Tunnel bearing capacity prediction model training method and device, equipment and medium

Abstract

The invention relates to the technical field of tunnel engineering, and discloses a training method and device of a tunnel bearing capacity prediction model, equipment and a medium, and the method comprises the steps: obtaining a training data set based on a finite element limit analysis model; iteratively training the reference models of different types according to the training data set, and determining a plurality of target reference models; iteratively training a plurality of different types of meta learners based on the plurality of target reference models and the training data set, and determining a target meta learner; and a fusion model is constructed on the basis of the target element learning device and the multiple target reference models, and the fusion model is used for carrying out soil bearing capacity prediction according to the characteristic parameters corresponding to the target tunnel engineering. Therefore, the fusion model for predicting the bearing capacity of the soil body is obtained, and the prediction efficiency is improved on the basis of ensuring the prediction accuracy of the bearing capacity of the soil body.

Description

Technical Field

[0001] The present invention relates to the technical field of tunnel engineering, and in particular to a training method, device, equipment and medium for a tunnel bearing capacity prediction model. Background Art

[0002] In the process of tunnel design and construction, accurate assessment of tunnel stability is crucial. The soil bearing capacity of the tunnel is a key indicator for tunnel stability assessment. When studying tunnel stability issues, classical theoretical methods such as the rigid body limit equilibrium method, the slip line method, and the limit analysis method are usually selected. However, the above methods require a significant simplification of the actual working conditions to reduce the difficulty of solving them. However, due to manpower limitations, the calculation efficiency is low.

[0003] With the continuous improvement of industrial level, computer computing power has developed rapidly, and the related finite element method has also begun to be widely used in the study of stability problems in geotechnical engineering. This method can select different constitutive models according to specific working conditions to adapt to various working conditions. Usually, the stability can be analyzed by load increment method and strength coefficient reduction method. However, the finite element method is not accurate in judging whether the rock and soil body is damaged through displacement load curve or strain and displacement cloud map. Based on this, the finite element limit analysis method is used to calculate the stability of the engineering field, which has a good calculation effect and has been widely used in the engineering field.

[0004] However, as a traditional numerical calculation method, the finite element limit analysis method has problems such as complex modeling and long calculation time when performing calculations and solutions. It is inefficient when calculating soil bearing capacity in the field of tunnel engineering. Summary of the invention

[0005] In view of this, the present invention provides a training method, device, equipment and medium for a tunnel bearing capacity prediction model to solve the problem of low efficiency when calculating soil bearing capacity in the field of tunnel engineering.

[0006] In a first aspect, the present invention provides a method for training a tunnel bearing capacity prediction model, the method comprising:

[0007] Acquire a training data set based on a finite element limit analysis model, wherein the sample data in the training data set includes: characteristic parameters corresponding to the tunnel project and soil bearing capacity labels;

[0008] Iteratively training a plurality of different types of benchmark models according to the training data set, and obtaining first performance evaluation results corresponding to each trained benchmark model, wherein the benchmark model is used to predict soil bearing capacity;

[0009] Determining a plurality of target benchmark models from a plurality of trained benchmark models based on the first performance evaluation result;

[0010] Iteratively training a plurality of meta-learners of different types based on the plurality of target benchmark models and the training data set, and obtaining a second performance evaluation result of each meta-learner after training;

[0011] Based on the second performance evaluation result, determining a target meta-learner from a plurality of trained meta-learners;

[0012] A fusion model is constructed based on the target meta-learner and the multiple target benchmark models, and the fusion model is used to predict the soil bearing capacity according to the characteristic parameters corresponding to the target tunnel project.

[0013] The method provided in this aspect calculates a large number of data sets that can be used for model training through a finite element limit analysis model, thereby training different types of benchmark models, and after the benchmark model training is completed, multiple target benchmark models with better effects are determined according to the performance evaluation results, so as to further train the meta-learner according to the multiple target benchmark models and the training data, and determine the best target meta-learner, and construct a fusion model in combination with the target benchmark model, so that the fusion model can replace the finite element limit analysis model to predict the bearing capacity of the tunnel. On the basis of taking into account the prediction accuracy, the advantages of the machine learning model with fast calculation speed and small application restrictions are brought into play to improve the efficiency of tunnel bearing capacity prediction.

