A method and system for evaluating the influence of immersed tube foundation trench excavation based on differentiable simulation, parameter optimization

Abstract

The application belongs to the technical field of the cross of artificial intelligence and computational mechanics, and particularly relates to a method and system for evaluating and optimizing parameters of immersed tube foundation trench excavation based on differentiable simulation, which comprises collecting initial data and constructing a finite element model based on the initial data; quantifying an engineering target as a loss function and setting an optimization target, solving the mechanical response of a neighboring structure under current construction parameters through forward simulation of the foundation trench excavation, and calculating the gradient of the loss function propagated to the loss function by the construction parameters to be optimized in reverse; then updating the construction parameters according to the gradient and a preset learning rate, repeating the forward solving and parameter updating process until the loss function converges and the optimal construction parameters are output, and then converting the optimal construction parameters into a specific implementation scheme. The method realizes accurate optimization of construction parameters with the help of differentiable simulation, can effectively evaluate the influence of excavation on the neighboring structure, and provides reliable technical support and scientific basis for the immersed tube foundation trench excavation.

Description

Technical Field

[0001] This invention belongs to the field of interdisciplinary technology of artificial intelligence and computational mechanics, and specifically relates to a method and system for impact assessment and parameter optimization of immersed tunnel trench excavation based on differentiable simulation. Background Technology

[0002] In the field of engineering design, finite element method (FEM) modeling is a commonly used physical simulation method for assessing the impact of new projects on existing structures. However, traditional FEM methods have a series of inherent limitations in impact analysis, which significantly restrict their effectiveness and efficiency in complex engineering problems. First, traditional FEM methods often have a relatively singular approach to parameter sensitivity analysis, making it difficult to comprehensively consider the interactions between multiple parameters and their combined impact on structural response. This leads to the need for multiple model adjustments and calculations during parameter optimization, increasing computational costs and extending the design cycle. Furthermore, because the calculation results lack clear guidance for design and construction, engineers' input during modeling often relies heavily on personal experience. All of these factors can easily lead to the optimization process getting stuck in local optima, making it impossible to guarantee finding the globally optimal solution.

[0003] During the excavation of the foundation trench for immersed tunnels, controlling the vibration and displacement of adjacent structures is crucial for ensuring construction safety and quality. Currently, model tests or finite element simulations are often used for research. However, parameter selection often relies on engineers' experience, which may lead to uncontrollable influences on adjacent structures. Although some studies combine traditional FEM with optimization algorithms (such as genetic algorithms or particle swarm optimization) for optimization, they do not directly utilize gradient optimization and require cross-platform data transfer, which is cumbersome and inefficient. Specifically, this manifests in the following ways: First, the optimization and simulation processes are decoupled, and cross-platform data exchange increases the risk of errors and reduces efficiency; second, optimization algorithms struggle to obtain parameter sensitivity information and rely on global searches or trial calculations, resulting in high computational costs and difficulty in timely completion when facing multi-parameter, high-dimensional problems; third, the initial range and sensitivity of parameters depend on experience, making it easy to get trapped in local optima, and the output results are not sufficiently instructive and cannot meet the needs of precise construction. Summary of the Invention

[0004] To address the aforementioned problems in existing technologies, this invention provides a method and system for assessing the impact of immersed tunnel excavation and optimizing parameters based on differential simulation. This solves the problem that traditional methods cannot directly obtain the sensitivity of physical response to construction parameters and can only rely on repeated and independent finite element calculations for parameter optimization.

[0005] The objective of this invention can be achieved through the following technical solution: a method for impact assessment and parameter optimization of immersed tunnel excavation based on differentiable simulation, characterized by comprising the following steps:

[0006] S1: Collect initial data related to the area to be excavated in the immersed tunnel trench and adjacent existing structures, set the control range of the construction parameters to be optimized, and construct a finite element model based on the collected initial data and the defined control range.

[0007] S2: Quantify the engineering objective into a loss function and set the optimization objective of the loss function. Perform a forward solution of the mechanical response of the adjacent structure under the current construction parameters by simulating the excavation of the foundation trench on the initial data. Based on the forward simulation process, calculate the construction parameters to be optimized and backpropagate them to the gradient of the loss function.

