Pile shoe foundation penetration risk early warning method and system based on Bayesian theory and optimization strategy and medium

Abstract

The invention relates to the field of ocean geotechnical engineering, in particular to a pile shoe foundation penetration risk early warning method and system based on the Bayesian theory and the optimization strategy and a medium, and the early warning method comprises the steps that prior distribution of soil layer uncertainty parameters is established, and real-time observation data in the penetration process is collected; screening an active parameter set by calculating the sensitivity of the model predicted value to the parameters; constructing a negative log-likelihood loss function, determining an active parameter optimal value by combining a Bayesian theory and an optimization search strategy, and updating posterior distribution of the active parameter optimal value; and a bearing capacity prediction value and a prediction interval are generated by adopting Monte Carlo sampling and Gaussian regression methods. Compared with the prior art, monitoring data can be dynamically integrated, the uncertainty of soil layer parameters is effectively reduced, continuous and accurate prediction of the bearing capacity in the whole penetration process is achieved, the pile pitching risk is controlled, a decision basis is provided for offshore pile standing, and the operation safety of a self-elevating offshore wind power installation platform is remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of marine geotechnical engineering, and in particular to a pile shoe foundation penetration risk early warning method, system and medium based on Bayesian theory and optimization strategy. BACKGROUND

[0002] With the rapid development of offshore wind power industry, the self-elevating offshore wind power installation platform has been widely used in the hoisting of wind turbine generators. Before the platform operation, pile insertion operation is needed, that is, the pile foot is placed on the seabed, and the pile foot is inserted into the seabed by loading water in the ship cabin. During the loading, the load of the entire platform is transmitted to the foundation soil through the pile shoe foundation. If the load applied to the pile shoe foundation exceeds the bearing capacity of the underlying soil, the pile shoe foundation will sink rapidly, that is, the penetration failure occurs, which is easy to cause serious inclination of the self-elevating offshore wind power installation platform, and further cause the pile leg to break, the platform to overturn, and thus cause serious accidents such as device damage and personnel injury.

[0003] The penetration of the pile shoe foundation mainly occurs in the layered foundation with hard upper layer and soft lower layer. The nearshore seabed soil layer is widely developed in the form of layered soil, and the complex layered stratum is easy to induce the penetration of the pile shoe. However, the existing research mainly focuses on the risk control of specific soil layers, and is limited to a certain penetration failure point, and does not evaluate the risk of the entire pile insertion process. In addition, the insertion and extraction of the pile of the self-elevating offshore wind power installation platform presents a high frequency characteristic due to the installation demand, and the geological exploration data is usually distributed in the center of the wind turbine position, which has a large spatial difference with the actual operation position of the pile shoe foundation, so that the geological exploration data is difficult to directly reflect the accurate geological conditions of the actual installation position of the pile shoe. Limited by the limited field test data, the soil layer parameters have a large uncertainty, which leads to insufficient prediction accuracy of the bearing capacity of the pile shoe foundation. In the current pile insertion process, the penetration depth and bearing capacity of the pile shoe foundation are monitored, but these monitoring data are difficult to predict the penetration risk that may exist in the subsequent pile insertion process.

[0004] However, the existing technology has obvious limitations in predicting the penetration risk of the pile shoe. For example, the existing technology CN105868481A proposes a marine platform pile shoe foundation installation risk control method based on Bayesian theory. The method constructs a load and depth prediction model for penetration failure, and updates the prediction results in real time by using monitoring data. However, this method mainly focuses on predicting a single penetration failure point, and does not continuously and dynamically predict the bearing capacity of the entire pile insertion process, which is difficult to comprehensively evaluate the risk evolution in the pile insertion process. In addition, the method still has insufficient treatment of the uncertainty of the soil layer parameters, especially when there is a spatial difference between the geological exploration data and the actual operation position of the pile shoe, the prediction accuracy is easily limited.

[0005] On the other hand, the prior art CN119578246A provides a pipe pile vertical bearing capacity estimation method based on Bayesian theory, which updates the posterior distribution through the prior distribution and the measured data to reduce the error caused by the uncertainty factor. However, this technology is aimed at pipe pile bearing capacity evaluation, not pile shoe foundation; its model does not involve penetration risk control, and lacks dynamic integration of monitoring data throughout the penetration process, and cannot be directly applied to real-time risk control of jack-up platform pile insertion operation.

[0006] How to use these monitoring data to reduce the uncertainty of soil layer parameters, predict the bearing capacity in the entire penetration process of the pile shoe foundation, and thus control the penetration risk existing in the entire pile insertion process is of great significance to ensure the safety during the installation process of the jack-up offshore wind power installation platform. SUMMARY

[0007] The purpose of the present application is to overcome the defects of the prior art and provide a pile shoe foundation penetration risk early warning method, system and medium based on Bayesian theory and optimization strategy, which can dynamically integrate monitoring data, effectively reduce the uncertainty of soil layer parameters, realize continuous and accurate prediction of the bearing capacity throughout the penetration process, and significantly improve the safety of jack-up offshore wind power installation platform operation.

[0008] The purpose of the present application can be achieved by the following technical solutions: The first aspect of the present application provides a pile shoe foundation penetration risk early warning method based on Bayesian theory and optimization strategy, comprising the following steps: S1, obtaining the shape parameters of the pile shoe foundation and the marine soil layer parameters, selecting the uncertainty parameters and establishing the prior distribution thereof; S2, obtaining the real-time observation data in the pile shoe penetration process, the observation data including the penetration depth and the corresponding bearing capacity, and assigning weights to each observation data point; S3, based on the pre-established pile shoe foundation bearing capacity calculation model and the weight assigned to each observation data point, calculating the sensitivity of the model prediction value to each uncertainty parameter, and selecting an active parameter set according to the sensitivity; S4, constructing a negative log-likelihood loss function according to the error between the prediction value of the pile shoe foundation bearing capacity calculation model and the observation data; S5, combining the prior distribution of the active parameter set and the negative log-likelihood loss function, applying Bayesian theory and optimization search strategy to determine the optimal value of the active parameter, and updating the posterior distribution of the active parameter, and the prior distribution of the non-active parameter remains unchanged; S6, Monte Carlo sampling is performed on the updated uncertainty parameter posterior distribution, the Gaussian regression method is used to correct the prediction mean and variance, the bearing capacity prediction value and prediction interval are generated, and according to the subsequent prediction bearing capacity change trend with depth, it is judged whether there is a puncture risk in the subsequent penetration process, and a warning signal is generated when the risk is identified.

