A knowledge-data driven method for predicting load-displacement curves of underground structures

Abstract

The present invention discloses a knowledge-data driven method for predicting underground structure load-displacement curves, the method comprising: obtaining underground structure engineering parameters, inputting the underground structure engineering parameters into a prediction model, obtaining intermediate parameter outputs of the prediction model; obtaining parameters of the load-displacement curve according to the intermediate parameter outputs; and establishing a load-displacement curve according to the parameters of the load-displacement curve. The present invention establishes a prediction model for a load-displacement curve through knowledge-data collaborative driving, and predicts four parameters required to establish a complete load-displacement curve function. The present invention verifies through experiments that using a neural network to predict the four parameters required for a load-displacement curve has a good effect, and has improved the error problem of obtaining a load-displacement curve in the prior art.

Description

Technical Field

[0001] The present invention relates to the technical field of underground structures, and in particular to a knowledge-data driven method for predicting load-displacement curves of underground structures. Background Art

[0002] How to accurately measure the overburden pressure of underground structures has always been a difficult problem in underground structure construction. After the existence of the "granary effect" was discovered, the existence of the soil arching effect in sand was proved through the trapdoor experiment. The shear failure and stress redistribution phenomenon of the soil during the descent of the trapdoor is usually called the soil arching effect. Based on this, the existing technology proposed a load-displacement curve, which is an important tool to link underground settlement deformation with the overburden pressure on underground structures. At present, most existing technologies mainly measure the load-displacement curve through trapdoor experiments and numerical simulations to obtain the change of the trapdoor pressure during the descent of the trapdoor.

[0003] In the prior art, most studies on load-displacement curves focus on the magnitude of specific points of the load-displacement curve, such as the minimum soil arch ratio and the final soil arch ratio. The methods used include trapdoor-based experiments and a series of centrifuge and 1g model experiments. Based on the trapdoor experiments, a calculation formula for trapdoor loads under different arch states was proposed according to the arch development on the trapdoor (from the initial curved structure to a triangle and finally a prismatic sliding body with vertical edges). A series of centrifuge and 1g model experiments were used to compare the minimum soil arch ratio and the minimum soil arch ratio displacement under different burial depth conditions.

[0004] But in fact, the load-displacement curve is a continuous curve. It is one-sided to predict the changes of the load-displacement curve based on certain specific points. With the development and research of technology, the existing technology points out that the changes of the load-displacement curve can be predicted by empirical functions through the study of geotechnical engineering experiments and the overall development of vertical loads in engineering projects. A simplified load-displacement curve is proposed, and three straight lines are used to describe the development of the stress on the trapdoor and the displacement of the trapdoor. Based on the three soil arch evolution models in the two-dimensional active trapdoor experiment, the stress distribution ratio in the entire load-displacement curve evolution process is calculated. This method of studying the load-displacement curve is more inclined to divide the load-displacement curve into different stages for research, which may lead to large errors in the calculation process. Therefore, the existing technology further proposes a composite function containing four parameters to describe the load-displacement curve based on the four-phase characteristics of the load-displacement curve, but the calculation of the parameters involves empirical values, which may lead to errors in the parameter acquisition process. This requires a more accurate calculation method to obtain the four parameters of the function model. Summary of the invention

[0005] The purpose of the present invention is to provide a knowledge-data driven method for predicting load-displacement curves of underground structures, aiming to solve the problem of large errors when establishing load-displacement curves in the prior art.

[0006] The technical solution adopted by the present invention to solve the technical problem is as follows:

[0007] The present invention provides a knowledge-data driven underground structure load-displacement curve prediction method, the knowledge-data driven underground structure load-displacement curve prediction method comprising:

[0008] Acquire underground structure engineering parameters, input the underground structure engineering parameters into a prediction model, and obtain intermediate parameter output of the prediction model;

[0009] Obtaining parameters of a load-displacement curve according to the intermediate parameter output;

[0010] A load-displacement curve is established according to the parameters of the load-displacement curve.

[0011] Furthermore, the obtaining of underground structure engineering parameters and inputting the underground structure engineering parameters into the prediction model specifically includes:

[0012] Obtain soil parameters, soil depth and underground structure width;

[0013] The soil parameters, soil burial depth and underground structure width are used as underground structure engineering parameter inputs into the prediction model.

