Primary reliability calculation method based on Sigmoid function adaptive step length adjustment

By introducing Sigmoid function and chaotic control factor in the first second-order moment method for adaptive step adjustment, the problems of low computational efficiency of high nonlinear extreme state functions and non-convergence of oscillation are solved, and higher iteration efficiency and robustness are achieved , suitable for the reliability calculation of complex engineering structures.

CN118964809APending Publication Date: 2024-11-15GUANGXI UNIVERSITY OF TECHNOLOGY
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411023027.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-29
Publication Date
2024-11-15

AI Technical Summary

Technical Problem

The existing primary second-order moment method is not effectively solved when dealing with complex engineering structures, especially when high nonlinear limit state functions, the problems of low computational efficiency and non-convergence of oscillations are not effectively solved, resulting in difficulty in balancing robustness and computational efficiency.

Method used

Adaptive step adjustment method based on Sigmoid function is adopted, and by introducing Sigmoid correction function and chaos control factor, the step length is adjusted according to the relationship between iterative angle and ideal step length, combined with standard normal conversion and monitoring coefficients of iterations, to achieve adaptive Adjustment of step size improves iteration efficiency and robustness.

Benefits of technology

It significantly improves the computing efficiency and convergence rate, ensures stable convergence results in complex engineering structures, and improves the robustness and wide application value of the algorithm.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118964809A_ABST
    Figure CN118964809A_ABST
Patent Text Reader

Abstract

The invention provides a one-time reliability calculation method based on Sigmoid function adaptive step length adjustment. The method comprises the following steps: determining an initial random variable and an initial parameter; carrying out standard normal transformation on the random variables and then completing first iteration; normal conversion of the random variables is completed through standard normal conversion; a Sigmoid function is introduced, and a value used for converting a coefficient of the Sigmoid function is determined; by calculating parameters such as f (uk), nk, pk and the like, calculating the size of a rotation angle theta k; and solving a chaos control factor lambda k in the iterative formula, and adaptively adjusting the step length of the finite step length method by combining the chaos control factor lambda k with a Sigmoid function. According to the method, a new reliable index solving method is constructed by adopting a Sigmoid function conforming to a mapping relation between an ideal iteration step length and an iteration rotation angle, so that the problems of low efficiency and non-convergence oscillation of a first-order second-order moment algorithm when a high-nonlinearity complex limit state function is solved are effectively solved; the robustness is improved, meanwhile, the calculation efficiency is improved, and the wide application and popularization value is achieved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of primary reliability analysis, and in particular to a primary reliability calculation method based on Sigmoid function adaptive step length adjustment. Background Art

[0002] As early as 1974, the concept of structural reliability was proposed by Freudenthal in the "Structural Safety" conference, but the computational tools and mathematical theories at that time were not yet mature, which was not conducive to the application and development of this concept. Later, Cornell established the connection and combination between the theory of first-order second moment and reliability analysis for the first time. Based on this concept, the first-order second moment method was born. In order to expand the application of the first-order second moment method, Hasofer and Lind proposed a new definition of reliability index, stipulating its geometric meaning as the shortest distance from the origin of the coordinates to the limit state surface in the standard normal space. This method makes a new definition of the reliability index from the level of the limit state surface, which greatly simplifies the calculation of structural reliability.

[0003] The above method has the advantages of clear concepts, simple iteration format and excellent convergence speed, especially when dealing with linear or low nonlinear function problems, it can give full play to its advantages of high efficiency and simplicity. However, when dealing with more complex engineering structures and the nonlinearity of the limit state function is high, there will be phenomena such as the iteration direction not intersecting with the limit state surface and the calculation process oscillating. This is unacceptable for most engineering problems with highly nonlinear limit state functions.

[0004] After years of development, many improved algorithms for the first-order second moment have optimized the computational performance of the classic first-order second moment method to a certain extent. However, they still cannot fully provide adaptive adjustment of the step size based on the specific properties of the limit state function and iterative information, nor have they formed a relatively complete and universal calculation method that takes into account both computational efficiency and accuracy. This method has certain limitations. Summary of the invention

[0005] In view of the above, it is necessary to provide a primary reliability calculation method based on the adaptive step size adjustment of the Sigmoid function to solve the common problems of low computational efficiency and non-convergence of oscillation when the primary reliability calculation method deals with complex and highly nonlinear limit state function problems, thereby achieving a good balance between the robustness and computational efficiency in the primary reliability algorithm.