[0009] Further, the pile shoe foundation penetration risk warning method based on Bayesian theory and optimization strategy further comprises: S7, the latest updated active parameter posterior distribution is taken as the prior distribution of the next round of iteration, steps S2 to S6 are repeated, and the prediction result is updated in a cycle by using the newly acquired observation data until the pile shoe foundation installation process is completed.

[0010] Further, in S1, the shape parameters and marine soil layer parameters of the pile shoe foundation are obtained, the specific process of selecting the uncertainty parameter category and establishing the prior distribution thereof includes: The geometric size of the pile shoe foundation is collected as the shape parameter, and the soil layer characteristics of each soil layer are obtained as the geological parameter; From the geological parameters, parameters that have a relatively significant impact on the bearing capacity calculation result are selected as the uncertainty parameters; Based on the engineering experience value, a design value is set for each uncertainty parameter, and a parameter interval is established as the initial prior distribution within a preset initial deviation range centered on the design value.

[0011] Further, in S2, the real-time observation data in the pile shoe penetration process is: in the process of penetrating the pile shoe into the seabed, the observation data sequence of the penetration depth and the bearing capacity provided by the ground soil reaction force at the penetration depth; The specific process of assigning weights to each observation data point in S2 includes: The depth distance of each observation point from the latest observation point is calculated, the weights are preliminarily assigned according to the distance, and finally the weights of all observation points are normalized.

[0012] Further, in S3, based on the pre-established pile shoe foundation bearing capacity calculation model and the weight assigned to each observation data point, the sensitivity of the model prediction value to each uncertainty parameter is calculated, and the specific process of screening the active parameter set according to the sensitivity includes: Each uncertainty parameter is disturbed, and the disturbed bearing capacity prediction curve is calculated; For each observation depth, the prediction bearing capacity change caused by parameter disturbance is calculated; According to the weight of each observation data point, the sensitivity is defined as the weighted bearing capacity change of each uncertainty parameter at all observation depths; Sort all parameters by sensitivity and filter those with a sensitivity higher than the maximum sensitivity preset ratio into the active parameter set; In S3, the pile shoe foundation bearing capacity calculation model is a physical model or empirical model used to calculate the bearing capacity corresponding to different depths during the pile shoe penetration process.

[0013] Furthermore, in S4, the specific process of constructing the negative log-likelihood loss function based on the error between the predicted value and the observed data of the pile foundation bearing capacity calculation model includes: Assume that the error between the model's predicted values ​​and the observed data follows a normal distribution with a mean of zero; The problem of maximizing the likelihood probability is transformed into the problem of minimizing the loss function. Combining the weights of the observation data allocated in step S2, a weighted negative log-likelihood loss function is constructed. The negative log-likelihood loss function represents the negative log value of the conditional probability of the observation data occurring under the current parameter values.

[0014] Furthermore, in S5, combining the prior distribution of the active parameter set and the negative log-likelihood loss function, Bayesian theory and an optimization search strategy are applied to determine the optimal values ​​of the active parameters, and the posterior distribution of the active parameters is updated. The specific process includes: Based on Bayesian theory, the posterior distribution is expressed as the product of the prior distribution and the likelihood function; Using the negative log-likelihood loss function as the objective function and the prior distribution range of the active parameters as the search boundary, an optimization algorithm is used to search for the parameter value with the minimum loss as the optimal value of the active parameters.

[0015] Keep the inactive parameters at the optimal value of the previous round, and together with the optimal value of the active parameters, form the current optimal parameter group; Calculate the weighted residual standard deviation of the predicted value and the observed data under the optimal parameter set, and compare it with a preset threshold. Adjust the distribution range of active parameters based on the comparison results: if the residual standard deviation meets the threshold requirement, shrink the distribution range; otherwise, expand the distribution range and re-optimize the parameters. The final determined parameter distribution interval is used as the updated posterior distribution of the active parameters.

[0016] Furthermore, the specific process of performing Monte Carlo sampling on the posterior distribution of the updated uncertainty parameters, and using Gaussian regression to correct the predicted mean and variance to generate the predicted bearing capacity value and prediction interval includes: Multiple sets of parameter samples are randomly drawn from the updated posterior distribution of the uncertainty parameters; For each set of parameter samples, the predicted bearing capacity at different penetration depths is obtained through the pile shoe foundation bearing capacity calculation model; The mean and variance of the predicted bearing capacity values ​​for all parameter samples at each depth point were statistically analyzed. A Gaussian process regression model is established by using the error between the predicted mean and the observed data as the training objective. A radial basis function kernel is used as the kernel function for Gaussian process regression to correct the error of the mean curve in the Monte Carlo simulation. The variance of the Gaussian process regression prediction is superimposed with the variance of the Monte Carlo simulation to obtain a comprehensive uncertainty measure; Based on the corrected mean and comprehensive variance, a bearing capacity prediction interval is generated at a specified confidence level.