[0014] In an optional implementation, the iteratively training a plurality of different types of meta-learners based on the plurality of target benchmark models and the training data set respectively includes:

[0015] Obtaining initial soil bearing capacity prediction values ​​calculated by each target benchmark model based on sample data in the training data set;

[0016] For each sample data, the initial soil bearing capacity prediction values ​​of each target benchmark model corresponding to the sample data are processed by each meta-learner to obtain the fused soil bearing capacity prediction values ​​corresponding to each meta-learner;

[0017] According to the comparison result between the fused soil bearing capacity prediction value and the soil bearing capacity label corresponding to the sample data, the internal parameters of the corresponding meta-learner are updated.

[0018] In this embodiment, the predicted values ​​of soil bearing capacity preliminarily determined by each target benchmark model are used as the input of the meta-learner, so that the meta-learner can fuse the outputs of each model to improve the accuracy of the final prediction result. In the process of training the meta-learner, the meta-learner can take into account the prediction results of each target benchmark model to ensure the prediction accuracy of the final constructed fusion model.

[0019] In an optional implementation, the acquiring of a training data set based on a finite element limit analysis model includes:

[0020] Call the constructed finite element limit analysis model;

[0021] Calculating multiple tunnel engineering parameter groups based on the finite element limit analysis model to obtain soil bearing capacity calculation values ​​corresponding to each tunnel engineering parameter group;

[0022] A plurality of characteristic parameters are determined in the tunnel engineering parameter group, and a training data set is constructed based on parameter values ​​corresponding to each characteristic parameter in each tunnel engineering parameter group and the calculated value of the soil bearing capacity.

[0023] In this embodiment, a pre-constructed finite element limit analysis model is used to calculate multiple groups of parameters involved in tunnel engineering to obtain the calculated values ​​of soil bearing capacity corresponding to each group of tunnel engineering parameters, thereby relatively quickly obtaining highly accurate training data to ensure the training effect of subsequent models.

[0024] In an optional implementation, the training data set includes: a first training data set and a second training data set;

[0025] The iterative training of multiple different types of benchmark models according to the training data set includes:

[0026] Iteratively training a plurality of different types of benchmark models according to the first training data set;

[0027] The iteratively training a plurality of different types of meta-learners based on the plurality of target benchmark models and the training data set comprises:

[0028] Based on the multiple target benchmark models and the second training data set, multiple meta-learners of different types are iteratively trained respectively.

[0029] In this embodiment, the training data is differentiated and the first training data set and the second training data set are used to train the benchmark model and the meta-learner respectively, so as to ensure the training effect of the final target benchmark model and the target meta-learner.

[0030] In an optional implementation, the training data set includes: a test set and a validation set;

[0031] Before obtaining the first performance evaluation results corresponding to each trained benchmark model, the method further includes:

[0032] Based on the validation set, determine the optimal hyperparameters corresponding to each benchmark model through grid search and random search;

[0033] Determine a trained benchmark model based on the model parameters determined by multiple rounds of iterative training and the optimal hyperparameters;

[0034] The obtaining of the first performance evaluation results of each benchmark model after training includes:

[0035] Predicting the initial soil bearing capacity of the sample data of the test set by using each trained benchmark model;

[0036] Determine the mean square error corresponding to each benchmark model based on the initial soil bearing capacity prediction value corresponding to each sample data, wherein the mean square error is used for performance evaluation;

[0037] The obtaining of the second performance evaluation results of each meta-learner after training includes:

[0038] Determine the fused soil bearing capacity prediction value corresponding to each sample data in the test set through each trained meta-learner and the target benchmark model;

[0039] The mean square error corresponding to each meta-learner is determined based on the fused soil bearing capacity prediction value corresponding to each sample data.

[0040] In this embodiment, the hyperparameters of each benchmark model after training are optimized through the validation set in the training data set, so that when the performance evaluation is performed subsequently, the prediction effect of the model is consistent with the actual performance of the model. At the same time, when the performance of the benchmark model and the meta-learner is evaluated, the corresponding mean square error is obtained through the test set, so as to perform performance evaluation on the learner and the benchmark model, so as to determine the best benchmark model and meta-learner, and ensure the prediction accuracy of the fusion model finally constructed.

[0041] In an optional embodiment, the method further includes:

[0042] Corresponding benchmark models are constructed based on linear regression algorithm, support vector regression algorithm, extreme gradient boosting algorithm, random forest algorithm and multilayer perceptron neural network respectively;

[0043] Construct the corresponding meta-learner based on the extreme gradient boosting algorithm and linear regression algorithm.