[0008] S3: Based on the gradient of the loss function and the preset learning rate, the optimizer updates the construction parameters, and re-inputs the updated construction parameters into the differentiable finite element solver. The solution process of S2 is repeated to calculate the mechanical response and loss function value corresponding to the updated parameters. The solution is iterated until the loss function converges and the optimal construction parameters are output.

[0009] S4: Visualize the mechanical response results after iterative convergence to generate an intuitive physical field analysis report. By comparing the response data before and after optimization, evaluate the impact of the foundation trench excavation on adjacent structures and the effect of parameter optimization, and transform the optimal construction parameters into specific implementation plans.

[0010] Preferably, the initial data includes the geometric dimensions of the project, material properties, boundary conditions, construction technology and construction parameters, and the construction parameters to be optimized are the single-shot impact energy generated during excavation and rock drilling;

[0011] The range of the single-shot impact energy P is determined by fitting the impact acceleration data as a function of impact time, obtained through on-site measurements using the rock drilling hammer, to an envelope function, and then integrating the result to obtain the envelope value. The fitting formula is as follows:

[0012] ;

[0013] In the formula: , For the start and end times of the impact; It is the amplitude coefficient; , The coefficients are used to describe the rate of descent and ascent of the envelope, respectively.

[0014] Preferably, S2 further includes the following sub-steps:

[0015] S21: Combining engineering objectives, the multi-objective requirements are transformed into a quantifiable comprehensive loss function. The optimization objective is set as minimizing the comprehensive loss function. A learner is used to adjust the construction parameters to minimize the loss value.

[0016] S22: Input the finite element model into the differentiable finite element solver and set the learning rate of the learner;

[0017] S23: Forward solution of mechanical response: Based on the initial data and current construction parameters in step S1, the forward simulation of the foundation trench excavation process is performed using a differentiable finite element solver to calculate the mechanical response at key locations during construction. Based on the forward simulation process, the construction parameters to be optimized are backpropagated to the gradient of the loss function.

[0018] S24: Calculate the gradient of the response loss relative to the construction parameters using the automatic differentiation function of the differentiable solver.

[0019] Preferably, the formula for calculating the comprehensive loss function in step S21 is as follows:

[0020] ;

[0021] Where E represents the total energy absorbed by the adjacent structure under the influence of construction, which is the sum of the structural kinetic energy and strain energy; D represents the norm of the maximum displacement of the specified monitoring node, used to control the deformation at key locations; C represents the construction efficiency required to complete the excavation task, defined as the number of impacts, which penalizes negative optimization to avoid generating meaningless solutions such as zero or minimal impact energy; λ1, λ2, and λ3 are the weight coefficients of each item.

[0022] Preferably, the formula for calculating the total energy E is:

[0023] ;

[0024] in, This is the kinetic energy term, used to represent the amount of motion of a single unit. It is the velocity vector of the i-th unit at time t. This is the transpose of the velocity vector. No. The quality matrix of each unit;

[0025] Used to represent the energy storage of a unit elastic deformation. The structure is respectively Each unit at time... The displacement vector, This is the transpose of the displacement vector. For the first The stiffness matrix of each element.

[0026] Preferably, the formula for calculating the norm D of the maximum displacement of the monitoring node is:

[0027] ;

[0028] in, Represents the observation node on the structure From time 0 to The furthest distance from the initial position; : Used to obtain the peak displacement in the time domain, representing a value at a non-fixed time. The norm representing the displacement is usually the Euclidean distance. This refers to the total number of observation nodes set up during construction.

[0029] A system for implementing an impact assessment and parameter optimization method for immersed tunnel trench excavation based on differentiable simulation, comprising:

[0030] The data processing module is used to read initial data, which includes the geometric dimensions, material properties, boundary conditions, construction technology and construction parameters of the actual project, and then transmits the initial data to the forward simulation module after standardization.

[0031] The forward simulation module is used to receive the data read by the data processing module, and use the data to build a finite element model and perform simulation. According to the engineering goal, the mechanical response is converted into a quantitative index, which is substituted into a preset comprehensive loss function to calculate the loss value under the current parameter combination, and the loss value is transmitted to the backward gradient propagation and optimization module.

