A physical-data dual driven method for evaluating disaster robustness of retaining wall

Abstract

The application discloses a physical-data dual driving retaining wall disaster robustness evaluation method, and relates to the technical field of structure disaster evaluation and computer science. In view of the problem that the existing retaining wall disaster robustness evaluation method cannot correct the physical model error in real time and is insufficient in accuracy, a retaining wall disaster process robustness evaluation method driven by fusion of a physical model and a neural network is provided. The steps are as follows: 1, time discretization and parameter vectorization are performed on the retaining wall disaster monitoring parameters; 2, a performance degradation function is constructed based on the physical model; 3, the hidden damage state is updated in combination with a neural network Cell module; 4, the error is corrected through a feedforward network to obtain a total performance degradation function; and 5, finally, a robustness index is calculated and the corresponding robustness grade is output, so that quantitative evaluation of the retaining wall disaster robustness is realized. The application can provide reliable quantitative support for retaining wall design optimization, operation monitoring and post-disaster recovery evaluation.

Description

Technical Field

[0001] This invention relates to the interdisciplinary fields of structural disaster assessment, civil engineering structural safety monitoring and computer science, and specifically to a physical-data dual-driven method for assessing the disaster robustness of retaining walls. Background Technology

[0002] Retaining walls, as commonly used support structures in geotechnical engineering, are prone to decreased anti-sliding stability, decreased overturning stability, and abrupt degradation of structural function under catastrophic disturbances such as earthquakes, rainstorms, floods, freeze-thaw cycles, erosion, and sudden increases in landslide thrust. Existing disaster robustness assessment methods mostly employ physical analysis models based on safety factors or data-driven statistical learning models. Methods based solely on physical models rely on idealized assumptions and parameter determinism, making it difficult to use real-time monitoring data to correct model errors online during disasters, resulting in insufficient assessment accuracy. While data-driven methods can fit complex nonlinearities, they lack stability mechanism constraints, easily leading to extrapolation distortion, weak interpretability, and insufficient engineering credibility. Summary of the Invention

[0003] Given the limitations of existing disaster robustness assessment methods, which cannot correct physical model errors in real time and suffer from insufficient accuracy, this invention aims to provide a physical-data dual-driven disaster robustness assessment method for retaining walls. This method utilizes stability mechanisms to construct physical priors throughout the entire disaster process, while simultaneously correcting physical model errors online through neural networks. This enables dynamic correction of the retaining wall performance degradation function during disasters, calculates the robustness gap area and robustness index, and outputs the robustness level. This provides accurate and reliable quantitative support for retaining wall design optimization, operational monitoring, and post-disaster recovery effect assessment.

[0004] To achieve the above objectives, this invention provides a physical-data dual-driven method for assessing the disaster robustness of retaining walls, comprising the following steps:

[0005] S1: Input the monitoring parameters of the retaining wall under disaster conditions, perform time discretization and parameter vectorization on the disaster process, and obtain the parameter vector of the disaster process. x k ;

[0006] S2, calculate the anti-skid stability coefficient at each time step based on the input parameters. Q 1( t k and overturning stability coefficient Q 2( t k The retaining wall performance degradation function is obtained through normalized function mapping. F phys ( t kPhysical model;

[0007] S3, construct the Cell module of the neural network that incorporates the physical mechanism, and input the parameter vector at the current time step. x k and the implicit damage state at the previous time step z k-1 Generate a current retaining wall performance degradation function that considers only physical mechanisms. F phys ( t k ) and candidate damage states fused with real-time data And update the implicit damage state at the current time step using a robust fusion gate. z k ;

[0008] S4, calculate the change in the corrected function function at the current time step using a feedforward network. ΔF ( t k ), superimposed performance degradation function F phys ( t k Then, the actual performance degradation function at the current time step is obtained. F total ( t k The total performance degradation function over the entire disaster period is calculated by summing the results. F total ( t );

[0009] S5, based on the total performance degradation function F total ( t ) Calculate the robustness gap area and the corresponding robustness index, output its robustness level, and realize the accurate quantitative assessment of the catastrophic robustness of the retaining wall.

[0010] Furthermore, in S1, the parameter vector of the catastrophe process x k It is used to fully characterize the stress state of the retaining wall at different times of disaster, and to provide parameter input for subsequent physical model calculations and neural network learning.

[0011] The time discretization uses equal time intervals Δ t Partitioning strategy, Δ t Determined based on the duration of the disaster's characteristics. This includes continuous disaster processes. t 0, t n Discretized into n The time step, the first kTime nodes at each time step t k satisfy:

[0012] (1)

[0013] In the formula t 0 represents the start time of the catastrophe; k From 1 to n Natural numbers between.