[0044] In this embodiment, corresponding machine learning models, namely, benchmark models and meta-learners, are constructed through a variety of different algorithms or architectures, so that the target benchmark model determined subsequently can predict the soil bearing capacity from multiple angles, thereby ensuring the prediction accuracy of the final fusion model.

[0045] In a second aspect, the present invention provides a method for predicting tunnel bearing capacity, the method comprising:

[0046] A fusion model constructed by calling the training method of the tunnel bearing capacity prediction model described in the first aspect or any one of its embodiments;

[0047] Obtaining target characteristic parameters corresponding to the target tunnel project, inputting the target characteristic parameters into the fusion model, and obtaining first prediction results corresponding to each target benchmark model in the fusion model;

[0048] The first prediction results corresponding to the respective target benchmark models are input into the target meta-learner of the fusion model to obtain the soil bearing capacity prediction result of the target tunnel project.

[0049] The method provided in this aspect predicts the soil bearing capacity of the tunnel by using the fusion model trained by the above-mentioned training method, which can take into account both prediction accuracy and prediction efficiency, and improve the soil bearing capacity prediction effect in actual application scenarios.

[0050] In a second aspect, the present invention provides a training device for a tunnel bearing capacity prediction model, the device comprising:

[0051] A data set acquisition module is used to acquire a training data set based on a finite element limit analysis model, wherein the sample data in the training data set includes: characteristic parameters corresponding to the tunnel engineering and soil bearing capacity labels;

[0052] A benchmark model training module, used for iteratively training a plurality of different types of benchmark models according to the training data set, and obtaining a first performance evaluation result corresponding to each trained benchmark model, wherein the benchmark model is used to predict soil bearing capacity;

[0053] a target model selection module, configured to determine a plurality of target benchmark models from a plurality of trained benchmark models based on the first performance evaluation result;

[0054] A meta-learner training module, configured to iteratively train a plurality of meta-learners of different types based on the plurality of target benchmark models and the training data set, and obtain a second performance evaluation result of each meta-learner after training;

[0055] a target meta-learner selection module, configured to determine a target meta-learner from among a plurality of trained meta-learners based on the second performance evaluation result;

[0056] A fusion model construction module is used to construct a fusion model based on the target meta-learner and the multiple target benchmark models, and the fusion model is used to predict the soil bearing capacity according to the characteristic parameters corresponding to the target tunnel project.

[0057] In a third aspect, the present invention provides a computer device, comprising: a memory and a processor, the memory and the processor are communicatively connected to each other, computer instructions are stored in the memory, and the processor executes the training method of the tunnel bearing capacity prediction model of the first aspect or any corresponding embodiment thereof or the tunnel bearing capacity prediction method corresponding to the second aspect by executing the computer instructions.

[0058] In a fourth aspect, the present invention provides a computer-readable storage medium having computer instructions stored thereon, the computer instructions being used to enable a computer to execute the training method for a tunnel bearing capacity prediction model of the first aspect or any corresponding embodiment thereof, or the tunnel bearing capacity prediction method corresponding to the second aspect. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0060] Figure 1 is a flow chart of a method for training a tunnel bearing capacity prediction model according to an embodiment of the present invention;

[0061] Figure 2 is a flow chart of another method for training a tunnel bearing capacity prediction model according to an embodiment of the present invention;

[0062] Figure 3 is an example diagram of characteristic parameters at a specific tunnel interface according to an embodiment of the present invention;

[0063] Figure 4 is a schematic flow chart of a method for predicting tunnel bearing capacity according to an embodiment of the present invention;

[0064] Figure 5 is an example diagram of a model training process according to an embodiment of the present invention;

[0065] Figure 6 is an example diagram of a model fusion process according to an embodiment of the present invention;

[0066] Figure 7This is an example diagram of an application process of a soil bearing capacity prediction model according to an embodiment of the present invention;

[0067] Figure 8 is a structural block diagram of a training device for a tunnel bearing capacity prediction model according to an embodiment of the present invention;

[0068] Fig. 9 is a structural block diagram of a tunnel bearing capacity prediction device according to an embodiment of the present invention;

[0069] Fig.10 It is a schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0070] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution 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 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 those skilled in the art without creative work are within the scope of protection of the present invention.

[0071] In the process of tunnel design and construction, accurate assessment of tunnel stability is crucial. The soil bearing capacity of the tunnel is a key indicator for tunnel stability assessment. When studying tunnel stability issues, classical theoretical methods such as the rigid body limit equilibrium method, the slip line method, and the limit analysis method are usually selected. However, the above methods require a significant simplification of the actual working conditions to reduce the difficulty of solving them. However, due to manpower limitations, the calculation efficiency is low.

