An intelligent design method for long-distance parallel tunnel lines

Abstract

The present invention discloses an intelligent design method for long-distance parallel tunnel routes. The method comprises: obtaining the starting and ending sections of existing and newly constructed tunnels, and constructing an independent variable value matrix; constructing an objective function and multiple constraints based on the tunnel variables of the existing and newly constructed tunnels and the independent variable value matrix; and constructing a multi-objective programming model based on the objective function and all the constraints; iterating the multi-objective programming model based on a pre-constructed improved particle swarm algorithm; terminating the iteration and outputting the iteration result of that iteration when the fitness value reaches a preset threshold, and generating a target design result based on the iteration result of the current iteration. The present invention combines multiple constraints and corresponding objective functions, enabling accurate target design results to be obtained.

Description

Technical Field

[0001] The present invention relates to the technical field of tunnel engineering, and in particular to a method, system, terminal and computer-readable storage medium for intelligent design of long-distance parallel tunnel lines. Background Art

[0002] With the increasing density of urban underground space development, the demand for long-distance, dual-track parallel tunnel projects is growing. Such designs must balance multiple factors, including surface settlement control, the safety of existing tunnel structures, construction sequence optimization, and complex geological conditions. Traditional design methods rely on engineer experience combined with finite element simulations. These methods suffer from low efficiency, high subjectivity, and difficulty in optimizing the global three-dimensional dynamic routing. However, with the advancement of computer technology, computer-aided design (CAD) can, to a certain extent, assist or even replace engineers' work.

[0003] However, there are significant limitations when abstracting the design of long-distance dual-track parallel tunnels into a multi-objective planning model and using algorithms to solve it. The current objective model is limited to short-distance tunnels and assumes that the two lines maintain a static parallel relationship. It cannot adapt to the three-dimensional characteristics of dynamic adjustment of burial depth, clearance, and relative position in long-distance scenarios. Existing methods also simplify three-dimensional line design into a two-dimensional plane problem, resulting in model distortion and an inability to reflect the changes in the spatial curvature of the tunnel axis in real projects, making it difficult to achieve accurate design of long-distance dual-track tunnels.

[0004] Therefore, the existing technology still needs to be improved and developed. Summary of the Invention

[0005] The main purpose of the present invention is to provide a method, system, terminal and computer-readable storage medium for the intelligent design of long-distance parallel tunnel lines, aiming to solve the significant limitations of the existing technology when abstracting the design of long-distance dual-line parallel tunnels into a multi-objective planning model and using algorithms to solve it. The current target model is limited to short-distance tunnels and assumes that the two lines maintain a static parallel relationship. It cannot adapt to the three-dimensional characteristics of dynamic adjustment of burial depth, clear distance and relative position in long-distance scenarios. In addition, the existing method simplifies the three-dimensional line design into a two-dimensional plane problem, resulting in model distortion and inability to reflect the changes in the spatial curvature of the tunnel axis in real projects, thereby making it difficult to achieve accurate design of long-distance dual-line tunnels.

[0006] To achieve the above-mentioned object, the present invention provides a method for intelligently designing a long-distance parallel tunnel line, the method comprising the following steps:

[0007] Obtaining the starting section and the ending section of the existing tunnel and the newly built tunnel, and constructing an independent variable value matrix based on the starting section and the ending section;

[0008] constructing an objective function and multiple constraints based on the tunnel variables of the existing tunnel and the newly built tunnel and the independent variable value matrix, and constructing a multi-objective programming model based on the objective function and all the constraints;

[0009] Iterating the multi-objective programming model according to a pre-constructed improved particle swarm algorithm, and after each iteration, generating target parameters based on the trained first generalized regression neural network and the trained second generalized regression neural network, generating lining displacement and surface settlement according to the target parameters, and calculating a current fitness value according to the lining displacement, the surface settlement, and a preset fitness function;

[0010] When the fitness value reaches the preset threshold, the iteration ends and the iteration result of the iteration is output, and the target design result is generated according to the iteration result of the current iteration.

[0011] Optionally, the iterating the multi-objective programming model according to the pre-built improved particle swarm algorithm further includes:

[0012] According to the objective function, a weighted total objective function is constructed using a weighted method;

[0013] According to all the constraints, a penalty function coefficient of each constraint is constructed using a penalty function method, and a penalty function is generated according to the penalty function coefficients of all the constraints and the weighted total objective function;

[0014] A fitness function is constructed according to the weighted total objective function and the penalty function.

[0015] Optionally, the iterating the multi-objective programming model according to the pre-built improved particle swarm algorithm further includes:

[0016] generating first training data and second training data based on the two-dimensional finite element model, wherein the first training data includes a first training tunnel variable and a training lining displacement parameter, and the second training data includes a second training tunnel variable and a training surface settlement parameter, and both the first training tunnel variable and the second training tunnel variable include a training tunnel diameter, a training tunnel burial depth, a training stratum loss rate, and a relative position of the training tunnel;

[0017] Training a first generalized regression neural network according to the first training data, and training a second generalized regression neural network according to the second training data;

[0018] After the training is completed, a trained first generalized regression neural network and a trained second generalized regression neural network are obtained.

[0019] Optionally, obtaining the starting section and the ending section of the existing tunnel and the newly built tunnel, and constructing an independent variable value matrix according to the starting section and the ending section, specifically includes:

[0020] Obtain the starting and ending sections of existing and new tunnels;

[0021] Dividing the existing tunnel and the newly built tunnel according to the starting section and the ending section to obtain an interval to be optimized;

[0022] The target independent variable is set, and an independent variable value matrix is ​​constructed according to the interval to be optimized and the target independent variable.

[0023] Optionally, constructing an objective function and multiple constraints based on the tunnel variables of the existing tunnel and the newly built tunnel and the independent variable value matrix, and constructing a multi-objective programming model based on the objective function and all the constraints specifically includes:

[0024] Constructing a surface settlement objective function, a lining displacement objective function, a tunnel clearance constraint condition, a slope constraint condition, a vertical curve radius constraint condition, and a circular curve radius constraint condition based on the tunnel variables of the existing tunnel and the new tunnel and the independent variable value matrix;

[0025] Constructing the objective function according to the surface settlement objective function and the lining displacement objective function;

[0026] Determine whether there is a target section between the existing tunnel and the newly built tunnel; if there is no target section between the existing tunnel and the newly built tunnel, construct a multi-objective programming model based on the objective function and all the constraints.

[0027] Optionally, the determining whether there is a target section between the existing tunnel and the newly-built tunnel further includes:

[0028] When there is a target section between the existing tunnel and the newly built tunnel, constructing a special area settlement constraint condition;

[0029] A multi-objective programming model is constructed based on the objective function and all current constraints.