[0014] The parameter vector x k as follows:

[0015] (2)

[0016] In the formula G The weight of the retaining wall itself; k From 1 to n Natural numbers between; x 0( t k ) is the first k Time nodes at each time step t k The distance from the toe of the retaining wall to the weight of the wall itself; E az ( t k )and x f ( t k ) are respectively the first k Time nodes at each time step t k The vertical component of the corresponding earth pressure and its distance from the toe of the wall; E ax ( t k )and z f ( t k ) are respectively the first k Time nodes at each time step t k The horizontal component of the earth pressure and its distance from the toe of the wall; μ ( t k ) is the first k Time nodes at each time step t k The base friction coefficient of the retaining wall can be monitored by pre-embedded sensors.

[0017] Furthermore, in S2, the anti-skid stability coefficient at each time step Q 1(t k The overturning stability coefficient at each time step characterizes the retaining wall's ability to resist sliding failure along the foundation. Q 2( t k The retaining wall's ability to resist overturning failure around its toe is characterized by a performance degradation function obtained through normalized functional mapping. F phys ( t k Physical model.

[0018] The anti-skid stability coefficient Q 1( t k The calculation formula is:

[0019] (3)

[0020] In the formula G The weight of the retaining wall itself; E az ( t k ) is the first k Time nodes at each time step t k The vertical component of the corresponding earth pressure; E ax ( t k ) is the first k Time nodes at each time step t k The horizontal component of the corresponding earth pressure; μ ( t k ) is the first k Time nodes at each time step t k The base friction coefficient corresponding to the retaining wall.

[0021] The overturning stability coefficient Q 2( t k The calculation formula is as follows:

[0022] (4)

[0023] In the formula G Due to the weight of the retaining wall itself, x 0( t k ) is the first k Time nodes at each time step t k The distance from the toe of the retaining wall to the weight of the wall itself; Eaz ( t k )and x f ( t k ) are respectively the first k Time nodes at each time step t k The vertical component of the corresponding earth pressure and its distance from the toe of the wall; E ax ( t k )and z f ( t k ) are respectively the first k Time nodes at each time step t k The horizontal component of the earth pressure and its distance from the toe of the wall.

[0024] The normalized function mapping is based on the critical failure state of the retaining wall, and the minimum value of the two stability coefficients is taken for normalization to construct a performance degradation function based solely on the physical model. F phys ( t k ):

[0025] (5)

[0026] In the formula [ Q ] max The theoretical limit value of the stability coefficient is determined by the upper limit of the safety factor specified in the retaining wall design code; Q 1( t k )and Q 2( t k These are the anti-skid stability coefficient and the anti-overturning stability coefficient, respectively.

[0027] Furthermore, in S3, the neural network Cell module, which integrates physical mechanisms, is used to characterize the temporal dependency relationship of "damage accumulation - state evolution - functional degradation" during the retaining wall catastrophe process. The core components of the neural network Cell module include a candidate damage state generation unit and a robust fusion gate; wherein, the candidate damage state generation unit is used in the first... k Each time step is based on a parameter vector. x k Implicit damage state compared to the previous time step z k 1. Generate candidate damage states Robust fusion gates are used for the performance degradation function obtained with S2. F phys ( t k Apply physical constraints to candidate damage states and update the latent damage states. z k .

[0028] The candidate damage state The formula for generating it is:

[0029] (6)

[0030] In the formula [ x k , z k-1 ] is the first k Parameter vectors at each time step x k With the k -1 time step of implicit damage state z k 1; k From 1 to n The natural numbers between; tanh is the hyperbolic tangent activation function, which can normalize the input features to the interval [-1,1], avoid numerical overflow caused by excessive differences in the numerical values ​​of different dimensions in the parameter vector, and ensure that the model can quickly learn the path dependency effect of damage accumulation; For the first k Candidate damage states fused from real-time data at each time step are used to characterize the potential damage evolution trend of retaining walls under catastrophic disturbances.

[0031] The gate control coefficient of the robust fusion gate β k The calculation formula is:

[0032] (7)

[0033] In the formula σ The Sigmoid activation function has an output range of [0,1] and exhibits good responsiveness to linear combinations of input features, ensuring the gating coefficients are stable. β k The changes are consistent with the stability degradation process characterized by physical mechanisms; b β The bias vector of the robust fusion gate is essentially the gating coefficient. β k Provide a "baseline offset" to prevent the model from relying too heavily on a linear combination of input features; k From 1 to nNatural numbers between; β k For the first k The gating coefficient of the robust fusion gate at each time step is used to adaptively adjust the contribution ratio between "historical damage memory" and "candidate damage update", and a physical performance decay function is introduced. F phys ( t k This implements physical consistency constraints for updating the damage state.