[0014] Furthermore, the soil parameters include soil density and internal friction angle.

[0015] Furthermore, the step of inputting the underground structure engineering parameters into the prediction model to obtain the intermediate parameter output of the prediction model specifically includes:

[0016] Build a neural network model;

[0017] Optimizing the parameters of the neural network model based on an optimization algorithm to obtain the prediction model;

[0018] The underground structure engineering parameters are input into a prediction model to obtain an intermediate parameter output of the prediction model.

[0019] Furthermore, the optimization algorithm is specifically a population intelligence algorithm, and the optimization algorithm is used to optimize the parameters of the neural network model to obtain the prediction model, which specifically includes:

[0020] Setting the search space dimension of the population intelligence algorithm to the number of the parameters, and setting the number of searchers to a set number of searchers, so that the searchers are randomly distributed in the search space;

[0021] Iterate the optimization. In each iteration, the searcher position is updated according to the population intelligence algorithm rules. The optimization parameters of each searcher are calculated according to the position of each searcher, and the optimization parameters are introduced into the generalized regression model to calculate the fitness value of each searcher. The global optimal solution and the personal optimal solution of each searcher are updated according to the fitness value of each searcher.

[0022] The global optimal solution obtained by iterative optimization is used as the parameter of the generalized regression neural network model.

[0023] Furthermore, the population intelligence algorithm is specifically one of a sparrow search algorithm and a particle swarm optimization algorithm.

[0024] Furthermore, the intermediate parameter output includes parameters , minimum soil arch ratio , minimum soil arch displacement and final soil arch ratio ;

[0025] The parameters of the load-displacement curve obtained according to the intermediate parameter output specifically include:

[0026] According to the final soil arch ratio Get Parameters :

[0027] ;

[0028] According to the parameters ,parameter , minimum soil arch ratio and minimum soil arch displacement Get Parameters :

[0029] ;

[0030] According to the parameters ,parameter ,parameter , minimum soil arch ratio and minimum soil arch displacement Get Parameters :

[0031] .

[0032] Furthermore, the parameters of the load-displacement curve include parameters ,parameter ,parameter and parameters ;

[0033] The load-displacement curve is established according to the parameters of the load-displacement curve, specifically according to the parameters ,parameter ,parameter and parameters Create a load-displacement curve:

[0034] ;

[0035] in, is the normalized pressure, is the normalized displacement of the underground structure.

[0036] The present invention adopts the above technical solution to achieve the following effects:

[0037] The present invention predicts the four parameters required to establish a complete load-displacement curve function by using a neural network. The present invention verifies through experiments that the use of a neural network to predict the four parameters required for the load-displacement curve has a good effect, and improves the error problem of obtaining the load-displacement curve in the prior art. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 It is a flow chart of the steps of a knowledge-data driven underground structure load-displacement curve prediction method in a preferred embodiment of the present invention;

[0039] Figure 2 It is a structural flow chart of a knowledge-data driven underground structure load-displacement curve prediction method in a preferred embodiment of the present invention;

[0040] Figure 3 is a schematic diagram of the structure of a neural network in a preferred embodiment of the present invention;

[0041] Figure 4 is a schematic diagram of a load-displacement curve in a preferred embodiment of the present invention;

[0042] Figure 5 This is a graph of the prediction effect of the generalized neural network model optimized by the particle swarm algorithm;

[0043] Figure 6 This is a prediction effect diagram of the generalized neural network model optimized by the sparrow search algorithm;

[0044] Figure 7 It is a schematic diagram comparing the test results of different models and centrifuge; DETAILED DESCRIPTION

[0045] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0046] The terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish different objects rather than to describe a specific order. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, device, product or end including a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units that are not listed, or may optionally include other steps or units inherent to these processes, methods, products or ends.

[0047] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present invention. The appearance of the phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0048] Embodiment 1

[0049] See also Figure 1 , Embodiment 1 of the present application is a knowledge-data driven underground structure load-displacement curve prediction method, wherein, Figure 1 The described method can be applied to obtain the load-displacement curve of a trapdoor.

[0050] Please refer to Figure 1 and Figure 2 As shown, Figure 1 and Figure 2 The method shown comprises the steps of:

[0051] S1. Obtain underground structure engineering parameters, input the underground structure engineering parameters into a prediction model, and obtain intermediate parameter output of the prediction model.