[0017] A second aspect of this invention provides a pile shoe foundation penetration risk early warning system based on Bayesian theory and optimization strategies, comprising: The data acquisition and preprocessing module is used to acquire the shape parameters of the pile shoe foundation, the marine soil layer parameters, and the real-time observation data during the pile shoe penetration process. It selects the uncertainty parameters and establishes their prior distribution, and assigns weights to each observation data point. The Bayesian update and parameter optimization module is used to calculate the sensitivity of the model prediction value to each uncertainty parameter based on the pile foundation bearing capacity calculation model and the weight of the observation data to screen the active parameter set. It constructs a negative log-likelihood loss function based on the error between the model prediction value and the observation data, and applies Bayesian theory and optimization strategy to determine the optimal value of the active parameter and update its posterior distribution. The uncertainty quantification module is used to perform Monte Carlo sampling on the posterior distribution of the updated uncertainty parameters, and to use Gaussian regression to correct the predicted mean and variance, thereby generating the predicted bearing capacity value and prediction interval. The visualization and early warning module is used to display the bearing capacity prediction results in real time and provide early warning prompts when a puncture risk is detected. Specifically, based on the subsequent trend of the predicted bearing capacity change with depth, it is determined whether there is a puncture risk in the subsequent penetration process, and an early warning signal is generated when the risk is identified to provide an early warning prompt.

[0018] A third aspect of the present invention provides a storage medium containing computer-executable instructions, which, when executed by a computer processor, is used to execute the pile shoe foundation penetration risk warning method described above based on Bayesian theory and optimization strategies.

[0019] Compared with existing technologies, this invention has the following advantages: Based on Bayesian theory and optimization strategies, this invention comprehensively considers the uncertainty of soil parameters, establishes a pile shoe foundation penetration risk early warning model, and updates the prediction results in real time based on monitoring data, providing an early warning method and system for identifying puncture risks. Compared with existing foundation installation risk control technologies, this invention is applicable to complex multi-layered marine soils and can control the load and depth location of puncture failure throughout the entire pile driving process. As monitoring data increases, the prediction results are continuously updated and approach the actual values, resulting in more precise risk control throughout the pile driving process. This invention can be applied to the installation of foundations for self-elevating offshore wind power installation platforms, which is of great significance for ensuring the safety of the platform's pile shoe foundation installation process. Attached Figure Description

[0020] Figure 1 A framework diagram for risk warning process of pile foundation penetration; Figure 2 Modular flowchart for risk warning system for pile foundation penetration; Figure 3 This is a diagram showing the shape parameters of the pile shoe foundation. Figure 4 shows some early warning diagrams during the pile shoe foundation penetration process. The bearing capacity and parameter updates during the penetration process are shown in Figure 4(a), Figure 4(b), and Figure 4(c). Detailed Implementation

[0021] The pile shoe foundation penetration risk early warning method based on Bayesian theory and optimization strategy in this invention includes the following steps: S1. Obtain the shape parameters of the pile shoe foundation and the marine soil layer parameters, select the uncertainty parameters and establish their prior distribution.

[0022] In S1, the specific process of obtaining the shape parameters of the pile shoe foundation and the marine soil layer parameters, selecting the category of uncertainty parameters, and establishing their prior distribution includes: Collect the geometric dimensions of the pile shoe foundation as shape parameters, and obtain the soil characteristics of each soil layer as geological parameters; Parameters that have a relatively significant impact on the bearing capacity calculation results are selected from the geological parameters as uncertainty parameters; Design values ​​are set for each uncertainty parameter based on engineering experience, and a parameter interval is established within a preset initial deviation range centered on these design values ​​as its initial prior distribution.

[0023] S2. Obtain real-time observation data during the pile shoe penetration process. The observation data includes the penetration depth and the corresponding bearing capacity, and assign weights to each observation data point.

[0024] In S2, the real-time observation data during the pile shoe penetration process is: the sequence of observation data formed by the penetration depth and the bearing capacity provided by the foundation soil reaction at the penetration depth during the process of the pile shoe penetrating the seabed. In S2, the specific process of assigning weights to each observation data point includes: Calculate the depth distance between each observation point and the latest observation point, initially assign weights based on the distance, and finally normalize the weights of all observation points.

[0025] S3. Based on the pre-established pile foundation bearing capacity calculation model and the weights assigned to each observation data point, calculate the sensitivity of the model's predicted values ​​to each uncertainty parameter, and select the set of active parameters based on the sensitivity.

[0026] In S3, based on the pre-established pile shoe foundation bearing capacity calculation model and the weights assigned to each observation data point, the sensitivity of the model's predicted values ​​to various uncertainty parameters is calculated. The specific process of selecting the set of active parameters based on the sensitivity includes: For each uncertainty parameter, a perturbation is applied, and the predicted bearing capacity curve after the perturbation is calculated; For each observation depth, calculate the change in predicted bearing capacity caused by parameter disturbance; Based on the weight of each observation data point, sensitivity is defined as the weighted bearing capacity change of each uncertainty parameter at all observation depths. Sort all parameters by sensitivity and filter those with a sensitivity higher than the maximum sensitivity preset ratio into the active parameter set; In S3, the pile shoe foundation bearing capacity calculation model is a physical model or empirical model used to calculate the bearing capacity corresponding to different depths during the pile shoe penetration process.

[0027] S4. Based on the error between the predicted value and the observed data of the pile foundation bearing capacity calculation model, construct a negative log-likelihood loss function.

[0028] In S4, the specific process of constructing the negative log-likelihood loss function based on the error between the predicted value and the observed data of the pile foundation bearing capacity calculation model includes: Assume that the error between the model's predicted values ​​and the observed data follows a normal distribution with a mean of zero; The problem of maximizing the likelihood probability is transformed into the problem of minimizing the loss function. Combining the weights of the observation data allocated in step S2, a weighted negative log-likelihood loss function is constructed. The negative log-likelihood loss function represents the negative log value of the conditional probability of the observation data occurring under the current parameter values.

