A Reliability Assessment Method for Offshore Pile-Bearing Structures Based on an Improved Kriging Model

Abstract

This invention discloses a reliability assessment method for nearshore pile-bearing structures based on an improved model. The method includes obtaining the failure risk weight of any sample in the initial high-risk sample set according to an active learning function to obtain a risk weight set; and obtaining a risk-oriented weighted FCM clustering model based on the fuzzy C-means clustering algorithm, according to the initial high-risk sample set and the risk weight set. This invention abandons the traditional pure geometric distance partitioning, and by mapping the limit state learning function to failure risk weights, and through the weighted FCM clustering model, it can effectively overcome the problems of information redundancy and local clustering defects that are unavoidable in the traditional parallel point addition mechanism. It can drive multiple parallel computing nodes to accurately and dispersedly cover the vicinity of the real limit state boundary, thereby avoiding the blind dispersion of a large amount of computing power to low-value safety areas during the data sampling stage.

Description

Technical Field

[0001] This invention relates to the field of structural reliability analysis technology, and more particularly to a method based on improved A reliability assessment method for near-shore pile-bearing structures based on the model. Background Technology

[0002] Offshore pile-supported structural systems (such as monopile offshore wind turbines) face multiple uncertainties during service, including extreme wind and wave loads and complex seabed geological conditions. Reliability analysis methods can effectively quantify the coupling effects of these multi-source random factors and quantitatively assess the failure probability of the structure in complex marine environments, thus providing a scientific basis for the safe design, life-cycle risk assessment, and operation and maintenance of offshore pile-supported structures.

[0003] Given that nearshore pile-supported structures typically involve high-dimensional random variables and highly nonlinear soil-structure dynamic interactions, obtaining the true ultimate limit state response often requires calling extremely time-consuming three-dimensional finite element simulation models. Traditional Monte Carlo simulation (MCS) is computationally prohibitively expensive and difficult to apply directly to assessments with minimal failure probabilities. Therefore, based on... Active learning reliability analysis methods for surrogate models have become mainstream in academia and engineering. This method adaptively finds high-value sample points through an active learning function, significantly reducing the number of calls to the finite element model. However, in practical engineering finite element simulations where a single run often takes several hours, traditional serial active learning strategies still suffer from a significant time bottleneck. To improve efficiency, various parallel point-addition strategies have been developed, such as K-means-based clustering. However, under complex, high-dimensional, and highly nonlinear boundary conditions, existing parallel clustering strategies have the following drawbacks: First, traditional clustering algorithms are easily misled by high-density safe zone samples in the global joint probability density distribution, causing the cluster centers to deviate from the true extremely high-risk failure boundary and resulting in a large number of invalid searches in the safe zone.

[0004] Secondly, when faced with extremely distorted nonlinear limit state surfaces in engineering, the spatial partitioning of traditional hard clustering is prone to causing cluster centers to detach from the boundary; and when the surrogate model is iteratively updated, the slight movement of boundary samples will cause cluster oscillations, resulting in parallel point selection deviating from the target, causing extremely serious waste of computing power and information redundancy in the later stages of iteration. Summary of the Invention

[0005] This invention provides a method based on improved A reliability assessment method for near-shore pile-bearing structures based on the model is proposed to overcome the aforementioned technical problems.

