Integrated prediction method for consolidation settlement of soft soil foundation based on BO-PINN method

Abstract

The application relates to a BO-PINN method-based soft soil foundation consolidation settlement integrated prediction method. The method comprises the following steps: obtaining a sample data set and a target detection sample; wherein the sample data set comprises a training sample set and a test sample set; an initial model of BO-PINN consolidation settlement forward and inverse is trained through the training sample set, model hyperparameters of the initial model of BO-PINN consolidation settlement forward and inverse and to-be-inverted soil layer parameters are iteratively screened, a trained BO-PINN consolidation settlement forward and inverse target model is obtained, and in-situ soil layer mechanical parameters are determined through inversion; the target detection sample is input into the BO-PINN consolidation settlement forward and inverse target model, and a soft soil foundation consolidation settlement prediction result corresponding to the target detection sample is obtained. In this way, the disadvantages of sample disturbance in traditional indoor soil test, high in-situ test cost and large data discreteness are reduced, the artificial cost and randomness of neural network parameter adjustment are reduced, and the requirements of multiple scene engineering demands such as construction period dynamic control, post-construction long-term settlement prediction and the like are met.

Description

Technical Field

[0001] This application relates to the fields of geotechnical mechanics and engineering technology, and in particular to an integrated prediction method for consolidation settlement of soft soil foundations based on the BO-PINN method. Background Technology

[0002] Soft soil foundations are widely distributed in coastal and inland river and lake areas of my country. In the construction of projects such as highways, port reclamation, and urban underground space development, the control of consolidation settlement deformation of soft soil foundations is a core issue determining the safety and long-term service performance of the project. Accurately predicting the development law of consolidation settlement and obtaining reliable in-situ mechanical parameters of the soil layer are key bases for foundation treatment scheme design, construction period optimization, and post-construction settlement assessment.

[0003] Currently, consolidation settlement calculations in engineering are mostly based on Terzaghi's one-dimensional consolidation theory and Biot's three-dimensional consolidation theory, combined with numerical methods such as finite element and finite difference methods for forward modeling. The prediction accuracy of such methods is highly dependent on the accuracy of soil mechanical parameters such as compression modulus, permeability coefficient, and consolidation coefficient. However, soft soil has significant heterogeneity and spatiotemporal variability. Indoor geotechnical tests suffer from size effects and sample disturbance problems. In-situ testing is costly and has large data dispersion, resulting in large deviations in soil parameter values. Ultimately, the predicted settlement results often have significant errors compared to the actual monitored values, making it difficult to meet the needs of refined engineering management. Summary of the Invention

[0004] Therefore, it is necessary to provide an integrated prediction method for consolidation settlement of soft soil foundations based on the BO-PINN method to address the aforementioned technical problems.

[0005] Firstly, this application provides an integrated prediction method for consolidation settlement of soft soil foundations based on the BO-PINN method, the method comprising: Obtain a sample dataset and object detection samples; wherein the sample dataset includes a training sample set and a test sample set; The initial model for forward and inverse BO-PINN consolidation settlement is trained using the training sample set. The model hyperparameters and soil parameters to be inverted are iteratively selected from the initial model for forward and inverse BO-PINN consolidation settlement. The trained target model for forward and inverse BO-PINN consolidation settlement and the in-situ mechanical parameters of the soil layer determined by inversion are obtained. The target detection sample is input into the BO-PINN consolidation settlement forward and inverse target model to obtain the consolidation settlement prediction result of the soft soil foundation corresponding to the target detection sample.

[0006] In one embodiment, the BO-PINN consolidation settlement forward and inverse target model is a two-layer coupled architecture: wherein, The outer layer is a Bayesian optimization module, which consists of a probabilistic surrogate model and a data acquisition function. It is used to iteratively optimize the hyperparameters of the physical information neural network and simultaneously invert the in-situ mechanical parameters of the soil layer. The inner layer is a physical information neural network module. The physical information neural network module uses a neural network as the fitting carrier to construct a target loss function that includes data fitting terms and physical constraint terms. The network weights are updated through backpropagation of the loss to complete the forward modeling prediction of pore water pressure and vertical displacement of soft soil foundation.

[0007] In one embodiment, the method for determining the physical constraint term includes: Based on the residual loss of the vertical force balance equation and the residual loss of the seepage continuity equation, and by summing the corresponding weighting coefficients, the loss term of the partial differential equation is determined. Based on the initial excess pore water pressure constraint residual loss and the initial vertical displacement constraint residual loss, and by summing the corresponding weight coefficients, the initial condition loss term is determined; The boundary condition loss term is determined by summing the residual losses of the seepage field boundary constraints and the displacement field boundary constraints, combined with the corresponding weighting coefficients. The physical constraint term is determined based on the partial differential equation loss term, the initial condition loss term, and the boundary condition loss term.

[0008] In one embodiment, the step of determining the initial condition loss term by summing the initial excess pore water pressure constraint residual loss and the initial vertical displacement constraint residual loss with corresponding weighting coefficients includes: The first residual is calculated based on the deviation between the first predicted excess pore water pressure and the actual excess pore water pressure at the initial moment. The norm of the first residual is taken and combined with the initial condition weighting coefficient to determine the initial excess pore water pressure constraint residual loss. The second residual is calculated based on the deviation between the first predicted vertical displacement and the actual vertical displacement at the initial moment. The norm of the second residual is taken and combined with the initial condition weighting coefficient to determine the initial vertical displacement constraint residual loss. The initial condition loss term is determined based on the initial excess pore water pressure constraint residual loss and the initial vertical displacement constraint residual loss.

[0009] In one embodiment, the seepage field boundary constraint residual loss includes a first constraint residual and a second constraint residual; the step of determining the boundary condition loss term by summing the seepage field boundary constraint residual loss and the displacement field boundary constraint residual loss with corresponding weighting coefficients includes: The first constraint residual is determined by combining the norm of the second predicted excess pore water pressure at the first boundary of the drainage boundary with the boundary weight coefficient. The second constraint residual is determined by combining the norm of the partial derivative of the second predicted excess pore water pressure along the depth at the second boundary of the non-drained boundary with the boundary weight coefficient. The residual loss of the displacement field boundary constraint is determined by combining the norm of the second predicted vertical displacement at the fixed displacement boundary with the boundary weight coefficient; wherein, the fixed displacement boundary indicates the boundary space location in the displacement field where a fixed vertical constraint is applied. The boundary condition loss term is determined based on the residual loss of the seepage field boundary constraint and the residual loss of the displacement field boundary constraint.