[0052] In this embodiment, the underground structure engineering parameters used specifically include soil parameters, soil burial depth (H) and underground structure width (B). The present invention draws the conclusion through analysis of the prior art that these three parameters significantly affect the relevant parameters of the load-displacement curve, wherein the soil parameters include soil density and internal friction angle, and the underground structure width corresponds to the width of different structures such as foundation pit width and tunnel width according to different construction types.

[0053] In this embodiment, the prediction model is a model composed of a neural network. The most important advantage of the neural network is that it can fully find the nonlinear relationship between input neurons and output neurons.

[0054] Neural networks have been widely used in various industries. Among them, the back propagation (BP) neural network is one of the earliest neural networks used in geotechnical engineering due to its strong nonlinear ability and learning ability. However, with the gradual expansion of back propagation applications, back propagation neural networks have exposed more and more shortcomings, such as local minimization problems, slow convergence speed, and the contradiction between different structural choices and application examples and network scale.

[0055] On this basis, radial basis function (RBF) neural network and its variant generalized regression neural network (GRNN) are widely used in geotechnical engineering and non-geotechnical engineering due to their high approximation ability, fast convergence speed and high training efficiency.

[0056] Please refer to Figure 2 and Figure 3 , the generalized regression neural network includes input layer, pattern layer, summation layer and output layer. The number of neurons in the input layer is equal to the dimension of the input vector in the learning sample. The input variable is directly transferred to the pattern layer. The number of neurons in the pattern layer is equal to the number of learning samples. The transfer function of the neurons in the pattern layer is:

[0057] ;

[0058] in, For the The output value of the pattern layer neuron, is the input variable, For the The learning samples of the neurons in the pattern layer, T represents the transposition, is the smoothing factor.

[0059] In the summation layer, two types of neurons are used for summation, and the outputs of all pattern layer neurons are subjected to arithmetic summation and weighted summation, respectively.

[0060] The summation layer includes a first type of summation neurons and a second type of summation neurons, wherein the number of the first type of summation neurons is 1, and the number of the second type of summation neurons is the same as the number of output parameters.

[0061] The first type of summation neurons perform arithmetic summation on the outputs of the pattern layer neurons:

[0062] ;

[0063] in, is the output of the first type of summing neuron, Indicates the total number of neurons in the layer where the neuron is located.

[0064] The second type of summation neurons perform weighted summation on the outputs of the pattern layer neurons:

[0065] ;

[0066] For the The output of the second type of summation neuron in the layer, For the The second type of summation neuron in the layer is The weighted values ​​assigned to the neurons in each pattern layer.

[0067] The output layer is used to obtain the predicted value of the parameter based on the output of the summation layer. The number of output neurons in the output layer is the same as the number of output parameters, and its output function is:

[0068] ;

[0069] in, For the output layer, The output value calculated by the output of the second type of summed neurons in the layer and the output of the first type of summed neurons is also the output value of the first The layer outputs the output of the neurons.

[0070] In generalized regression neural networks, how to reasonably select smoothing factors plays an important role in the establishment of generalized regression models, and has a significant impact on the performance of generalized regression neural networks. In the process of generalized regression model design, it is necessary to try different smoothing factors to determine the optimal value.

[0071] In this embodiment, two swarm intelligence algorithms are used to optimize the generalized regression neural network. The swarm intelligence algorithm can be a particle swarm optimization algorithm (PSO) or a sparrow search algorithm (SSA).

[0072] The particle swarm optimization algorithm is a group intelligence optimization algorithm in the field of computational intelligence. The particle swarm optimization algorithm is the basis for the study of bird predation behavior. In this embodiment, it is applied to the optimization of neural networks to achieve individual optimization in the solvable space. The particle swarm optimization algorithm first initializes a group of particles in the solvable space. Each particle represents a potential optimal solution to the extreme value optimization problem. The particle characteristics are represented by three indicators: position, speed, and fitness value. The fitness value is calculated by the fitness function, and its value represents the quality of the particle.

[0073] Assuming that in a In the dimensional solvable space, there are The population of searchers, here particles are used to represent searchers, , represents the position of particle number 1, represents the position of particle number 2, Indicates the number The position of each particle is represented by a Dimensional vector express, , and It indicates the The position of a particle in different search dimensions represents the position of the particle in the solvable space and also represents a solution.