[0032] The backward gradient propagation and optimization module is used to calculate the gradient and optimize the construction parameters based on the gradient. The built-in optimizer is used to perform directional propagation and parameter iteration.

[0033] The data storage module is used to store the running data generated by the finite element model during the simulation process. The running data includes the model's state data, gradient data, and simulation result data.

[0034] The analysis and visualization module, after receiving the optimal parameters output by the backward gradient propagation module, retrieves multi-round simulation data from the data storage module, compares key indicators before and after optimization, clarifies the influence weight of each parameter on the engineering objective, and generates diverse visualization results.

[0035] The beneficial effects of this invention are as follows:

[0036] This invention constructs a finite element model by collecting initial construction data, setting the control range of construction parameters to be optimized, and then quantifying the engineering objective into a loss function and setting an optimization target. The mechanical response of adjacent structures under the current construction parameters is solved through trench excavation simulation, while simultaneously calculating the gradient of the construction parameters to be optimized propagating back to the loss function. Subsequently, based on this gradient and a preset learning rate, the construction parameters are updated, and the forward solution and parameter update process is repeated until the loss function converges and the optimal construction parameters are output. Finally, the converged mechanical response results are visualized to generate a physics field analysis report, comparing the data before and after optimization to assess the impact and optimization effect, and transforming the optimal construction parameters into a specific implementation plan. This method achieves precise optimization of construction parameters through differentiable simulation, effectively assessing the impact of immersed tunnel trench excavation on adjacent structures, and providing reliable technical support and scientific basis for the smooth implementation of immersed tunnels. Attached Figure Description

[0037] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings.

[0038] Figure 1 This is a flowchart of the method steps of the present invention;

[0039] Figure 2 This is a flowchart of the sub-steps executed in step S2 of the present invention;

[0040] Figure 3 This is a structural block diagram of the system of the present invention. Detailed Implementation

[0041] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided.

[0042] Please see Figures 1-3 This embodiment provides a method for impact assessment and parameter optimization of immersed tunnel trench excavation based on differentiable simulation, including the following detailed steps:

[0043] S1: Initial Data Acquisition and Finite Element Model Construction: Initial data related to the area to be excavated in the immersed tunnel trench and adjacent existing structures are collected, and the control range of the construction parameters to be optimized is set. Based on the collected data, a finite element model is constructed. Specifically, the initial data includes information on the area to be excavated in the immersed tunnel trench and relevant information on adjacent existing structures. The information on the area to be excavated in the immersed tunnel trench includes geometric dimensions, material properties, boundary conditions, construction technology, and construction parameters, specifically including: soil layer distribution, soil mechanical parameters (such as cohesion, internal friction angle, and elastic modulus), and groundwater depth. Relevant information on adjacent existing structures includes structural type (bridge foundation, underground pipeline, rail transit, and existing immersed tunnel sections, etc.), material, dimensions, burial depth, and current structural status (such as settlement and crack distribution, etc.).

[0044] Based on this, the construction parameters to be optimized and their control range are clearly defined. In this system example, the optimized construction parameters only use single-click impact energy, and the drilling efficiency is calculated from the single-click impact energy and time. Given that the only construction parameter to be optimized is single-click impact energy, the control range of single-click impact energy is mainly determined by the impact energy of commonly available rock drilling equipment models after preliminary conversion and calculation to avoid excessive optimization difficulty. The single-click impact energy P is determined by fitting the impact acceleration data as a function of impact time to an envelope function obtained from on-site measurements of commonly used excavation rock drilling hammers, and then integrating the envelope value. The fitting formula is:

[0045] ;

[0046] In the formula: , For the start and end times of the impact; It is the amplitude coefficient; , The fitting coefficients describe the rate of descent and ascent of the envelope to a certain extent. The control range of the single-click impact energy P needs to be determined in conjunction with specific engineering specifications and the performance of on-site equipment. Finally, the above data is converted into a numerical model using finite element analysis software (such as ANSYS or ABAQUS), mesh elements are generated, and the constitutive relationship, contact conditions, and boundary constraints (such as displacement constraints and hydraulic boundaries) between the soil and the structure are defined to construct a finite element model that can reflect the interaction between the foundation trench excavation and the existing structure.