[0034] The hidden damage state z k The update formula is:

[0035] (8)

[0036] In the formula z k For the first k The updated implicit damage state at each time step is used to characterize the cumulative damage level and path dependence effect of the retaining wall throughout the entire disaster process. k From 1 to n Natural numbers between.

[0037] Furthermore, in S3, the neural network Cell module further considers only the current retaining wall performance degradation function based on physical mechanisms. F phys ( t k ) as a physical prior constraint, and related to the candidate damage state Jointly participate in updating the latent damage state to ensure the latent damage state z k Consistent with the stability degradation process represented by the physical model; wherein, the consistency constraint is achieved by... F phys ( t k The implementation of the robust fusion gate as an input term ensures that the fusion gate weights increase as the rate of functional degradation determined by the physical model increases. β k The system responds accordingly to enhance the absorption of candidate damage states and reflect damage transitions under sudden catastrophic events.

[0038] Furthermore, in S4, a feedforward neural network is used to establish the residual mapping relationship between the parameter vector and the latent damage state to the change in the functional function, so as to correct the physical model error in real time. k The change in the corrected function function Δ at each time step F ( t k )satisfy:

[0039] (9)

[0040] In the formula g θ [,] represents the mapping function of a feedforward neural network (FNN). Feedforward neural networks excel at learning complex nonlinear mapping relationships. The error between the physical model and the actual system response is essentially a nonlinear residual caused by the coupling of multiple factors. Feedforward neural networks can learn to accurately fit this residual through multiple layers of weights; Δ F ( t k ) is the first k The change in the function after correction at each time step, i.e., the change relative to the physical performance degradation function. F phys ( t k The residual correction magnitude is used to characterize the systematic deviation and nonlinear error between the physical model and the actual system response.

[0041] The actual performance degradation function F total ( t k By ΔF ( t k ) superimposed to F phys ( t k )get:

[0042] (10)

[0043] In the formula Δ F ( t k ) is the first k The change in the function after correction at each time step; F phys ( t k ) is the first k The performance degradation function of the retaining wall based solely on the physical model at each time step.

[0044] The total performance degradation function of the entire disaster process F total ( t The result is obtained by summing over discrete time steps:

[0045] (11)

[0046] In the formula N The number of time steps in the disaster process.F total ( t ) is the total performance degradation function over the entire disaster period, used for subsequent robustness index calculations.

[0047] Furthermore, in S5, the robust gap area is used to comprehensively characterize the coupling effect of "functional loss amplitude - duration" of the retaining wall throughout the entire disaster process. The robust gap area... A g The calculation formula is:

[0048] (12)

[0049] In the formula T Total duration of the disaster F ( t ) represents the normal service performance function of the retaining wall when no disaster occurs. F total ( t ) is the total performance degradation function during the entire disaster period.

[0050] The retaining wall disaster robustness index R The formula used to normalize the catastrophic robustness of retaining walls is as follows:

[0051] (13)

[0052] The formula contains the catastrophic robustness index of the retaining wall. R ∈(0,1), R The closer the value is to 1, the higher the catastrophic robustness of the retaining wall. R The closer the value is to 0, the lower the catastrophic robustness of the retaining wall.

[0053] The retaining wall's catastrophic robustness level is based on the robustness index. R Size is graded: when R ≥ R At level 1, it can be classified as high robustness. High robustness indicates that the retaining wall experiences minimal functional loss during a disaster, maintains its core load-bearing capacity, and can continue to serve without repair; when R 2≤ R <R At level 1, it can be classified as moderately robust, indicating that the retaining wall has some functional loss, but the core bearing capacity is not affected, and it can be restored to normal service after minor repairs; when R ≤ R A value of 2 indicates low robustness, meaning the retaining wall suffers severe functional loss and its core bearing capacity is impaired, requiring large-scale reinforcement or demolition and reconstruction. R 1 and R2 is the threshold for disaster robustness classification, which is determined based on the limits in the retaining wall design specifications, historical disaster statistics, or expert scoring and evaluation.

[0054] Compared with existing technologies, this invention integrates the core advantages of physical mechanisms and neural networks, breaks through the limitations of traditional single evaluation paradigms, and proposes a physical-data dual-driven method for evaluating the catastrophic robustness of retaining walls, which has the following beneficial effects:

[0055] (1) The physical mechanism and data-driven approach are deeply coupled. Physical prior constraints are constructed by anti-slip and anti-overturning stability coefficients. Combined with the neural network Cell module, the temporal dependency relationship of "damage accumulation-state evolution-functional degradation" is captured. This ensures both the engineering interpretability and traceability of the evaluation results and the ability of the data-driven model to fit complex nonlinear responses. This solves the core problems of pure physical models being difficult to correct online and pure data models lacking physical constraints.

