Non-probabilistic reliability analysis method for tunnels based on symbolic regression algorithm
The limit state function in tunnel engineering is explicitly modified through the symbol regression algorithm, which solves the problem that implicit limit state function is difficult to calculate in traditional methods, realizes non-probability reliability analysis, and improves the accuracy and simplicity of tunnel reliability evaluation.
Patent Information
- Application Number
- CN202410585338.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-05-13
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2044-05-13
AI Technical Summary
In tunnel engineering, traditional probabilistic reliability analysis methods are difficult to obtain relevant statistical information of physical and mechanical parameters of geotechnical bodies, making the analysis complex and difficult to calculate non-probability reliability indexes, especially when the variability of physical and mechanical parameters of geotechnical bodies and In uncertainty, the limit state function is often a highly nonlinear implicit function and cannot be effectively calculated.
Using the non-probability reliability analysis method of tunnels based on symbol regression algorithm, the mechanical characteristic function of soft rock formation tunnel is constructed, the limit state function is explicitly calculated, the non-probability reliability index is calculated, and the response interval is processed by truncating coefficients to construct non-probability Reliability analysis evaluates metrics to evaluate the reliability of the tunnel.
Without a large amount of statistical information on physical and mechanical parameters of rock and soil, the symbol regression algorithm simplifies the fitting of the limit state function, reduces the complexity of the nested structure of formulas, improves the fitting accuracy, and reveals that uncertain parameters are stable to the tunnel structure. The effect of nature avoids the large and distorted intervals of the response interval calculation results.
Smart Images

Figure CN118520554B_ABST
Abstract
Description
Technical Field
[0001] The present application belongs to the technical field of tunnel engineering, and specifically relates to a tunnel non-probabilistic reliability analysis method based on a symbolic regression algorithm. Background Art
[0002] When analyzing the reliability and stability of tunnel engineering, the traditional probabilistic reliability analysis method has high accuracy and a relatively mature calculation and analysis system. However, in actual engineering, it is not easy to obtain a large amount of statistical information related to the physical and mechanical parameters of rock and soil. Based on this, the analysis indicators and evaluation methods of non-probabilistic reliability are used to characterize the uncertainty parameters of the tunnel structure in the form of interval variables. Only the variable distribution boundary needs to be obtained for analysis, which can effectively solve the defects in the traditional probabilistic reliability analysis method.
[0003] However, when using non-probabilistic reliability methods to study and analyze tunnel engineering problems, due to the variability and uncertainty of the physical and mechanical parameters of the rock mass itself, multiple uncertain factors affect the stability of the tunnel surrounding rock, which leads to complex analysis models and further increases the difficulty of analysis. In many cases, the limit state function of the tunnel structure will appear in the form of a highly nonlinear implicit function, and the corresponding non-probabilistic reliability index cannot be calculated.
[0004] Therefore, it is necessary to provide a tunnel non-probabilistic reliability analysis method based on a symbolic regression algorithm to solve the problems raised in the above background technology. Summary of the invention
[0005] The present application provides a tunnel non-probabilistic reliability analysis method based on a symbolic regression algorithm. The symbolic regression algorithm is introduced to explicitly represent the implicit function of the limit state function, thereby solving the problem of implicit limit state function in the prior art. The non-probabilistic reliability index can be calculated using the explicitly represented limit state function.
[0006] In order to solve the above technical problems, the technical solution of this application is:
[0007] A tunnel non-probabilistic reliability analysis method based on symbolic regression algorithm comprises the following steps:
[0008] Step S1, based on the convergence constraint method of modern surrounding rock and support action principle, construct the mechanical characteristic function of the soft rock stratum tunnel under the action of shotcrete support during excavation, and obtain the limit state function of the soft rock stratum tunnel without losing stability based on the mechanical characteristic function. The limit state function is about the deformation pressure p around the tunnel. s Implicit function of ;
[0009] Step S2, dividing the parameters in the implicit function into uncertain parameters and determined parameters, setting the value range of the uncertain parameters and the value of the determined parameters, using a symbolic regression algorithm to explicitly represent the limit state function, and calculating the original response range of the limit state function;
[0010] Step S3, calculating the maximum relative deviation of each uncertain parameter and the center value of its interval, then assigning a weight to each uncertain parameter according to the proportion of the maximum deviation of each uncertain parameter in the relative deviations of all uncertain parameters, constructing a truncation coefficient with the weighted sum of all uncertain parameters, and using the truncation coefficient to truncate the original response interval;
[0011] Step S4, based on the interval non-probabilistic reliability theory, construct a non-probabilistic reliability analysis and evaluation index from the truncated response interval, and use the non-probabilistic reliability analysis and evaluation index to evaluate the tunnel reliability.
[0012] Preferably, in step S2, the limit state function after parameter division is expressed as:
[0013]
[0014] In the formula, p s is the deformation pressure around the tunnel; c rm is the cohesion of the surrounding rock after shotcrete support; is the internal friction angle of the surrounding rock; E is the elastic modulus of the surrounding rock; p0 is the original rock stress; t c is the thickness of the concrete spraying layer; lim is the ultimate strain rate of surrounding rock stability; c rm , E, p0, t c , ε lim These are all uncertain parameters.