[0074] No. Particles The speed is , , and It indicates the The speed of particles in different search dimensions, individuals have individual extreme values , similarly, , and It indicates the The extreme values ​​of particles in different search dimensions, the solvable space has a global extreme value , , and They represent the global extreme values ​​in different search dimensions respectively.

[0075] The particle swarm optimization algorithm iterates through the above settings. In each iteration, the particles update their speed and position through individual extreme values ​​and global extreme values:

[0076] ;

[0077] ;

[0078] in, Indicates speed, represents the extreme value, represents the position of the particle, and the superscripts of each value represent the number of iterations. Indicates Iterations, Indicates iterations, the first letter of the subscript represents the particle corresponding to the value, Indicates Particles, It means global, and the second letter represents the corresponding dimension. Indicates dimensions, is the inertia weight, and is a non-negative constant, called the acceleration factor, and is a random number distributed between [0, 1].

[0079] The sparrow search algorithm is inspired by the social intelligence, foraging, and predator avoidance behaviors observed in sparrows. By simulating the social behavior and adaptive strategies of sparrows, the sparrow search algorithm balances exploration and exploitation, effectively avoids local optimal traps, and ensures efficient and extensive search coverage.

[0080] The sparrow search algorithm divides the sparrow population into three groups: finders, entrants, and scouts. The part with high fitness is selected as the finder to provide optimization direction for most entrants; the scout is used to judge that the alarm value exceeds the optimization range and will make the entrant enter the optimization range. The sparrow search algorithm will make group decisions based on the positions and goals of the finders, entrants, and scouts. Specifically, the key implementation steps of the sparrow search algorithm are as follows:

[0081] Similar to population optimization, in this embodiment, it is assumed that In the dimensional solvable space, there are A searcher, here sparrow is used to represent the searcher.

[0082] At each iteration, the fitness value of each individual is calculated and the fitness is sorted. Mark the maximum fitness And its corresponding position , and the minimum and maximum fitness And its corresponding position .

[0083] The individual sparrows will decide whether to eat first based on their fitness values. The individual sparrows with better fitness values ​​are considered discoverers and will guide the entire group closer to the food source. The update formula for the discoverer's position is as follows:

[0084] ;

[0085] in, Represents a random number in the range [0, 1], represents the maximum number of iterations, Indicates the position of the particle. In the sparrow search algorithm, the particle refers to the sparrow. The first letter of the subscript represents the value corresponding to the sparrow, and the second letter represents the corresponding dimension. is the alarm value in the range [0, 1], is a safe value in the range [0.5, 1], is a random number following a normal distribution, which is used to determine the step size, is the identity matrix.

[0086] The meaning of the above formula is, This indicates that there is no threat of natural enemies around the foraging areas of the sparrow population. Therefore, the discoverer will expand the search area and conduct a more extensive search campaign; Meaning that some members of the sparrow group detect the presence of predators around the feeding area, which will send out a warning signal, instructing the entire group to move to other safe areas to continue foraging.

[0087] The positions of the contestants are updated as follows:

[0088] ;

[0089] In the formula, express The optimal position of the finder in the iteration, express The global worst position of the iteration, Indicates a random 1× Matrix, element data is 1 or -1, when , it means that the contestants cannot get food and need to fly to other areas to find food.

[0090] In the SSA algorithm, SD sparrows are randomly selected from the population for early warning reconnaissance in each generation, and the position update formula is as follows:

[0091] ;

[0092] In the formula, express The global optimal position of the iteration. is a random number between [-1, 1] affected by uniform distribution, which is used to control the movement direction of the sparrow. is the step size control parameter, which is a random number that follows a standard normal distribution. is a small constant with a non-zero denominator. is the current fitness value of the individual researcher.

[0093] when , indicating that the sparrow individuals are located at the edge of the entire population, are vulnerable to the total number of natural enemies, and need to be updated. , suggesting that sparrows at the center of the population are aware of the danger and need to move closer to sparrows in other areas.

[0094] In this embodiment, the search space dimension of the population intelligence algorithm is set to 1, the number of searchers is set to 20, and the number of iterations of each searcher is set to 30. The position of each searcher should satisfy the range [0, 1]. The speed and position of the searcher are randomly initialized so that the searchers are randomly distributed in the search space.