[0030] Optionally, after each iteration, generating target parameters based on the trained first generalized regression neural network and the trained second generalized regression neural network, generating lining displacement and surface settlement according to the target parameters, and calculating a current fitness value according to the lining displacement, the surface settlement, and a preset fitness function, specifically includes:

[0031] Obtaining the tunnel diameter, tunnel burial depth, and stratum loss rate of the existing tunnel and the newly built tunnel;

[0032] After each iteration, tunnel variables are obtained according to the tunnel diameters, tunnel burial depths, stratum loss rates of the existing tunnel and the newly built tunnel and the iteration results obtained in the iteration;

[0033] Inputting the tunnel variables into the trained first generalized regression neural network and the trained second generalized regression neural network respectively to generate target parameters;

[0034] The lining displacement and the ground settlement are generated according to the target parameters, and the current fitness value is calculated according to the lining displacement, the ground settlement and a preset fitness function.

[0035] In addition, to achieve the above-mentioned object, the present invention further provides a long-distance parallel tunnel line intelligent design system, wherein the long-distance parallel tunnel line intelligent design system includes:

[0036] An independent variable value matrix construction module is used to obtain the starting section and the end section of the existing tunnel and the newly built tunnel, and to construct an independent variable value matrix based on the starting section and the end section;

[0037] a multi-objective programming model construction module, configured to construct an objective function and a plurality of constraints based on the tunnel variables of the existing tunnel and the newly built tunnel and the independent variable value matrix, and to construct a multi-objective programming model based on the objective function and all the constraints;

[0038] an iteration module, configured to iterate the multi-objective programming model according to a pre-constructed improved particle swarm algorithm, and after each iteration, generate target parameters based on the trained first generalized regression neural network and the trained second generalized regression neural network, generate lining displacement and surface settlement according to the target parameters, and calculate a current fitness value according to the lining displacement, the surface settlement, and a preset fitness function;

[0039] The result generation module is used to end the iteration and output the iteration result of the iteration when the fitness value reaches a preset threshold, and generate the target design result according to the iteration result of the current iteration.

[0040] In addition, to achieve the above-mentioned purpose, the present invention also provides a terminal, wherein the terminal includes: a memory, a processor, and a long-distance parallel tunnel line intelligent design program stored in the memory and runnable on the processor, and when the long-distance parallel tunnel line intelligent design program is executed by the processor, the steps of the long-distance parallel tunnel line intelligent design method as described above are implemented.

[0041] In addition, to achieve the above-mentioned purpose, the present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a long-distance parallel tunnel line intelligent design program, and when the long-distance parallel tunnel line intelligent design program is executed by a processor, the steps of the long-distance parallel tunnel line intelligent design method as described above are implemented.

[0042] In the present invention, the starting section and the end section of the existing tunnel and the newly built tunnel are obtained, and an independent variable value matrix is ​​constructed according to the starting section and the end section; an objective function and multiple constraints are constructed according to the tunnel variables of the existing tunnel and the newly built tunnel and the independent variable value matrix, and a multi-objective programming model is constructed according to the objective function and all the constraints; the multi-objective programming model is iterated according to a pre-constructed improved particle swarm algorithm, and after each iteration, target parameters are generated based on a trained first generalized regression neural network and a trained second generalized regression neural network, lining displacement and surface settlement are generated according to the target parameters, and a current fitness value is calculated according to the lining displacement, the surface settlement and a preset fitness function; when the fitness value reaches a preset threshold, the iteration is terminated and the iterative result of the iteration is output, and a target design result is generated according to the iterative result of the current iteration. The present invention constructs an independent variable value matrix through discrete section coordinate parameterization, thereby obtaining a three-dimensional dynamic optimization model, breaking through the traditional two-dimensional simplified assumptions and quantifying complex constraints such as slope and vertical curve; integrating curve theory with generalized regression neural networks, establishing a multi-objective planning model, realizing full-section settlement distribution prediction and lining multi-index displacement analysis, replacing finite element simulation, and combining directional penalty terms to improve the particle swarm algorithm, significantly improving the convergence speed of high-dimensional variables and the quality of Pareto solution sets, so that the present invention can obtain accurate target design results. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 It is a flow chart of a preferred embodiment of the intelligent design method for long-distance parallel tunnel lines of the present invention;

[0044] Figure 2 This is a schematic cross-sectional view of the intelligent design method for long-distance parallel tunnel lines of the present invention;

[0045] Figure 3 This is a schematic diagram of the proportion of surface settlement levels in the intelligent design method for long-distance parallel tunnel lines of the present invention;

[0046] Figure 4 This is a schematic diagram of the lining displacement levels in the intelligent design method for long-distance parallel tunnel lines of the present invention;

[0047] Figure 5 This is a schematic diagram of the tunnel clearance in the intelligent design method for long-distance parallel tunnel lines of the present invention;

[0048] Figure 6 This is a schematic diagram of settlement control in a special section of the intelligent design method for long-distance parallel tunnel lines of the present invention;

[0049] Figure 7 This is a schematic diagram of the meaning of tunnel variables in the intelligent design method for long-distance parallel tunnel lines of the present invention;

[0050] Figure 8 It is a schematic diagram of a finite element model in the intelligent design method for long-distance parallel tunnel lines of the present invention;

[0051] Figure 9 This is a schematic diagram of the lining displacement prediction effect in the intelligent design method for long-distance parallel tunnel lines of the present invention;

[0052] Figure 10 This is a schematic diagram of the improved particle swarm algorithm in the intelligent design method for long-distance parallel tunnel lines of the present invention;

[0053] Figure 11 This is a structural diagram of a preferred embodiment of the intelligent design system for long-distance parallel tunnel lines of the present invention;

[0054] Figure 12 FIG. 4 is a structural diagram of a preferred embodiment of the terminal of the present invention. DETAILED DESCRIPTION

[0055] In order to make the purpose, technical solutions and advantages of the present invention more clear and distinct, the present invention is further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0056] With the increasing density of urban underground space development, the demand for long-distance, dual-track parallel tunnel projects is growing. Such designs must balance multiple factors, including surface settlement control, the safety of existing tunnel structures, construction sequence optimization, and complex geological conditions. Traditional design methods rely on engineer experience combined with finite element simulations. These methods suffer from low efficiency, high subjectivity, and difficulty in optimizing the global three-dimensional dynamic routing. However, with the advancement of computer technology, computer-aided design (CAD) can, to a certain extent, assist or even replace engineers' work.

[0057] In the field of intelligent optimization algorithms, due to their global search capabilities and adaptability to multi-objective problems, intelligent optimization technology has been widely applied to tunnel route design. Currently, complex engineering problems are typically abstracted into multi-objective planning models and solved using algorithms. However, this process has significant limitations: First, multi-objective models are limited to short-distance tunnels and assume a static parallel relationship between two lines, making them unable to adapt to the three-dimensional characteristics of long-distance scenarios, where depth, clearance, and relative position are dynamically adjusted. Second, existing methods simplify three-dimensional route design into a two-dimensional planar problem, resulting in model distortion and an inability to reflect the changes in spatial curvature of the tunnel axis in real projects. These shortcomings make it difficult for traditional methods to achieve dynamic clearance control in long-distance dual-track tunnels, graded settlement in special areas (such as millimeter-level settlement limits in hospital sections), and coordinated optimization of high-dimensional variables (discrete section coordinates).