[0010] In one embodiment, determining the partial differential equation loss term by summing the residual losses from the vertical force balance equation and the seepage continuity equation, combined with the corresponding weighting coefficients, includes: The residual of the first equation is calculated based on the undrained elastic modulus, Poisson's ratio, the second-order partial derivative of the third predicted vertical displacement along the depth, the effective stress coefficient, and the first-order partial derivative of the third predicted excess pore water pressure along the depth. The norm of the residual of the first equation is taken and combined with the weighting coefficient of the first partial differential equation to determine the residual loss of the vertical force balance equation. The residual of the second equation is calculated based on the first-order partial derivative of the third predicted excess pore water pressure along time, the undrained elastic modulus, the effective stress coefficient, the permeability coefficient, and the second-order partial derivative of the third predicted vertical displacement along depth. The norm of the residual of the second equation is taken and combined with the weighting coefficient of the second partial differential equation to determine the residual loss of the seepage continuity equation. The loss term of the partial differential equation is determined based on the residual loss of the vertical force balance equation and the residual loss of the seepage continuity equation.

[0011] In one embodiment, the process of training an initial BO-PINN consolidation settlement forward and inverse model using the training sample set, iteratively filtering the model hyperparameters and soil layer parameters to be inverted from the initial BO-PINN consolidation settlement forward and inverse model, and obtaining the trained BO-PINN consolidation settlement forward and inverse target model, as well as the inverted and determined in-situ mechanical parameters of the soil layer, includes: The foundation depth and time coordinates in the training sample set are input into the neural network of the initial model for forward and inverse modeling of BO-PINN consolidation settlement. The excess pore water pressure and vertical displacement are estimated through forward propagation to obtain the estimated values ​​of excess pore water pressure and vertical displacement at the corresponding points. Based on the estimated value of excess pore water pressure and the estimated value of vertical displacement, the target loss function is calculated by combining the data fitting term and the physical constraint term, and the total loss value of the target loss function is obtained. Based on the total loss value, the weights and biases of the neural network are updated by backpropagation using the chain rule, and the estimated values ​​of the undrained elastic modulus and permeability coefficient are generated synchronously and iteratively. When the total loss value meets the convergence criterion, the BO-PINN consolidation settlement forward and inverse model is obtained, and the in-situ mechanical parameters of the soil layer are determined by inversion.

[0012] Secondly, this application also provides an integrated prediction device for consolidation settlement of soft soil foundation based on the BO-PINN method, comprising: An acquisition module is used to acquire a sample dataset and object detection samples; wherein, the sample dataset includes a training sample set and a test sample set; The construction module is used to train the initial model of BO-PINN consolidation settlement forward and inverse through the training sample set, iteratively filter the model hyperparameters and soil layer parameters to be inverted of the initial model of BO-PINN consolidation settlement forward and inverse, and obtain the trained target model of BO-PINN consolidation settlement forward and inverse, as well as the in-situ mechanical parameters of the soil layer determined by inversion. The detection module is used to input the target detection sample into the BO-PINN consolidation settlement forward and inverse target model to obtain the consolidation settlement prediction result of the soft soil foundation corresponding to the target detection sample.

[0013] Thirdly, this application also provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, it implements the steps of the method described in any embodiment of this application.

[0014] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the method described in any embodiment of this application.

[0015] The aforementioned integrated prediction method for consolidation settlement of soft soil foundations based on the BO-PINN method constructs an integrated forward and inverse model of BO-PINN consolidation settlement. This allows for the reverse derivation of in-situ mechanical parameters of the soil layer that closely match the actual working conditions of the soil based on field-measured monitoring data. This effectively reduces the drawbacks of traditional indoor geotechnical tests, such as sample disturbance, size effect deviations, high costs of in-situ testing, and large data dispersion. It addresses the core problem of inaccurate parameter values ​​leading to distorted prediction results in traditional consolidation calculation methods, significantly reducing the deviation between predicted settlement results and actual field monitoring values. Furthermore, compared to traditional purely physics-driven numerical methods such as finite element and finite difference methods, it exhibits stronger adaptability to working conditions and higher prediction accuracy. Meanwhile, by jointly optimizing the model hyperparameters and the parameters of the soil layer to be inverted globally, the manual cost and randomness of neural network parameter tuning are effectively reduced, the convergence efficiency of model training and the stability of inversion results are improved, and the trained model can support all-time and space-time consolidation settlement prediction at any depth and at any time. It can flexibly meet the engineering needs of multiple scenarios such as dynamic control during construction and long-term settlement prediction after construction, and provide reliable technical support for the refined control of consolidation settlement of soft soil foundations. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating an integrated prediction method for consolidation and settlement of soft soil foundations based on the BO-PINN method, according to an exemplary embodiment.

[0017] Figure 2 This is a graph showing the total loss value of the initial model for forward and inverse modeling of BO-PINN consolidation settlement as a function of the number of training iterations, according to an exemplary embodiment.

[0018] Figure 3 This is a schematic diagram illustrating the inversion results of the permeability coefficient of the target model for forward and inverse modeling of BO-PINN consolidation settlement according to an exemplary embodiment;

[0019] Figure 4 This is a schematic diagram illustrating the inversion results of the undrained elastic modulus of the target model for forward and inverse modeling of BO-PINN consolidation settlement according to an exemplary embodiment;

[0020] Figure 5 This is a schematic diagram showing the result of predicting excess pore water pressure using the BO-PINN consolidation settlement forward and inverse target model according to an exemplary embodiment.

[0021] Figure 6 This is a schematic diagram illustrating the vertical displacement prediction results of the BO-PINN consolidation settlement forward and inverse target model according to an exemplary embodiment.

[0022] Figure 7This is a structural block diagram of an integrated prediction device for consolidation and settlement of soft soil foundation based on the BO-PINN method, according to an exemplary embodiment.

[0023] Figure 8 This is an internal structural diagram of an electronic device according to an exemplary embodiment. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0025] The terms "first," "second," and "third" used in the embodiments of this application are for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined as "first," "second," or "third" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0026] The BO-PINN-based integrated prediction method for consolidation and settlement of soft soil foundations provided in this application embodiment can be applied to electronic devices. The electronic device can establish a communication connection with a device, apparatus, or combination thereof used to collect data related to the integrated prediction of consolidation and settlement of soft soil foundations based on the BO-PINN method, via wired or wireless means. The electronic device can be any mobile terminal or fixed terminal. The terminal can be a device that provides voice and / or data connectivity to the user. For example, the terminal can be an Internet of Things (IoT) terminal, such as a sensor device, a mobile phone or so-called "cellular" phone, and a computer with an IoT terminal; for example, it can be a fixed, portable, pocket-sized, handheld, or computer-built-in device. In related technologies, electronic devices have interactive components, such as touch screens or non-touch screens.