[0047] S2: The engineering objective is transformed into a quantifiable loss function. Using a differentiable finite element solver and gradient descent optimization algorithm, the construction parameters are automatically optimized to minimize the comprehensive loss function. The specific execution process is as follows:

[0048] S21: Combining engineering objectives, the multi-objective requirements are transformed into a quantifiable comprehensive loss function. The optimization objective is set as minimizing the comprehensive loss function. A learner is used to adjust the construction parameters to minimize the loss value.

[0049] Using the maximum settlement of the existing structure, the stability coefficient of the foundation trench slope, the total excavation time, and the amount of support material used as inputs, a loss function is constructed through linear weighted summation. ,

[0050] loss function The calculation formula is:

[0051]

[0052] Where: E is the physical safety term, which is used to define the total energy absorbed by the adjacent structure under the influence of construction. The total energy is the sum of the structural kinetic energy and strain energy. Minimizing E can reduce the risk of structural damage.

[0053] D is the deformation control term, which defines the norm of the maximum displacement of the specified monitoring node and is used to control the deformation at key locations;

[0054] C is the economic cost term, used to define the construction efficiency required to complete the excavation task. It is usually quantified as the number of mechanical operations (such as the number of impacts or the number of drillings) or the material loss rate. It is used to penalize negative optimization to avoid producing meaningless solutions such as zero or minimal impact energy.

[0055] λ1, λ2, and λ3 are the weight coefficients of each item. They can be determined using empirical assignment or multi-objective optimization algorithms, or they can be set according to the project priority (e.g., when the adjacent structure is an important bridge, the settlement weight can be increased to 0.6).

[0056] The optimization objective is to minimize this comprehensive loss function, that is, to reduce the loss value to a minimum by adjusting the construction parameters;

[0057] S22: Input the finite element model into the differentiable finite element solver and set the learner's learning rate. Specifically, import the constructed finite element model into the differentiable finite element solver (such as PyTorch-FE or TensorFlow-FEM based on automatic differentiation technology). FEM can retain the computational graph during the simulation process, supporting subsequent gradient solving. Meanwhile, the optimizer's default learning rate... The initial learning rate is set to 0.2. The learning rate determines the step size of parameter updates and needs to be adjusted according to the sensitivity of the parameters (for example, for excavation speed parameters that have a significant impact on settlement, a smaller learning rate can be used to avoid iterative oscillations).

[0058] S23: Forward Solution of Mechanical Response: Based on the initial data and current construction parameters in step S1, a forward simulation of the immersed tunnel excavation process is performed using a differentiable finite element solver to calculate the mechanical response at key locations during construction. Specifically, this includes:

[0059] Forward solution of mechanical response: Based on the initial data in S1 and the current construction parameters (such as the initially set excavation speed, layer thickness, and zoning range), a differentiable finite element solver is used to perform a forward simulation of the foundation trench excavation process, calculating the mechanical response at key locations during construction. This includes the displacement field and stress distribution of the existing structure, the deformation and stability of the foundation trench slope, changes in pore water pressure, and the internal forces of the support structure. The differentiable finite element solver can be optimized using GPUs to accelerate the solution and automatically differentiates and solves the gradient during the forward solution process.

[0060] In the forward solution process, the excavation and drilling simulation assumes an excavation and drilling volume. With click energy and rock strength Related, The calculation expression is:

[0061] ;

[0062] in, This is a correction constant; The crushing energy threshold is the minimum impact energy required to effectively crush rocks. For energy sensitivity, Typically less than 1, it is used to describe the diminishing marginal effect of rock failure response caused by increasing impact energy as drilling volume increases; This is the strength influence index. In excavation and rock drilling simulations, this formula is used to passivate the removed elements by multiplying the element material properties by a minimum coefficient. This minimizes the impact of the drilled portion on the simulation, approximating failure, while maintaining the mesh topology and preserving the differentiability of gradient calculations.