[0058] In the field of machine learning, current proxy models, with their efficient computing power, quantitative output, and high prediction accuracy, have become a key technology replacing traditional finite element simulation. For the design of dual-track parallel tunnels, existing models have significant limitations: First, they are limited in functionality. Most models focus only on maximum settlement or peak bending moment values ​​and are unable to output full-section settlement distribution curves within the tunnel's influence area. Second, they lack physical interpretability. Existing purely data-driven models ignore prior engineering knowledge such as the Peck curve, resulting in predictions that are disconnected from the actual mechanical mechanisms of the project. These shortcomings make it difficult for traditional proxy models to support the drawing of three-dimensional settlement contours and the calculation of graded settlement control objective functions in the dynamic route design of long-distance dual-track tunnels.

[0059] In response to one or more of the above problems, the present invention obtains the starting section and the end section of the existing tunnel and the newly built tunnel, and constructs an independent variable value matrix based on the starting section and the end section; constructs an objective function and multiple constraints based on the tunnel variables of the existing tunnel and the newly built tunnel and the independent variable value matrix, and constructs a multi-objective programming model based on the objective function and all the constraints; iterates the multi-objective programming model according to a pre-constructed improved particle swarm algorithm, and after each iteration, generates target parameters based on a trained first generalized regression neural network and a trained second generalized regression neural network, generates lining displacement and surface settlement based on the target parameters, and calculates the current fitness value based on the lining displacement, the surface settlement and a preset fitness function; when the fitness value reaches a preset threshold, ends the iteration and outputs the iterative result of the iteration, and generates a target design result based on the iterative result of the current iteration.

[0060] The intelligent design method for long-distance parallel tunnel lines described in the preferred embodiment of the present invention is as follows: Figure 1 As shown, the intelligent design method for long-distance parallel tunnel lines includes the following steps:

[0061] Step S10: Obtain the starting section and the ending section of the existing tunnel and the newly-built tunnel, and construct an independent variable value matrix according to the starting section and the ending section.

[0062] Specifically, the present invention addresses the intelligent design of long-distance, dual-track tunnels, addressing the issue of their three-dimensional alignment. This is then addressed through intelligent optimization algorithms. Therefore, for both existing and newly constructed tunnels, the present invention discretizes the horizontal and vertical cross-sectional data of long-distance, dual-track tunnels into a 4k-dimensional matrix of numerical variables. Dynamic coordinate parameterization is then used to characterize the spatial variations in tunnel slope, vertical curve radius, and planar alignment, overcoming the limitations of traditional two-dimensional parallel tunnel models.

[0063] Specifically, obtaining the starting section and the ending section of the existing tunnel and the newly built tunnel, and constructing an independent variable value matrix according to the starting section and the ending section, specifically includes:

[0064] Obtain the starting and ending sections of existing and new tunnels;

[0065] Dividing the existing tunnel and the newly built tunnel according to the starting section and the ending section to obtain an interval to be optimized;

[0066] The target independent variable is set, and an independent variable value matrix is ​​constructed according to the interval to be optimized and the target independent variable.

[0067] In the present invention, Figure 2 As shown, ET represents an existing tunnel, and NT represents a newly built tunnel. Indicates the existing tunnel k The center point of the cross section x coordinate, Indicates the existing tunnel k The center point of the cross section z coordinate, Indicates the newly built tunnel k The center point of the cross section x coordinate, Indicates the newly built tunnel k The center point of the cross section z coordinate, k for y 0 section to y n+1 The serial number of a section in the section is constructed based on the starting and ending sections of the existing tunnel and the new tunnel. Figure 2 The longitudinal section shown in (a) in the figure is used to obtain the double-track tunnel to be optimized, that is, y 0 section built to y n+1 cross-section, where ySection 0 is the starting section of the existing tunnel and the new tunnel. y n+1 The cross section is the terminal section of the existing tunnel and the new tunnel. Figure 2 (b) in the figure shows the corresponding plan view. O is the corresponding coordinate origin, Figure 2 (c) in the figure indicates the relative positions of the existing tunnel and the new tunnel.

[0068] in, y 0 section and y n+1 On the cross section, the coordinates of the tunnel center point are known. y 0 section to y n+1 The area between the sections is the area to be optimized and requires computer-aided design. The total length of the computer-aided design is L ,Will L Divided into n+1 intervals. y k For example, the cross section of the two tunnels x Coordinates and z The coordinates are unknown, so there are 4 independent variables on each section. For the entire line, there are 4 k Therefore, after discretizing the coordinates of the tunnel center point, the engineering problem of 3D design optimization is transformed into 4 k The specific independent variable value matrix is ​​as follows:

[0069] ;

[0070] in, Represents the x-coordinate set of the center points of n sections of the existing tunnel, Represents the z-coordinate set of the center points of n sections of the existing tunnel, Represents the x-coordinate set of the center points of the n sections of the newly built tunnel, Represents the z coordinate set of the center points of the n sections of the newly built tunnel. Regarding the coordinates of the discretized tunnel center point, In the interval, both tunnels are straight segments, starting from y k-1 Cross section, end point y k Section; the number of sections to be optimized should be determined according to actual needs; the discretized 3D double-track tunnel can be represented in the form of n 2D section combinations; that is, the above representation, in the present invention, a 2D machine learning agent model can be established in the subsequent solution. The surface settlement and lining displacement within the section can be accurately predicted by the proxy model y k-1 andy k The values ​​on the cross section are approximated by interpolation.

[0071] Step S20: construct an objective function and multiple constraints based on the tunnel variables of the existing tunnel and the newly built tunnel and the independent variable value matrix, and construct a multi-objective programming model based on the objective function and all the constraints.

[0072] Specifically, in the present invention, for the independent variable value matrix, the present invention constructs the objective function and constraint conditions accordingly, thereby obtaining a multi-objective planning model, and converting the corresponding long-distance dual-line parallel tunnel design problem into a further model solving problem.

[0073] Furthermore, constructing an objective function and multiple constraints based on the tunnel variables of the existing tunnel and the newly built tunnel and the independent variable value matrix, and constructing a multi-objective programming model based on the objective function and all the constraints, specifically includes:

[0074] Constructing a surface settlement objective function, a lining displacement objective function, a tunnel clearance constraint condition, a slope constraint condition, a vertical curve radius constraint condition, and a circular curve radius constraint condition based on the tunnel variables of the existing tunnel and the new tunnel and the independent variable value matrix;

[0075] Constructing the objective function according to the surface settlement objective function and the lining displacement objective function;

[0076] Determine whether there is a target section between the existing tunnel and the newly built tunnel; if there is no target section between the existing tunnel and the newly built tunnel, construct a multi-objective programming model based on the objective function and all the constraints.

[0077] Specifically, in the present invention, corresponding functions and constraints are generated according to the tunnel variables of the existing tunnel and the newly-built tunnel and the independent variable value matrix.