[0006] In order to achieve the above object, the technical solution adopted by the present invention is:

[0007] A primary reliability calculation method based on Sigmoid function adaptive step size adjustment includes the following steps:

[0008] Step 1: Determine the random variable X of the limit state function = (x1, x2, x3, ..., x n ) and the statistical characteristics of random variables, and set the coefficient k for monitoring the number of iterations and the lower bound λ of the algorithm l , where statistical characteristics include but are not limited to mean and standard deviation;

[0009] Step 2: Use standard normal transformation to normalize the random variables, then perform the first reliability analysis iteration and collect iteration information, including but not limited to iteration failure point coordinates and reliability indicators;

[0010] Step 3: Use standard normal transformation to transform the known random variable x k Transform the random variable u into the standard normal space k ;

[0011] Step 4: Introduce the Sigmoid function, and determine the step adjustment method of the Sigmoid function according to the relationship between the turning angle and the ideal iteration step length; transform the Sigmoid function and introduce coefficients A, a, B, b, and c to correct it, and obtain the Sigmoid correction function. By determining the values ​​of coefficients A, a, B, b, and c, the Sigmoid correction function conforms to the relationship between the turning angle and the ideal iteration step length;

[0012] Step 5: Calculate the function value f(u k ), the direction vector n of the previous iteration direction k , the direction vector p to be iteratively calculated this time k Parameters, thereby calculating the iterative rotation angle θ corresponding to the kth iteration k size;

[0013] Step 6: Update the chaos control factor λ k , the chaos control factor λ k Combined with the Sigmoid correction function, the step size of the finite step size method is adaptively adjusted; and the variable u that obeys the normal distribution is k+1 Convert to random variable x k+1 ;

[0014] Step 7: Set the allowable error ε′ and calculate the relative error ε=|∥x between the two iteration points. k+1 ∥-∥x k ∥| / ∥x k ∥, if the convergence condition ε<ε′ is met, stop the iteration and output the result, otherwise repeat steps 3 to 6 until the relative error between the two iteration points before and after the calculation meets the convergence condition.

[0015] Furthermore, in step 1, the coefficient k of monitoring the number of iterations is 0, and the lower bound of the algorithm is l =0.1.

[0016] Furthermore, in step 4, the relationship between the rotation angle and the ideal iteration step length is an inversely proportional relationship.

[0017] Furthermore, in step 4, the Sigmoid correction function is:

[0018]

[0019] In the formula, t is

[0020] Furthermore, in step 5, the iterative rotation angle θ k The size is calculated by the following formula:

[0021] θ k =arccos(ζ k )

[0022] in,

[0023] In the formula, ζ k It represents the cosine value of the rotation angle between the two iterations.

[0024] Furthermore, in step 6, the chaos control factor λ k Combined with the Sigmoid correction function, the following adjustment function is obtained:

[0025]

[0026] In the formula, λ k-1 is the chaos control factor of the previous iteration.

[0027] The primary reliability calculation method based on the adaptive step size adjustment of the Sigmoid function proposed in the present invention applies the Sigmoid function to the existing primary reliability algorithm, which can well solve the common problems of low computational efficiency and non-convergence of oscillation when the primary reliability calculation method handles complex and highly nonlinear limit state function problems, thereby balancing the robustness and computational efficiency of the primary reliability algorithm. Specifically,

[0028] 1. The present invention, through in-depth research on the drawbacks of the existing primary reliability algorithm, proposes an intrinsic nonlinear mapping relationship between the iteration angle and the ideal step length, and proposes the principles that should be followed when adjusting the step length. In this way, the mapping relationship is used as the basis for adjusting the algorithm step length, making the adjustment of the step length more scientific and effective.