[0015] Preferably, the symbolic regression algorithm is selected from one of the ALPS-GP algorithm, the OSGP algorithm, and the GE algorithm.
[0016] Preferably, the calculation process of the original response interval includes the following steps:
[0017] The uncertain parameters are standardized:
[0018] x=x c +δx r ;
[0019] In the formula, x c 、x r are the interval mean and deviation of the uncertainty parameter x respectively; δ is the standardized value of the uncertainty parameter x, δ∈[-1,1], δ=δ1,δ2,…δ i …δn ;
[0020] Get the new limit state function g(x):
[0021] g(x)=G(δ1,δ2,…δ i …δ n )=0;
[0022] Calculate the response interval of the limit state function g(x) [M l , M u ], which is the original response interval:
[0023] M u = max G(δ1, δ2, ... δ n );
[0024] M l =min G(δ1, δ2, ...δ n ).
[0025] Preferably, the truncation coefficient is expressed as:
[0026]
[0027] Where T is the cutoff coefficient; t i represents any uncertain parameter x i Maximum relative deviation; n represents the number of uncertain parameters; ω i Represents the uncertain parameter x i The weight of ; where:
[0028]
[0029] In the formula, They represent the upper and lower limits of the uncertainty parameter value interval respectively; x0 represents the center value of the uncertainty parameter value interval;
[0030]
[0031] Preferably, the truncated response interval is expressed as: in:
[0032] when and hour,
[0033] when and hour,
[0034] when and hour,
[0035] when and hour,
[0036] Wherein, Δ1 represents the upper limit deviation of the response interval relative to the center point of the interval; Δ2 represents the lower limit deviation of the response interval relative to the center point of the interval; M0 represents the center value of the response interval;
[0037]
[0038] Preferably, the non-probabilistic reliability analysis evaluation index is expressed as:
[0039]
[0040] Wherein, η is a non-probabilistic reliability analysis and evaluation index; when η>1, the tunnel structure is judged to be stable and reliable; when η<1, the tunnel structure is judged to be unstable; when η=1, the tunnel structure is judged to be in a critical state.
[0041] Preferably, in the process of fitting the limit state function using a symbolic regression algorithm, the MSE index, MAE index and R 2 The index is used as an evaluation indicator to assess the impact of the number of samples and the combination of operation symbols on the fitting accuracy, adjust the number of samples and the combination of operation symbols, and output the optimal fitting result.
[0042] Preferably, the following steps are also included after step S5:
[0043] The influence of uncertainty parameters on tunnel reliability is analyzed by defining two indicators: fluctuation amplitude and sensitivity factor.
[0044] The fluctuation amplitude λ is defined as:
[0045]
[0046] The fluctuation amplitude λ is used to indicate the change in the uncertainty parameter value interval in the non-probabilistic case;
[0047] Sensitivity factor θ i Defined as:
[0048]
[0049] In the formula, Δg i When the i-th uncertainty parameter is volatile, the response interval of the limit state function is but Δg is the response range of the structural limit state function when all uncertainty parameters are volatile. In this case, the response range is [g l , gu ], then Δg=g u -g l θ i ∈[0,1]; sensitivity factor θ i It is used to analyze the influence of the uncertainty parameter value range on the change of the structural response value range, θ i The larger it is, the more sensitive the structural response is to the uncertainty parameter, and the greater the influence of the uncertainty parameter on the structure; otherwise, the smaller it is.
[0050] Preferably, the influence of various uncertainty parameters on tunnel reliability is analyzed, specifically including:
[0051] Analyze the impact of fluctuations in uncertainty parameters on non-probabilistic reliability indicators of tunnel structures;
[0052] Analyze the influence of uncertainty parameter value range on the response range of tunnel structure limit state function;
[0053] Analyze the relationship between the fluctuation amplitude of uncertainty parameters and the sensitivity factor.
[0054] The beneficial effects of this application are:
[0055] (1) The non-probabilistic reliability analysis method based on interval theory provided in this application does not require obtaining a large amount of statistical information related to the physical and mechanical parameters of the rock and soil mass;
[0056] (2) The symbolic regression algorithm is introduced to solve the problem that the limit state function of the tunnel structure in the non-probabilistic reliability analysis is highly implicitly nonlinear. It is easier to fit the explicit function expression for reliability analysis and reveals the impact of the change in the range of uncertainty parameters on the stability of the tunnel structure. Compared with other machine learning algorithms, the expression obtained by the symbolic regression algorithm not only greatly reduces the nested structure and complexity of the formula, but also ensures a high fitting accuracy, making the fitting result closer to the actual engineering situation.
[0057] (3) Using the truncation coefficient to truncate the response interval of the limit state function can effectively solve the interval expansion phenomenon and avoid the distortion of the response interval calculation results due to the large interval. BRIEF DESCRIPTION OF THE DRAWINGS
[0058] Figure 1 Represents the surrounding rock characteristics and support curves;
[0059] Figure 2 A schematic diagram showing tunnel excavation and support;
[0060] Figure 3 It is a curve graph of the change of MSE index with the number of samples;
[0061] Figure 4 It is a curve chart of MAE index changing with the number of samples.