[0078] Among them, for the surface subsidence objective function, this case adopts the grade proportion weighted method to quantify it. The specific implementation method of the "grade proportion weighted method" is as shown in Table 1. s Indicates surface settlement, and all settlement values ​​within a certain range are expressed in levels; then, Figure 3 As shown in the figure, a surface subsidence classification map is drawn, where x indivual y Indicates the coordinate axis constructed in the tunnel cross section. The surface settlement classification map is actually a special settlement contour map. Calculate the area of ​​each grade in the contour map, and then calculate the area of ​​the first grade in turn. si The proportion of a given area Asi ; Finally, the proportion of each level in a given area A si Perform weighted summation, and the summation result is the objective function of surface subsidence, that is, the objective function of surface subsidence O s .

[0079] Table 1: Surface subsidence grades

[0080]

[0081] A si Satisfy the following formula:

[0082] ;

[0083] Correspondingly, the surface subsidence objective function O s Expressed as:

[0084] ;

[0085] in, A s1 Indicates the proportion of surface settlement with level 1, A s2 Indicates the proportion of surface settlement of level 2, A s3 Indicates the proportion of surface subsidence of level 3, A s4 Indicates the proportion of surface subsidence of level 4, i Indicates the corresponding level. It should be noted that the "value range" column in Table 1 of the present invention can be adjusted according to the actual project situation, and the number of levels can also be arbitrary. Table 1 of the present invention only lists one possible situation. In the surface settlement objective function, A si The coefficients before can also be adjusted according to the actual situation of the project. For example, in order to avoid excessive settlement in the actual project, it is recommended to increase A s1 The coefficient before.

[0086] For lining displacement, similar to surface settlement, the overall idea of ​​constructing the lining displacement objective function is to use lining displacement (lining displacement refers to the displacement of the existing tunnel lining), and the same level proportion weighting method is used. However, the specific construction process of the lining displacement objective function is different from that of surface settlement. The specific construction process is: first, according to the set independent variable, that is, the maximum horizontal displacement u x , maximum vertical displacement u y and maximum radial convergence u rThree indicators to find the most dangerous displacement u , expressed as the following formula:

[0087] ;

[0088] Afterwards, as shown in Table 2, all lining displacements within a certain value range are expressed in grades. u As an indicator, a lining displacement classification diagram is drawn, such as Figure 4 As shown, u The most dangerous displacement, and y represents the longitudinal coordinate of the tunnel. The lining displacement classification diagram is actually one dimension lower than the surface settlement classification diagram. This is because the surface settlement is in the form of a curve on each section, while the lining displacement has only one single value. u , the lining displacement classification diagram is actually another form of yu Curve (tunnel longitudinal coordinate - most dangerous displacement curve); then calculate the cumulative length of each level during the entire tunnel construction process (different from the area in the surface settlement map), and then calculate the first ui The proportion of grade to total tunnel length A ui ; Finally, the proportion of each level in a given area A ui Perform weighted summation, and the summation result is the lining displacement objective function O u , expressed as the following formula:

[0089] ;

[0090] in, A u1 Indicates the proportion of the first level to the total length of the tunnel, A u2 Indicates the proportion of the second level to the total length of the tunnel, A u3 Indicates the proportion of the third level to the total length of the tunnel, A u4 Indicates the proportion of the 4th level to the total length of the tunnel.

[0091] Table 2: Lining displacement grades

[0092]

[0093] Regarding tunnel clearance constraints, a too small clearance between tunnels can cause harmful effects on the structure between the two tunnels, such as stress concentration in the surrounding rock and structural deformation, thus affecting the long-term stability and safety of the tunnels. Furthermore, a too small clearance between tunnels can increase construction difficulty and project costs. Therefore, a minimum clearance between tunnels should be limited.

[0094] In the present invention, it is assumed that between any two adjacent sections, both tunnels are straight lines. y k Section to y k+1 Cross section, tunnel clearance at position y P(y) Satisfy the following formula:

[0095] ;

[0096] in, y k ≤ y ≤ y k+1 , D represents the tunnel diameter, and 、 、 and is an intermediate variable, and the calculation formula is as follows:

[0097] ;

[0098] ;

[0099] ;

[0100] ;

[0101] At this point, the problem of minimum clearance between any two adjacent sections of the tunnel is transformed into a quadratic function maximum value problem. Figure 5 As shown, the quadratic function P(y) Minimum MP k It can be obtained in three situations, that is, for two tunnels with a given starting point and end point, the minimum clear distance between the tunnels may appear in three positions: starting position I, that is Figure 5 The section where a is located, the end position is III, that is, Figure 5 The section where e is located, and the middle position Ⅱ, that is Figure 5 The middle position of the section where c is located is Figure 5 Between the sections b and d.

[0102] For the starting position, at this time, y = y k , we can get the following formula:

[0103] ;

[0104] or ;

[0105] For the end position, at this time, y = y k+1 , we can get the following formula:

[0106] ;

[0107] or ;

[0108] For the middle position, MP k If we take the quadratic function on its axis of symmetry, we can get the following formula:

[0109] ;

[0110] ;

[0111] In the present invention, restrictions are imposed on the clearance distance of tunnels in all sections, and the minimum clearance distance of the two tunnels should satisfy the formula corresponding to the terminal position; and due to the discretization of the tunnel sections, the following formula is satisfied:

[0112] ;

[0113] ;

[0114] in, Indicates the minimum value set. y k ≤ y ≤ y k+1 .

[0115] Regarding the slope constraint, steep slopes increase the need for ventilation within the tunnel, and larger slopes can affect vehicle handling and ride comfort. This is especially true in long tunnels, where the slope's impact on ventilation and comfort is more significant. To address multiple factors, including environmental impact, ventilation requirements, economy, safety, and comfort, it's necessary to limit the maximum slope of the tunnel. Therefore, the slope in the tunnel must satisfy the following equation:

[0116] ;

[0117] Among them, Gr Indicates the actual slope of the tunnel, Gr max Indicates the maximum slope allowed in the tunnel. The slope is specifically expressed by the absolute value of the ratio of the longitudinal change to the lateral change between the two sections. Gr When >0, it indicates uphill; GrWhen <0, it indicates a downhill slope. In the present invention, since the longitudinal coordinates of the long tunnel have been discretized, the maximum slope limit can also be expressed as the following formula:

[0118] ;

[0119] in, Indicates the actual slope of the existing tunnel, Indicates the actual slope of the newly built tunnel.

[0120] Regarding vertical curve radius constraints, when a vehicle travels on a vertical curve in a tunnel, especially at night, if the radius is too small, the headlight illumination distance will be too short, affecting driving safety and speed. Furthermore, when a vehicle travels on a concave vertical curve, it will be affected by centrifugal force. The smaller the radius, the greater the centrifugal force. Therefore, it is necessary to set a limit on the minimum vertical curve radius in the tunnel. The vertical curve radius should satisfy the following formula:

[0121] ;

[0122] in, Rv Indicates the vertical curve radius of the tunnel, Indicates the minimum radius of a concave vertical curve, Indicates the minimum radius of a convex vertical curve. Rv When >0, it means that the segment is a convex vertical curve; when Rv When <0, the curve is a concave vertical curve. "convex" represents a convex vertical curve, while "concava" represents a concave vertical curve. Concave and convex curves have different physical properties. Concave curves focus primarily on the effects of centrifugal force and headlight illumination distance, while convex curves focus more on sight distance and the driver's line of sight. Therefore, the minimum radius restrictions for concave and convex vertical curves are also different.