[0056] (2) The robust fusion gate and the residual feedforward network work together. The fusion gate adjusts the contribution ratio of historical damage memory and candidate damage update through adaptive gating coefficient. The residual feedforward network corrects the systematic bias and nonlinear error of the physical model in real time, which significantly improves the evaluation stability and generalization ability under the scenarios of parameter uncertainty and catastrophic change.

[0057] (3) The multi-dimensional robustness quantification system accurately characterizes the impact of disasters. The robustness gap area comprehensively reflects the coupling effect of "functional loss amplitude - duration". The robustness index is normalized and quantified. Combined with the three-level robustness classification, it provides intuitive and clear assessment results for the disaster status of retaining walls, solving the problems of single traditional assessment indicators and fuzzy quantification.

[0058] (4) Based on the full-process quantitative evaluation results, the performance degradation law and damage evolution path of retaining wall during disaster can be systematically analyzed, providing multi-dimensional and reliable quantitative support for retaining wall design optimization, operation monitoring and early warning and post-disaster recovery effect evaluation, and helping to improve engineering safety control and resilience. Attached Figure Description

[0059] Figure 1 This is a schematic diagram of the scenario and system for the embodiment;

[0060] Figure 2 This is a flowchart of the retaining wall disaster robustness assessment method driven by the fusion of physical mechanism and neural network of the present invention.

[0061] Figure 3 This is a schematic diagram of the neural network Cell module structure that integrates physical mechanisms in an embodiment of the present invention;

[0062] Figure 4This is a schematic diagram of the residual feedforward network architecture design in an embodiment of the present invention;

[0063] Figure 5 This is a schematic diagram of the robust gap area and the retaining wall performance function in an embodiment of the present invention. Detailed Implementation

[0064] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. These embodiments are implemented based on the technical solution of the present invention, providing detailed implementation methods and specific operating procedures. However, the scope of protection of the present invention is not limited to the following embodiments.

[0065] Example

[0066] This embodiment provides a physical-data dual-driven method for assessing the catastrophic robustness of retaining walls. It is used to correct the performance degradation of retaining walls in real time during catastrophic disturbances and output robustness index and robustness level, enabling real-time quantitative assessment of the robustness of retaining walls during catastrophic events.

[0067] Figure 1 This is a schematic diagram of the implementation scenario and system. Fiber optic strain gauges are deployed along the depth of the wall on-site, vibrating wire earth pressure sensors are deployed in the back area of ​​the wall, and pre-embedded friction coefficient sensors are deployed at the bottom of the wall. All sensors are directly connected to an industrial-grade data acquisition box deployed at the edge of the site via dedicated data cables. The acquisition box collects the monitoring data from each sensor in real time and transmits the monitoring data of the soil wall disaster process at the disaster site to a remote server. This provides continuous and complete raw data input for subsequent time discretization, parameter vectorization, and robustness assessment processes, and the subsequent implementation steps S1-S5 are run on the server.

[0068] like Figure 2 As shown, a physical-data dual-driven method for assessing the catastrophic robustness of retaining walls includes the following steps:

[0069] S1: Input the monitoring parameters of the retaining wall under disaster conditions, perform time discretization and parameter vectorization on the disaster process, and obtain the parameter vector of the disaster process. x k ;

[0070] S2, calculate the anti-skid stability coefficient at each time step based on the input parameters. Q 1( t k and overturning stability coefficient Q 2( t k The retaining wall performance degradation function is obtained through normalized function mapping. F phys ( t k Physical model;

[0071] S3, construct the Cell module of the neural network that incorporates the physical mechanism, and input the parameter vector at the current time step. x k and the implicit damage state at the previous time step z k-1 Generate a current retaining wall performance degradation function that considers only physical mechanisms. F phys ( t k ) and candidate damage states fused with real-time data And update the implicit damage state at the current time step using robust gates. z k ;

[0072] S4, calculate the change in the corrected function function at the current time step using a feedforward network. ΔF ( t k ), superimposed performance degradation function F phys ( t k Then, the actual performance degradation function at the current time step is obtained. F total ( t k The total performance degradation function over the entire disaster period is calculated by summing the results. F total ( t );

[0073] S5, based on the total performance degradation function F total ( t ) Calculate the robustness gap area and the corresponding robustness index, output its robustness level, and realize the accurate quantitative assessment of the catastrophic robustness of the retaining wall.

[0074] In this embodiment, the above implementation steps are described as follows:

[0075] S1: Time discretization and parameter vectorization of the catastrophe process.