[0062] Figure 5 YesR 2 The curve graph of the indicator changing with the number of samples;
[0063] Figure 6 It is a comparison chart of non-probabilistic reliability index under the change of each uncertainty parameter interval;
[0064] Figure 7 It is a comparison chart of the changes in uncertainty parameter sensitivity factors under changes in fluctuation amplitude. DETAILED DESCRIPTION
[0065] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.
[0066] Please refer to Figure 1-Figure 7 The present application proposes a tunnel non-probabilistic reliability analysis method based on a symbolic regression algorithm, comprising the following steps:
[0067] Step S1, based on the convergence constraint method of modern surrounding rock and support action principle, construct the mechanical characteristic function of the soft rock stratum tunnel under the action of shotcrete support during excavation, and obtain the limit state function of the soft rock stratum tunnel without losing stability based on the mechanical characteristic function. The limit state function is about the deformation pressure p around the tunnel. s Implicit function of .
[0068] For the convenience of analysis, the surrounding rock is regarded as an isotropic medium existing in a natural stress field and subjected to hydrostatic pressure. According to the MC yield criterion and the plastic flow law, the formation characteristic function is derived and expressed as:
[0069]
[0070] In the formula, M is the radial displacement of the surrounding rock towards the tunnel after excavation; ep is the stress difference of the elastic-plastic boundary; μ is the Poisson's ratio of the surrounding rock; G is the shear modulus of the surrounding rock; R p is the radius of the plastic zone formed after the surrounding rock is disturbed, determined according to the Kastner H principle; r0 is the tunnel radius; p a is the deformation pressure of the surrounding rock around the tunnel; p0 is the original rock stress, where:
[0071]
[0072]
[0073]
[0074] In the formula, is the internal friction angle of the surrounding rock; c0 is the cohesion of the surrounding rock; E is the elastic modulus of the surrounding rock.
[0075] During tunnel excavation, shotcrete support is used as temporary support. The shotcrete layer can improve the strength of the surrounding rock in the excavation area, and the anchor rod will increase the cohesion of the surrounding rock in the anchoring area to a certain extent, while having little effect on other mechanical parameters such as the internal friction angle. The cohesion of the surrounding rock after shotcrete support is calculated as follows:
[0076]
[0077] In the formula, c rm is the cohesion of surrounding rock after shotcrete support; τ α A is the shear strength of the anchor rod; s is the cross-sectional area of the anchor rod; S a , S b They are the longitudinal and transverse spacing of anchor rod arrangement respectively.
[0078] According to the modern mechanical viewpoint that surrounding rock and support bear the load together and interact with each other, the concrete spraying layer, as an auxiliary bearing part, bears a small load pressure and is in the elastic deformation working stage, which conforms to the thick-walled cylinder bearing principle in elastic mechanics. Therefore, at a certain moment after the concrete spraying layer is applied, the surrounding rock is deformed and contacts with the concrete spraying layer and generates u s The deformation of the tunnel periphery is p s For: p s =K s u s ;
[0079] In the formula, K s is the stiffness coefficient of the concrete spraying layer, which satisfies the following formula:
[0080]
[0081] In the formula, E c is the elastic modulus of concrete; μ c is the Poisson's ratio of concrete; r1 is the inner radius of the thick-walled cylinder; t c is the thickness of the concrete spray layer, t c =r0-r1.
[0082] After the support is applied, when the deformation u s Reach u c When the surrounding rock and support reach a state of equilibrium, Figure 1The total deformation u generated around the tunnel is shown as point D on the surrounding rock and support characteristic curve. all and deformation pressure p s satisfy:
[0083] u all =u0+u c ;
[0084] p s =p a ;
[0085] Where u0 is the deformation of the tunnel periphery when support is applied, which depends on the timing of support application. Suppose deformation u0 is equal to the total deformation u′ of the tunnel periphery without support. all The ratio is α, then:
[0086] u0=αu′ all ;
[0087] Put p a =0 Substituting into the above formula, we can get u′ all The calculation expression is:
[0088]
[0089] Then the deformation pressure around the tunnel is s The expression is:
[0090]
[0091] From the above formula, we can see that p s The expression contains p on both sides. s , and the function is complex, it is a highly nonlinear implicit function.