[0123] In actual engineering, the vertical curve radius refers to the curvature radius of a curve set at the turning point of two slope sections on the longitudinal section. In the present invention, the longitudinal coordinates of the long tunnel have been discretized, so the vertical curve radius can be approximately calculated. For the vertical curve radius near a certain section position, the calculation method is as follows: first, obtain three points on the same curve near the section, and express the curves corresponding to these three points with a quadratic function. Then, use the coordinates of these three points to find the analytical expression of the quadratic function; finally, find the curvature radius of the quadratic function at the target point, and use it to approximate the vertical curve radius of the tunnel at that point. In the present invention, restrictions are placed on the vertical curve radius of all positions (including the starting position and the receiving position), and each section should satisfy the following formula:

[0124] ;

[0125] in, Indicates the vertical curve radius of the existing tunnel, Indicates the vertical curve radius of the new tunnel.

[0126] For the circular curve radius constraint, when the circular curve radius Rc If the radius is too small, vehicles will be subjected to greater centrifugal force when passing through the curve, which may lead to deterioration of driving conditions and even endanger driving safety. In addition, the radius of the circular curve will also affect driving comfort and technical and economic efficiency. Therefore, the minimum circular curve radius of the tunnel should be restricted and should meet the following formula:

[0127] ;

[0128] in, Rc min Represents the minimum circular curve radius. It should be noted that, similar to the vertical curve radius, in actual engineering, the circular curve radius refers to the curvature radius of a curve set at the turning point of two slope sections on the plan view. In the present invention, the longitudinal coordinates of the long tunnel have been discretized, so the circular curve radius can be approximately calculated. The calculation method of the circular curve radius is the same as the method for calculating the vertical curve; in the present invention, restrictions are made on the circular curve radius of all positions (including the starting position and the receiving position), and each section should satisfy the following formula:

[0129] ;

[0130] in, represents the circular curve radius of the existing tunnel, Indicates the circular curve radius of the new tunnel.

[0131] Furthermore, the determining whether there is a target section between the existing tunnel and the newly-built tunnel further includes:

[0132] When there is a target section between the existing tunnel and the newly built tunnel, constructing a special area settlement constraint condition;

[0133] A multi-objective programming model is constructed based on the objective function and all current constraints.

[0134] Specifically, when there is a target section between the existing tunnel and the newly built tunnel, the settlement constraint condition of the special area should be considered and the settlement constraint condition of the special area should also be used as a constraint condition to construct a multi-objective planning model.

[0135] Among them, the target areas are pre-set special areas, such as hospitals and universities.

[0136] That is, when the tunnel approaches certain special sections, it is necessary to strictly control the settlement for reasons such as ensuring safety and maintaining precision instruments. This control adopts different control standards according to the specific sections, such as Figure 6 As shown, settlement control of the tunnel near the hospital, where x 1 and x 2 represents the horizontal coordinate of the hospital's location, y 1 and y 2 represents the vertical coordinate of the hospital's location. The specific control requirements are shown in the following formula:

[0137] ;

[0138] in, s ( x , y ) represents the settlement at the corresponding position, s s represents the restrained settlement, x and y Indicates the corresponding position.

[0139] After obtaining the objective function and all the constraints, a multi-objective planning model is constructed accordingly. In the present invention, for the case where there is no target section between the existing tunnel and the newly built tunnel, the multi-objective planning model is expressed as follows:

[0140] ;

[0141] ;

[0142] Among them, st (subject to) means constrained, and when considering the target area, the corresponding special area settlement constraint conditions are added to the conditions.

[0143] Step S30: Iterate the multi-objective programming model according to the pre-constructed improved particle swarm algorithm, and after each iteration, generate target parameters based on the trained first generalized regression neural network and the trained second generalized regression neural network, generate lining displacement and surface settlement according to the target parameters, and calculate the current fitness value according to the lining displacement, the surface settlement and a preset fitness function.

[0144] In this invention, based on actual engineering requirements, a dual objective function is constructed: a surface settlement objective function and a lining displacement objective function. Simultaneously, rigid civil engineering code requirements such as slope limits, vertical curve radius, circular curve radius, and dynamic clearance are converted into mathematical constraints to ensure that the spacing between tunnels in any adjacent sections meets safety thresholds. This multi-objective planning problem and the mathematical model optimization challenge of high-dimensional variables (4k dimensions) are modeled as a target programming model and solved using a particle swarm algorithm. The objective function and constraints in the multi-objective planning are used to calculate the fitness values ​​of the particles in the particle swarm algorithm, thereby linking the algorithm to the mathematical model. The particle swarm algorithm iterates continuously. If the result generated by a particular iteration meets the required design, the result is output; otherwise, if it does not meet the requirements, the tunnel route design is adjusted and the iteration continues. During the intelligent optimization algorithm's solution process, each iteration generates a large number of different variable values ​​(i.e., different horizontal and vertical sections of the dual-track tunnel), and the surface settlement and lining displacement corresponding to each horizontal and vertical section vary. Therefore, to rapidly calculate surface settlement and lining displacement under different horizontal and vertical sections, a machine learning proxy model—the first generalized regression neural network and the second generalized regression neural network—should be established. The primary function of the proxy model is to replace the original complex model with a simplified model with less computational effort, thereby improving computational efficiency and accelerating the decision-making process.

[0145] Furthermore, the multi-objective programming model is iterated according to the pre-built improved particle swarm algorithm, and the method further includes:

[0146] According to the objective function, a weighted total objective function is constructed using a weighted method;

[0147] According to all the constraints, a penalty function coefficient of each constraint is constructed using a penalty function method, and a penalty function is generated according to the penalty function coefficients of all the constraints and the weighted total objective function;

[0148] A fitness function is constructed according to the weighted total objective function and the penalty function.

[0149] Specifically, for the fitness function used in the process of solving the multi-objective programming model, the present invention considers using the particle swarm algorithm to solve the multi-objective programming model. There are two problems: multi-objective programming problems are often difficult to achieve the optimal solution for multiple objectives at the same time, and there is no unique solution; and the particle swarm algorithm cannot directly solve constrained optimization problems.