[0029] S5. Combining the prior distribution of the active parameter set and the negative log-likelihood loss function, apply Bayesian theory and optimization search strategy to determine the optimal value of the active parameters, and update the posterior distribution of the active parameters, while keeping the prior distribution of the inactive parameters unchanged.

[0030] In S5, the specific process of determining the optimal values ​​of the active parameters and updating the posterior distribution of the active parameters by combining the prior distribution of the active parameter set and the negative log-likelihood loss function, applying Bayesian theory and an optimization search strategy, includes: Based on Bayesian theory, the posterior distribution is expressed as the product of the prior distribution and the likelihood function; Using the negative log-likelihood loss function as the objective function and the prior distribution range of the active parameters as the search boundary, an optimization algorithm is used to search for the parameter value with the minimum loss as the optimal value of the active parameters.

[0031] Keep the inactive parameters at the optimal value of the previous round, and together with the optimal value of the active parameters, form the current optimal parameter group; Calculate the weighted residual standard deviation of the predicted value and the observed data under the optimal parameter set, and compare it with a preset threshold. Adjust the distribution range of active parameters based on the comparison results: if the residual standard deviation meets the threshold requirement, shrink the distribution range; otherwise, expand the distribution range and re-optimize the parameters. The final determined parameter distribution interval is used as the updated posterior distribution of the active parameters.

[0032] S6. Perform Monte Carlo sampling on the posterior distribution of the updated uncertainty parameters, and use Gaussian regression to correct the predicted mean and variance, generating the predicted bearing capacity value and prediction interval.

[0033] In S6, the specific process of performing Monte Carlo sampling on the updated posterior distribution of uncertainty parameters, using Gaussian regression to correct the predicted mean and variance, and generating the predicted bearing capacity value and prediction interval includes: Multiple sets of parameter samples are randomly drawn from the updated posterior distribution of the uncertainty parameters; For each set of parameter samples, the predicted bearing capacity at different penetration depths is obtained through the pile shoe foundation bearing capacity calculation model; The variance of the bearing capacity prediction values ​​of all parameter samples at each depth point is statistically analyzed, and the predicted value corresponding to the parameter group with the maximum a posteriori probability is used to represent the statistical mean. A Gaussian process regression model is established by using the error between the predicted mean and the observed data as the training objective. A radial basis function kernel is used as the kernel function for Gaussian process regression to correct the error of the mean curve in the Monte Carlo simulation. The variance of the Gaussian process regression prediction is superimposed with the variance of the Monte Carlo simulation to obtain a comprehensive uncertainty measure; Based on the corrected mean and comprehensive variance, a bearing capacity prediction interval is generated at a specified confidence level.

[0034] S7. Use the latest updated posterior distribution of active parameters as the prior distribution for the next iteration, and repeat steps S2 to S6. Use the newly acquired observation data to cyclically update the prediction results until the pile foundation installation process is completed.

[0035] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. Component models, material names, connection structures, circuit structures, control methods, algorithms, and other features not explicitly described in this technical solution are considered common technical features disclosed in the prior art.

[0036] Example 1 The pile shoe foundation penetration risk early warning method based on Bayesian theory and optimization strategy in this embodiment includes the following steps: Step 1: Obtain the parameters required to calculate the penetration bearing capacity of the pile shoe foundation, including the shape parameters of the pile shoe foundation and the geological parameters of each soil layer, determine the uncertainty parameters, and construct their initial prior distribution; Step 2: Obtain the bearing capacity and penetration depth of the real-time observation points during the penetration process as observation data, and assign weights to the data of each observation point. Step 3: Calculate the sensitivity of the model's predicted values ​​to changes in uncertain parameters, and select parameters whose sensitivity is higher than the maximum sensitivity by a certain proportion as the active parameter set; Step 4: Construct a negative log-likelihood loss function based on the error between the predicted value calculated by the pile foundation bearing capacity calculation model and the observed data; Step 5: Combining the prior distribution of active parameters and the negative log-likelihood loss function, determine the optimal value of active parameters under the maximum likelihood probability according to Bayesian theory and optimization search strategy, and update the posterior distribution of active parameters, while keeping the prior distribution of inactive parameters unchanged. Step 6: For the posterior distribution of the uncertainty parameters, use Monte Carlo sampling to sample N sets of parameters, calculate the bearing capacity results under each set of parameters, and calculate their variance. Use the bearing capacity prediction value under the maximum likelihood probability to represent the mean, and use Gaussian regression to correct the mean and variance, while generating the prediction interval. Step 7: Use the posterior distribution of the active parameters as the prior distribution, and repeat steps 2 to 6 until the pile shoe foundation installation is completed and no new observation data is generated.

[0037] As a further improvement of the present invention, the initial prior distribution in step 1 P (θ) is assumed to be a parameter range established with the design value as the center within a preset initial deviation range.

[0038] As a further improvement of the present invention, in step 2, the weights of each observation point data are assigned according to the distance from the depth of the latest observation point, with higher weights assigned to those closer to the observation point. The weights of all observation points are calculated and normalized to ensure that the sum remains consistent.

[0039] As a further improvement of the present invention, the sensitivity in step 3, for each uncertainty parameter θ m The system performs a perturbation and calculates the predicted bearing capacity curve after the perturbation. For each observation depth, it calculates the change in predicted bearing capacity caused by the parameter perturbation. Based on the weight of each observation data point, it defines the sensitivity as the weighted change in bearing capacity of each uncertainty parameter across all observation depths. .

[0040] Select a sensitivity higher than the maximum sensitivity ratio ζ (like ζ= The parameters representing 10% are used as the active parameter set. .