[0095] The smoothing factor of each searcher is calculated and introduced into the generalized regression model. The fitness value of each smoothing factor is calculated. If the fitness of the searcher is better than the single optimal solution, the fitness replaces the single optimal solution. If the fitness of the searcher is better than the global optimal solution, the fitness replaces the global optimal solution.

[0096] Until the iterative maximum value is reached. The global optimal solution of the smoothing factor is introduced into the generalized regression model.

[0097] S2. Obtaining parameters of the load-displacement curve according to the intermediate parameter output.

[0098] In this embodiment, the intermediate parameter output includes the initial arch modulus ; Minimum soil arch ratio , minimum soil arch displacement The final soil arch ratio .

[0099] Specifically, according to the research of the prior art, please refer to Figure 4 , the Y axis of the load-displacement curve is the normalized pressure , the x-axis is the normalized displacement of the underground structure , It is numerically equal to the support stress of the underground structure divided by the static ground stress. The value is equal to the downward displacement of the underground structure Divide by the width of the structure The prior art measures the load-displacement curve by measuring the load and displacement on the trapdoor in a trapdoor experiment, such as Figure 4 As shown, the load-displacement curve can be divided into four stages, which are described as follows:

[0100] Phase 1 (before point A): Normalized pressure With the normalized displacement of the underground structure In this stage, the soil around the trapdoor remains in an elastic state, and an arch phenomenon occurs in the soil, but no shear plane is formed. The pressure drop rate in this stage is called the initial modulus of the arch in the prior art.

[0101] Phase 2 (between points A and B): With the normalized displacement of the underground structure The increase in normalized pressure The deceleration rate gradually decreases. With the formation of the first shear surface above the shear gate, the normalized pressure reaches the minimum at point B, and the area below the shear plane and between the shear gates is a local failure area, where the soil displacement increases significantly.

[0102] Stage 3 (between points B and C): As the displacement of the trap door increases, the shear surface gradually fails upward, and the normalized pressure With the normalized displacement of the underground structure The soil above the trapdoor has plastic deformation and the slope of the load-displacement curve at this stage is Represents the load recovery capability.

[0103] Stage 4 (after point C): As the shear plane resolves to the ground, the soil above the trapdoor is completely destroyed and the soil above the trapdoor is in a loose state, however, as the normalized displacement of the trapdoor increases, the normalized pressure remains unchanged.

[0104] From the above analysis, it can be seen that the load-displacement curve has the following characteristics:

[0105] The load-displacement curve has a starting point and changes with the increase of normalized door displacement, that is, when the plate door displacement is 0, the normalized pressure is 1;

[0106] As the displacement of the trap door increases, the load-displacement curve shows a trend of decreasing first and then increasing, gradually reaching the final value. When the normalized trap door displacement is small, the load-displacement curve decreases linearly (i.e., the first stage), and when the load-displacement curve reaches its minimum value, the load-displacement curve has a rebound characteristic (i.e., the second and third stages);

[0107] When the normalized trapdoor displacement is large, the load-displacement curve shows linear invariance (i.e., stage 4, where the first derivative of the load-displacement curve is 0).

[0108] Based on the above characteristics, the composite function of the load-displacement curve established in the prior art is as follows:

[0109] ;

[0110] in, , , and are all function related parameters, among which, is the first-order derivative in the linear stage, is the normalized pressure at which the load-displacement curve shows linear invariance.

[0111] When displacement When the value is small, the composite function can be written as , represents the stage displacement of linear reduction When it is 0, the value of the linear part of the function and the exponential part has the value 1, the composite function has the value 1. When , the exponential part is 0. The value of the composite function is (i.e. final soil arch ratio ).

[0112] According to the above analysis, the composite function can effectively realize the characteristics shown by the load-displacement curve. Subsequently, the present invention will analyze how to obtain the four parameters of the load-displacement curve and their related physical meanings.

[0113] First, define the parameters represents the exponential part of the function, that is , then the composite function can be expressed as:

[0114] ;

[0115] Take the independent variable in the equation The first-order derivative of gives the following first-order derivative formula:

[0116] ;

[0117] Afterwards, please refer to Figure 4 , Figure 4 Point B is the extreme point of the load-displacement curve and also the minimum point. Therefore, the first-order derivative of the composite function at point B is 0. At the coordinates of point B, the first-order derivative formula can be expressed as:

[0118] ;

[0119] in, is a parameter The value at point B is is the normalized displacement of the underground structure at point B, is the normalized pressure at point B.