[0076] S1.1 on the continuous disaster process [ t 0, t n Using equal time intervals Δ t The partitioning strategy is discretized into n The time step, the first k Time nodes at each time step t k satisfy:

[0077] (1)

[0078] In the formula t 0 represents the start time of the catastrophe; Δ t The discrete time step; k From 1 to n Natural numbers between.

[0079] S1.2 further vectorizes the monitored disaster condition parameters and uses them as a unified input to the S2 physical model and the S3 neural network, where the first... k Parameter vectors at each time step x k for:

[0080] (2)

[0081] In the formula G Due to the weight of the retaining wall itself, x 0( t k ) is the first k Time nodes at each time step t k The distance from the toe of the retaining wall to the weight of the wall itself; E az ( t k )and x f ( t k ) are respectively the first k Time nodes at each time step t k The vertical component of the corresponding earth pressure and its distance from the toe of the wall; E ax ( t k )and z f ( t k ) are respectively the first k Time nodes at each time step t k The horizontal component of the earth pressure and its distance from the toe of the wall; μ ( t k ) is the first k Time nodes at each time step t k The base friction coefficient corresponding to the retaining wall.

[0082] S2: Stability calculation and performance degradation function construction of retaining wall based on physical model.

[0083] S2.1 Calculates the first... k Anti-skid stability coefficient at each time step Q 1( t k ):

[0084] (3)

[0085] In the formula G The weight of the retaining wall itself; E az ( t k ) is the first k Time nodes at each time step t k The vertical component of the corresponding earth pressure; E ax ( t k ) is the first k Time nodes at each time step t k The horizontal component of the corresponding earth pressure; μ ( t k ) is the first k Time nodes at each time step t k The base friction coefficient corresponding to the retaining wall.

[0086] S2.2 Simultaneously calculate the first k Overturning stability coefficient at each time step Q 2( t k ):

[0087] (4)

[0088] In the formula G Due to the weight of the retaining wall itself, x 0( t k ) is the first k Time nodes at each time step t k The distance from the toe of the retaining wall to the weight of the wall itself; E az ( t k )and x f ( t k ) are respectively the first k Time nodes at each time step t k The vertical component of the corresponding earth pressure and its distance from the toe of the wall;E ax ( t k )and z f ( t k ) are respectively the first k Time nodes at each time step t k The horizontal component of the earth pressure and its distance from the toe of the wall.

[0089] S2.3 Further, taking the critical failure state of the retaining wall as a benchmark, the minimum value of the two stability coefficients is used for normalized function mapping to construct a performance degradation function given only by physical mechanisms:

[0090] (5)

[0091] In the formula [ Q ] max The theoretical limit value of the stability coefficient is determined by the upper limit of the safety factor specified in the retaining wall design code; Q 1( t k )and Q 2( t k These are the anti-skid stability coefficient and the anti-overturning stability coefficient, respectively.

[0092] S3: Construction of Neural Network Cell Modules Integrating Physical Mechanisms and Update of Latent Damage States.

[0093] like Figure 3 As shown, this embodiment constructs a neural network Cell module that integrates physical mechanisms to characterize the temporal dependency relationship of "damage accumulation - state evolution - functional degradation" in the catastrophic process of a retaining wall. The neural network Cell module includes candidate damage state generation units and robust fusion gates.

[0094] The candidate damage state generation unit generates candidate damage states. :

[0095] (6)

[0096] In the formula [ x k , z k-1 ] is the first k Parameter vectors at each time step x k With the k -1 time step of implicit damage state z k 1; k From 1 to n The natural numbers between; tanh is the hyperbolic tangent activation function; For the first k Candidate damage states fused from real-time data at each time step are used to characterize the potential damage evolution trend of retaining walls under catastrophic disturbances.

[0097] The robust fusion gate is used to calculate the robust fusion gate gating coefficient. β k :

[0098] (7)

[0099] In the formula σ () is the Sigmoid activation function; b β The bias vector of the robust fusion gate; β k For the first k The gating coefficient of the robust fusion gate at each time step is used to adaptively adjust the contribution ratio between "historical damage memory" and "candidate damage update", and a physical performance decay function is introduced. F phys ( t k This implements physical consistency constraints for updating the damage state.

[0100] Furthermore, update the implicit damage state at the current time step. z k :

[0101] (8)

[0102] In the formula z k For the first k The updated implicit damage state at each time step is used to characterize the cumulative damage level and path dependence effect of the retaining wall throughout the entire disaster process.

[0103] Furthermore, the neural network Cell module will F phys ( t k This serves as a physical prior constraint in the gating input, ensuring that the robustness of the fusion gate gating coefficient increases as the rate of functional degradation determined by the physical model increases. β k A corresponding response is generated to enhance the absorption of candidate damage states and reflect damage transitions under sudden catastrophic events, thereby ensuring the implicit damage state at the current time step. z k Consistency with the stability degradation process.