[0041] As a further improvement of the present invention, the negative log-likelihood loss function in step 4, for ease of practical numerical optimization, transforms the problem of maximizing the likelihood probability into the problem of minimizing the loss, assuming that the error follows a normal distribution, and its formula is: in, q obs For the load-bearing capacity of the observation, q pred For the bearing capacity predicted by the model, σ 2 This represents the variance of the error. Considering depth weights, the negative log-likelihood loss function can be expressed as: As a further improvement to the present invention, the Bayesian theory in step 5 is expressed as follows: in, θ This represents the uncertainty parameter to be estimated; X This represents the observed data, namely the depth and the corresponding penetration resistance; P ( θ | X ) is the posterior distribution, that is, given the observed data X Post-parameter θ The probability distribution; P ( X |θ) is the likelihood function, that is, under the assumption that the parameter is θ At that time, the observed dataX The probability of occurrence; P ( θ ) is the prior distribution; P ( X () represents the marginal likelihood.

[0042] The parameter update objective is to determine the maximum likelihood probability. P ( X |θ), i.e., minimum negative log-likelihood loss L ( θ This process is used to find the optimal parameter set, thereby updating the posterior distribution. Considering the boundedness of the parameter range in practical problems, the negative log-likelihood loss function and the prior distribution of the active parameters are used as the objective function and parameter boundary, respectively. An optimization algorithm is employed to search the parameter range and obtain the optimal value Φ of the active parameters that maximizes the likelihood probability and minimizes the loss. opt .

[0043] The inactive parameters are taken from the previous optimal value (initially the design value). The inactive parameters are combined to form the optimal parameter set. The weighted standard deviation of the predicted bearing capacity curve under the optimal parameter set and the normalized residual of the current observation data are calculated. σ pred and judge σ pred Compared with the preset accuracy threshold R The relationship. If σ pred ≤ R If the model prediction matches the observation well, the parameter interval is shrunk centered on the optimal vector of active parameters. σ pred > R If the optimal parameter vector is used as the center, the parameter interval is expanded, and the process reverts to the optimization strategy. The search for the optimal value of the active parameters after expanding the parameter boundary is repeated. If the loop exceeds the limit and the optimal value is still not satisfied... σ pred ≤ R If the condition is met, then the parameter interval after the last cycle is taken as the posterior distribution of the active parameters.

[0044] As a further improvement of the present invention, the Gaussian regression in step 6 uses the error between the predicted bearing capacity value and the observed data under the parameter set (i.e., the optimal parameter set) at the maximum likelihood probability as the target value, the observation depth as the input value, and outputs the correction error and the correction standard deviation, establishing the Gaussian model as follows: in, k ( x , x’ ) is the kernel function.

[0045] The corrected error value is superimposed on the predicted value to obtain the final predicted value. The corrected variance is superimposed on the Monte Carlo simulation variance, and a 95% confidence interval is used as the prediction interval.

[0046] Example 2 Based on the above method, this embodiment also provides a pile shoe foundation penetration risk early warning system based on Bayesian theory and optimization strategies, including: The data acquisition and preprocessing module is used to acquire information on pile shoe shape, soil layer parameters, and monitoring data, generate an initial prior distribution of uncertainty parameters, and assign weights to the data at each observation point.

[0047] The Bayesian update and parameter optimization module is used to obtain the set of active parameters, construct the negative log-likelihood loss function, determine the optimal value of the active parameters under the maximum likelihood probability based on Bayesian theory and optimization search strategy, and update the posterior distribution of the uncertain parameters.

[0048] The uncertainty quantification module is used to sample the posterior distribution of uncertainty parameters using Monte Carlo methods, and then corrects the sampled statistical mean and variance using Gaussian regression to generate confidence intervals.

[0049] The visualization and early warning module is used to display the bearing capacity prediction curve, prediction interval and soil layer parameter changes in real time, and provide early warning of potential puncture risks. Specifically, based on the subsequent predicted bearing capacity change trend, it determines whether there is a puncture risk in the subsequent penetration process, and generates an early warning signal when the risk is identified.

[0050] In practical implementation, the data acquisition and preprocessing module is responsible for collecting real-time monitoring data on pile shoe geometry, marine soil characteristics, and the penetration process. Based on engineering experience, it establishes initial prior probability distributions for key uncertainty parameters and assigns differentiated weights based on the distance between the observation point and the current penetration depth to reflect the temporal correlation of the data. The Bayesian update and parameter optimization module uses the bearing capacity calculation model and weighted data to calculate the sensitivity of each parameter, selects the set of active parameters that significantly affect the prediction, constructs a weighted negative log-likelihood loss function to measure the difference between the model prediction and the measured data, and applies an optimization search strategy within the Bayesian framework to find the optimal values ​​of the active parameters that minimize the loss, thereby updating the posterior probability distribution of the parameters. The uncertainty quantification module performs Monte Carlo random sampling on the updated posterior distribution, simulates the statistical characteristics of the bearing capacity prediction through a large number of parameter samples, and uses Gaussian process regression to correct the error of the prediction mean. Simultaneously, it integrates the model's own variance and the regression variance to generate prediction confidence intervals characterizing the uncertainty. The visualization and early warning module visualizes the corrected bearing capacity prediction curve and prediction range, and identifies whether the subsequent predicted bearing capacity continuously decreases with depth or does not show significant growth over a long depth range. If so, it determines that there is a puncture risk and immediately triggers an early warning prompt, thereby completing a closed-loop early warning process from data collection, model update, uncertainty quantification to risk decision-making.

[0051] Example 3 This embodiment also includes a storage medium for computer-executable instructions, which, when executed by a computer processor, is used to perform the pile foundation penetration risk warning method based on Bayesian theory and optimization strategies described above. The storage medium can be an electronic medium, magnetic medium, optical medium, electromagnetic medium, infrared medium, or a semiconductor system or propagation medium. The storage medium can also include semiconductor or solid-state memory, magnetic tape, removable computer disk, random access memory (RAM), read-only memory (ROM), hard disk, and optical disc. Optical discs can include optical disc-read-only memory (CD-ROM), optical disc-read / write (CD-RW), and DVD.