[0120] Transforming this formula we can get the conclusion:

[0121] ;

[0122] Afterwards, according to the parameters The definition of can be obtained, the parameter Relative to the independent variable The first-order derivative of is:

[0123] ;

[0124] In addition, by changing the parameter The definition of is logarithmic, and the parameter Written as:

[0125] ;

[0126] In summary, the first-order derivative formula through point B can be expressed as:

[0127] ;

[0128] Among them, the parameters The value at point B for:

[0129] ;

[0130] By combining the above formulas, we can get and The formula is, where the parameters The formula is:

[0131] ;

[0132] parameter The formula is:

[0133] ;

[0134] Obviously, to obtain a composite function of the load-displacement curve, only the initial arch modulus needs to be obtained. , minimum soil arch ratio , minimum soil arch displacement and final soil arch ratio , you can use the parameter Formula and parameters The formula gets the final function parameters.

[0135] In this embodiment, according to the final soil arch ratio Get Parameters :

[0136] ;

[0137] According to the initial arch modulus ,parameter , minimum soil arch ratio and minimum soil arch displacement Get Parameters :

[0138] ;

[0139] According to the initial arch modulus ,parameter ,parameter , minimum soil arch ratio and minimum soil arch displacement Get Parameters :

[0140] .

[0141] S3. Establishing a load-displacement curve according to the parameters of the load-displacement curve.

[0142] After the load-displacement curve is established, the underground structure project can be evaluated based on the load-displacement curve.

[0143] The assessment can be a prediction of the results after construction before construction, or it can be a continuous monitoring of the underground structure project after the construction is completed, and an assessment of the current or future status of the underground structure project based on the monitoring results.

[0144] The present invention collected a total of 116 sets of experimental data, of which 90 sets were used as training sets for neural network data learning and 26 sets were used as test sets for neural network verification. However, it is worth noting that the soil used in the trapdoor experiment cannot be simply represented by two parameters, and since the experiment is artificial, experimental errors are bound to exist.

[0145] Please refer to Figure 5 , Figure 6 , Figure 5 This is the prediction effect diagram of the generalized neural network model (PSO-GRNN) optimized by the particle swarm algorithm. Figure 6 This is a prediction effect diagram of the generalized neural network model (SSA-GRNN) optimized by the sparrow search algorithm.

[0146] Figure 5 and Figure 6 (a) is the initial arch modulus The prediction effect diagram, Figure 5 and Figure 6 (b) is the minimum soil arch displacement The prediction effect diagram, Figure 5 and Figure 6(c) is the minimum soil arch ratio The prediction effect diagram, Figure 5 and Figure 6 (d) is the final soil arch ratio Prediction effect diagram.

[0147] Each data point in the scatter plot represents the distribution of a set of measured and predicted values. The results show that the swarm intelligence optimization algorithm has good prediction performance for the four parameters of the load-displacement curve. At the same time, it can be found that the parameters For most points in the scatter plot, when the value is less than 0.05, the minimum soil arch ratio displacement should be about 5% of the trapdoor width.

[0148] In order to further verify the proposed GRC function model and evaluate the accuracy of the output parameters predicted by the GRNN model, the GRC obtained by the three two-dimensional trapdoor model tests was compared with the GRC function model proposed in this study. The comparative analysis aims to evaluate the effectiveness of the proposed model.

[0149] The present invention uses the centrifuge model results with H / B= values ​​of 0.7, 1.7 and 2, respectively, and the experimental results using fused silica sand as the filling sample to verify the proposed GRC model.

[0150] Please refer to Table 1 for the predicted parameter values:

[0151] Table 1. Load-displacement curve parameter values

[0152]

[0153] Figure 7 This is a comparison diagram of different models and centrifuge test results. Figure 7 (a) is a comparison diagram of different models and centrifuge test results at H / B=0.7. Figure 7 (b) is a comparison diagram of different models with H / B=1.7 and the centrifuge test results. Figure 7 (c) is a schematic diagram comparing the different models with H / B=2 and the centrifuge test results.

[0154] The experimental results were compared with the results of the GRC function model predicted by the neural network. It can be seen from the experimental results that with the increase of the burial depth ratio H / B, the minimum soil arch ratio and the final soil arch ratio both decrease. From the shape of the curve drawn by the function model, it can be seen that the proposed function model can effectively display the four stages of GRC characteristics, indicating that the establishment of the GRC function model is effective.

