable state parameters within the constraints of variable state parameters. It calls an artificial neural network to predict the corresponding normalized small strain shear modulus of the candidate combinations of variable state parameters and uses an intelligent optimization algorithm to obtain the variable state parameters of the corresponding indoor reshaping sand. The database stores variable state parameters and inherent physical parameters of standard sand and outdoor undisturbed natural sand under different working conditions, as well as normalized small strain shear modulus, and supports engineering application calls. The database communicates with the data processing module, the indoor-in-situ intelligent reshaping model, and the display module through standardized interfaces. The data processed by the data processing module is stored in the database. The indoor-in-situ intelligent reshaping model can call the data in the database to dynamically update and iteratively optimize the indoor-in-situ intelligent reshaping model. The display module supports the visualization of parameters. The construction process of the indoor-in-situ intelligent remodeling model is as follows: Input the normalized small strain shear modulus of the outdoor undisturbed natural sand, and denote it as the target value; set the constraint range of each variable state parameter, and generate candidate combinations of variable state parameters through random linear mapping within the constraint range. Then, the normalized small strain shear modulus prediction model is called, and the corresponding normalized small strain shear modulus is predicted using the candidate combination of variable state parameters, which is denoted as the predicted value. Fitness is calculated by summing the absolute value of the difference between the predicted value and the target value with the penalty constraint term; The optimal variable state parameter is obtained by performing an iterative process to minimize the absolute value of the difference between the predicted value and the target value. The penalty constraint term includes six constraint violation activation functions, which correspond to six violation conditions. The six violation conditions are: sample saturation liquefaction, undefined moisture content, excessive compaction energy, loose and collapsed sand sample, overload of universal testing machine, and inability to simulate shallow soil stress state. The components of the "sand gene" are defined as inherent physical parameters and variable state parameters. Among them, particle size, particle size distribution, maximum void ratio, and minimum void ratio are used as inherent physical parameters, while vertical effective stress, water content, and relative density are used as variable state parameters. Gene compilation is the compilation of variable state parameters through intelligent optimization algorithms.

4. The dynamic characteristic-driven sand remodeling gene compilation system according to claim 3, characterized in that, The indoor bending element device includes: a protective sleeve, and a rectangular experimental box formed by four side plates and a bottom plate that have been treated with anti-corrosion. Pre-drilled holes for installing protective sleeves are provided on a pair of opposite side plates, and threaded holes are provided on at least one remaining side plate. The threaded holes are filled with permeable metal stones, and a solenoid valve for controlling drainage conditions during the test is installed on the threaded holes. The protective sleeve includes a protective sleeve cover and a protective sleeve body. The protective sleeve cover and the protective sleeve body are fixed together by bolts to form a protective sleeve with a pre-reserved encapsulation groove at the front end and a closed rear end. The encapsulation groove is used to fill the bending element sensor, and the front end of the bending element sensor is exposed outside the encapsulation groove. The protective sleeve is T-shaped in general, including a horizontal part and a vertical part. The vertical part of the protective sleeve is passed through the reserved hole of the experimental box and exits outside the experimental box. At this time, the horizontal part is attached to the inside of the reserved hole. The horizontal part is fixed to the inner wall of the experimental box by bolts, and the vertical part is exposed outside the experimental box. The space between the encapsulation recess and the bent element sensor is sealed with silicone. The vertical section is provided with a number of adjustment holes along the height direction to adjust the area of ​​the bending element sensor exposed at the front end of the protective sleeve. Two bending element sensors are installed on the two side plates: one is a transmitting bending element sensor, and the other is a receiving bending element sensor. The outdoor bending element device includes a horizontal fixed rod and two vertical fixed rods with conical hammers symmetrically and movably mounted on the horizontal fixed rod. The vertical fixed rods are engraved with a depth scale. Bending element sensors are installed at the lower part of the two vertical fixed rods. One bending element sensor is a receiving bending element sensor, and the other bending element sensor is a transmitting bending element sensor. Protective iron plates are installed on the vertical fixed rods above and below the bending element sensors. After the bending element sensors are encapsulated with epoxy resin, they are sealed with silicone between themselves and the vertical fixed rods.

5. The dynamic characteristic-driven sand remodeling gene compilation system according to claim 4, characterized in that, Heavy-duty feet are installed at the bottom of the experimental chamber.

Images