[0027] In some embodiments, such as Figure 1 As shown, an integrated prediction method for consolidation settlement of soft soil foundations based on the BO-PINN method is provided. The method includes the following steps:

[0028] S101, Obtain the sample dataset and object detection samples; wherein, the sample dataset includes a training sample set and a test sample set.

[0029] In this embodiment, the sample dataset can be a collection of data used for training and performance verification of the BO-PINN consolidation settlement forward and inverse model, including a training sample set and a test sample set.

[0030] In this embodiment, the training sample set can be the core data subset in the sample dataset used to drive model iteration and parameter optimization, including the monitoring dataset and the collocation dataset.

[0031] In this embodiment, the monitoring dataset is a labeled dataset derived from on-site measurements, used to constrain the degree of fit between the model output and the actual engineering state.

[0032] Optionally, the monitoring dataset may include, but is not limited to, at least one of pore water pressure monitoring data and vertical displacement monitoring data collected by sensors.

[0033] For example, monitoring data can be collected from monitoring equipment such as pore water pressure sensors, settlement plates, and stratified settlement meters at the cross-section of soft soil foundation.

[0034] Optionally, the monitoring dataset can be acquired through at least one of the following methods: actual data collection and generation from on-site sensors, and numerical forward modeling synthesis.

[0035] In some embodiments, on-site sensors can be used to collect and generate monitoring datasets. Pore water pressure sensors and vertical displacement monitoring devices are buried at preset points and depths in the target soft soil foundation. Pore water pressure monitoring data and vertical displacement monitoring data at the corresponding points are collected according to a preset sampling period. After noise reduction and time-series processing, a real monitoring dataset is formed.

[0036] In some embodiments, numerical forward modeling can be used to acquire synthetic monitoring datasets. First, a soft soil foundation calculation model with defined boundary conditions is constructed. After specifying the physical and mechanical properties of the soil layers, forward modeling is used to obtain the reference values ​​of excess pore water pressure and vertical displacement across the entire computational domain. Then, the deployment logic and sampling rules of the field sensors are simulated to extract the corresponding data. The data is then organized in chronological order to generate a synthetic monitoring dataset.

[0037] For example, the electronic device can construct a soft soil foundation model with a depth of 10m. The top layer of the seepage field is set as a drainage boundary and the bottom surface adopts a non-drainage boundary condition. The bottom of the displacement field is set as a completely fixed constraint and the top is allowed to have free vertical deformation. After obtaining the benchmark value through forward modeling, the simulation field sampling logic randomly extracts 100 sets of data and regularizes them according to the time sequence to obtain a synthetic monitoring dataset.

[0038] In this embodiment, the collocation point dataset is mainly applied to the fields of scientific computing and physical information neural networks; where collocation points refer to discrete points sampled in the spatial or temporal domain for evaluating the residuals of partial differential equations.

[0039] Optionally, the sampling method for the points may include, but is not limited to, at least one of Latin hypercube sampling, uniform random sampling, and grid sampling.

[0040] In this embodiment, the test sample set is a subset of the sample dataset that does not participate in the model training process and is only used to verify the model's prediction accuracy and generalization performance.

[0041] In this embodiment, the target detection sample is the input data to be analyzed, which is input into the trained BO-PINN consolidation settlement forward and inverse target model and used to output the corresponding consolidation settlement prediction results.

[0042] For example, the target detection sample can be a combination of spatial coordinates and time nodes of the section to be predicted, or it can be the subsequent extended prediction time node data of the monitored section.

[0043] S102, train the initial model for forward and inverse BO-PINN consolidation settlement using the training sample set, iteratively screen the model hyperparameters and soil parameters to be inverted of the initial model for forward and inverse BO-PINN consolidation settlement, and obtain the trained target model for forward and inverse BO-PINN consolidation settlement, as well as the in-situ mechanical parameters of the soil layer determined by inversion.

[0044] In this embodiment, the initial model for BO-PINN consolidation settlement forward and inverse modeling indicates an initial network model that has not yet been trained. The basic network structure of this initial model may include, but is not limited to, at least one of a fully connected neural network and a residual neural network.

[0045] In this embodiment, the model hyperparameters can be network configuration parameters that are pre-set before model training and automatically learned and updated, used to control the model's convergence efficiency and prediction accuracy.

[0046] Optionally, model hyperparameters may include, but are not limited to, at least one of the following: number of network layers, number of neurons in a single hidden layer, initial learning rate, weight coefficients of the loss term, and training batch size.

[0047] In this embodiment, the soil layer parameters to be inverted can be soil physical and mechanical property parameters whose initial values ​​are unknown and need to be solved by inverse calculation through model training in combination with monitoring data. These are the intrinsic core factors controlling the development law of consolidation settlement.

[0048] Optionally, the soil parameters to be inverted include the permeability coefficient and the undrained elastic modulus.

[0049] In some embodiments, iterative screening can be performed based on a Bayesian optimization framework, with the parameter sampling direction for the next round updated in each round based on the evaluation results of historical parameters, until the convergence condition is met or the preset number of iterations is reached; for example, the convergence condition can be that the total loss value of the target loss function is less than the reference standard.

[0050] In this embodiment, the in-situ mechanical parameters of the soil layer can be parameter values ​​determined by model inversion that reflect the actual physical and mechanical properties of the target soft soil foundation on site, and can be used to characterize the consolidation deformation characteristics and seepage characteristics of the soil layer.

[0051] In some embodiments, the BO-PINN consolidation settlement forward and inverse target model is a two-layer coupled architecture: wherein, The outer layer is a Bayesian optimization module, which consists of a probabilistic surrogate model and a data acquisition function. It is used to iteratively optimize the hyperparameters of the physical information neural network and simultaneously invert the in-situ mechanical parameters of the soil layer. The inner layer is a physical information neural network module. The physical information neural network module uses a neural network as the fitting carrier to construct a target loss function that includes data fitting terms and physical constraint terms. The network weights are updated through backpropagation of the loss to complete the forward modeling prediction of pore water pressure and vertical displacement of soft soil foundation.