[0029] 2. The present invention introduces a Sigmoid function model and performs necessary translation and contraction transformation on it to better fit the relationship between the iterative turning angle and the ideal step length, and establishes a specific adaptive adjustment method between the ideal step length and the iterative turning angle.

[0030] 3. The present invention inserts key parameters A, a, B, b, and c into the Sigmoid function model, so that the modified Sigmoid function is perfectly combined with the primary reliability calculation method, and the function model's adjustment of the adaptive step size is closer to the ideal effect.

[0031] 4. The present invention applies the modified parameterized Sigmoid function model to the step size adjustment of a reliability iteration calculation, combines the Sigmoid function with a reliability algorithm, and generates an adaptive ideal step size suitable for the current iteration, replacing the original fixed step size. This makes the algorithm proposed in the present invention have higher iteration efficiency and robustness, and greatly improves its convergence rate while ensuring a stable convergence result, so that the algorithm can be more widely and stably applied to relatively complex projects, and has broad promotion and application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 is a flow chart of the method given by the present invention;

[0033] Figure 2 is the curve of the Sigmoid correction function model of the present invention when c changes;

[0034] Figure 3 It is an iterative path diagram of the iterative points of the method given by the present invention and other methods in Example 1; in the figure, (a) represents the classic first-order second moment method, (b) represents the chaos control algorithm, and (c) represents the method of the present invention;

[0035] Figure 4 It is an iterative history diagram of the reliability index β of the method given by the present invention and other methods in Example 1;

[0036] Figure 5 A schematic diagram of a curved slender column provided in the second embodiment of the present invention;

[0037] Figure 6 It is an iterative process diagram of the reliability index β of the method given by the present invention and other methods in Example 2;

[0038] The following specific implementation manner will further illustrate the present invention in conjunction with the above-mentioned drawings. DETAILED DESCRIPTION

[0039] See also Figure 1In a preferred embodiment of the present invention, a primary reliability calculation method based on Sigmoid function adaptive step size adjustment comprises the following steps:

[0040] Step 1: Determine the random variable X of the limit state function = (x1, x2, x3, ..., x n ) and the statistical characteristics of random variables, and set the coefficient k for monitoring the number of iterations and the lower bound λ of the algorithm l , where the statistical features include but are not limited to mean and standard deviation. That is, determine the initial random variables and initial parameters. In the present invention, the coefficient k for monitoring the number of iterations is set to 0, and the lower bound λ of the algorithm is l =0.1.

[0041] Step 2: Use standard normal transformation to normalize random variables, then perform the first reliability analysis iteration and collect iteration information. The collected iteration information includes but is not limited to iteration failure point coordinates and reliability indicators.

[0042] Step 3: Use standard normal transformation to transform the known random variable x k Transform the random variable u into the standard normal space k .

[0043] Step 4: Introduce the Sigmoid function, and determine the step size adjustment method of the Sigmoid function according to the relationship between the turning angle and the ideal iterative step size; transform the Sigmoid function and introduce coefficients A, a, B, b, and c to correct it, and obtain the Sigmoid correction function. By determining the values ​​of coefficients A, a, B, b, and c, the Sigmoid correction function conforms to the relationship between the turning angle and the ideal iterative step size.

[0044] Among them, the Sigmoid function is: The relationship between the rotation angle and the ideal iteration step size is inversely proportional, that is, when the iteration angle θ k When the iteration angle θ is small, a larger relaxation step size should be considered to speed up convergence. k When the angle is large, a smaller step size should be considered to stabilize the iteration and avoid oscillation. In order to transform the Sigmoid function to conform to the relationship between the above turning angle and the ideal iteration step size, the Sigmoid function is transformed and corrected. Specifically, the prototype of the Sigmoid function is first transformed into the following function: In order to achieve optimal control of the iteration step size through the sigmoid function, the above converted function is corrected, and coefficients A, a, B, b, and c are introduced to obtain the following Sigmoid correction function: To determine the coefficients in the correction function, it is necessary to compare the characteristics of the function when the parameters take different values. That is, when determining the values ​​of coefficients A, a, B, b, and c, it is necessary to make the function conform to the relationship between the rotation angle and the ideal iteration step. For example, the coefficients A, a, B, b, and c of the Sigmoid correction function are A=1.5, a=1, B=2, b=0.5, and c=3 respectively. Taking parameter c as an example, Figure 2 The function graph of the Sigmoid correction function under different values ​​of parameter c. It can be seen from the figure that when c is 3, the function best conforms to the above relationship, and in the interval [0,1], when the function variable t is small, the function value can be sufficiently amplified, and when t is large enough, the function value is small. When applied to step size adjustment, the step size can be fully and reasonably adjusted to make the algorithm achieve stable convergence. Therefore, the coefficient c is 3. The determination of other coefficients will not be elaborated in detail here.