[0006] To achieve the above objectives, the technical solution of the present invention is as follows: A method based on improvement The reliability assessment method for near-shore pile-bearing structures using the model includes the following steps: S1: An initial training sample set is generated by sampling within the design space containing the parameters of the nearshore pile-bearing structure; and the parameters of the nearshore pile-bearing structure include material parameters, soil parameters, and structural dimension parameters; S2: Based on the parameter samples in the initial training sample set, establish the finite element numerical model of the corresponding nearshore pile-bearing structure, and obtain the displacement response value of the corresponding nearshore pile-bearing structure through finite element analysis; construct the limit state function based on the displacement response value and the preset response threshold. S3: Based on the initial training sample set and displacement response values, the preset... The proxy model is trained to obtain the initial... Proxy model; S4: Generate a candidate sample pool based on the design space of nearshore pile-bearing structure parameters, and then... After obtaining the displacement response prediction value from the candidate sample pool, the surrogate model obtains the prediction mean and prediction standard deviation of each candidate sample according to the limit state function. S5: Construct an active learning function based on the predicted mean and predicted standard deviation, and remove safe samples from the candidate sample pool based on the active learning function to obtain a preliminary high-risk sample set. Based on the active learning function, the failure risk weight of any sample in the initial high-risk sample set is obtained to obtain the risk weight set. S6: Based on the fuzzy C-means clustering algorithm, obtain a risk-oriented weighted FCM clustering model according to the initial high-risk sample set and the risk weight set; According to the weighted FCM clustering model, the set of high-risk samples in the initial screening is divided into several sub-clusters, and the sample closest to the cluster center is extracted within each sub-cluster as a parallel new sampling point; S7: Based on the newly added parallel sampling points, call the finite element numerical model of the near-shore pile bearing structure in S2 to solve for the real response, and expand the solution results to the initial training sample set; Based on the expanded initial training sample set and the corresponding displacement response values, repeat steps S3 to S6 to iteratively update the initial training sample set. The proxy model continues until all candidate samples in the candidate sample pool meet the preset stopping condition to obtain the final result. Proxy model; S8: According to the final The surrogate model performs final response prediction for each candidate sample, obtains the limit state function value based on the final response prediction, and takes the candidate sample with the corresponding limit state function value less than 0 as the failure sample. The proportion of the number of failure samples to the total number of candidate samples is defined as the failure probability, thereby realizing the reliability assessment of the near-shore pile-bearing structure.

[0007] Furthermore, the limit state function constructed in S2 is:

[0008] In the formula: This represents the value of the function in the limit state; This indicates the preset response threshold; This represents the displacement response value of a near-shore pile-bearing structure.

[0009] Furthermore, S5 specifically includes the following steps: S51: Construct the active learning function based on the predicted mean and predicted standard deviation as follows:

[0010] In the formula: This represents an active learning function; Indicates candidate samples The corresponding predicted mean; Indicates candidate samples The corresponding standard deviation of the forecast; S52: Eliminate safe samples from the candidate sample pool based on the active learning function to obtain a preliminary high-risk sample set. And the rejection condition is: , Indicates a preset threshold; S53: Based on the active learning function, obtain the failure risk weight of any sample in the initial high-risk sample set to obtain the risk weight set; The formula for obtaining the failure risk weight is:

[0011] In the formula: This indicates the samples in the initial screening high-risk sample set. Failure risk weight; ξ represents the design constant; Indicates sample The active learning function value.

[0012] Furthermore, step S6 specifically includes the following steps: S61: Based on the fuzzy C-means clustering algorithm, obtain a risk-oriented weighted FCM clustering model according to the initial high-risk sample set and the risk weight set; Furthermore, the weighted objective function of the risk-oriented weighted FCM clustering model... for:

[0013] In the formula: This indicates the number of candidate points in the initial high-risk sample set; K This indicates the current parallel batch size, which is the number of sub-clusters divided by the weighted FCM clustering model; Indicates the first j Cluster center vectors; Indicates the first i Each sample belongs to a cluster. j The membership degree, and satisfying ; m Indicates the fuzzy index; Solving the weighted objective function using the Lagrange multiplier method Obtain cluster centers membership degree The iterative update formula is:

[0014]

[0015] Based on cluster centers membership degree The iterative update formula is used to update and obtain the latest subclusters; S62: Based on the weighted FCM clustering model described in S61, the initial high-risk sample set is divided into several sub-clusters and updated. The sample closest to its cluster center is extracted from each updated sub-cluster and used as a parallel new sampling point.

[0016] Furthermore, in S1, the method for generating an initial training sample set by sampling within the design space containing the parameters of the near-shore pile-bearing structure adopts the Latin hypercube sampling method.

[0017] Furthermore, in S4, the method for generating a candidate sample pool based on the design space of nearshore pile-bearing structure parameters adopts the Monte Carlo sampling method.