[0150] In this regard, the present invention adopts the weighted method to solve the multi-objective programming problem, which is often difficult to achieve the optimal solution in multiple objectives at the same time, and there is no unique solution. The weighted method is a classic method for solving multi-objective programming problems. Its basic idea is to convert multiple objective functions into a single weighted sum by assigning different weights to each objective function, thereby converting it into a single-objective optimization problem. When the weighted method is applied in the present invention, appropriate weights are assigned to each objective function according to the needs of the actual problem; then, by optimizing this weighted sum objective function, a solution is obtained, and the objective function value corresponding to this solution is the optimal solution under the given weight. However, since the weighted method only focuses on one weighted objective through weighted sum, the result of the solution is often a point on a specific Pareto (primary and secondary graph) frontier. Therefore, in order to obtain the entire Pareto frontier, the present invention usually needs to run the weighted method multiple times, adjusting different weight combinations each time. Specifically, the weighted total objective function obtained by the weighted method is shown in the following formula:

[0151] ;

[0152] in, , O represents the function value of the weighted total objective function, w s and w u Indicates the corresponding weight.

[0153] To address the inability of the particle swarm algorithm to directly solve constrained optimization problems, the present invention employs a penalty function approach. In this invention, intelligent tunnel route design requires satisfying multiple constraints, and for calculating these constraints, the present invention employs a penalty function approach. The basic idea behind the penalty function approach is to transform a constrained problem into an unconstrained optimization problem by adding penalties for constraint violations to the objective function. Specifically, by adding an additional penalty term to the objective function, the solution is forced to satisfy the constraints. Taking tunnel slope as an example, the following equation can be obtained:

[0154] ;

[0155] ;

[0156] in, For ET The penalty function coefficient for slope within the interval. When the slope is less than the specified limit slope, the coefficient is 0. is the penalty function coefficient of the slope of ET in the entire optimization interval L. The calculation of the penalty function coefficients of other constraints is similar to that of the slope. Therefore, the value of the penalty function is expressed as follows:

[0157] ;

[0158] in, It represents the penalty function coefficient of the slope of the newly built tunnel in the entire optimization interval L; It represents the penalty function coefficient of the vertical curve radius of the existing tunnel in the entire optimization interval L; It represents the penalty function coefficient of the vertical curve radius of the newly built tunnel in the entire optimization interval L; It represents the penalty function coefficient of the circular curve radius of the newly built tunnel in the entire optimization interval L; It represents the penalty function coefficient of the circular curve radius of the existing tunnel in the entire optimization interval L; It represents the penalty function coefficient of the existing tunnel clearance in the entire optimization interval L; It represents the penalty function coefficient of the net distance of the newly built tunnel within the entire interval L to be optimized.

[0159] Adding the objective function O and the penalty function P gives the value of the fitness function F, as shown in the following formula:

[0160] .

[0161] Furthermore, the multi-objective programming model is iterated according to the pre-built improved particle swarm algorithm, and the method further includes:

[0162] generating first training data and second training data based on the two-dimensional finite element model, wherein the first training data includes a first training tunnel variable and a training lining displacement parameter, and the second training data includes a second training tunnel variable and a training surface settlement parameter, and both the first training tunnel variable and the second training tunnel variable include a training tunnel diameter, a training tunnel burial depth, a training stratum loss rate, and a relative position of the training tunnel;

[0163] Training a first generalized regression neural network according to the first training data, and training a second generalized regression neural network according to the second training data;

[0164] After the training is completed, a trained first generalized regression neural network and a trained second generalized regression neural network are obtained.

[0165] In this paper, two types of proxy models are constructed. For lining displacement prediction, a generalized regression neural network (GRNN) is used to directly establish a nonlinear mapping relationship between input variables (such as tunnel diameter D and existing tunnel depth C) and output targets (maximum horizontal displacement, vertical displacement, and radial convergence), with the maximum of these three variables used as the most dangerous displacement indicator. For surface settlement prediction, another GRNN is used to predict the unknown parameters in the surface settlement curve formula for a dual-track tunnel. The predicted values ​​of these parameters are substituted back into the settlement curve formula to generate the entire surface settlement curve. Through a knowledge-data dual-driven strategy, the proxy model retains the interpretability of physical laws while overcoming the reliance on experience of traditional methods, providing high-precision, low-computational cost objective function feedback for the optimization algorithm.

[0166] Specifically, in the process of establishing the data set, the present invention sets 6 key variables, including tunnel diameter D , Existing tunnel burial depth C , Existing tunnel stratum loss rate C ref ET , new tunnel stratum loss rate C ref NT , horizontal distance between the existing tunnel and the new tunnel Δx , vertical distance between the existing tunnel and the new tunnel Δz One way to take values ​​for these variables is shown in Table 3. The meanings of the variables are as follows: Figure 7 shown.

[0167] Table 3: Variable settings

[0168]

[0169] Among these variables, C , Δx and Δz It simulates the different center point coordinate values ​​of the two tunnels; C ref ET and C ref NT This is to simulate the difference in ground loss rate caused by different shield machines or excavation conditions in the project; D The setting is to adapt to the differences of different types of shield tunnels such as subways and underground expressways. Such 6 variables can cover the different situations of actual projects as much as possible. It is worth noting in the table that due to the limitations of burial depth and tunnel clearance, Δx and Δz Sometimes it is not possible to obtain all the values ​​listed in the table.

[0170] According to the above method, multiple sets of different working conditions can be obtained from the data set. Subsequently, a 2D finite element simulation method will be used to establish a preliminary data set, including the first training tunnel variable and the second training tunnel variable; finite element simulation will be performed using plaxis2D (a general geotechnical finite element calculation software). The model diagram is shown in the figure below. Figure 8 After finite element simulation, surface settlement data and lining displacement data were extracted as the content of the data set, thus obtaining the first and second training data. 80% of the data in the data set was used as the training set, and 20% of the data was used as the test set. The corresponding training was performed to obtain the first and second trained generalized regression neural networks.

[0171] At the same time, in order to verify whether the proxy model, that is, the first generalized regression neural network, is reliable, each data point in the test set is plotted on the horizontal axis with the true value as the horizontal axis and the predicted value as the vertical axis. Figure 9 In. With the coefficient of determination R 2 The error between the true value and the predicted value is quantified by the mean square error RMSE, R 2 The closer it is to 1 or the closer RMSE is to 0, the closer the true value is to the predicted value. Figure 9 It can be seen that for the three output lining displacements, the R 2 All are above 0.90 and RMSE are less than 2.00mm. In addition, Figure 9 The dashed line in indicates that the true value is equal to the predicted value. Figure 9 As can be seen in the corresponding figures (a), (b), and (c), with the exception of a few points, most predictions are close to the dashed line, indicating a small difference between the true and predicted values. These results demonstrate that the prediction error is small and the accuracy of the proxy model meets the requirements of this invention.

[0172] Furthermore, after each iteration, target parameters are generated based on the trained first generalized regression neural network and the trained second generalized regression neural network, lining displacement and surface settlement are generated according to the target parameters, and a current fitness value is calculated according to the lining displacement, the surface settlement, and a preset fitness function, specifically including:

[0173] Obtaining the tunnel diameter, tunnel burial depth, and stratum loss rate of the existing tunnel and the newly built tunnel;

[0174] After each iteration, tunnel variables are obtained according to the tunnel diameters, tunnel burial depths, stratum loss rates of the existing tunnel and the newly built tunnel and the iteration results obtained in the iteration;

[0175] Inputting the tunnel variables into the trained first generalized regression neural network and the trained second generalized regression neural network respectively to generate target parameters;

[0176] The lining displacement and the ground settlement are generated according to the target parameters, and the current fitness value is calculated according to the lining displacement, the ground settlement and a preset fitness function.