[0052] In this embodiment, the probabilistic surrogate model can be a probabilistic prediction model in the Bayesian optimization module used to fit the mapping relationship between parameter values ​​and model performance, and used to predict the performance distribution corresponding to unknown parameter combinations based on historical evaluation results.

[0053] Optionally, the probabilistic surrogate model may include, but is not limited to, at least one of the Gaussian process model, random forest surrogate model, and deep neural network surrogate model.

[0054] In this embodiment, the acquisition function can be an evaluation function in the Bayesian optimization module used to guide the direction of parameter sampling in the next round, used to balance exploration and utilization in the parameter optimization process, and to screen potential optimal parameter combinations.

[0055] Optionally, the acquisition function may include, but is not limited to, at least one of the following: the expectation improvement function, the probability improvement function, and the confidence interval upper bound function.

[0056] For example, the neural network can adopt a fully connected network architecture, with a target number of hidden layers (e.g., 5 layers), and each hidden layer is configured with the target number of neurons (e.g., 30). The activation function of the hidden layer can be the hyperbolic tangent function (Tanh), and the iterative update of the network weights and biases can be achieved using the Adam adaptive optimizer.

[0057] In this embodiment, the data fitting term can be a component of the target loss function used to measure the degree of matching between the model's predicted value and the monitoring data, and is used to constrain the model output to fit the measured response pattern.

[0058] In some embodiments, the data fitting term can be constructed by calculating the mean square error of the predicted pore water pressure, the predicted vertical displacement, and the corresponding monitoring data.

[0059] In this embodiment, the physical constraint term can be a component of the target loss function used to measure the degree to which the model output satisfies the consolidation mechanics control equation, thereby ensuring that the model prediction results conform to the basic laws of consolidation mechanics.

[0060] In some embodiments, physical constraint terms can be constructed based on the residual calculation of the consolidation control equations, corresponding to the physical constraint calculation results of the collocation dataset.

[0061] In some embodiments, the Bayesian optimization module can define a joint hyperparameter search space based on soft soil foundation consolidation engineering experience, geological survey data, and neural network training rules. This space includes model hyperparameters such as neural network learning rate, weights of each loss term, and training batch size, as well as soil mechanical parameters such as undrained elastic modulus and permeability coefficient. Subsequently, a number of initial parameter samples are generated in the search space using a Latin hypercube sampling strategy. Each set of initial parameters is then substituted into the BO-PINN consolidation settlement forward and inverse initial model to complete one round of training. The Gaussian process regression model is used as the probabilistic surrogate model, and the expected improvement function is used as the acquisition function to continuously iterate and optimize until the hyperparameter combination with the highest accuracy (or lowest error) is obtained.

[0062] For example, an electronic device can first construct a joint hyperparameter search space, fit the mapping relationship between each parameter combination in the search space and the model performance through a probabilistic surrogate model, predict the model prediction error corresponding to different hyperparameter combinations, and then guide the iterative sampling direction through a sampling function to continuously select the parameter combination with the lowest prediction error in multiple iterations. The hyperparameter search space can be set to cover both continuous and discrete parameters, and the learning rate can continuously take values ​​in the range of [0.0001, 0.1]. The number of hidden layers is an integer variable, and the search range can be 1 to 10 layers. The number of neurons in a single layer can range from 5 to 500. The activation function can be discretely selected from three types of functions: Sin, Sigmoid, and Tanh. Under the Bayesian optimization framework, a Gaussian process regression model is used as a probabilistic surrogate model, combined with an expected improvement sampling function to balance the exploration and utilization in the parameter search process, thereby achieving efficient global parameter optimization. After multiple rounds of iterative optimization and convergence, the optimal network architecture and training configuration for the BO-PINN consolidation settlement forward and inverse model were finally determined as follows: a 6-layer fully connected neural network with 234 neurons per layer and a learning rate of 1×10⁻⁶. -4 The activation function chosen is the Sin function.

[0063] In some embodiments, the process of training an initial BO-PINN consolidation settlement forward and inverse model using the training sample set, iteratively filtering the model hyperparameters and soil layer parameters to be inverted from the initial BO-PINN consolidation settlement forward and inverse model, and obtaining the trained BO-PINN consolidation settlement forward and inverse target model, as well as the inverted and determined in-situ mechanical parameters of the soil layer, includes: The foundation depth and time coordinates in the training sample set are input into the neural network of the initial model for forward and inverse modeling of BO-PINN consolidation settlement. The excess pore water pressure and vertical displacement are estimated through forward propagation to obtain the estimated values ​​of excess pore water pressure and vertical displacement at the corresponding points. Based on the estimated value of excess pore water pressure and the estimated value of vertical displacement, the target loss function is calculated by combining the data fitting term and the physical constraint term, and the total loss value of the target loss function is obtained. Based on the total loss value, the weights and biases of the neural network are updated by backpropagation using the chain rule, and the estimated values ​​of the undrained elastic modulus and permeability coefficient are generated synchronously and iteratively. When the total loss value meets the convergence criterion, the BO-PINN consolidation settlement forward and inverse model is obtained, and the in-situ mechanical parameters of the soil layer are determined by inversion.

[0064] In some embodiments, the electronic device can organize the foundation depth coordinates and time coordinates corresponding to each point in the training sample set into a standardized input vector, input it into the multi-layer fully connected neural network of the initial model, and complete the forward transmission layer by layer through the linear transformation of the hidden layer and the activation function operation. Finally, the output layer outputs two target physical quantities corresponding to the input points, namely the estimated value of excess pore water pressure and the estimated value of vertical displacement. The input points simultaneously cover the measured points corresponding to the monitoring dataset and the computational domain points corresponding to the coordinate point dataset.

[0065] In some embodiments, the electronic device can calculate the mean square error of the estimated excess pore water pressure, the estimated vertical displacement, and the corresponding measured monitoring data for the points corresponding to the monitoring dataset, forming a data fitting term; for the points corresponding to the collocation dataset, substitute the estimated excess pore water pressure and the estimated vertical displacement into the Biot consolidation control equation, calculate the mean square value of the equation residual, forming a physical constraint term; and perform a weighted summation of the data fitting term and the physical constraint term according to a preset loss weight coefficient to obtain the total loss value corresponding to the target loss function.