[0018] Beneficial effects: This invention provides a method based on improved The present invention proposes a reliability assessment method for nearshore pile-bearing structures. This method abandons the traditional pure geometric distance partitioning and instead maps the limit state learning function to failure risk weights. These weights are then deeply integrated into the fuzzy C-means clustering algorithm to obtain a risk-oriented weighted FCM clustering model. The weighted FCM clustering model effectively overcomes the unavoidable information redundancy and local clustering defects in the traditional parallel point addition mechanism. It can drive multiple parallel computing nodes to accurately and dispersedly cover the vicinity of the real limit state boundary, thereby avoiding the blind dispersion of a large amount of computing power to low-value safety areas during the data sampling stage. This effectively achieves high accuracy and efficiency in the reliability assessment of nearshore pile-bearing structures. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This invention is based on improvements A flowchart of the reliability assessment method for near-shore pile-bearing structures in the model; Figure 2 This is a schematic diagram of the numerical model of the monopile offshore wind turbine structure in this embodiment. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] This embodiment provides a method based on improvement. The reliability assessment method for near-shore pile-bearing structures in the model, such as Figure 1 As shown, the specific steps include: S1: The Latin hypercube sampling method is used to generate an initial training sample set by sampling within the design space containing the parameters of the nearshore pile-bearing structure; and the parameters of the nearshore pile-bearing structure include material parameters, soil parameters, and structural dimension parameters. Specifically, the design space for nearshore pile-bearing structure parameters (including different material parameters, soil parameters, structural dimensional parameters, etc.) The initial training sample set is generated using Latin hypercube sampling. Sample size Typically, the design variable dimensions are taken as 5-10 times. S2: Based on the parameter samples in the initial training sample set, establish a finite element numerical model of the corresponding nearshore pile-bearing structure, and obtain the displacement response values ​​of the corresponding nearshore pile-bearing structure through finite element analysis. The limit state function is constructed based on the displacement response value and a preset response threshold as follows: In the formula: This represents the value of the function in the limit state; This indicates the preset response threshold; The displacement response value of the nearshore pile-bearing structure is represented; wherein, the method for establishing the finite element numerical model of the corresponding nearshore pile-bearing structure in this embodiment is a well-known existing technical means, and will not be described in detail here; S3: Based on the initial training sample set and displacement response values, i.e. For preset Proxy model ( The surrogate model is trained using a well-known and mature mathematical method to obtain an initial... Proxy Model In this embodiment, the preset... The process of training the proxy model is a well-known existing technique and will not be elaborated on here. S4: Using the Monte Carlo sampling method, a candidate sample pool is generated based on the design space of nearshore pile-bearing structure parameters, and then... (The sentence is incomplete and requires further context to be fully translated.) After obtaining the displacement response prediction value from the candidate sample pool, the surrogate model obtains the prediction mean and prediction standard deviation of each candidate sample according to the limit state function. Specifically, a candidate sample pool is generated spatially based on the parameters of the nearshore pile-bearing structure using the traditional Monte Carlo sampling method (MCS). , usually take To ensure probability coverage; utilize the currently trained agent model. For each sample in the candidate sample pool The displacement response value is quickly predicted and obtained for each sample. value. The model can not only provide Predicted mean (i.e., the prediction results) can also provide Predictive standard deviation (That is, the degree of uncertainty of the prediction result); S5: Construct an active learning function based on the predicted mean and predicted standard deviation, and remove safe samples from the candidate sample pool according to the active learning function to obtain a preliminary high-risk sample set; based on the active learning function, obtain the failure risk weight of any sample in the preliminary high-risk sample set to obtain a risk weight set, specifically including the following steps: S51: To identify the sample most likely to be disrupted and with the greatest prediction uncertainty, an active learning function is constructed based on the prediction mean and prediction standard deviation:

[0023] In the formula: This represents an active learning function; Indicates candidate samples The corresponding predicted mean; Indicates candidate samples The corresponding standard deviation of the forecast; where the absolute value of the forecast mean is... It is positively correlated with symbol accuracy. Then it shows a negative correlation trend; The numerical characteristics indicate that when As the value approaches zero, the sample space location approaches the limit state interface. ; and when When a large value is observed, it indicates that the surrogate model has high predictive uncertainty in that region; S52: Eliminate safe samples from the candidate sample pool based on the active learning function to obtain a preliminary high-risk sample set. And the rejection condition is: , This represents a preset threshold; in this embodiment, the entire field is obtained. Then, to ensure computational efficiency, we first eliminated... Safe samples were used to construct an initial screening set. ; S53: Based on the active learning function, obtain the failure risk weight of any sample in the initial high-risk sample set to obtain a risk weight set; the formula for obtaining the failure risk weight is:

[0024] In the formula: This indicates the samples in the initial screening high-risk sample set. The failure risk weight; ξ represents the design constant, a small constant that ensures the denominator is non-zero, taken as 10. -6 ; Indicates sample The active learning function value, in denominator form, ensures that the sample is closer to the limit state surface.