[0063] During the simulation process, it is necessary to perform step-by-step calculations according to the excavation stage (e.g., first excavate the first layer of soil and rock, solve the mechanical response of that stage, and then simulate the excavation of the next layer) to ensure consistency with the actual construction steps.

[0064] Based on the forward simulation process, the construction parameters to be optimized are backpropagated to the gradient of the loss function:

[0065] Calculate the gradient of the response loss with respect to construction parameters: Using the automatic differentiation function of the differentiable solver, based on the mechanical response and comprehensive loss function obtained from the forward simulation, calculate the partial derivative (gradient) of the loss value with respect to each construction parameter to be optimized. For example, calculate "the rate of change of the loss value with excavation speed" and "the rate of change of the loss value with support stiffness". The sign of the gradient indicates the direction of parameter adjustment (i.e., a positive gradient means that increasing the parameter will increase the loss and the parameter needs to be decreased, while a negative gradient is the opposite). The absolute value of the gradient indicates the degree of influence of the construction parameter on the loss value (the larger the absolute value, the more sensitive the parameter).

[0066] Step S2 establishes the correlation between construction parameters and engineering response through "target quantization - forward simulation - gradient solution".

[0067] S3: Parameter Iterative Optimization and Convergence Judgment

[0068] Based on the gradient obtained in S23 and the preset learning rate, the construction parameters are updated using an optimizer. The optimizer can employ any optimization algorithm, including Adam, RMSprop, or traditional gradient descent. The optimization path is: New parameter value = Current parameter value - Learning rate × Gradient value. For example, if the gradient of the excavation speed is positive, it indicates that the current speed is too high, leading to increased losses. Therefore, the excavation speed is reduced according to "New speed = Current speed - Learning rate × Gradient".

[0069] Re-enter the updated construction parameters into the differentiable finite element solver and repeat the forward solution process in S23. Calculate the mechanical response and loss function values ​​under the new parameters until the loss function converges or the preset iteration conditions are met. Here, the iteration conditions are the conditions after the update. Less than the predetermined threshold .

[0070] Continue iterating through the above process of "parameter update - forward solution - loss calculation" until the loss function value tends to stabilize (e.g., the change in loss value over 5 consecutive iterations is less than 1e-5) or the preset maximum number of iterations is reached (e.g., 100 times). The construction parameters obtained at this point are the better solution for the current optimization stage.

[0071] S4: Results Visualization and Implementation Plan Transformation

[0072] The mechanical response results (such as settlement contour maps, stress distribution contour lines, and displacement time series curves) after iterative convergence are visualized to generate an intuitive physical field analysis report. By comparing the response data before and after optimization (such as reducing the maximum settlement of the existing structure from 15mm to 8mm after optimization), the impact of the foundation trench excavation on adjacent structures and the effect of parameter optimization are evaluated. At the same time, the engineering applicability analysis of the converged optimal construction parameters (such as optimal excavation speed of 0.8m / h, layer thickness of 2m, and support stiffness of 30GPa) is conducted. Combined with the on-site construction conditions (such as equipment model and geological variations), the parameters are transformed into specific implementation plans. For example, the model and operating efficiency of the rock drill are determined based on the optimal excavation speed, the specific elevation of the excavation layer is determined based on the layer thickness, and the support materials (such as steel sheet pile model and grouting ratio) are selected based on the support stiffness requirements.

[0073] Among them, the optimal excavation parameters are analyzed and transformed into specific methods for the scheme. Based on the obtained optimal single-click energy and simulation step size, the corresponding excavation frequency is calculated in reverse. Combined with the total excavation volume, quantitative guidance is provided for selecting or matching suitable excavation equipment while minimizing the impact on adjacent structures and completing the predetermined rock drilling excavation task.