[0066] In some embodiments, the electronic device can take minimizing the total loss value as the optimization objective, calculate the gradient of the total loss value with respect to the weights and biases of each layer of the neural network layer by layer from the output layer to the input layer based on the chain rule, and update the weights and biases of the neural network through the gradient descent optimizer; at the same time, the undrained elastic modulus and the permeability coefficient are embedded into the model as trainable variables, and the gradient of the total loss value with respect to the two parameters to be inverted is calculated and updated synchronously to generate the estimated values ​​of the undrained elastic modulus and the estimated values ​​of the permeability coefficient to be inverted in the current iteration.

[0067] In some embodiments, the electronic device can verify the value and decrease of the current total loss value after each round of updating the weights, biases and parameters to be inverted. When the total loss value drops below the reference standard, or the decrease is less than the set value for multiple consecutive iterations, or the preset maximum number of iterations is reached, it is determined that the convergence criterion is met. At this time, the weights and biases of the neural network are fixed to obtain the trained BO-PINN consolidation settlement forward and inverse target model. At the same time, the estimated values ​​of the undrained elastic modulus and permeability coefficient obtained in the current iteration are used as the in-situ mechanical parameters of the soil layer determined by the final inversion.

[0068] For example, such as Figure 2 As shown, Figure 2 The graph shows the total loss of the initial model for BO-PINN consolidation settlement forward and inverse modeling as a function of the number of training iterations. The training loss corresponds to the total loss calculated from the training sample set, which is used to drive the inverse update of the network weights. The test loss corresponds to the total loss calculated from the test sample set. The test sample set is independently divided using the same point composition rules as the training sample set, and also includes monitoring data points, internal collocation points, initial condition points, and boundary condition points, which are used to verify the generalization and convergence performance of the model.

[0069] For example, such as Figure 3 As shown, Figure 3 This is a schematic diagram showing the inversion results of the permeability coefficient of the target model for forward and inverse modeling of BO-PINN consolidation settlement.

[0070] For example, such as Figure 4 As shown, Figure 4 This is a schematic diagram of the undrained elastic modulus inversion results for the target model of BO-PINN consolidation settlement forward and inverse modeling.

[0071] In this way, both measured data constraints and physical law constraints can be simultaneously incorporated into the process of forward fitting of the consolidation response by the neural network, taking into account both the measured fit and mechanical rationality of the model output; the mode of simultaneously updating the network parameters and the parameters of the soil layer to be inverted can realize the coupled solution of forward prediction and parameter inversion, which helps to reduce the additional computational overhead brought by independent forward and inversion processes; based on the termination judgment mechanism of residual convergence, the training iteration scale can be reasonably controlled while ensuring model performance, thereby improving the overall operating efficiency of model training and parameter inversion.

[0072] S103, input the target detection sample into the BO-PINN consolidation settlement forward and inverse target model to obtain the consolidation settlement prediction result of the soft soil foundation corresponding to the target detection sample.

[0073] In one embodiment, the target detection sample can be a combination of spatial coordinates and time nodes of the section to be predicted. The target detection sample is input into the BO-PINN consolidation settlement forward and inverse target model to predict the excess pore water pressure and vertical displacement of the soft soil foundation corresponding to the target detection sample; for example... Figure 5 and Figure 6 As shown, Figure 5 A schematic diagram showing the results of predicting excess pore water pressure using the BO-PINN consolidation settlement forward and inverse target model. Figure 6 A schematic diagram showing the results of vertical displacement prediction for the BO-PINN consolidation settlement forward and inverse target model.

[0074] The aforementioned integrated prediction method for consolidation settlement of soft soil foundations based on the BO-PINN method constructs an integrated forward and inverse model of BO-PINN consolidation settlement. This allows for the reverse derivation of in-situ mechanical parameters of the soil layer that closely match the actual working conditions of the soil based on field-measured monitoring data. This effectively reduces the drawbacks of traditional indoor geotechnical tests, such as sample disturbance, size effect deviations, high costs of in-situ testing, and large data dispersion. It addresses the core problem of inaccurate parameter values ​​leading to distorted prediction results in traditional consolidation calculation methods, significantly reducing the deviation between predicted settlement results and actual field monitoring values. Furthermore, compared to traditional purely physics-driven numerical methods such as finite element and finite difference methods, it exhibits stronger adaptability to working conditions and higher prediction accuracy. Meanwhile, it achieves joint global optimization of model hyperparameters and soil layer parameters to be inverted, effectively reducing the manual cost and randomness of neural network parameter tuning, improving the convergence efficiency of model training and the stability of inversion results. Moreover, the trained model can support all-time consolidation settlement prediction at any depth and any time, which can flexibly meet the engineering needs of multiple scenarios such as dynamic control during construction and long-term settlement prediction after construction, providing reliable technical support for the refined control of consolidation settlement of soft soil foundations.

[0075] In some embodiments, the method for determining the physical constraint term includes: Based on the residual loss of the vertical force balance equation and the residual loss of the seepage continuity equation, and by summing the corresponding weighting coefficients, the loss term of the partial differential equation is determined. Based on the initial excess pore water pressure constraint residual loss and the initial vertical displacement constraint residual loss, and by summing the corresponding weight coefficients, the initial condition loss term is determined; The boundary condition loss term is determined by summing the residual losses of the seepage field boundary constraints and the displacement field boundary constraints, combined with the corresponding weighting coefficients. The physical constraint term is determined based on the partial differential equation loss term, the initial condition loss term, and the boundary condition loss term.

[0076] In some embodiments, determining the initial condition loss term by summing the initial excess pore water pressure constraint residual loss and the initial vertical displacement constraint residual loss with corresponding weighting coefficients includes: The first residual is calculated based on the deviation between the first predicted excess pore water pressure and the actual excess pore water pressure at the initial moment. The norm of the first residual is taken and combined with the initial condition weighting coefficient to determine the initial excess pore water pressure constraint residual loss. The second residual is calculated based on the deviation between the first predicted vertical displacement and the actual vertical displacement at the initial moment. The norm of the second residual is taken and combined with the initial condition weighting coefficient to determine the initial vertical displacement constraint residual loss. The initial condition loss term is determined based on the initial excess pore water pressure constraint residual loss and the initial vertical displacement constraint residual loss.

[0077] For example, the formula for the initial conditional loss term is as follows: ; ; ; in, Indicates the initial excess pore water pressure constraint residual loss; Indicates the initial vertical displacement constraint residual loss; Indicates the actual excess pore water pressure distribution at the initial moment; Indicates the first predicted excess pore water pressure at the initial moment; Indicates the initial time, i.e., t=0; Indicates excess pore water pressure; Indicates the foundation depth; Indicates vertical displacement; Indicates the first predicted vertical displacement at the initial moment; Indicates the actual vertical displacement at the initial moment; Indicates the initial condition weighting coefficients.