[0025] S6: Based on the fuzzy C-means clustering algorithm, a risk-oriented weighted FCM clustering model is obtained according to the initial high-risk sample set and the risk weight set; according to the weighted FCM clustering model, the initial high-risk sample set is divided into several sub-clusters, and the sample closest to the cluster center within each sub-cluster is extracted as a parallel new sampling point, specifically including the following steps: S61: Based on the fuzzy C-means clustering algorithm, obtain a risk-oriented weighted FCM clustering model according to the initial high-risk sample set and the risk weight set; Furthermore, the weighted objective function of the risk-oriented weighted FCM clustering model... for:

[0026] In the formula: This indicates the number of candidate points in the initial high-risk sample set; K This indicates the current parallel batch size, which is the number of sub-clusters divided by the weighted FCM clustering model. It is usually set to 6. Indicates the first j Cluster center vectors; Indicates the first i Each sample belongs to a cluster. j The membership degree, and satisfying ; m The fuzzy index is typically taken as... ; The algorithm aims to minimize the following weighted objective function, which is solved using the Lagrange multiplier method. Obtain cluster centers membership degree The iterative update formula is:

[0027]

[0028] Based on cluster centers membership degree The iterative update formula is used to update and obtain the latest subclusters; S62: Based on the weighted FCM clustering model described in S61, the set of high-risk samples in the initial screening is divided into several sub-clusters and updated. The sample closest to its cluster center is extracted from each updated sub-cluster as a parallel new sampling point. S7: Based on the newly added parallel sampling points, call the finite element numerical model of the near-shore pile bearing structure in S2 to solve for the real response, and expand the solution results to the initial training sample set; Based on the expanded initial training sample set and the corresponding displacement response values, repeat steps S3 to S6 to iteratively update the initial training sample set. The proxy model continues until all candidate samples in the candidate sample pool meet the preset stopping condition to obtain the final result. Proxy model; In this embodiment, after obtaining the cluster centers, the samples closest to the cluster centers are extracted from each sub-cluster and added to the initial training sample set generated using Latin hypercube sampling. The updated sample set is then used to solve the numerical model of the actual near-shore pile-bearing structure. The model repeats steps S3 to S6 until the set convergence condition is met. All samples in the middle satisfy ; S8: According to the final The surrogate model performs final response prediction for each candidate sample, obtains the limit state function value based on the final response prediction, and takes the candidate sample with the corresponding limit state function value less than 0 as the failure sample. The proportion of the number of failure samples to the total number of candidate samples is defined as the failure probability, thereby realizing the reliability assessment of the near-shore pile-bearing structure.

[0029] This embodiment also includes the following application examples: (1) Numerical Model Establishment: The example will conduct a reliability assessment of a monopile offshore wind turbine (MOWT) structure under complex wind and wave combined loads. The specific parameter distribution is detailed in Tables 1 and 2. Based on Numerical models are established using finite element calculation platforms, such as Figure 2 As shown, to balance solution accuracy and computational efficiency under large-scale sampling, the nonlinear method recommended by the American Petroleum Institute (API) is adopted. py , tz and Qz The foundation springs simulate complex pile-soil interactions (SSI), and a schematic diagram of its numerical model is shown below. Figure 1 As shown. Furthermore, according to relevant structural design specifications, for the Serviceability Limit State (SLS), the permissible horizontal displacement threshold at the top of the wind turbine tower is set to 1% of the tower hub height. Therefore, the limit state function of this system can be defined as follows: In the formula: This indicates the permissible horizontal displacement threshold; This indicates the horizontal displacement of the top of the wind turbine tower; Table 1. Main parameters of soil layers around and at the pile tip

[0030] Table 2. Main structural parameters

[0031] To comprehensively evaluate the reliability of monopile offshore wind turbines under real-world service conditions, this example constructs a high-dimensional random variable space, which encompasses geotechnical parameters (internal friction angles of the upper and lower sand layers). φ 1, φ 2 and severe γ 1, γ 2) Structural material properties (elastic modulus of steel) E Single pile wall thickness ), and marine environmental loads (average wind speed) With significant wave height The data exhibits multi-source uncertainty characteristics. The marine environmental load parameters are based on annual extreme wind and wave data from the Weizhou Island area of ​​the South my country Sea. The statistical distribution characteristics of each random variable are shown in Table 3.