[0045] Step 5: Calculate the function value f(u k ), the direction vector n of the previous iteration direction k , the direction vector p to be iteratively calculated this time k Parameters, thereby calculating the iterative rotation angle θ corresponding to the kth iteration k size.

[0046] In this step, f(u k ) is the function value of the limit state function of the reliability index to be determined, which can be directly obtained. That is to say, for each type of reliability calculation problem, there is a limit state function of the reliability index to be determined. This function belongs to known conditions, so it can be directly obtained by substituting known parameters. The iterative rotation angle θ k The size is calculated by the following formula:

[0047] θ k =arccos(ζ k )

[0048] in,

[0049] In the formula, ζ k It represents the cosine value of the rotation angle between the two iterations.

[0050] Step 6: Update the chaos control factor λ k , the chaos control factor λ k Combined with the Sigmoid correction function, the step size of the finite step size method is adaptively adjusted; and the variable u that obeys the normal distribution is k+1 Convert to random variable x k+1 Among them, the chaos control factor λ kCombined with the Sigmoid correction function, the following adjustment function is obtained:

[0051]

[0052] In the formula, λ k-1 is the chaos control factor of the previous iteration.

[0053] Step 7: Set the allowable error ε′ and calculate the relative error ε=|∥x between the two iteration points. k+1 ∥-∥x k ∥| / ∥x k ∥, if the convergence condition ε<ε′ is met, then stop the iteration and output the result, otherwise repeat steps 3 to 6 until the relative error between the two iteration points before and after the calculation meets the convergence condition, that is, the relative error meets the accuracy requirement. It should be noted that the setting of the allowable error ε′ can be specifically set according to needs, for example, ε′=10 -4 .

[0054] In order to further illustrate and explain the primary reliability calculation method provided by the present invention and reflect the advantages of the method provided by the present invention, the present invention provides the following two embodiments for further explanation, which are as follows.

[0055] Example 1: Two-dimensional high nonlinear example

[0056] The limit state function of a structure is: g = x1 3 +x2 3 -4, in the equation, x1 and x2 are both normally distributed and independent of each other, with means μ x1 =3.0 and μ x2 =2.9, standard deviation is σ x1 =σ x2 =1.0, solve for the reliability index β of the limit state function.

[0057] Step 1: Set the random variables x1 and x2, as well as the distribution type and mean μ of the random variables x1 =3.0, μ x2 =2.9 and standard deviation σ x1 =σ x2 =1.0, determines the parameter k=0 for monitoring the number of iterations, and the lower bound of the algorithm is λ l =0.1.

[0058] Step 2: Use standard normal transformation to normalize the random variable, then complete the first iteration and obtain the failure point coordinate x of the first iteration. (1) and the reliability index β.

[0059] Step 3: Use standard normal transformation to normalize the random variables. In this example, the random variables are x1 and x2, the distribution type and parameters are given, they satisfy the normal distribution and are independent of each other, so normalization and variable transformation are not required.

[0060] Step 4: After converting and correcting the Sigmoid function prototype, the Sigmoid correction function is obtained as Among them, the coefficients A=1.5, a=1, B=2, b=0.5, and c=3 are selected.

[0061] Step 5: Calculate the direction vector n of the previous iteration direction using the following formula k , the direction vector p to be iteratively calculated this time k , the cosine value of the angle between the two iterations ζ k parameter:

[0062]

[0063] Among them, f(u k ) is obtained directly.

[0064] Then, the iterative rotation angle θ is calculated by the following formula: k Size:

[0065] θ k =arccos(ζ k ).