[0092] After the support installation is completed, the total deformation around the tunnel is u all for:
[0093]
[0094] Combination Figure 1 The surrounding rock and support characteristic curve shown in the figure shows that the surrounding rock and support reach critical stability at point E, and the corresponding tunnel perimeter allowable limit deformation is u lim , which includes the elastic deformation u e 、Plastic deformation p and expansion deformation u d . Allowable limit deformation u lim Depends on the ultimate strain rate ε of surrounding rock stability lim , construction technology, process and related safety requirements during tunnel excavation and construction. lim is the allowable limit deformation of the tunnel u limThe ratio of the radius of the tunnel to r0 is expressed as:
[0095]
[0096] At the same time, based on the analysis of the control effect of shotcrete support on the tunnel, it can be seen that the total deformation when the surrounding rock and support reach equilibrium should include the deformation u0 of the tunnel surrounding rock before the support is applied and the deformation u generated by the surrounding rock and support after the shotcrete support is installed. c If the surrounding rock is to be stable and not destabilized, the total deformation u of the system in equilibrium must be ensured. all Not exceeding the allowable limit deformation u around the tunnel lim , that is, the limit state function of the soft rock tunnel is:
[0097]
[0098] Step S2, dividing the parameters in the implicit function into uncertain parameters and determined parameters, setting the value range of the uncertain parameters and the value of the determined parameters, using a symbolic regression algorithm to explicitly represent the limit state function, and calculating the original response range of the limit state function;
[0099] Because p s The expression of is an implicit function, while the limit state function is related to p s Related functions, the limit state function is also an implicit function. For the basic parameters contained in the function, c rm , E, p0, t c , ε lim It has high variability and uncertainty, so it is considered as an uncertain parameter. The variability of other parameters is relatively small, so in order to simplify the analysis, they are all considered as deterministic parameters and directly assigned values based on relevant literature or experience. The limit state function of soft rock tunnel can also be expressed as:
[0100] The symbolic regression algorithm is used to calculate the limit state function The symbolic regression algorithm is selected from one of the ALPS-GP (Age-Layered Population Structure Genetic Programming) algorithm, the OSGP (Offspring Selection Genetic Programming) algorithm, and the GE (Grammatical Evolution) algorithm.
[0101] In the symbolic regression algorithm, each formula can be expressed as an expression or syntax tree consisting of basic functions, independent variables and coefficients. The problem of fitting formulas in the symbolic regression algorithm can be regarded as the problem of generating expression trees. The root node of each expression tree is an independent variable. By continuously selecting replaceable independent variable nodes, they are randomly replaced with other symbolic subtrees, thereby generating new expression tree branches and forming new formulas. The complexity of the expression tree obtained by symbolic regression is described by depth and length. Depth indicates the hierarchical depth of functions or expressions in the symbolic regression model. Deeper models often have more nested structures; length indicates the number of nodes, that is, the length of functions or expressions. Longer models often have more parameters and operators and can represent more complex functional relationships. The expression tree adjusts its fitness to the optimal value through continuous mutation and evolution, thereby obtaining the optimal formula.
[0102] The explicit expression of the limit state function is:
[0103] g(x) = g(x1, x2, ... x i …x n )=0;
[0104] Where x = x1, x2, … x i …x n Indicates an uncertain parameter.
[0105] Since the uncertain parameters in the function have a certain range of variation, the solved limit state function will also fluctuate within a certain range, and its fluctuation range is expressed as the response interval M = [M l , M u ], where M u 、M l Represent the upper and lower limits of the response interval respectively.
[0106] For situations where the structural limit state function is more complex and the nonlinearity is higher, the approximate solution method of optimization calculation can be used. After the value range of the series of uncertainty parameters is determined, these uncertainty parameters are standardized as follows:
[0107] x=x c +δx r ;
[0108] In the formula, x c 、x r are the interval mean and deviation of the uncertainty parameter x respectively; δ is the standardized value of the uncertainty parameter x, δ∈[-1,1], δ=δ1,δ2,…δ i …δ n ;
[0109] Get the new standardized limit state function:
[0110] g(x)=G(δ1,δ2,…δ i …δ n )=0;
[0111] Calculate the response interval of the limit state function g(x) [M l , M u ], which is the original response interval: M u = max G(δ1, δ2, ... δ n );
[0112] M l =min G(δ1, δ2, ...δ n ).
[0113] Step S3, calculate the maximum relative deviation of each uncertain parameter and the center value of its interval, then assign a weight to each uncertain parameter according to the proportion of the maximum deviation of each uncertain parameter in the relative deviation of all uncertain parameters, construct a truncation coefficient with the weighted sum of all uncertain parameters, and use the truncation coefficient to truncate the original response interval.
[0114] During the interval calculation process, the correlation error of the same parameter appears multiple times and the rounding error caused by the "outward truncation" value selection method of the interval algorithm makes the interval calculation prone to interval expansion, making the final response interval calculation result larger and distorted. In order to solve the interval expansion problem, this application uses the truncation coefficient to truncate the interval result.
[0115] The truncation coefficient is expressed as:
[0116]
[0117] Where T is the cutoff coefficient; t i represents any uncertain parameter x i Maximum relative deviation; n represents the number of uncertain parameters; ω i Represents the uncertain parameter x i The weight of ; where:
[0118]
[0119] In the formula, They represent the upper and lower limits of the uncertainty parameter value interval respectively; x0 represents the center value of the uncertainty parameter value interval;
[0120]
[0121] The upper and lower limit deviations of the original response interval are the sum of the upper and lower limits of all input uncertainty parameter errors. Therefore, the error value of each uncertain parameter has a different impact on the response interval. Assigning a weight to each uncertain parameter based on the proportion of the maximum deviation of each uncertain parameter in the relative deviation of all uncertain parameters can make the truncated response interval more accurate and form a better response to the fluctuation of the uncertain parameter.