[0078] In some embodiments, the seepage field boundary constraint residual loss includes a first constraint residual and a second constraint residual; the step of determining the boundary condition loss term by summing the seepage field boundary constraint residual loss and the displacement field boundary constraint residual loss with corresponding weighting coefficients includes: The first constraint residual is determined by combining the norm of the second predicted excess pore water pressure at the first boundary of the drainage boundary with the boundary weight coefficient. The second constraint residual is determined by combining the norm of the partial derivative of the second predicted excess pore water pressure along the depth at the second boundary of the non-drained boundary with the boundary weight coefficient. The residual loss of the displacement field boundary constraint is determined by combining the norm of the second predicted vertical displacement at the fixed displacement boundary with the boundary weight coefficient; wherein, the fixed displacement boundary indicates the boundary space location in the displacement field where a fixed vertical constraint is applied. The boundary condition loss term is determined based on the residual loss of the seepage field boundary constraint and the residual loss of the displacement field boundary constraint.

[0079] For example, the expression formula for the boundary condition loss term is as follows: ; ; ; in, Indicates residual loss due to boundary constraints in the seepage field; Indicates residual loss due to boundary constraints in the displacement field; Indicates the residual of the first constraint; Indicates the second predicted excess pore water pressure; Indicates the foundation depth; Indicates excess pore water pressure; Indicator boundary weight coefficients; Indicates time; Indicates the residual of the second constraint; Indicates the partial derivative of the second predicted excess pore water pressure along depth; Indicates the second predicted vertical displacement; Indicates vertical displacement.

[0080] In some embodiments, determining the partial differential equation loss term by summing the residual losses from the vertical force balance equation and the seepage continuity equation, combined with the corresponding weighting coefficients, includes: The residual of the first equation is calculated based on the undrained elastic modulus, Poisson's ratio, the second-order partial derivative of the third predicted vertical displacement along the depth, the effective stress coefficient, and the first-order partial derivative of the third predicted excess pore water pressure along the depth. The norm of the residual of the first equation is taken and combined with the weighting coefficient of the first partial differential equation to determine the residual loss of the vertical force balance equation. The residual of the second equation is calculated based on the first-order partial derivative of the third predicted excess pore water pressure along time, the undrained elastic modulus, the effective stress coefficient, the permeability coefficient, and the second-order partial derivative of the third predicted vertical displacement along depth. The norm of the residual of the second equation is taken and combined with the weighting coefficient of the second partial differential equation to determine the residual loss of the seepage continuity equation. The loss term of the partial differential equation is determined based on the residual loss of the vertical force balance equation and the residual loss of the seepage continuity equation.

[0081] For example, the loss term in a partial differential equation can be calculated as follows: ; ; ; in, Indicator of loss terms in partial differential equations; Indicates the residual loss in the vertical force balance equation; Indicator of residual loss in seepage continuity equation; Indicates the undrained elastic modulus; Indicator Poisson's ratio; The second-order partial derivative of the third predicted vertical displacement along the depth is indicated; Indicates vertical displacement; Indicates the foundation depth; Indicator of effective stress (Biot) coefficient; The third prediction of excess pore water pressure is indicated by the first-order partial derivative along depth. Indicates the weighting coefficients of the first partial differential equation; Indicates the weighting coefficients of the second partial differential equation; Indicator permeability coefficient; The third prediction of excess pore water pressure is indicated by the first-order partial derivative along time. The second-order partial derivative of the third predicted vertical displacement along the depth is indicated.

[0082] For example, the target loss function can be expressed as follows: ; in, Indicates physical constraints; Indicates data loss items.

[0083] For example, the physical constraint term can be expressed as follows: ; in, Indicator of loss terms in partial differential equations; Indicates the initial condition loss term; Indicates the boundary condition loss term.

[0084] For example, the formula for expressing data constraints can be as follows: ; in, Indicates excess pore water pressure; Indicates the foundation depth; Indicates vertical displacement; t indicates time; Indicates the true value (observed value) of excess pore water pressure measured by the on-site sensor; Indicates the true value (observed value) of the vertical displacement measured by the on-site sensors; The fourth indicator predicts excess pore water pressure; Indicates the fourth predicted vertical displacement; Weighting coefficients indicating the loss of monitoring data.

[0085] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0086] Based on the same inventive concept, this application also provides an integrated prediction device for consolidation and settlement of soft soil foundations based on the BO-PINN method, used to implement the aforementioned integrated prediction method for consolidation and settlement of soft soil foundations based on the BO-PINN method. The solution provided by this device is similar to the solution described in the above method. Therefore, the specific limitations of one or more embodiments of the integrated prediction device for consolidation and settlement of soft soil foundations based on the BO-PINN method provided below can be found in the limitations of the integrated prediction method for consolidation and settlement of soft soil foundations based on the BO-PINN method described above, and will not be repeated here.

[0087] In one embodiment, such as Figure 7 As shown, an integrated prediction device for consolidation settlement of soft soil foundation based on the BO-PINN method is provided. The device includes: The acquisition module 10 is used to acquire a sample dataset and target detection samples; wherein, the sample dataset includes a training sample set and a test sample set; Module 20 is used to train the initial model for forward and inverse BO-PINN consolidation settlement using the training sample set, iteratively filter the model hyperparameters and soil parameters to be inverted of the initial model for forward and inverse BO-PINN consolidation settlement, and obtain the trained target model for forward and inverse BO-PINN consolidation settlement, as well as the in-situ mechanical parameters of the soil layer determined by inversion. The detection module 30 is used to input the target detection sample into the BO-PINN consolidation settlement forward and inverse target model to obtain the consolidation settlement prediction result of the soft soil foundation corresponding to the target detection sample.

[0088] In some embodiments, the BO-PINN consolidation settlement forward and inverse target model is a two-layer coupled architecture: wherein, The outer layer is a Bayesian optimization module, which consists of a probabilistic surrogate model and a data acquisition function. It is used to iteratively optimize the hyperparameters of the physical information neural network and simultaneously invert the in-situ mechanical parameters of the soil layer. The inner layer is a physical information neural network module. The physical information neural network module uses a neural network as the fitting carrier to construct a target loss function that includes data fitting terms and physical constraint terms. The network weights are updated through backpropagation of the loss to complete the forward modeling prediction of pore water pressure and vertical displacement of soft soil foundation.