[0066] Step 6: Set the chaos control factor λ k Combined with the Sigmoid correction function, the following adjustment function is obtained:

[0067]

[0068] In the formula, λ k-1 is the chaos control factor of the previous iteration.

[0069] Then, the step size of the finite step size method is adaptively adjusted through the adjustment function.

[0070] Step 7: Set ε′ = 10 -4 is the allowable error, and the relative error between the two iteration points is calculated ε=|∥x k+1 ∥-∥x k ∥| / ∥x k ∥, if the convergence condition ε<ε′ is met, then stop the iteration and output the result, otherwise repeat steps 3 to 6, that is, the convergence condition is not met at this time, so return to step 3 to continue the calculation. Until the error meets the convergence condition and output the result.

[0071] Based on the above steps, the results shown in Table 1 are obtained. In order to better illustrate the effectiveness and advantages of the method given by the present invention (SSA method), the present invention also compares the classical first-order second moment method (HL-RF method) and the chaos control algorithm (CC algorithm) with the method given by the present invention. The comparison results are also listed in Table 1. Table 1 compares the Monte Carlo results by 10 6 The sampling is obtained and the result is regarded as the exact solution for comparison of the calculation error. In addition, in order to fully compare and illustrate, the specification of the present invention is attached. Figure 3 The iteration path diagram of each method iteration point is also given. Figure 4 The iterative process diagrams of the reliability index β of each method are given.

[0072] Table 1 Comparison of results of various methods in Example 1

[0073]

[0074] The example given in Example 1 of the present invention is a highly nonlinear power-type two-dimensional example. Figure 3 It can be seen that the nonlinearity of the example is relatively high near the failure point, so the classical second-order moment method will produce periodic oscillations and cannot converge to a stable result. Both the chaos control algorithm and the method proposed in the present invention can converge to the failure point, and the reliability index β results are consistent. The convergence efficiency of the chaos control algorithm is low, and the number of iterations required is too large, requiring 123 iterations to complete. The method proposed in the present invention requires 15 iterations, which is approximately 1 / 8 of the number of iterations of the chaos control algorithm. Therefore, the method given in the present invention shows an advantage in computational efficiency. This Example 1 verifies the unique advantages of the method proposed in the present invention in terms of computational efficiency and robustness.

[0075] Example 2: Curved Slender Column Example

[0076] like Figure 5 The slender column shown is an elastic slender column that is subjected to horizontal load H and vertical load P. The elastic modulus E, moment of inertia I, and length L of the slender elastic column are affected by the horizontal load H and the vertical load P. The bottom of the column is connected by a nonlinear rotation spring, which provides a magnitude of The present embodiment involves six basic random variables, and the statistical properties of the six random variables are given in Table 2.

[0077] Table 2 Statistical characteristics of random variables in Example 2

[0078]

[0079] The limit state function of Example 2 is: g = 10-δ, where δ is the horizontal displacement of the column top measured in inches under the applied load, calculated as follows:

[0080] in,

[0081] n is the lateral stiffness. When solving the reliability index β of the limit state function of Example 2, the specific steps of the calculation method proposed by the present invention are the same as those of Example 1, except that the number of random variables and the corresponding statistical characteristics in step 1 are different, and the other steps are the same and will not be elaborated here.

[0082] The calculation results obtained by the calculation method proposed by the present invention are shown in Table 3. In order to further illustrate the feasibility and effectiveness of the method given by the present invention, this embodiment, like Example 1, also compares the classical first-order second moment method (HL-RF method), chaos control algorithm (CC algorithm) and the method given by the present invention. The comparison results are listed in Table 3. Similarly, Table 3 also compares the Monte Carlo results by 10 6 The sampling is obtained and the result is regarded as the exact solution for comparison of calculation error. In addition, in order to fully compare and illustrate, and considering that the number of variables in Example 2 is large and the MPP search path cannot be presented in the U space, this example only gives the iterative process diagram of the reliability index β of each method.