[0032] Table 3. Characteristics of Random Parameter Distribution

[0033] (2) Load modeling: The overall aerodynamic thrust acting on the rotor hub ( F B ) and the local drag force acting on a unit height of the tower ( F T The solution formula is:

[0034] In the formula: To represent air density, take 1.225 kg / m³. 3 ; R T Indicates the rotor radius; Indicates the dimensionless thrust coefficient; This represents the drag coefficient, taken as 1.2; Indicates height z The outer diameter of the tower body; , These represent average wind speed and fluctuating wind speed, respectively; where the average wind speed is in the vertical direction. Following an exponential distribution, the solution formula is:

[0035] In the formula: Indicates standard height Average wind speed at a location typically 10m. Indicates the calculated height; α This represents the surface roughness coefficient.

[0036] For pulsating wind speed Time-history simulation was performed using spectral representation combined with the Davenport spectrum, which is widely used in engineering. The specific calculation expression for the power spectral density is as follows:

[0037]

[0038] In the formula: Represents the fluctuating wind speed spectrum; Indicates frequency; K This represents the correlation coefficient of ground roughness; For monopile offshore wind turbine foundations, to improve computational efficiency, the wave diffraction effect on the pile is ignored, and the Morison equation is used to simplify the calculation of the wave force it bears. The specific calculation formula is as follows:

[0039] In the formula: Indicates the drag coefficient; Indicates the quality coefficient; Indicates the density of seawater; D Indicates the pile diameter; , Let represent the horizontal acceleration and velocity components of a water particle, respectively; where the horizontal acceleration and velocity components of the water particle can be solved using the formulas of linear wave theory:

[0040]

[0041] In the formula: , H , g These represent wave number, wave height, and gravitational acceleration, respectively. Indicates the calculated height; h Indicates water depth; , t Indicates position and time; ω represents the wave's circular frequency; Finally, based on the random wave spectrum, the irregular wave field is discretized into a linear superposition of multiple small-amplitude regular waves using spectral representation combined with linear wave theory. These are then substituted into the Morison equation to obtain the final random wave load time history. Considering the strong systematicity and high fitting accuracy of the JONSWAP spectrum, this spectrum is selected for load simulation in this embodiment, and its expression is:

[0042]

[0043] In the formula: This represents the peak factor, which is set to 3.3. Indicates the frequency of the spectral peak; Indicates the period of the spectral peak.

[0044] (3) Results analysis: First, an initial training sample set of 48 samples with a random vector dimension of 6 times was selected for the numerical model described above. Using the initial samples as basic input parameters, finite element numerical models of offshore wind turbines were established, and the time history of the turbine tower top displacement was calculated as the output response. The final model was constructed by relating the input and output. Proxy model. Based on this, 100,000 candidate samples are generated using MCS. The average statistical results of the method described in this embodiment, which was run independently 10 times with the same 48 initial sample points, are detailed in Table 4.

[0045] Table 4. Reliability Analysis Results

[0046] Compared with the prior art, the beneficial effects of the method described in this embodiment are as follows: This invention addresses the efficient solution of reliability in complex nonlinear engineering structures, based on active learning. Based on surrogate model theory, this paper proposes a parallel active learning method using fuzzy clustering with risk weights. The core advantage of this method lies in its departure from traditional pure geometric distance partitioning. It innovatively maps the limit state learning function to failure risk weights and deeply integrates them into the fuzzy C-means (FCM) clustering algorithm. This strategy effectively overcomes the unavoidable information redundancy and local clustering defects inherent in traditional parallel point-addition mechanisms. It enables multiple parallel computing nodes to accurately and dispersedly cover the vicinity of the true limit state boundary, thus avoiding the blind dispersion of a large amount of computing power to low-value safety areas during the data sampling phase. A flexible fuzzy membership mechanism is used to fit the nonlinear failure boundary, achieving adaptive parallel point-addition reliability analysis.