[0074] Furthermore, in step S21, the formula for calculating the physical safety item (total energy) E is:

[0075] ;

[0076] in, This is the kinetic energy term, used to represent the amount of motion of a single element, and is sensitive to impact loads, such as explosions. It is the velocity vector of the i-th unit at time t. This is the transpose of the velocity vector. No. The element mass matrix of each element, which includes material density and geometric properties;

[0077] Used to represent the energy storage of an element's elastic deformation, sensitive to static conditions (such as soil pressure). The structure is respectively Each unit at time... The displacement vector, This is the transpose of the displacement vector; For the first The stiffness matrix of each element includes the elastic modulus and topology;

[0078] By calculating the physical energy term E, the construction dynamic process is decomposed into unit-level kinetic energy and strain time integrals. Compared with the traditional static assumption, the integral term over time can better reflect its changes in the time dimension. Using the maximum offset distance in the time domain as the safety criterion is more in line with the risk control logic of the project and provides a basis for the subsequent quantification of construction parameters by the loss function.

[0079] The formula for calculating the displacement term (deformation control term) D is:

[0080] ;

[0081] in, Represents the observation node on the structure From time 0 to The furthest distance from the initial position; Used to obtain the peak displacement in the time domain, representing a value at a non-fixed time. The norm representing the displacement is usually the Euclidean distance; This refers to the total number of observation nodes set up during construction.

[0082] The displacement term is used to control the deformation and damage risk of the monitoring points by using the maximum average displacement of the monitoring nodes throughout the construction process. Compared with the traditional static loss function, the displacement peak value in the time domain is more sensitive to locally generated deformations.

[0083] Please see Figure 3 A system for implementing an impact assessment and parameter optimization method for immersed tunnel trench excavation based on differentiable simulation, the system comprising:

[0084] The data processing module reads initial data, standardizes it, and then transmits it to the forward simulation module. The initial data includes the geometric dimensions, material properties, boundary conditions, construction techniques, and construction parameters of the actual project. The standardization process before transmitting the initial data to the forward simulation module specifically includes:

[0085] After the system starts, the data processing module reads the initial data of the immersed tunnel trench excavation through the set interface. This includes the project's geometric dimensions, material properties, boundary conditions, construction technology, construction parameters, and initial state data, as well as the control range of the construction parameters to be optimized. Material parameters include soil cohesion, elastic modulus, and the degree of settlement of the existing structure. The simulated environment includes groundwater level, ground temperature, and boundary water pressure. Construction technology data includes excavation equipment model and support method description. In this optimization scenario, the initial state corresponds to the "intermediate / final optimization state." For subsequent visualization and display of the optimization process, the state of the initial optimization parameters (here, only the impact energy is needed) is required. The optimization is reflected by a loss function (three terms). The initial values ​​of the loss function and the initial values ​​of the optimization parameters, etc., related to subsequent optimization, constitute the initial state data. For example, the original topography of the trench (geometric dimensions) and the initial value of the existing structure's settlement participate in the calculation of the subsequent loss function, which is one way to represent the initial state data.

[0086] The initial data is standardized and passed to the forward simulation module to provide a foundation for the construction of the finite element model. The standardization process here includes format unification, missing value imputation, and outlier removal.

[0087] The forward simulation module receives the initial data from the data processing module, uses the data to build a finite element model and perform simulation, calculates the loss function, and visually evaluates the impact of the current combination of data parameters on the adjacent structure; specifically, it includes:

[0088] After receiving the initial data, the forward simulation module calls the built-in modeling engine to automatically generate a finite element model of the immersed tunnel trench excavation: it divides the mesh based on topographic data (e.g., a denser mesh is used at the bottom of the trench, and a sparser mesh is used far from the excavation area), defines the element constitutive relations based on material parameters (e.g., the Mohr-Coulomb model is used for soil and rock, and a linear elastic model is used for concrete structures), and sets boundary conditions based on simulation environment parameters (e.g., pore water pressure is applied at the groundwater level, and fixed displacement constraints are set at the model edges). After the finite element model is built, the mesh quality is automatically checked to ensure simulation accuracy.

[0089] Based on the constructed finite element model, the module simulates the foundation trench excavation process according to the construction process data: starting from the initial state, the grid units are gradually "removed" according to the preset excavation sequence (such as layered excavation thickness and segmented advancement distance), and the mechanical response after each step of excavation is calculated simultaneously (displacement of existing structures, horizontal displacement of foundation trench slope, stability, and internal forces of support structures).