[0052] Application Example 1 Reference Figure 1 This application example presents a pile shoe foundation penetration risk early warning method based on Bayesian theory and optimization strategies. Application Example 1 is selected from a set of centrifuge model tests. The pile shoe foundation information is as follows: Figure 3 As shown in Table 1, the design values ​​of soil layer parameters are shown in Table 2, and some monitoring data are shown in Table 3. This example, combined with... Figure 1 The method of the present invention includes the following steps: Step 1: Obtain the parameters required to calculate the penetration bearing capacity of the pile shoe foundation, including the shape parameters of the pile shoe foundation and the geological parameters of each soil layer, determine the uncertainty parameters, and construct their initial prior distribution.

[0053] Due to the uncertainty of geological parameters, soil layer parameters that have a significant impact on bearing capacity are extracted as uncertainty parameters. An initial prior distribution is generated based on the design values ​​of the uncertainty parameters and considering the deviation range.

[0054] The deviation range can be adjusted according to the actual engineering needs. In this example, a deviation of Δ=30% is taken. The thickness of the last soil layer is indirectly updated by fixing the total thickness of the soil layers and updating the changes in the thickness of the remaining soil layers. Therefore, its prior distribution is not considered. The posterior distribution of the uncertainty parameters is shown in Table 1.

[0055] Step 2: Obtain the bearing capacity and penetration depth of the real-time observation points during the penetration process as observation data, and assign weights to the data of each observation point.

[0056] The weights are assigned based on the inverse of the distance to the depth of the latest observation point, with higher weights given to those closer to the observation point. The weights of all observation points are calculated and then normalized to ensure consistency.

[0057] In this example, the weights are kept at 100. When the observation depth reaches 4.1 m, the weights of each observation point are [3.56, 4.18, 5.21, 6.54, 13.65, 66.88].

[0058] Step 3: Calculate the sensitivity of the model's predicted values ​​to changes in uncertain parameters, and select parameters whose sensitivity is higher than the maximum sensitivity by a certain proportion as the active parameter set.

[0059] For each uncertainty parameter θ m The perturbation is performed, and the bearing capacity prediction curve after the perturbation is calculated. For each observation depth, the change in predicted bearing capacity caused by the parameter perturbation is calculated.

[0060] Based on the weights of each observation data point, sensitivity is defined as the weighted variation of bearing capacity of each uncertainty parameter across all observation depths. .

[0061] Select a sensitivity higher than the maximum sensitivity ratio ζ (like ζ= The parameters (10%) are used as the active parameter set. .

[0062] In this example, when the observation depth reaches 4.1 m, the active parameter set obtained by sorting according to the sensitivity of uncertainty parameters is Φ={ h 1, Dr,φ , s ut , k}

[0063] in: h 1: Defined as the thickness of the first soil layer, representing the vertical distance of the upper soil layers involved in the pile shoe penetration process.

[0064] Dr: Defined as the relative density of sand, reflecting the compaction state of the soil and affecting its shear strength and bearing capacity.

[0065] φ : Defined as the internal friction angle of soil, it represents the frictional characteristics between soil particles and is an important indicator for calculating shear resistance.

[0066] s ut : Defined as the undrained shear strength of clay, characterizing the shear capacity of soil under rapid loading conditions.

[0067] k : Defined as the soil strength gradient parameter, describing the ratio of shear strength change with depth, and used for model correction.

[0068] Step 4: Construct a negative log-likelihood loss function based on the error between the predicted value calculated by the pile foundation bearing capacity calculation model and the observed data.

[0069] To facilitate practical numerical optimization, the maximum likelihood problem is equivalent to the minimum loss problem. It is assumed that the error follows a normal distribution, and its formula is: in, q obs For the load-bearing capacity of the observation, q pred For the bearing capacity predicted by the model, σ 2 This represents the variance of the error. Considering depth weights, the negative log-likelihood loss function can be expressed as: Step 5: Combining the prior distribution of active parameters and the negative log-likelihood loss function, determine the optimal value of active parameters under the maximum likelihood probability according to Bayesian theory and optimization search strategy, and update the posterior distribution of active parameters, while keeping the prior distribution of inactive parameters unchanged.

[0070] Bayesian theory, for example, is expressed as follows: in, θ This represents the uncertainty parameter to be estimated; X This represents the observed data, namely the depth and the corresponding penetration resistance;P ( θ | X ) is the posterior distribution, that is, given the observed data X Post-parameter θ The probability distribution; P ( X |θ) is the likelihood function, that is, under the assumption that the parameter is θ At that time, the observed data X The probability of occurrence; P ( θ ) is the prior distribution; P ( X () represents the marginal likelihood.

[0071] The parameter update objective is to determine the maximum likelihood probability. P ( X |θ), i.e., minimum negative log-likelihood loss L ( θ This process is used to find the optimal parameter set, thereby updating the posterior distribution. Considering the boundedness of the parameter range in practical problems, the negative log-likelihood loss function and the prior distribution of the active parameters are used as the objective function and parameter boundary, respectively. An optimization algorithm is employed to search the parameter range and obtain the optimal value Φ of the active parameters that maximizes the likelihood probability and minimizes the loss. opt .

[0072] The inactive parameters are taken from the previous optimal value (initially the design value). The inactive parameters are combined to form the optimal parameter set. The weighted standard deviation of the predicted bearing capacity curve under the optimal parameter set and the normalized residual of the current observation data are calculated. σ pred and judge σ pred Compared with the preset accuracy threshold R The relationship.