[0047] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method based on improvement The reliability assessment method for near-shore pile-bearing structures based on the model is characterized by, Specifically, the following steps are included: S1: An initial training sample set is generated by sampling within the design space containing the parameters of the nearshore pile-bearing structure; and the parameters of the nearshore pile-bearing structure include material parameters, soil parameters, and structural dimension parameters; S2: Based on the parameter samples in the initial training sample set, establish the finite element numerical model of the corresponding nearshore pile-bearing structure, and obtain the displacement response value of the corresponding nearshore pile-bearing structure through finite element analysis; construct the limit state function based on the displacement response value and the preset response threshold. S3: Based on the initial training sample set and displacement response values, the preset... The proxy model is trained to obtain the initial... Proxy model; S4: Generate a candidate sample pool based on the design space of nearshore pile-bearing structure parameters, and then... After obtaining the displacement response prediction value from the candidate sample pool, the surrogate model obtains the prediction mean and prediction standard deviation of each candidate sample according to the limit state function. S5: Construct an active learning function based on the predicted mean and predicted standard deviation, and remove safe samples from the candidate sample pool based on the active learning function to obtain a preliminary high-risk sample set. Based on the active learning function, the failure risk weight of any sample in the initial high-risk sample set is obtained to obtain the risk weight set. S6: Based on the fuzzy C-means clustering algorithm, obtain a risk-oriented weighted FCM clustering model according to the initial high-risk sample set and the risk weight set; According to the weighted FCM clustering model, the set of high-risk samples in the initial screening is divided into several sub-clusters, and the sample closest to the cluster center is extracted within each sub-cluster as a parallel new sampling point; S7: Based on the newly added parallel sampling points, call the finite element numerical model of the near-shore pile bearing structure in S2 to solve for the real response, and expand the solution results to the initial training sample set; Based on the expanded initial training sample set and the corresponding displacement response values, repeat steps S3 to S6 to iteratively update the initial training sample set. The proxy model continues until all candidate samples in the candidate sample pool meet the preset stopping condition to obtain the final result. Proxy model; S8: According to the final The surrogate model performs final response prediction for each candidate sample, obtains the limit state function value based on the final response prediction, and takes the candidate sample with the corresponding limit state function value less than 0 as the failure sample. The proportion of the number of failure samples to the total number of candidate samples is defined as the failure probability, thereby realizing the reliability assessment of the near-shore pile-bearing structure.

2. A method based on improvement according to claim 1 The reliability assessment method for near-shore pile-bearing structures based on the model is characterized by, The limit state function constructed in S2 is: In the formula: This represents the value of the function in the limit state. This indicates the preset response threshold; This represents the displacement response value of a near-shore pile-bearing structure.

3. A method based on improvement according to claim 2 The reliability assessment method for near-shore pile-bearing structures based on the model is characterized by, S5 specifically includes the following steps: S51: Construct the active learning function based on the predicted mean and predicted standard deviation as follows: In the formula: This represents an active learning function; Indicates candidate samples The corresponding predicted mean; Indicates candidate samples The corresponding standard deviation of the forecast; S52: Eliminate safe samples from the candidate sample pool based on the active learning function to obtain a preliminary high-risk sample set. And the rejection condition is: , Indicates a preset threshold; S53: Based on the active learning function, obtain the failure risk weight of any sample in the initial high-risk sample set to obtain the risk weight set; The formula for obtaining the failure risk weight is: In the formula: This indicates the samples in the initial screening high-risk sample set. Failure risk weight; ξ represents the design constant; Indicates sample The active learning function value.

4. A method based on improvement according to claim 3 The reliability assessment method for near-shore pile-bearing structures based on the model is characterized by, S6 specifically includes the following steps: S61: Based on the fuzzy C-means clustering algorithm, obtain a risk-oriented weighted FCM clustering model according to the initial high-risk sample set and the risk weight set; Furthermore, the weighted objective function of the risk-oriented weighted FCM clustering model... for: In the formula: This indicates the number of candidate points in the initial high-risk sample set; K This indicates the current parallel batch size, which is the number of sub-clusters divided by the weighted FCM clustering model; Indicates the first j Cluster center vectors; Indicates the first i Each sample belongs to a cluster. j The membership degree, and satisfying ; m Indicates the fuzzy index; Solving the weighted objective function using the Lagrange multiplier method Obtain cluster centers membership degree The iterative update formula is: Based on cluster centers membership degree The iterative update formula is used to update and obtain the latest subclusters; S62: Based on the weighted FCM clustering model described in S61, the initial high-risk sample set is divided into several sub-clusters and updated. The sample closest to its cluster center is extracted from each updated sub-cluster and used as a parallel new sampling point.

5. A method based on improvement according to claim 1 The reliability assessment method for near-shore pile-bearing structures based on the model is characterized by, The method in S1 for generating an initial training sample set by sampling within the design space containing parameters of near-shore pile-bearing structures employs the Latin hypercube sampling method.

6. A method based on improvement according to claim 1 The reliability assessment method for near-shore pile-bearing structures based on the model is characterized by, The method for generating a candidate sample pool in S4 based on the design space of nearshore pile-bearing structure parameters adopts the Monte Carlo sampling method.

Images