[0104] S4: Residual feedforward neural network correction and calculation of total performance decay function.

[0105] The residual feedforward neural network is a network architecture consisting of an input layer, hidden layers, and an output layer. Figure 4 A schematic diagram of its specific architecture design is provided. To correct the physical model errors in S3 in real time, this embodiment sets up a residual feedforward neural network to establish the "parameter vector". x k and latent damage state z k "and "change in function Δ" F ( t k The residual mapping relationship of ")" is designed as follows: The input layer receives the first ")" and its specific architecture is as follows: k time steps ( k From 1 to n The parameter vector of natural numbers between x k and latent damage state z k Feature fusion is performed to provide basic input for subsequent feature extraction. The hidden layer consists of two fully connected layers, each with 64 neurons, both using the ReLU activation function. The ReLU activation function can effectively alleviate the gradient vanishing problem and enhance the network's ability to learn complex nonlinear features. The output of hidden layer 1 is transmitted to hidden layer 2 after weight updates. Through progressive nonlinear transformation, the extraction accuracy of complex residual features between the physical model and the real system response is further enhanced. The output layer uses a linear activation function to avoid introducing additional nonlinear bias and ensure that the network can accurately output the function function change Δ that matches the residual size. F ( t k ).

[0106] Specifically, the change in the function Δ F ( t k Calculated using the following formula:

[0107] (9)

[0108] In the formula g θ [,] represents the mapping function of the residual feedforward neural network; Δ F ( t k ) is the first k The change in the function after correction at each time step, i.e., the change relative to the physical performance degradation function. F phys (t k The residual correction magnitude is used to characterize the systematic deviation and nonlinear error between the physical model and the actual system response.

[0109] The residual correction term Δ F ( t k Superimposed on the physical performance degradation function F phys ( t k The actual performance degradation function is obtained. F total ( t k ):

[0110] (10)

[0111] In the formula Δ F ( t k ) is the first k The change in the corrected function at each time step F phys ( t k ) is the first k The performance degradation function of the retaining wall based solely on the physical model at each time step.

[0112] The total performance degradation function of the entire catastrophe process is obtained by summing over discrete time steps. F total ( t ):

[0113] (11)

[0114] In the formula N The number of time steps in the disaster process. F total ( t ) is the total performance degradation function over the entire disaster period, used for subsequent robustness index calculations.

[0115] S5: Robust notch area A g Robustness index R Output with grade.

[0116] S5.1 This embodiment uses robust notch area A g (like Figure 5 (As shown) The robust gap area comprehensively characterizes the coupling effect of "functional loss amplitude - duration". A g The calculation formula is:

[0117] (12)

[0118] In the formula T Total duration of the disaster F ( t ) represents the normal service performance function of the retaining wall when no disaster occurs. F total ( t ) is the total performance degradation function during the entire disaster period.

[0119] S5.2 further, the retaining wall catastrophe robustness index R The formula used to normalize the catastrophic robustness of retaining walls is as follows:

[0120] (13)

[0121] The formula contains the catastrophic robustness index of the retaining wall. R ∈(0,1), R The closer the value is to 1, the higher the catastrophic robustness of the retaining wall. R The closer the value is to 0, the lower the catastrophic robustness of the retaining wall.

[0122] The S5.3 retaining wall catastrophe robustness level is based on the robustness index in S5.2. R Size is graded: when R ≥ R At 1, it can be classified as high robustness; when R 2≤ R <R At 1, it can be classified as moderately robust; when R ≤ R At level 2, it is classified as low robustness. Among them, R 1 and R 2 is the threshold for disaster robustness classification, which is determined based on the limits in the retaining wall design specifications, historical disaster statistics, or expert scoring and evaluation.

[0123] This invention proposes a disaster robustness assessment framework driven by a fusion of physical mechanisms and neural networks, targeting scenarios involving the identification of functional degradation and the quantitative assessment of robustness of retaining walls under catastrophic disturbances such as earthquakes, torrential rain erosion, and sudden increases in landslide thrust. This framework takes the time-discretization and vectorization of disaster process monitoring parameters as input. First, it constructs a normalized physical performance decay function based on the anti-sliding stability coefficient and the anti-overturning stability coefficient. Then, it constructs a neural network Cell module incorporating physical priors and a robustness fusion gate to achieve temporal updates of implicit damage states. Finally, it uses a feedforward residual network to correct physical model errors online, obtaining the total performance decay function. The robustness gap area and robustness index are then used as the final assessment parameters. RIt outputs the disaster robustness level, realizing a unified quantitative expression of "functional loss amplitude - duration - recovery process", providing an interpretable, updatable and engineerable assessment path for retaining wall disaster assessment and risk control.