[0177] Specifically, although the particle swarm algorithm has unique advantages in processing constrained optimization, its convergence speed, accuracy and global search ability may not be outstanding in the performance of certain problems. Therefore, the present invention uses an improved particle swarm algorithm in the process of using the particle swarm algorithm to solve the multi-objective programming mathematical model. Compared with the traditional particle swarm algorithm, the improved particle swarm algorithm adds an operator that is oriented away from the boundary of the constraint condition, which can accelerate the convergence speed, improve the accuracy and enhance the global search ability. And when the first i The particle d When the generation is updated, the update formula of its position remains unchanged, and the update formula of its speed is as follows:

[0178] ;

[0179] The position update formula is as follows:

[0180] ;

[0181] in, Indicates the i The particle d The speed of generation, ω represents the inertia weight, Indicates the i The particle d -1 generation speed, represents the individual learning factor, represents the social learning factor, represents the correction factor, Indicates the i The particle d Generation's personal best position, Indicates the d The global best position of the generation, Indicates the i The particle d The position of the generation, No. i The particle d The punish function value of the generation, Individual learning radius, represents the social learning radius, represents the correction radius, Indicates the i The particle d -1 generation position.

[0182] like Figure 10 As shown in the improved particle swarm algorithm flow chart, the part of calculating the punish function value and correction direction of each particle is the improved part of the particle swarm algorithm.

[0183] Furthermore, in the present invention, after each iteration of the multi-objective programming model using the improved particle swarm algorithm, the tunnel diameter, tunnel burial depth and stratum loss rate of the existing tunnel and the newly built tunnel are obtained, thereby obtaining tunnel variables based on the tunnel diameter, tunnel burial depth, stratum loss rate of the existing tunnel and the newly built tunnel and the iterative result obtained in this iteration; the tunnel variables are then input into the trained first generalized regression neural network and the trained second generalized regression neural network respectively to generate target parameters, wherein the target parameters include parameters for calculating lining displacement and surface settlement; thereafter, the lining displacement and surface settlement are generated, and the fitness value is calculated to determine whether to end the iteration.

[0184] Among them, for surface settlement, that is, the prediction of surface settlement curve, the present invention is based on the Peck curve theory and combined with the theoretical derivation of the surface settlement curve caused by double-track tunnel construction, and applies the following formula to the surface settlement curve involved in the present invention:

[0185] ;

[0186] in, S ( x ) is the general formula of the double Peck curve, where S 0,ET , l ET , i ET , S 0，NT , l NT , i NT There are 6 undetermined parameters. In practical application, the second generalized regression neural network is used to predict these 6 undetermined parameters under the current situation, and the corresponding surface settlement curve can be obtained by bringing them into the general formula of the double Peck curve. D , C , C ref ET , C ref NT , Δx , Δz), the second generalized regression neural network method is used to calculate these six undetermined parameters, resulting in the corresponding surface settlement curve. This method uses a combination of knowledge-driven and data-driven approaches to predict the settlement curve for a specific section. The data-driven approach uses a generalized regression neural network (GRNN) as the foundational algorithm for constructing a surface settlement proxy model, resulting in the second generalized regression neural network. Overall, the second generalized regression neural network is first used to predict the values ​​of the six undetermined parameters. These parameters are then substituted into the surface settlement curve application formula to derive the entire surface settlement curve.

[0187] Compared with surface settlement, the method used to construct the machine learning agent model of lining displacement, that is, the first generalized regression neural network, is relatively simple; it does not require knowledge-driven, but only uses data-driven methods to predict lining displacement (maximum horizontal displacement u x , maximum vertical displacement u y , maximum convergence u r ).

[0188] Step S40: When the fitness value reaches a preset threshold, the iteration is terminated and the iteration result of the iteration is output, and the target design result is generated according to the iteration result of the current iteration.

[0189] During the iteration process, when the fitness value reaches the preset threshold, it means that the result obtained at this time meets all the constraints, then the iteration ends and the iteration result of this iteration is output, and the target design result is generated according to the iteration result of the current iteration.

[0190] The present invention obtains the starting section and the ending section of the existing tunnel and the newly built tunnel, and constructs an independent variable value matrix based on the starting section and the ending section; constructs an objective function and multiple constraints based on the tunnel variables of the existing tunnel and the newly built tunnel and the independent variable value matrix, and constructs a multi-objective programming model based on the objective function and all the constraints; iterates the multi-objective programming model based on a pre-constructed improved particle swarm algorithm, and after each iteration, generates target parameters based on a trained first generalized regression neural network and a trained second generalized regression neural network, generates lining displacement and surface settlement based on the target parameters, and calculates a current fitness value based on the lining displacement, the surface settlement, and a preset fitness function; when the fitness value reaches a preset threshold, ends the iteration and outputs the iteration result of the iteration, and generates a target design result based on the iteration result of the current iteration. The present invention constructs an independent variable value matrix through discrete section coordinate parameterization, thereby obtaining a three-dimensional dynamic optimization model, breaking through the traditional two-dimensional simplified assumptions and quantifying complex constraints such as slope and vertical curve; integrating curve theory with generalized regression neural networks, establishing a multi-objective planning model, realizing full-section settlement distribution prediction and lining multi-index displacement analysis, replacing finite element simulation, and combining directional penalty terms to improve the particle swarm algorithm, significantly improving the convergence speed of high-dimensional variables and the quality of Pareto solution sets, so that the present invention can obtain accurate target design results.

[0191] Furthermore, if Figure 11 As shown, based on the above-mentioned long-distance parallel tunnel line intelligent design method, the present invention also provides a long-distance parallel tunnel line intelligent design system, wherein the long-distance parallel tunnel line intelligent design system includes:

[0192] An independent variable value matrix construction module 111 is used to obtain the starting section and the end section of the existing tunnel and the newly built tunnel, and to construct an independent variable value matrix according to the starting section and the end section;

[0193] A multi-objective programming model construction module 112 is configured to construct an objective function and multiple constraints based on the tunnel variables of the existing tunnel and the newly built tunnel and the independent variable value matrix, and to construct a multi-objective programming model based on the objective function and all the constraints;

[0194] an iterative module 113, configured to iterate the multi-objective programming model according to a pre-constructed improved particle swarm algorithm, and after each iteration, generate target parameters based on the trained first generalized regression neural network and the trained second generalized regression neural network, generate lining displacement and surface settlement according to the target parameters, and calculate a current fitness value according to the lining displacement, the surface settlement, and a preset fitness function;

[0195] The result generation module 114 is used to end the iteration and output the iteration result of the iteration when the fitness value reaches a preset threshold, and generate the target design result according to the iteration result of the current iteration.