[0122] The truncated response interval is expressed as: in:
[0123] when and hour,
[0124] when and hour,
[0125] when and hour,
[0126] when and hour,
[0127] Wherein, Δ1 represents the upper limit deviation of the response interval relative to the center point of the interval; Δ2 represents the lower limit deviation of the response interval relative to the center point of the interval; M0 represents the center value of the response interval;
[0128]
[0129] Comparing the cutoff coefficient with the deviation of the response interval to determine the final response interval can avoid the cutoff coefficient being greater than 1 and further expanding the response interval.
[0130] Step S4, based on the interval non-probabilistic reliability theory, construct a non-probabilistic reliability analysis and evaluation index from the truncated response interval, and use the non-probabilistic reliability analysis and evaluation index to evaluate the tunnel reliability.
[0131] The non-probabilistic reliability analysis evaluation index is expressed as:
[0132]
[0133] When η>1, the tunnel structure limit state function curve does not intersect with the failure domain, and the tunnel structure is stable and reliable; when η<-1, the tunnel structure limit state function curve intersects with the failure domain, and the tunnel structure becomes unstable; when -1<η<1, the tunnel structure is in a critical state, which may be safe or unstable. Strictly speaking, the possibility of instability is unsafe. Therefore, when η<1, the tunnel structure can be regarded as unstable. According to the above criteria, reliability analysis and stability judgment can be carried out on the actual engineering of tunnel structure.
[0134] Example 1
[0135] The excavation section of a tunnel is circular, and the excavation radius r0 is 5.3m. According to the on-site engineering rock mass classification and physical and mechanical parameter evaluation, the surrounding rock in the excavation area is V-grade soft rock, the Poisson's ratio of the surrounding rock is μ=0.4, the cohesion c0 of the surrounding rock is about 0.276MPa~0.358MPa, and the internal friction angle of the surrounding rock is The range is 23.84°~28.56°, and the elastic modulus E of the surrounding rock ranges from 1139.2MPa to 1420.8MPa.
[0136] According to the on-site survey monitoring data and combined with relevant theoretical speculation, the original rock stress field p0 in the tunnel area is roughly between 6.44MPa and 7.56MPa. According to geological information and excavation conditions, shotcrete support is adopted as temporary support. The relevant support parameters are as follows: concrete strength grade is C30, concrete elastic modulus E c =31GPa, Poisson's ratio of concrete μ c =0.2. Due to construction reasons, the thickness of the concrete spraying layer is c It is easy to fluctuate within a certain range, and its value is considered to be 0.255m~0.345m; the anchor rod diameter is 22mm, the length is 4m, and the shear strength τ α =312MPa, cross-sectional area A s =380.13mm 2 The vertical and horizontal spacing of the installation is 1m. The range of each uncertainty parameter is shown in Table 1, where the surrounding rock internal friction angle is in radian form during symbolic regression fitting.
[0137] Table 1 Value ranges of tunnel uncertainty parameters
[0138]
[0139]
[0140] Regarding the values of other parameters: Regarding the ratio α, it is related to the timing of support construction. According to relevant specifications, it is recommended to be applied in the range of 0.5 to 0.8. For soft rock tunnels, considering the principles of supporting softness first and hardness later, and timely closure, a conservative value of 0.5 is more appropriate.
[0141] First, when the number of samples is 1000, the three SR algorithm models of ALPS-GP, OSGP and GE are used to calculate the limit state function. When extracting sample points, in order to avoid the sampling points from being concentrated and to reduce the number of iterations while ensuring the sampling accuracy, Latin hypercube sampling is considered.
[0142] The ALPS-GP algorithm, OSGP algorithm, and GE algorithm are used for fitting respectively. The fitting results of the three algorithms are compared as follows:
[0143] (1) ALPS-GP algorithm fitting results:
[0144]
[0145] Among them, a0=0.5135, a1=2.217, a2=1.8866, a3=0.0061193, a4=0.71466,
[0146] a5=-1.7294, a6=-13.535, a7=-0.65992, a8=-0.69901;
[0147] (2) OSGP algorithm fitting results:
[0148]
[0149] Among them, a0=1.1483, a1=1.0357, a2=1.4329, a3=0.30578, a4=-0.49576, a5=-1.3957, a6=1.0039, a7=-1.144, a8=0.1004, a9=-5.2131, a 10 =2.5075, a 11 =5.1071, a 12 =6.8954, a 13 =0.27977, a 14 =1.7985, a 15 =-1.7083, a 16 =-0.31656, a 17 =13.85, a 18 =0.79145, a 19 =0.080574.
[0150] (3) GE algorithm fitting results:
[0151]
[0152] Among them, a0=-585.54, a1=-2901, a2=-2462.1, a3=0.00058015, a4=-0.0003344, a5=1801.2, a6=-1.0, a7=-0.10198, a8=0.80125.