[0090] After the simulation is completed, the module converts the mechanical response into a quantitative index based on the engineering objectives (such as controlling the maximum settlement of the existing structure to ≤10mm and the slope safety factor to ≥1.3), substitutes it into the preset comprehensive loss function, and calculates the loss value under the current parameter combination.

[0091] The backward gradient propagation and optimization module is used to calculate the gradient and optimize the construction parameters based on the gradient. The built-in optimizer is used to execute the complete optimization loop of directional propagation and parameter iteration. Specifically, after the forward simulation module outputs the loss value, the backward gradient propagation and optimization module starts the automatic differentiation function: based on the differentiable finite element theory, it traces back along the simulation calculation graph to calculate the gradient of the loss function with respect to each construction parameter (such as excavation speed, support stiffness, grouting pressure). The gradient calculation results are transmitted to the data storage module in real time.

[0092] The updated parameters are fed back to the forward simulation module, triggering a new round of finite element simulation and loss function calculation. The module repeats the cycle of "gradient calculation - parameter update - forward simulation" until the loss function converges, at which point the optimal solution for construction parameters is output.

[0093] The data storage module is used to synchronously start the recording function during system operation: during the simulation process, it stores the state data of the unit in real time (such as the stress and strain values ​​of the finite element mesh element and nodal displacement), the gradient data generated by the backward gradient propagation (such as the partial derivatives of the construction parameters with respect to the loss function), and the simulation result data of each stage (such as the settlement prediction value and slope stability coefficient of each iteration). The storage module adopts a time-series database (such as InfluxDB) structure and stores data in categories according to "simulation stage - parameter combination - response index", supporting data backtracking and comparison query of multi-round simulation results.

[0094] The analysis and visualization module utilizes differentiable simulation to output the computational optimization results and generate visualized physical fields and analytical charts. After receiving the optimal parameters output from the gradient propagation module, the module retrieves multi-round simulation data from the data storage module, compares key indicators before and after optimization, and clarifies the influence weight of each parameter on the engineering objective. It generates diverse visualization results, drawing settlement cloud maps, displacement vector maps, or stress distribution maps to describe the corresponding physical fields in the mechanical response calculation. It also generates loss function iteration curves to intuitively present the convergence process of construction parameter optimization, and the generated charts are displayed using an interactive interface.

[0095] The system automates the entire process from initial data input to visualized output of optimized parameters: the data processing module ensures data continuity, the forward simulation module digitizes the physical process, the backward gradient propagation and optimization module enables intelligent parameter iteration, and the analysis and visualization module transforms simulation results into engineering practice basis, ultimately implementing a method for impact assessment and parameter optimization of immersed tunnel trench excavation based on differentiable simulation.

[0096] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for impact assessment and parameter optimization of immersed tunnel trench excavation based on differentiable simulation, characterized in that: Includes the following steps: S1: Collect initial data related to the area to be excavated in the immersed tunnel trench and adjacent existing structures, set the control range of the construction parameters to be optimized, and construct a finite element model based on the collected initial data and the defined control range. S2: Quantify the engineering objective into a loss function and set the optimization objective of the loss function. Perform a forward solution of the mechanical response of the adjacent structure under the current construction parameters by simulating the excavation of the foundation trench on the initial data. Based on the forward simulation process, calculate the construction parameters to be optimized and backpropagate them to the gradient of the loss function. S3: Based on the gradient of the loss function and the preset learning rate, the optimizer updates the construction parameters, and re-inputs the updated construction parameters into the differentiable finite element solver. The solution process of S2 is repeated to calculate the mechanical response and loss function value corresponding to the updated parameters. The solution is iterated until the loss function converges and the optimal construction parameters are output. S4: Visualize the mechanical response results after iterative convergence to generate an intuitive physical field analysis report. By comparing the response data before and after optimization, evaluate the impact of the foundation trench excavation on the adjacent structure and the effect of parameter optimization, and transform the optimal construction parameters into specific implementation plans. S2 also includes the following sub-steps: S21: Combining engineering objectives, the multi-objective requirements are transformed into a quantifiable comprehensive loss function. The optimization objective is set as minimizing the comprehensive loss function. A learner is used to adjust the construction parameters to minimize the loss value. S22: Input the finite element model into the differentiable finite element solver and set the learning rate of the learner; S23: Forward solution of mechanical response: Based on the initial data and current construction parameters in step S1, the forward simulation of the foundation trench excavation process is performed using a differentiable finite element solver to calculate the mechanical response at key locations during construction. Based on the forward simulation process, the construction parameters to be optimized are backpropagated to the gradient of the loss function. S24: Calculate the gradient of the response loss relative to the construction parameters using the automatic differentiation function of the differentiable solver.