[0089] In some embodiments, the construction module 20 includes: The first determining unit is used to determine the loss term of the partial differential equation by summing the residual loss of the vertical force balance equation and the residual loss of the seepage continuity equation, combined with the corresponding weight coefficients. The second determining unit is used to determine the initial condition loss term by summing the initial excess pore water pressure constraint residual loss and the initial vertical displacement constraint residual loss with the corresponding weight coefficients. The third determining unit is used to determine the boundary condition loss term by summing the residual loss of the seepage field boundary constraint and the residual loss of the displacement field boundary constraint, combined with the corresponding weight coefficients. The fourth determining unit is used to determine the physical constraint term based on the partial differential equation loss term, the initial condition loss term, and the boundary condition loss term.

[0090] In some embodiments, the second determining unit is configured to perform the following steps: The first residual is calculated based on the deviation between the first predicted excess pore water pressure and the actual excess pore water pressure at the initial moment. The norm of the first residual is taken and combined with the initial condition weighting coefficient to determine the initial excess pore water pressure constraint residual loss. The second residual is calculated based on the deviation between the first predicted vertical displacement and the actual vertical displacement at the initial moment. The norm of the second residual is taken and combined with the initial condition weighting coefficient to determine the initial vertical displacement constraint residual loss. The initial condition loss term is determined based on the initial excess pore water pressure constraint residual loss and the initial vertical displacement constraint residual loss.

[0091] In some embodiments, the seepage field boundary constraint residual loss includes a first constraint residual and a second constraint residual; the first determining unit is configured to perform the following steps: The first constraint residual is determined by combining the norm of the second predicted excess pore water pressure at the first boundary of the drainage boundary with the boundary weight coefficient. The second constraint residual is determined by combining the norm of the partial derivative of the second predicted excess pore water pressure along the depth at the second boundary of the non-drained boundary with the boundary weight coefficient. The residual loss of the displacement field boundary constraint is determined by combining the norm of the second predicted vertical displacement at the fixed displacement boundary with the boundary weight coefficient; wherein, the fixed displacement boundary indicates the boundary space location in the displacement field where a fixed vertical constraint is applied. The boundary condition loss term is determined based on the residual loss of the seepage field boundary constraint and the residual loss of the displacement field boundary constraint.

[0092] In some embodiments, the first determining unit is configured to perform the following steps: The residual of the first equation is calculated based on the undrained elastic modulus, Poisson's ratio, the second-order partial derivative of the third predicted vertical displacement along the depth, the effective stress coefficient, and the first-order partial derivative of the third predicted excess pore water pressure along the depth. The norm of the residual of the first equation is taken and combined with the weighting coefficient of the first partial differential equation to determine the residual loss of the vertical force balance equation. The residual of the second equation is calculated based on the first-order partial derivative of the third predicted excess pore water pressure along time, the undrained elastic modulus, the effective stress coefficient, the permeability coefficient, and the second-order partial derivative of the third predicted vertical displacement along depth. The norm of the residual of the second equation is taken and combined with the weighting coefficient of the second partial differential equation to determine the residual loss of the seepage continuity equation. The loss term of the partial differential equation is determined based on the residual loss of the vertical force balance equation and the residual loss of the seepage continuity equation.

[0093] In some embodiments, the construction module 20 includes: The foundation depth and time coordinates in the training sample set are input into the neural network of the initial model for forward and inverse modeling of BO-PINN consolidation settlement. The excess pore water pressure and vertical displacement are estimated through forward propagation to obtain the estimated values ​​of excess pore water pressure and vertical displacement at the corresponding points. Based on the estimated value of excess pore water pressure and the estimated value of vertical displacement, the target loss function is calculated by combining the data fitting term and the physical constraint term, and the total loss value of the target loss function is obtained. Based on the total loss value, the weights and biases of the neural network are updated by backpropagation using the chain rule, and the estimated values ​​of the undrained elastic modulus and permeability coefficient are generated synchronously and iteratively. When the total loss value meets the convergence criterion, the BO-PINN consolidation settlement forward and inverse model is obtained, and the in-situ mechanical parameters of the soil layer are determined by inversion.

[0094] Each module in the above-mentioned integrated prediction device for consolidation and settlement of soft soil foundation based on the BO-PINN method can be implemented in whole or in part through software, hardware, or a combination thereof. Each module can be embedded in the processor of the electronic device in hardware form or independent of the processor, or stored in the memory of the electronic device in software form, so that the processor can call and execute the corresponding operations of each module.

[0095] In one embodiment, an electronic device is provided, the internal structure of which can be shown as follows: Figure 8 As shown, the electronic device includes a processor, memory, communication interface, display unit, and input device connected via a method bus. The processor provides computational and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores operating methods and computer programs. The internal memory provides an environment for the operation of the operating methods and computer programs stored in the non-volatile storage medium. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, NFC (Near Field Communication), or other technologies. When the computer program is executed by the processor, it implements an integrated prediction method for consolidation and settlement of soft soil foundations based on the BO-PINN method. The display screen of the electronic device can be a liquid crystal display (LCD) or an e-ink display. The input device can be a touch layer covering the display screen, or buttons, a trackball, or a touchpad mounted on the device's casing.

[0096] Those skilled in the art will understand that Figure 8The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the electronic device to which the present application is applied. The specific electronic device may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0097] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the steps in the above method embodiments.

[0098] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Furthermore, any references to memory, storage, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory.

[0099] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as the combination of these technical features does not contradict each other, it should be considered within the scope of this specification. The above embodiments only illustrate several implementation methods of this application, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of this application. It should be noted that for those skilled in the art, several modifications and improvements can be made without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for predicting consolidation settlement of soft soil foundations based on the BO-PINN method, characterized in that, The method includes: Obtain a sample dataset and object detection samples; wherein the sample dataset includes a training sample set and a test sample set; The initial model for forward and inverse BO-PINN consolidation settlement is trained using the training sample set. The model hyperparameters and soil parameters to be inverted are iteratively selected from the initial model for forward and inverse BO-PINN consolidation settlement. The trained target model for forward and inverse BO-PINN consolidation settlement and the in-situ mechanical parameters of the soil layer determined by inversion are obtained. The target detection sample is input into the BO-PINN consolidation settlement forward and inverse target model to obtain the consolidation settlement prediction result of the soft soil foundation corresponding to the target detection sample.