[0153] The fitting results of the three algorithms are summarized in Table 2
[0154] Table 2 Fitting results using ALPS-GP algorithm, OSGP algorithm, and GE algorithm
[0155] algorithm ALPS-GP algorithm OSGP Algorithm GE Algorithm <![CDATA[p s Response Range]]> [0.8090,1.0249] [0.8142,1.0315] [0.8199,1.0388] Non-probabilistic reliability index η 1.2834 1.2829 1.2823 Model Depth 6 11 8 Model length 18 39 20 <![CDATA[The number of model parameter a i > 9 20 9 Iterations 500 25 30~50 MSE 1.188E-05 1.602E-06 1.952E-05 MAE 2.516E-03 8.984E-04 3.004E-03 <![CDATA[R 2 ]]> 0.99805 0.99996 0.99718 Calculate execution time 15s 1min 2min40s
[0156] By comparing the fitting results of the three algorithms, it is found that the three algorithms can obtain roughly the same p under the premise of ensuring high accuracy and low error. s The interval range and the non-probabilistic reliability index η value. However, in terms of the complexity of the model, when the model depth is not very different, the OSGP fitting formula is longer, with more parameters, and its formula tree also has more branches, which is more complicated and cumbersome; the GE fitting formula is simpler than OSGP, but it takes longer to calculate when the number of iterations is similar. Compared with the first two, ALPS-GP can quickly complete a higher number of iterations and obtain a simpler expression, which can effectively reduce the computational workload of subsequent non-probabilistic reliability analysis, and the effect is better.
[0157] In order to further verify the accuracy of the symbolic regression model, the sample number is also selected as 1000, and the fitting results of the three algorithms, ALPS-GP algorithm, OSGP algorithm, and GE algorithm, are compared with the results of the response surface method (RS) and support vector regression (SVMR) in traditional technologies. The evaluation indicators are MSE, MAE and R 2 The fitting performance of the three algorithms, ALPS-GP algorithm, OSGP algorithm and GE algorithm, is evaluated by using the indicators. The comparison results are shown in Table 3.
[0158] Table 3 Comparison of fitting performance of various algorithms
[0159] MSE MAE <![CDATA[R 2 ]]> <![CDATA[p s Calculation interval]]> ALPS-GP 1.188E-05 2.516E-03 0.99805 [0.8090,1.0249] OSGP 1.602E-06 8.984E-04 0.99996 [0.8142,1.0315] GE 1.952E-05 3.004E-03 0.99718 [0.8199,1.0388] RS 2.253E-07 3.138E-04 0.99995 [1.6010,2.0281] SVMR 3.423E-05 1.820E-03 0.99918 NaN
[0160] In the table, NaN indicates that the function being calculated is too complex to obtain its exact expression, and thus the p under this algorithm cannot be obtained. s Calculate the interval.
[0161] It can be seen from Table 3 that the ALPS-GP algorithm, OSGP algorithm, and GE algorithm can all achieve the fitting regression of the implicit function with high accuracy. s The function expression of p is calculated s The prediction interval of ,comparison shows that among the symbolic regression algorithms, the results of ALPS-GP algorithm, OSGP algorithm and GE algorithm are relatively close.
[0162] The response surface method (RS) in traditional technology can ensure high accuracy, but the p s The interval is offset and there are errors. Compared with symbolic regression, the expression obtained by fitting is a fixed quadratic function polynomial, and the symbolic function form cannot be flexibly adjusted.
[0163] The support vector regression (SVMR) in traditional technology can also guarantee high accuracy. The final function expression is related to the number of training samples in the provided data set. Each support vector sample corresponds to its Lagrange parameter difference. Therefore, the fitting function expression should be represented by all support vectors and their corresponding Lagrange parameter differences and kernel function parameters γ and bias b. It is very complex and lengthy, has many parameters, and has a very high computational cost.
[0164] Comprehensive comparison shows that the symbolic regression algorithm can maintain a high fitting accuracy, and compared with the traditional response surface method (RS), the symbolic function form can be flexibly adjusted; compared with the traditional support vector regression (SVMR), the number of parameters can be reduced, the length of the function can be shortened, and the calculation cost can be controlled.
[0165] When using the symbolic regression algorithm for fitting, it is also necessary to consider the impact of the number of samples and the combination of symbolic functions on the model accuracy.
[0166] First, we analyze the effect of sample size on model accuracy. When selecting the symbol range for symbolic regression fitting, we consider four basic operations: +, -, ×, and ÷. The results of sample size and model accuracy are shown in Table 4 and Figure 3-5 shown.
[0167] Table 4 Comparison of fitting accuracy under different sample numbers
[0168]
[0169]
[0170] Through comparison, it can be seen that the accuracy of the three models is generally high, especially when the number of samples is 1000, it reaches the optimal value and the fitting effect is better.
[0171] Next, the influence of symbolic function combination on fitting accuracy is discussed. One of the characteristics of symbolic regression algorithm is that the form of fitting target expression can be determined by flexible adjustment according to actual needs. In addition to the basic four arithmetic operations, power function, logarithmic function and other types can also be selected for fitting analysis.