[0155] In summary, the present invention predicts the four parameters required to establish a complete load-displacement curve function by using a neural network. The present invention has verified through experiments that the use of a neural network to predict the four parameters required for the load-displacement curve has a good effect, and has improved the error problem of obtaining the load-displacement curve in the prior art.

[0156] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or terminal including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or terminal. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the presence of other identical elements in the process, method, article or terminal including the element.

[0157] Of course, those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing related hardware (such as a processor, a controller, etc.) through a computer program, and the program can be stored in a computer-readable storage medium, and the program can include the processes of the above-mentioned method embodiments when executed. The storage medium can be a memory, a disk, an optical disk, etc.

[0158] It should be understood that the application of the present invention is not limited to the above examples. For ordinary technicians in this field, improvements or changes can be made 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 knowledge-data driven underground structure load-displacement curve prediction method, characterized in that: The knowledge-data driven underground structure load-displacement curve prediction method includes: Acquire underground structure engineering parameters, input the underground structure engineering parameters into a prediction model, and obtain intermediate parameter output of the prediction model; Obtaining parameters of a load-displacement curve according to the intermediate parameter output; Establishing a load-displacement curve according to the parameters of the load-displacement curve; The intermediate parameter output includes the initial arch modulus , minimum soil arch ratio , minimum soil arch displacement and final soil arch ratio ; The parameters of the load-displacement curve obtained according to the intermediate parameter output specifically include: According to the final soil arch ratio Get Parameters : ； According to the initial arch modulus ,parameter , minimum soil arch ratio and minimum soil arch displacement Get Parameters : ； According to the initial arch modulus ,parameter ,parameter , minimum soil arch ratio and minimum soil arch displacement Get Parameters : 。 2. A knowledge-data driven underground structure load-displacement curve prediction method according to claim 1, characterized in that: The obtaining of underground structure engineering parameters and inputting the underground structure engineering parameters into the prediction model specifically includes: Obtain soil parameters, soil depth and underground structure width; The soil parameters, soil burial depth and underground structure width are used as underground structure engineering parameter inputs into the prediction model.

3. A knowledge-data driven underground structure load-displacement curve prediction method according to claim 2, characterized in that: The soil parameters include soil density and internal friction angle.

4. The knowledge-data driven underground structure load-displacement curve prediction method according to claim 1, characterized in that: The step of inputting the underground structure engineering parameters into the prediction model to obtain the intermediate parameter output of the prediction model specifically includes: Build a neural network model; Optimizing the parameters of the neural network model based on an optimization algorithm to obtain the prediction model; The underground structure engineering parameters are input into a prediction model to obtain an intermediate parameter output of the prediction model.

5. A knowledge-data driven underground structure load-displacement curve prediction method according to claim 4, characterized in that: The optimization algorithm is specifically a population intelligence algorithm, and the optimization algorithm is used to optimize the parameters of the neural network model to obtain the prediction model, which specifically includes: Setting the search space dimension of the population intelligence algorithm to the number of the parameters, and setting the number of searchers to a set number of searchers, so that the searchers are randomly distributed in the search space; Iterate the optimization. In each iteration, the searcher position is updated according to the population intelligence algorithm rules. The optimization parameters of each searcher are calculated according to the position of each searcher, and the optimization parameters are introduced into the neural network model to calculate the fitness value of each searcher. The global optimal solution and the personal optimal solution of each searcher are updated according to the fitness value of each searcher. The global optimal solution obtained by iterative optimization is used as the parameter of the neural network model.

6. A knowledge-data driven underground structure load-displacement curve prediction method according to claim 5, characterized in that: The population intelligence algorithm is specifically one of a sparrow search algorithm and a particle swarm optimization algorithm.

7. The knowledge-data driven underground structure load-displacement curve prediction method according to claim 1, characterized in that: The parameters of the load-displacement curve include the initial arch modulus ,parameter ,parameter and parameters ; The load-displacement curve is established according to the parameters of the load-displacement curve, specifically according to the initial arch modulus ,parameter ,parameter and parameters Create a load-displacement curve: ； in, is the normalized pressure, is the normalized displacement of the underground structure.

Images