[0073] like σ pred ≤ R If the model prediction matches the observation well, the parameter interval is shrunk centered on the optimal vector of active parameters. σ pred > R If the optimal parameter vector is used as the center, the parameter interval is expanded, and the process reverts to the optimization strategy. The search for the optimal value of the active parameter after expanding the parameter boundary is repeated. If the optimal value is still not satisfied after the loop count exceeds the limit, the process continues. σ pred If the condition is less than or equal to the threshold, then the parameter interval after the last cycle is taken as the posterior distribution of the active parameters.

[0074] In this example, when the observation depth reaches 4.1 m, the posterior distribution of the uncertainty parameter is updated to { h 1∈[4.46,5.44],γ 1∈[7.89,13.23], Dr∈[81.11,99.13], φ ∈[27.46,33.56], γ 2∈[5.31,9.61], s ut ∈[1.99, 2.45], k =∈[1.99,2.45], St∈[1.58,2.32]}.

[0075] Step 6: For the posterior distribution of the uncertainty parameters, use Monte Carlo sampling to sample N sets of parameters, calculate the bearing capacity results under each set of parameters, and calculate their variance. Use the bearing capacity prediction value under the maximum likelihood probability to represent the mean, and use Gaussian regression to correct the mean and variance, while generating the prediction interval.

[0076] Using the error between the predicted bearing capacity value and the observed data under the parameter set with the maximum a posteriori probability (i.e., the optimal parameter set) as the target value, and the observation depth as the input value, the Gaussian model is established by outputting the correction error and correction variance as the output: in, k ( x , x’ ) is the kernel function.

[0077] The corrected error value is superimposed on the mean to obtain the predicted bearing capacity value for this round, and the corrected variance is superimposed on the Monte Carlo simulation variance. A 95% confidence interval is used as the prediction interval.

[0078] In this example, 100 sets of parameter samples were collected. When the observation depth reached 4.1 m, the final bearing capacity prediction curve and prediction interval are shown in Figure 4(b).

[0079] Step 7: Use the posterior distribution of active parameters as the prior distribution, and keep the prior distribution of inactive parameters unchanged. Repeat steps 2 to 6 until the pile shoe foundation installation is completed and no new observation data is generated.

[0080] During the bearing capacity prediction process based on the monitoring data in this case, some bearing capacity prediction curves show changes as shown in Figures 4(a), 4(b), and 4(c).

[0081] The above description is a further detailed explanation of the present invention in conjunction with specific ordered embodiments, and is not intended to limit the invention. Any modifications and variations made to the present invention within the spirit and scope of the claims fall within the protection scope of the present invention.

[0082] Table 1 Instance Parameter Table Table 2 Monitoring Data Table The above description of the embodiments is provided to enable those skilled in the art to understand and use the invention. It will be apparent to those skilled in the art that various modifications can be made to these embodiments, and the general principles described herein can be applied to other embodiments without inventive effort. Therefore, the present invention is not limited to the above embodiments, and any improvements and modifications made by those skilled in the art based on the disclosure of the present invention without departing from the scope of the invention should be within the protection scope of the present invention.

Claims

1. A method for early warning of penetration risk in pile shoe foundations based on Bayesian theory and optimization strategies, characterized in that, Includes the following steps: S1. Obtain the shape parameters of the pile shoe foundation and the marine soil layer parameters, select the uncertainty parameters and establish their prior distribution; S2. Obtain real-time observation data during the pile shoe penetration process. The observation data includes the penetration depth and the corresponding bearing capacity, and assign weights to each observation data point. S3. Based on the pre-established pile foundation bearing capacity calculation model and the weights assigned to each observation data point, calculate the sensitivity of the model's predicted values ​​to each uncertainty parameter, and select the set of active parameters based on the sensitivity. S4. Based on the error between the predicted value and the observed data of the pile foundation bearing capacity calculation model, construct a negative log-likelihood loss function; S5. Combining the prior distribution of the active parameter set and the negative log-likelihood loss function, apply Bayesian theory and optimization search strategy to determine the optimal value of the active parameters, and update the posterior distribution of the active parameters, while keeping the prior distribution of the inactive parameters unchanged. S6. Perform Monte Carlo sampling on the posterior distribution of the updated uncertainty parameters, use Gaussian regression to correct the predicted mean and variance, generate the predicted bearing capacity value and prediction interval, and determine whether there is a puncture risk in the subsequent penetration process based on the subsequent predicted bearing capacity change trend, and generate an early warning signal when the risk is identified.

2. The method for early warning of pile shoe foundation penetration risk based on Bayesian theory and optimization strategy according to claim 1, characterized in that, The pile shoe foundation penetration risk early warning method based on Bayesian theory and optimization strategy also includes: S7. Use the latest updated posterior distribution of active parameters as the prior distribution for the next iteration, and repeat steps S2 to S6. Use the newly acquired observation data to cyclically update the prediction results until the pile foundation installation process is completed.

3. The method for early warning of pile shoe foundation penetration risk based on Bayesian theory and optimization strategy according to claim 1, characterized in that, In S1, the specific process of obtaining the shape parameters of the pile shoe foundation and the marine soil layer parameters, selecting the category of uncertainty parameters, and establishing their prior distribution includes: Collect the geometric dimensions of the pile shoe foundation as shape parameters, and obtain the soil characteristics of each soil layer as geological parameters; Parameters that have a relatively significant impact on the bearing capacity calculation results are selected from the geological parameters as uncertainty parameters; Design values ​​are set for each uncertainty parameter based on engineering experience, and a parameter interval is established within a preset initial deviation range centered on these design values ​​as its initial prior distribution.

4. The method for early warning of pile shoe foundation penetration risk based on Bayesian theory and optimization strategy according to claim 1, characterized in that, In S2, the real-time observation data during the pile shoe penetration process is: the sequence of observation data formed by the penetration depth and the bearing capacity provided by the foundation soil reaction at the penetration depth during the process of the pile shoe penetrating the seabed. In S2, the specific process of assigning weights to each observation data point includes: Calculate the depth distance between each observation point and the latest observation point, initially assign weights based on the distance, and finally normalize the weights of all observation points.