[0172] The following will discuss the impact of the combination of operator symbols on the symbolic regression fitting results based on the case where the sample points are 1000. The results are shown in Table 5.
[0173] Table 5 Comparison of fitting accuracy under different combinations of operation symbols and functions
[0174]
[0175]
[0176] The comparison shows that, whether it is a different combination of symbols or increasing the number of symbol functions, the accuracy of the fitting results of the symbolic regression algorithm is higher, p s There is no significant difference in the interval calculation results, and the effect is good. This shows that compared with other implicit function fitting methods, symbolic regression can better choose symbol combinations and function expression forms according to the needs of researchers.
[0177] After determining the non-probabilistic reliability analysis evaluation index, it is also necessary to analyze the influence of various uncertainty parameters on tunnel reliability in order to better guide actual engineering construction. In this application, two indicators, fluctuation amplitude and sensitivity factor, are used to analyze the influence of uncertainty parameters on tunnel reliability.
[0178] Among them, the fluctuation amplitude λ is defined as:
[0179]
[0180] The fluctuation amplitude λ is used to indicate the change in the uncertainty parameter value range in non-probabilistic situations.
[0181] Sensitivity factor θ i Defined as:
[0182]
[0183] In the formula, Δg i When the i-th uncertainty parameter is volatile, the response interval of the limit state function is but Δg is the response range of the structural limit state function when all uncertainty parameters are volatile. In this case, the response range is [g l , g u ], then Δg=g u -gl θ i ∈[0, 1].
[0184] Sensitivity factor θ i The local gradient information is used to analyze the influence of the uncertainty parameter value interval on the change of the structural response value interval, θ i The larger it is, the more sensitive the structural response is to the uncertainty parameter, and the greater the influence of the uncertainty parameter on the structure; otherwise, the smaller it is.
[0185] Analyze the impact of various uncertainty parameters on tunnel reliability, including:
[0186] Analyze the impact of fluctuations in uncertainty parameters on non-probabilistic reliability indicators of tunnel structures;
[0187] Analyze the influence of uncertainty parameter value range on the response range of tunnel structure limit state function;
[0188] Analyze the relationship between the fluctuation amplitude of uncertainty parameters and the sensitivity factor.
[0189] The analysis results are as follows Figure 6-Figure 7 As shown, from Figure 6 and Figure 7 It can be seen that:
[0190] Analyze from the perspective of fluctuation amplitude. Let k be the slope of the secant line of each curve, and we can get: As the uncertainty parameter range increases, the non-probabilistic reliability index of the tunnel structure decreases. Among the various uncertain parameters, the internal friction angle of the surrounding rock has the greatest impact on the non-probabilistic reliability index, and the thickness of the concrete spraying layer has the least impact on the index. The effects of cohesion, elastic modulus and initial ground stress on the index are similar.
[0191] The sensitivity factors of various uncertainty parameters are calculated, as shown in Table 6. From the perspective of sensitivity factors, it can be seen that the sensitivity factors of the six uncertainty parameters have the following size relationship: Therefore, it is believed that the tunnel structure is more sensitive to the internal friction angle of the surrounding rock, and its influence on the stability of the tunnel structure is stronger; on the contrary, the thickness of the concrete spraying layer has the least influence on the stability of the tunnel structure. The sensitivity factors of the initial ground stress, elastic modulus and cohesion are slightly different, so it can be considered that the influence of these three uncertain parameters on the stability of the tunnel structure is similar, which is roughly the same as the conclusion obtained by analyzing the fluctuation amplitude.
[0192] Table 6 Sensitivity factors of various uncertainty parameters
[0193]
[0194]
[0195] Considering the impact between the two indicators, the changes in the sensitivity factors of each uncertainty parameter with the fluctuation range are summarized in Table 7.
[0196] Table 7 Variation of sensitivity factors of uncertainty parameters with fluctuation amplitude
[0197]
[0198]
[0199] It can be seen from Table 7 that as the fluctuation amplitude increases, the sensitivity factor of each uncertainty parameter also increases, indicating that the larger the range of the uncertainty parameter, the more likely the tunnel structure stability will be affected. Secondly, the increase in the sensitivity factor of the uncertainty parameter is: This shows that compared with other uncertain parameters, the tunnel structure is more sensitive to the internal friction angle of the surrounding rock, and the stability of the tunnel structure is most affected by it, while the thickness of the concrete spraying layer has the least impact. The changes of the three parameters, initial ground stress, elastic modulus and cohesion, are similar.
[0200] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present application, ordinary technicians in this field can also make many forms without departing from the purpose of the present application and the scope of protection of the claims, all of which are within the protection of the present application.