2. The method according to claim 1, characterized in that, The BO-PINN consolidation settlement forward and inverse model is a two-layer coupled architecture: where... The outer layer is a Bayesian optimization module, which consists of a probabilistic surrogate model and a data acquisition function. It is used to iteratively optimize the hyperparameters of the physical information neural network and simultaneously invert the in-situ mechanical parameters of the soil layer. The inner layer is a physical information neural network module. The physical information neural network module uses a neural network as the fitting carrier to construct a target loss function that includes data fitting terms and physical constraint terms. The network weights are updated through backpropagation of the loss to complete the forward modeling prediction of pore water pressure and vertical displacement of soft soil foundation.

3. The method according to claim 2, characterized in that, The method for determining the physical constraint terms includes: Based on the residual loss of the vertical force balance equation and the residual loss of the seepage continuity equation, and by summing the corresponding weighting coefficients, the loss term of the partial differential equation is determined. Based on the initial excess pore water pressure constraint residual loss and the initial vertical displacement constraint residual loss, and by summing the corresponding weight coefficients, the initial condition loss term is determined; The boundary condition loss term is determined by summing the residual losses of the seepage field boundary constraints and the displacement field boundary constraints, combined with the corresponding weighting coefficients. The physical constraint term is determined based on the partial differential equation loss term, the initial condition loss term, and the boundary condition loss term.

4. The method according to claim 3, characterized in that, The initial condition loss term is determined by summing the initial excess pore water pressure constraint residual loss and the initial vertical displacement constraint residual loss with corresponding weighting coefficients, including: The first residual is calculated based on the deviation between the first predicted excess pore water pressure and the actual excess pore water pressure at the initial moment. The norm of the first residual is taken and combined with the initial condition weighting coefficient to determine the initial excess pore water pressure constraint residual loss. The second residual is calculated based on the deviation between the first predicted vertical displacement and the actual vertical displacement at the initial moment. The norm of the second residual is taken and combined with the initial condition weighting coefficient to determine the initial vertical displacement constraint residual loss. The initial condition loss term is determined based on the initial excess pore water pressure constraint residual loss and the initial vertical displacement constraint residual loss.

5. The method according to claim 3, characterized in that, The seepage field boundary constraint residual loss includes the first constraint residual and the second constraint residual; the boundary condition loss term is determined by summing the seepage field boundary constraint residual loss and the displacement field boundary constraint residual loss with corresponding weight coefficients, including: The first constraint residual is determined by combining the norm of the second predicted excess pore water pressure at the first boundary of the drainage boundary with the boundary weight coefficient. The second constraint residual is determined by combining the norm of the partial derivative of the second predicted excess pore water pressure along the depth at the second boundary of the non-drained boundary with the boundary weight coefficient. The residual loss of the displacement field boundary constraint is determined by combining the norm of the second predicted vertical displacement at the fixed displacement boundary with the boundary weight coefficient; wherein, the fixed displacement boundary indicates the boundary space location in the displacement field where a fixed vertical constraint is applied. The boundary condition loss term is determined based on the residual loss of the seepage field boundary constraint and the residual loss of the displacement field boundary constraint.

6. The method according to claim 3, characterized in that, The method for determining the partial differential equation loss term by summing the residual losses from the vertical force balance equation and the seepage continuity equation, combined with the corresponding weighting coefficients, includes: The residual of the first equation is calculated based on the undrained elastic modulus, Poisson's ratio, the second-order partial derivative of the third predicted vertical displacement along the depth, the effective stress coefficient, and the first-order partial derivative of the third predicted excess pore water pressure along the depth. The norm of the residual of the first equation is taken and combined with the weighting coefficient of the first partial differential equation to determine the residual loss of the vertical force balance equation. The residual of the second equation is calculated based on the first-order partial derivative of the third predicted excess pore water pressure along time, the undrained elastic modulus, the effective stress coefficient, the permeability coefficient, and the second-order partial derivative of the third predicted vertical displacement along depth. The norm of the residual of the second equation is taken and combined with the weighting coefficient of the second partial differential equation to determine the residual loss of the seepage continuity equation. The loss term of the partial differential equation is determined based on the residual loss of the vertical force balance equation and the residual loss of the seepage continuity equation.

7. The method according to claim 1, characterized in that, The process involves training an initial BO-PINN consolidation settlement forward and inverse model using the training sample set, iteratively filtering the model hyperparameters and soil layer parameters to be inverted from the initial BO-PINN consolidation settlement forward and inverse model, and obtaining the trained BO-PINN consolidation settlement forward and inverse target model, as well as the in-situ mechanical parameters of the soil layer determined by the inversion, including: The foundation depth and time coordinates in the training sample set are input into the neural network of the initial model for forward and inverse modeling of BO-PINN consolidation settlement. The excess pore water pressure and vertical displacement are estimated through forward propagation to obtain the estimated values ​​of excess pore water pressure and vertical displacement at the corresponding points. Based on the estimated value of excess pore water pressure and the estimated value of vertical displacement, the target loss function is calculated by combining the data fitting term and the physical constraint term, and the total loss value of the target loss function is obtained. Based on the total loss value, the weights and biases of the neural network are updated by backpropagation using the chain rule, and the estimated values ​​of the undrained elastic modulus and permeability coefficient are generated synchronously and iteratively. When the total loss value meets the convergence criterion, the BO-PINN consolidation settlement forward and inverse model is obtained, and the in-situ mechanical parameters of the soil layer are determined by inversion.

8. An integrated prediction device for consolidation settlement of soft soil foundation based on the BO-PINN method, characterized in that, The device includes: An acquisition module is used to acquire a sample dataset and object detection samples; wherein, the sample dataset includes a training sample set and a test sample set; The construction module is used to train the initial model of BO-PINN consolidation settlement forward and inverse through the training sample set, iteratively filter the model hyperparameters and soil layer parameters to be inverted of the initial model of BO-PINN consolidation settlement forward and inverse, and obtain the trained target model of BO-PINN consolidation settlement forward and inverse, as well as the in-situ mechanical parameters of the soil layer determined by inversion. The detection module is used to input the target detection sample into the BO-PINN consolidation settlement forward and inverse target model to obtain the consolidation settlement prediction result of the soft soil foundation corresponding to the target detection sample.

9. An electronic device, characterized in that, The method includes a memory and a processor, wherein the memory stores a computer program, which, when executed by the processor, implements the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it performs the steps of the method according to any one of claims 1 to 7.

Images