5. The method for early warning of pile shoe foundation penetration risk based on Bayesian theory and optimization strategy according to claim 1, characterized in that, In S3, based on the pre-established pile shoe foundation bearing capacity calculation model and the weights assigned to each observation data point, the sensitivity of the model's predicted values ​​to various uncertainty parameters is calculated. The specific process of selecting the set of active parameters based on the sensitivity includes: For each uncertainty parameter, perturb it near its current optimal value, and calculate the bearing capacity prediction curve after the perturbation respectively; For each observation depth, calculate the change in predicted bearing capacity caused by parameter disturbance; Based on the weight of each observation data point, calculate the weighted average sensitivity of each uncertainty parameter across all observation depths; Sort all parameters by sensitivity and filter those with a sensitivity higher than the maximum sensitivity preset ratio into the active parameter set; In S3, the pile shoe foundation bearing capacity calculation model is a physical model or empirical model used to calculate the bearing capacity corresponding to different depths during the pile shoe penetration process.

6. The method for early warning of pile shoe foundation penetration risk based on Bayesian theory and optimization strategy according to claim 1, characterized in that, In S4, the specific process of constructing the negative log-likelihood loss function based on the error between the predicted value and the observed data of the pile foundation bearing capacity calculation model includes: Assume that the error between the model's predicted values ​​and the observed data follows a normal distribution with a mean of zero; The problem of maximizing the likelihood probability is transformed into the problem of minimizing the loss function. Combining the weights of the observation data allocated in step S2, a weighted negative log-likelihood loss function is constructed. The negative log-likelihood loss function represents the negative log value of the conditional probability of the observation data occurring under the current parameter values.

7. The method for early warning of pile shoe foundation penetration risk based on Bayesian theory and optimization strategy according to claim 1, characterized in that, In S5, the specific process of determining the optimal values ​​of the active parameters and updating the posterior distribution of the active parameters by combining the prior distribution of the active parameter set and the negative log-likelihood loss function, applying Bayesian theory and an optimization search strategy, includes: Based on Bayesian theory, the posterior distribution is expressed as the product of the prior distribution and the likelihood function; The negative log-likelihood loss function is used as the objective function, and the prior distribution range of the active parameters is used as the search boundary. An optimization algorithm is used to search within the boundary to determine the optimal value of the active parameter that minimizes the loss function. Keep the inactive parameters at the optimal value of the previous round, and together with the optimal value of the active parameters, form the current optimal parameter group; Calculate the weighted residual standard deviation of the predicted value and the observed data under the optimal parameter set, and compare it with a preset threshold. Adjust the distribution interval of the active parameters based on the comparison results: if the residual standard deviation meets the threshold requirement, shrink the distribution interval with the optimal parameter value as the center; otherwise, expand the distribution interval with the optimal parameter value as the center and re-optimize the parameters. The final determined parameter distribution interval is used as the updated posterior distribution of the active parameters.

8. The method for early warning of pile shoe foundation penetration risk based on Bayesian theory and optimization strategy according to claim 1, characterized in that, The specific process of performing Monte Carlo sampling on the posterior distribution of the updated uncertainty parameters, correcting the predicted mean and variance using Gaussian regression, and generating the predicted bearing capacity values ​​and prediction intervals includes: Multiple sets of parameter samples are randomly drawn from the updated posterior distribution of the uncertainty parameters; For each set of parameter samples, the predicted bearing capacity at different penetration depths is obtained through the pile shoe foundation bearing capacity calculation model; The variance of the bearing capacity prediction values ​​of all parameter samples at each depth point is statistically analyzed, and the predicted value corresponding to the parameter group with the maximum a posteriori probability is used to represent the statistical mean. A Gaussian process regression model is established by using the error between the predicted mean and the observed data as the training objective. A radial basis function kernel is used as the kernel function for Gaussian process regression to correct the error of the mean curve in the Monte Carlo simulation. The variance of the Gaussian process regression prediction is superimposed with the variance of the Monte Carlo simulation to obtain a comprehensive uncertainty measure; Based on the corrected mean and comprehensive variance, a bearing capacity prediction interval is generated at a specified confidence level.

9. A pile shoe foundation penetration risk early warning system based on Bayesian theory and optimization strategy, characterized in that, include: The data acquisition and preprocessing module is used to acquire the shape parameters of the pile shoe foundation, the marine soil layer parameters, and the real-time observation data during the pile shoe penetration process. It selects the uncertainty parameters and establishes their prior distribution, and assigns weights to each observation data point. The Bayesian update and parameter optimization module is used to calculate the sensitivity of the model prediction value to each uncertainty parameter based on the pile foundation bearing capacity calculation model and the weight of the observation data to screen the active parameter set. It constructs a negative log-likelihood loss function based on the error between the model prediction value and the observation data, and applies Bayesian theory and optimization strategy to determine the optimal value of the active parameter and update its posterior distribution. The uncertainty quantification module is used to perform Monte Carlo sampling on the posterior distribution of the updated uncertainty parameters, and to use Gaussian regression to correct the predicted mean and variance, thereby generating the predicted bearing capacity value and prediction interval. The visualization and early warning module is used to display the bearing capacity prediction results in real time and provide early warning prompts when a puncture risk is detected. Specifically, it determines whether there is a puncture risk in the subsequent penetration process based on the subsequent predicted bearing capacity change trend, and generates an early warning signal when the risk is identified.

10. A storage medium containing computer-executable instructions, characterized in that, When the storage medium containing the computer-executable instructions is executed by a computer processor, it is used to perform the pile shoe foundation penetration risk warning method based on Bayesian theory and optimization strategy as described in any one of claims 1 to 8.

Images