Claims
1. A non-probabilistic reliability analysis method for tunnels based on symbolic regression algorithm, It is characterized in that The steps include: Step S1, based on the convergence constraint method of modern surrounding rock and support action principle, construct the mechanical characteristic function of the soft rock stratum tunnel under the action of shotcrete support during excavation, and obtain the limit state function of the soft rock stratum tunnel without losing stability based on the mechanical characteristic function. The limit state function is about the deformation pressure p around the tunnel. s Implicit function of ; Step S2, dividing the parameters in the implicit function into uncertain parameters and determined parameters, setting the value range of the uncertain parameters and the value of the determined parameters, using a symbolic regression algorithm to explicitly represent the limit state function, and calculating the original response range of the limit state function; The limit state function after parameter division is expressed as: In the formula, p s is the deformation pressure around the tunnel; c rm is the cohesion of the surrounding rock after shotcrete support; is the internal friction angle of the surrounding rock; E is the elastic modulus of the surrounding rock; p 0 is the original rock stress; t c is the thickness of the concrete spraying layer; lim is the ultimate strain rate of surrounding rock stability; c rm , E.p 0 ,t c , ε lim All are uncertain parameters; The calculation process of the original response interval includes the following steps: The uncertain parameters are standardized: x=x c +δx r ; In the formula, x c 、x r are the interval mean and deviation of the uncertainty parameter x respectively; δ is the standardized value of the uncertainty parameter x, δ∈[-1,1], δ=δ 1 ,δ 2 ,…δ i …δ n ; Get the new limit state function g(x): g(x)=G(δ 1 ,d 2 ,…d i …d n )=0; Calculate the response interval of the limit state function g(x) [M l ,M u ], which is the original response interval: M u =max G(δ 1 ,d 2 ,…d n ); M l =min G(δ 1 ,d 2 ,…d n ); Step S3, calculating the maximum relative deviation of each uncertain parameter and the center value of its interval, then assigning a weight to each uncertain parameter according to the proportion of the maximum deviation of each uncertain parameter in the relative deviations of all uncertain parameters, constructing a truncation coefficient with the weighted sum of all uncertain parameters, and using the truncation coefficient to truncate the original response interval; The truncation coefficient is expressed as: Where T is the cutoff coefficient; t i represents any uncertain parameter x i Maximum relative deviation; n represents the number of uncertain parameters; ω i Represents the uncertain parameter x i The weight of ; where: In the formula, They represent the upper and lower limits of the uncertain parameter value interval respectively; x 0 Indicates the central value of the interval of uncertain parameter values; The truncated response interval is expressed as: in: when and hour, when and hour, when and hour, when and hour, In the formula, Δ 1 Indicates the upper limit deviation of the response interval relative to the center point of the interval; Δ 2 Indicates the lower limit deviation of the response interval relative to the center point of the interval; M 0 represents the central value of the response interval; Step S4, based on the interval non-probabilistic reliability theory, construct a non-probabilistic reliability analysis and evaluation index from the truncated response interval, and use the non-probabilistic reliability analysis and evaluation index to evaluate the tunnel reliability; The non-probabilistic reliability analysis evaluation index is expressed as: Wherein, η is a non-probabilistic reliability analysis and evaluation index; when η>1, the tunnel structure is judged to be stable and reliable; when η<1, the tunnel structure is judged to be unstable; when η=1, the tunnel structure is judged to be in a critical state.
2. According to the tunnel non-probabilistic reliability analysis method based on symbolic regression algorithm in claim 1, It is characterized in that The symbolic regression algorithm is selected from one of the ALPS-GP algorithm, the OSGP algorithm, and the GE algorithm.
3. The tunnel non-probabilistic reliability analysis method based on symbolic regression algorithm according to claim 1, It is characterized in that In the process of fitting the limit state function using the symbolic regression algorithm, the MSE index, MAE index and R 2 The index is used as an evaluation indicator to assess the impact of the number of samples and the combination of operation symbols on the fitting accuracy, adjust the number of samples and the combination of operation symbols, and output the optimal fitting result.
4. The tunnel non-probabilistic reliability analysis method based on symbolic regression algorithm according to claim 1, It is characterized in that After step S4, the following steps are also included: The influence of uncertainty parameters on tunnel reliability is analyzed by defining two indicators: fluctuation amplitude and sensitivity factor. The fluctuation amplitude λ is defined as: The fluctuation amplitude λ is used to indicate the change in the uncertainty parameter value interval in the non-probabilistic case; Sensitivity factor θ i Defined as: In the formula, Δg i When the i-th uncertainty parameter is volatile, the response interval of the limit state function is but Δg is the response range of the structural limit state function when all uncertainty parameters are volatile. In this case, the response range is [g l ,g u ], then Δg=g u -g l θ i ∈[0,1]; sensitivity factor θ i It is used to analyze the influence of the uncertainty parameter value range on the change of the structural response value range, θ i The larger it is, the more sensitive the structural response is to the uncertainty parameter, and the greater the influence of the uncertainty parameter on the structure; otherwise, the smaller it is.
5. The tunnel non-probabilistic reliability analysis method based on symbolic regression algorithm according to claim 4, It is characterized in that Analyze the impact of various uncertainty parameters on tunnel reliability, including: Analyze the impact of fluctuations in uncertainty parameters on non-probabilistic reliability indicators of tunnel structures; Analyze the influence of uncertainty parameter value range on the response range of tunnel structure limit state function; Analyze the relationship between the fluctuation amplitude of uncertainty parameters and the sensitivity factor.