[0124] Compared with existing technologies, this invention breaks through the traditional evaluation paradigm that relies solely on safety factors or simple data fitting, while possessing both physical interpretability and data-driven high precision. On the one hand, it explicitly constrains the performance degradation evolution with a stability mechanism, making the evaluation results consistent with engineering criteria and traceable. On the other hand, through gating fusion and residual correction mechanisms, it transforms real-time monitoring information into adaptive compensation for deviations in the physical model, significantly improving the evaluation stability and generalization ability under conditions of parameter uncertainty, catastrophic changes, and multi-stage evolution. This enables more reliable output of robustness indices and levels, providing more scientific and intelligent technical support and calculation methods for rapid post-disaster assessment, reinforcement decisions, and toughness improvement of retaining walls.

Claims

1. A method for evaluating the disaster robustness of a physical-data dual-driven retaining wall, characterized in that, Includes the following steps: S1, input monitoring parameters of the retaining wall under the disaster condition, time-disperse the disaster process and parameter-vectorize the disaster process to obtain a parameter vector of the disaster process x k ; S2, calculate the anti-sliding stability factor at each time step according to the input parameters Q 1( t k ) and the anti-overturning stability factor Q 2( t k ), get the retaining wall performance degradation function through the normalized function mapping F phys ( t k ) physical model; S3, construct the Cell module of the neural network that incorporates the physical mechanism, and input the parameter vector at the current time step. x k and the implicit damage state at the previous time step z k-1 Generate a current retaining wall performance degradation function that considers only physical mechanisms. F phys ( t k ) and candidate damage states fused with real-time data And update the implicit damage state at the current time step using a robust fusion gate. z k ; S4, calculate the change in the corrected function function at the current time step using a feedforward network. ΔF ( t k ), superimposed performance degradation function F phys ( t k Then, the actual performance degradation function at the current time step is obtained. F total ( t k The total performance degradation function over the entire disaster period is calculated by summing the results. F total ( t ); S5, based on the total performance degradation function F total ( t ) Calculate the robustness gap area and the corresponding robustness index, output its robustness level, and realize the accurate quantitative assessment of the catastrophic robustness of the retaining wall.

2. The physical-data dual-driven method for assessing the catastrophic robustness of retaining walls as described in claim 1, characterized in that, The method to implement S1 includes the following steps in sequence: S1.1 on the continuous disaster process [ t 0, t n Using equal time intervals Δ t The partitioning strategy is discretized into n The time step, the first k Time nodes at each time step t k satisfy: （1） In the formula t 0 represents the start time of the catastrophe; Δ t The discrete time step; k From 1 to n Natural numbers between; S1.2 vectorizes the monitored disaster condition parameters and uses them as a unified input to the S2 physical model and the S3 neural network, where the first... k Parameter vectors at each time step x k for: （2） In the formula G Due to the weight of the retaining wall itself, x 0( t k ) is the first k Time nodes at each time step t k The distance from the toe of the retaining wall to the weight of the wall itself; E az ( t k )and x f ( t k ) are respectively the first k Time nodes at each time step t k The vertical component of the corresponding earth pressure and its distance from the toe of the wall; E ax ( t k )and z f ( t k ) are respectively the first k Time nodes at each time step t k The horizontal component of the earth pressure and its distance from the toe of the wall; μ ( t k ) is the first k Time nodes at each time step t k The base friction coefficient corresponding to the retaining wall.

3. The physical-data dual-driven method for assessing the catastrophic robustness of retaining walls as described in claim 1, characterized in that, The method to implement S2 includes the following steps in sequence: S2.1 Calculates the first... k Anti-skid stability coefficient at each time step Q 1( t k ): （3） In the formula G The weight of the retaining wall itself; E az ( t k ) is the first k Time nodes at each time step t k The vertical component of the corresponding earth pressure; E ax ( t k ) is the first k Time nodes at each time step t k The horizontal component of the corresponding earth pressure; μ ( t k ) is the first k Time nodes at each time step t k The base friction coefficient corresponding to the retaining wall; S2.2 Simultaneously calculate the first k Overturning stability coefficient at each time step Q 2( t k ): （4） In the formula G Due to the weight of the retaining wall itself, x 0( t k ) is the first k Time nodes at each time step t k The distance from the toe of the retaining wall to the weight of the wall itself; E az ( t k )and x f ( t k ) are respectively the first k Time nodes at each time step t k The vertical component of the corresponding earth pressure and its distance from the toe of the wall; E ax ( t k )and z f ( t k ) are respectively the first k Time nodes at each time step t k The horizontal component of the earth pressure and its distance from the toe of the wall; S2.3 uses the critical failure state of the retaining wall as a benchmark, takes the minimum value of the two stability coefficients for normalized function mapping, and constructs a performance degradation function given only by physical mechanisms: （5） In the formula [ Q ] max The theoretical limit value of the stability coefficient is determined by the upper limit of the safety factor specified in the retaining wall design code; Q 1( t k )and Q 2( t k These are the anti-skid stability coefficient and the anti-overturning stability coefficient, respectively.