2. The method for impact assessment and parameter optimization of immersed tunnel trench excavation based on differentiable simulation according to claim 1, characterized in that: The initial data includes the geometric dimensions, material properties, boundary conditions, construction technology and construction parameters of the project, and the construction parameters to be optimized are the single-shot impact energy generated during excavation and rock drilling. The range of the single-shot impact energy P is determined by fitting the impact acceleration data as a function of impact time, obtained through on-site measurements using the rock drilling hammer, to an envelope function, and then integrating the result to obtain the envelope value. The fitting formula is as follows: ； In the formula: , For the start and end times of the impact; It is the amplitude coefficient; , The coefficients are used to describe the rate of descent and ascent of the envelope, respectively.

3. The method for impact assessment and parameter optimization of immersed tunnel trench excavation based on differentiable simulation as described in claim 1, characterized in that: The formula for calculating the comprehensive loss function in step S21 is as follows: ； Where E represents the total energy absorbed by the adjacent structure under the influence of construction, which is the sum of the structural kinetic energy and strain energy; D represents the norm of the maximum displacement of the specified monitoring node, used to control the deformation at key locations; C represents the construction efficiency required to complete the excavation task, defined as the number of impacts, which penalizes negative optimization to avoid generating meaningless solutions with zero or minimal impact energy. λ 1. λ 2. λ 3 represents the weighting coefficient for each item.

4. The method for impact assessment and parameter optimization of immersed tunnel trench excavation based on differentiable simulation according to claim 3, characterized in that: The formula for calculating the total energy E is: ； in, This is the kinetic energy term, used to represent the amount of motion of a single unit. It is the velocity vector of the i-th unit at time t. This is the transpose of the velocity vector. For the first The quality matrix of each unit; Used to represent the energy storage of a unit elastic deformation. The structure is respectively Each unit at time... The displacement vector, This is the transpose of the displacement vector. For the first The stiffness matrix of each element.

5. The method for impact assessment and parameter optimization of immersed tunnel trench excavation based on differentiable simulation according to claim 4, characterized in that: The formula for calculating the norm D of the maximum displacement of the monitoring node is: ； in, Represents the observation node on the structure From time 0 to The furthest distance from the initial position; Used to obtain the peak displacement in the time domain, representing a value at a non-fixed time. The norm representing the displacement is usually the Euclidean distance; This refers to the total number of observation nodes set up during construction.

6. A system for implementing the method according to any one of claims 1-5, characterized in that: include: The data processing module is used to read initial data, which includes the geometric dimensions, material properties, boundary conditions, construction technology and construction parameters of the actual project, and then transmits the initial data to the forward simulation module after standardization. The forward simulation module is used to receive the data read by the data processing module, and use the data to build a finite element model and perform simulation. According to the engineering goal, the mechanical response is converted into a quantitative index, which is substituted into a preset comprehensive loss function to calculate the loss value under the current parameter combination, and the loss value is transmitted to the backward gradient propagation and optimization module. The backward gradient propagation and optimization module is used to calculate the gradient and optimize the construction parameters based on the gradient. The built-in optimizer is used to perform directional propagation and parameter iteration. The data storage module is used to store the running data generated by the finite element model during the simulation process. The running data includes the model's state data, gradient data, and simulation result data. The analysis and visualization module, after receiving the optimal parameters output by the backward gradient propagation module, retrieves multi-round simulation data from the data storage module, compares key indicators before and after optimization, clarifies the influence weight of each parameter on the engineering objective, and generates diverse visualization results.

Images