[0083] Table 3 Comparison of results of Example 2

[0084]

[0085] from Figure 6 It can be seen from the calculation results of Table 3 that all four algorithms can reach convergence. However, since this embodiment 2 is a complex and highly nonlinear example containing 6 normally distributed random variables, its complexity is relatively high. The use of the classical first-order second-order moment method and the chaos control algorithm will generate more calculations, and the changes in the reliability index β during the iteration process of the classical first-order second-order moment method fluctuate greatly, and the convergence trend is unstable. In comparison, the method proposed in the present invention requires fewer iterations and has higher convergence efficiency. The number of iterations is about 1 / 2 of the chaos control algorithm, and the convergence process presented is more stable and ideal than the classical first-order second-order moment method. In summary, the adaptive step size adjustment method based on the Sigmoid function provided by the present invention is feasible in the application of complex problems and shows excellent computing performance.

[0086] The above description is a detailed description of the preferred feasible embodiments of the present invention, but the embodiments are not intended to limit the scope of the patent application of the present invention. All equivalent changes or modified changes completed under the technical spirit suggested by the present invention should fall within the patent scope covered by the present invention.

Claims

1. A primary reliability calculation method based on Sigmoid function adaptive step size adjustment, characterized in that: The steps include: Step 1: Determine the random variable X of the limit state function = (x1, x2, x3, ..., x n ) and the statistical characteristics of random variables, and set the coefficient k for monitoring the number of iterations and the lower bound λ of the algorithm l , where statistical characteristics include but are not limited to mean and standard deviation; Step 2: Use standard normal transformation to normalize the random variables, then perform the first reliability analysis iteration and collect iteration information, including but not limited to iteration failure point coordinates and reliability indicators; Step 3: Use standard normal transformation to transform the known random variable x k Transform the random variable u into the standard normal space k ; Step 4: Introduce the Sigmoid function, and determine the step adjustment method of the Sigmoid function according to the relationship between the turning angle and the ideal iteration step length; transform the Sigmoid function and introduce coefficients A, a, B, b, and c to correct it, and obtain the Sigmoid correction function. By determining the values ​​of coefficients A, a, B, b, and c, the Sigmoid correction function conforms to the relationship between the turning angle and the ideal iteration step length; Step 5: Calculate the function value f(u k ), the direction vector n of the previous iteration direction k , the direction vector p to be iteratively calculated this time k Parameters, thereby calculating the iterative rotation angle θ corresponding to the kth iteration k size; Step 6: Update the chaos control factor λ k , the chaos control factor λ k Combined with the Sigmoid correction function, the step size of the finite step size method is adaptively adjusted; and the variable u that obeys the normal distribution is k+1 Convert to random variable x k+1 ; Step 7: Set the allowable error ε′ and calculate the relative error ε=|∥x between the two iteration points. k+1 ∥-∥x k ∥| / ∥x k ∥, if the convergence condition ε<ε′ is met, stop the iteration and output the result, otherwise repeat steps 3 to 6 until the relative error between the two iteration points before and after the calculation meets the convergence condition.

2. A primary reliability calculation method based on Sigmoid function adaptive step size adjustment as claimed in claim 1, characterized in that: In step 1, the coefficient k of monitoring the number of iterations is 0, and the lower bound of the algorithm is l =0.

1.

3. A primary reliability calculation method based on Sigmoid function adaptive step size adjustment as claimed in claim 1, characterized in that: In step 4, the relationship between the rotation angle and the ideal iteration step length is inversely proportional.

4. A primary reliability calculation method based on Sigmoid function adaptive step size adjustment as claimed in claim 1, characterized in that: In step 4, the Sigmoid correction function is: In the formula, t is 5. A primary reliability calculation method based on Sigmoid function adaptive step size adjustment as claimed in claim 1, characterized in that: In step 5, the iterative rotation angle θ k The size is calculated by the following formula: i k =arccos(ζ k ) in, In the formula, ζ k It represents the cosine value of the rotation angle between the two iterations.

6. A primary reliability calculation method based on Sigmoid function adaptive step size adjustment as claimed in claim 1, characterized in that: In step 6, the chaos control factor λ k Combined with the Sigmoid correction function, the following adjustment function is obtained: In the formula, λ k-1 is the chaos control factor of the previous iteration.