4. The physical-data dual-driven method for assessing the catastrophic robustness of retaining walls as described in claim 1, characterized in that, Methods for implementing S3 include: A neural network Cell module integrating physical mechanisms is constructed to characterize the temporal dependency of "damage accumulation - state evolution - functional degradation" in the catastrophic process of retaining walls; the neural network Cell module includes candidate damage state generation units and robust fusion gates; The candidate damage state generation unit generates candidate damage states. : （6） In the formula [ x k , z k-1 ] is the first k Parameter vectors at each time step x k With the k -1 time step of implicit damage state z k 1; k From 1 to n The natural numbers between; tanh is the hyperbolic tangent activation function; For the first k Candidate damage states that fuse real-time data at each time step are used to characterize the potential damage evolution trend of retaining walls under catastrophic disturbances. The robust fusion gate is used to calculate the robust fusion gate gating coefficient. β k : （7） In the formula σ () is the Sigmoid activation function; b β The bias vector of the robust fusion gate; β k For the first k The gating coefficient of the robust fusion gate at each time step is used to adaptively adjust the contribution ratio between "historical damage memory" and "candidate damage update", and a physical performance decay function is introduced. F phys ( t k Achieve physical consistency constraints for updating damage states; Update the implicit damage state at the current time step. z k : （8） In the formula z k For the first k The updated implicit damage state at each time step is used to characterize the cumulative damage level and path dependence effect of the retaining wall throughout the entire disaster process.

5. The physical-data dual-driven method for assessing the catastrophic robustness of retaining walls as described in claim 1, characterized in that, Methods for implementing S4 include: Feedforward neural networks are used to establish the residual mapping relationship between parameter vectors and latent damage states to changes in the function, in order to correct physical model errors in real time. k The change in the corrected function function Δ at each time step F ( t k )satisfy: （9） In the formula g θ [,] represents the mapping function of the feedforward neural network, which learns to accurately fit the residual through multiple layers of weights; Δ F ( t k ) is the first k The change in the function after correction at each time step, i.e., the change relative to the physical performance degradation function. F phys ( t k The magnitude of the residual correction is used to characterize the systematic deviation and nonlinear error between the physical model and the actual system response; The actual performance degradation function F total ( t k By ΔF ( t k ) superimposed to F phys ( t k )get: （10） In the formula Δ F ( t k ) is the first k The change in the function after correction at each time step; F phys ( t k ) is the first k The performance degradation function of the retaining wall based solely on the physical model at each time step; Total performance degradation function throughout the entire disaster process F total ( t The result is obtained by summing over discrete time steps: （11） In the formula N The number of time steps in the disaster process. F total ( t ) is the total performance degradation function over the entire disaster period, used for subsequent robustness index calculations.

6. The physical-data dual-driven method for assessing the catastrophic robustness of retaining walls as described in claim 1, characterized in that, The method to implement S5 includes the following steps in sequence: S5.1 using robust gap area A g The robust gap area comprehensively characterizes the coupling effect of "functional loss amplitude-duration". A g The calculation formula is: （12） In the formula T Total duration of the disaster F ( t ) represents the normal service performance function of the retaining wall when no disaster occurs. F total ( t ) represents the total performance degradation function over the entire disaster period; S5.2 Retaining Wall Catastrophic Robustness Index R The formula used to normalize the catastrophic robustness of retaining walls is as follows: （13） The formula contains the catastrophic robustness index of the retaining wall. R ∈(0,1), R The closer the value is to 1, the higher the catastrophic robustness of the retaining wall. R The closer the value is to 0, the lower the catastrophic robustness of the retaining wall. The S5.3 retaining wall catastrophe robustness level is based on the robustness index in S5.

2. R Sizes are classified.

7. The physical-data dual-driven method for assessing the catastrophic robustness of retaining walls as described in claim 6, characterized in that, In S5.3, the catastrophic robustness level of the retaining wall is based on the robustness index in S5.

2. R Size is classified into categories, specifically: when R ≥ R At level 1, it can be classified as high robustness; when R 2≤ R <R At level 1, it can be classified as moderately robust; when R ≤ R At level 2, it is classified as low robustness; in, R 1 and R 2 is the threshold for disaster robustness classification, which is determined based on the limits in the retaining wall design specifications, historical disaster statistics, or expert scoring and evaluation.

Images