[0196] Furthermore, if Figure 12 As shown, based on the above-mentioned long-distance parallel tunnel line intelligent design method and system, the present invention also provides a terminal, which includes a processor 10, a memory 20 and a display 30. Figure 12 Only some of the components of the terminal are shown, but it should be understood that implementation of all of the shown components is not required, and more or fewer components may be implemented instead.

[0197] In some embodiments, the memory 20 may be an internal storage unit of the terminal, such as the terminal's hard drive or memory. In other embodiments, the memory 20 may also be an external storage device of the terminal, such as a plug-in hard drive, a Smart Media Card (SMC), a Secure Digital (SD) card, a flash memory card, etc. Furthermore, the memory 20 may include both the terminal's internal storage unit and an external storage device. The memory 20 is used to store application software installed in the terminal and various data, such as program code installed in the terminal. The memory 20 may also be used to temporarily store data that has been output or is about to be output. In one embodiment, the memory 20 stores a long-distance parallel tunnel line intelligent design program 40, which can be executed by the processor 10, thereby implementing the long-distance parallel tunnel line intelligent design method of the present invention.

[0198] In some embodiments, the processor 10 may be a central processing unit (CPU), a microprocessor, or other data processing chip, configured to execute program codes or process data stored in the memory 20, such as executing the intelligent design method for long-distance parallel tunnel lines.

[0199] In some embodiments, the display 30 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, an OLED (Organic Light-Emitting Diode) touch screen, etc. The display 30 is used to display information on the terminal and to display a visual user interface.

[0200] In one embodiment, when the processor 10 executes the long-distance parallel tunnel line intelligent design program 40 in the memory 20 , the steps of the above-mentioned long-distance parallel tunnel line intelligent design method are implemented.

[0201] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a long-distance parallel tunnel line intelligent design program, and when the long-distance parallel tunnel line intelligent design program is executed by a processor, the steps of the long-distance parallel tunnel line intelligent design method as described above are implemented.

[0202] It should be noted that, in this document, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or terminal comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or terminal. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or terminal comprising the element.

[0203] Of course, those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing related hardware (such as a processor, controller, etc.) through a computer program. The program can be stored in a computer-readable storage medium that can be read by a computer. When executed, the program can include the processes in the above-described method embodiments. The computer-readable storage medium can be a memory, a magnetic disk, an optical disk, etc.

[0204] It should be understood that the application of the present invention is not limited to the above examples. For ordinary technicians in this field, they can make improvements or changes based on the above description. All these improvements and changes should fall within the scope of protection of the claims attached to the present invention.

Claims

1. A method for intelligent design of long-distance parallel tunnel lines, characterized in that: The intelligent design method for long-distance parallel tunnel lines includes: Obtaining the starting section and the ending section of the existing tunnel and the newly built tunnel, and constructing an independent variable value matrix based on the starting section and the ending section; constructing an objective function and multiple constraints based on the tunnel variables of the existing tunnel and the newly built tunnel and the independent variable value matrix, and constructing a multi-objective programming model based on the objective function and all the constraints; The step of constructing an objective function and multiple constraints based on the tunnel variables of the existing tunnel and the newly built tunnel and the independent variable value matrix, and constructing a multi-objective programming model based on the objective function and all the constraints, specifically includes: Constructing a surface settlement objective function, a lining displacement objective function, a tunnel clearance constraint condition, a slope constraint condition, a vertical curve radius constraint condition, and a circular curve radius constraint condition based on the tunnel variables of the existing tunnel and the new tunnel and the independent variable value matrix; Constructing the objective function according to the surface settlement objective function and the lining displacement objective function; determining whether there is a target section between the existing tunnel and the newly built tunnel, and if no target section exists between the existing tunnel and the newly built tunnel, constructing a multi-objective programming model according to the objective function and all the constraints; Iterating the multi-objective programming model according to a pre-constructed improved particle swarm algorithm, and after each iteration, generating target parameters based on the trained first generalized regression neural network and the trained second generalized regression neural network, generating lining displacement and surface settlement according to the target parameters, and calculating a current fitness value according to the lining displacement, the surface settlement, and a preset fitness function; After each iteration, target parameters are generated based on the trained first generalized regression neural network and the trained second generalized regression neural network, lining displacement and surface settlement are generated according to the target parameters, and a current fitness value is calculated according to the lining displacement, the surface settlement, and a preset fitness function, specifically including: Obtaining the tunnel diameter, tunnel burial depth, and stratum loss rate of the existing tunnel and the newly built tunnel; After each iteration, tunnel variables are obtained according to the tunnel diameters, tunnel burial depths, stratum loss rates of the existing tunnel and the newly built tunnel and the iteration results obtained in the iteration; Inputting the tunnel variables into the trained first generalized regression neural network and the trained second generalized regression neural network respectively to generate target parameters; generating lining displacement and ground settlement according to the target parameters, and calculating a current fitness value according to the lining displacement, the ground settlement and a preset fitness function; When the fitness value reaches the preset threshold, the iteration ends and the iteration result of the iteration is output, and the target design result is generated according to the iteration result of the current iteration.

2. The intelligent design method for long-distance parallel tunnel lines according to claim 1, characterized in that: The multi-objective programming model is iterated according to the pre-built improved particle swarm algorithm, and the method further includes: According to the objective function, a weighted total objective function is constructed using a weighted method; According to all the constraints, a penalty function coefficient of each constraint is constructed using a penalty function method, and a penalty function is generated according to the penalty function coefficients of all the constraints and the weighted total objective function; A fitness function is constructed according to the weighted total objective function and the penalty function.

3. The intelligent design method for long-distance parallel tunnel lines according to claim 1, characterized in that: The multi-objective programming model is iterated according to the pre-built improved particle swarm algorithm, and the method further includes: generating first training data and second training data based on the two-dimensional finite element model, wherein the first training data includes a first training tunnel variable and a training lining displacement parameter, and the second training data includes a second training tunnel variable and a training surface settlement parameter, and both the first training tunnel variable and the second training tunnel variable include a training tunnel diameter, a training tunnel burial depth, a training stratum loss rate, and a relative position of the training tunnel; Training a first generalized regression neural network according to the first training data, and training a second generalized regression neural network according to the second training data; After the training is completed, a trained first generalized regression neural network and a trained second generalized regression neural network are obtained.

4. The intelligent design method for long-distance parallel tunnel lines according to claim 1, characterized in that: The step of obtaining the starting section and the ending section of the existing tunnel and the newly built tunnel, and constructing an independent variable value matrix based on the starting section and the ending section, specifically includes: Obtain the starting and ending sections of existing and new tunnels; Dividing the existing tunnel and the newly built tunnel according to the starting section and the ending section to obtain an interval to be optimized; The target independent variable is set, and an independent variable value matrix is ​​constructed according to the interval to be optimized and the target independent variable.

5. The intelligent design method for long-distance parallel tunnel lines according to claim 1, characterized in that: The determining whether there is a target section between the existing tunnel and the newly built tunnel further includes: When there is a target section between the existing tunnel and the newly built tunnel, constructing a special area settlement constraint condition; A multi-objective programming model is constructed based on the objective function and all current constraints.

Images