[0072] With the continuous improvement of industrial level, computer computing power has developed rapidly, and the related finite element method has also begun to be widely used in the study of stability problems in geotechnical engineering. This method can select different constitutive models according to specific working conditions to adapt to various working conditions. Usually, the stability can be analyzed by load increment method and strength coefficient reduction method. However, the finite element method is not accurate in judging whether the rock and soil body is damaged through displacement load curve or strain and displacement cloud map. Based on this, the finite element limit analysis method is used to calculate the stability of the engineering field, which has a good calculation effect and has been widely used in the engineering field.

[0073] However, as a traditional numerical calculation method, the finite element limit analysis method has problems such as complex modeling and long calculation time when performing calculations and solutions. It is inefficient when calculating soil bearing capacity in the field of tunnel engineering.

[0074] To this end, an embodiment of the present invention provides a training method for a tunnel bearing capacity prediction model, which uses a finite element limit analysis model to calculate a large number of data sets that can be used for model training, thereby training different types of benchmark models, and after the benchmark model training is completed, multiple target benchmark models with better effects are determined according to performance evaluation results, so that a meta-learner is further trained according to the multiple target benchmark models and training data, and the best target meta-learner is determined. In combination with the target benchmark model, a fusion model is constructed, so that the fusion model can act as a proxy for the finite element limit analysis model to predict the tunnel bearing capacity. On the basis of taking into account the prediction accuracy, the advantages of the machine learning model with fast calculation speed and small application restrictions are brought into play, thereby improving the efficiency of tunnel bearing capacity prediction.

[0075] According to an embodiment of the present invention, an embodiment of a training method for a tunnel bearing capacity prediction model is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in an order different from that shown here.

[0076] In this embodiment, a method for training a tunnel bearing capacity prediction model is provided, which can train the machine learning model for predicting the soil bearing capacity. Figure 1 is a flow chart of a method for training a tunnel bearing capacity prediction model according to an embodiment of the present invention, such as Figure 1 As shown, the process includes the following steps:

[0077] Step S101, obtaining a training data set based on a finite element limit analysis model, wherein sample data in the training data set includes: characteristic parameters corresponding to the tunnel project and soil bearing capacity labels.

[0078] The finite element limit analysis method is a commonly used method for calculating rock stability in the engineering field. In the field of tunnel construction, it is usually necessary to construct a constitutive model and combine it with the finite element limit calculation method to calculate the soil bearing capacity at different positions of the tunnel. This calculation method can relatively accurately determine the maximum soil bearing capacity of the entire tunnel. Specifically, when performing the calculation, it is necessary to obtain specific departure survey data and tunnel data, such as soil parameters corresponding to different positions in the entire tunnel project, tunnel depth, tunnel diameter and other numbers. The finite element limit analysis model is used to perform numerical calculations on these data to obtain the soil bearing capacity of each corresponding position in the tunnel.

[0079] However, for the finite element limit analysis model, it is necessary to use a complex nonlinear model when calculating the bearing capacity, and it often takes a lot of time to perform nonlinear simulation, but its calculation accuracy is relatively high. Therefore, in actual construction, the use of the finite element limit analysis model to predict the bearing capacity usually takes a lot of time, and because it uses traditional numerical methods, the threshold for parameter adjustment during actual prediction is also relatively high, affecting the prediction efficiency. It should be noted that the bearing capacity prediction involved in the present invention refers to the maximum bearing load that can be borne by each specific position of the tunnel.

[0080] However, the finite element limit analysis model that has been built can be used to calculate multiple sets of manually set tunnel engineering parameters in advance to obtain a large number of calculated values, which can be used as training data sets to train machine learning models. The trained machine learning model is used as a proxy model for the finite element limit analysis model to perform soil bearing capacity prediction instead of it, giving full play to the advantages of the machine learning model with low threshold and fast calculation speed. The characteristic parameters may include: tunnel parameters or rock parameters such as tunnel depth, tunnel diameter, rock density, rock strength, etc.

[0081] Each sample data in the training data set can be understood as a set of tunnel engineering parameters and corresponding soil bearing capacity labels, where a set of tunnel engineering parameters can be an engineering parameter group composed of relevant parameters corresponding to each specific position in a tunnel project, and the corresponding soil bearing capacity label is the calculated soil bearing capacity of each specific position. In some cases, in order to simplify the training of the model, each sample data can also be the relevant parameters and corresponding bearing capacity label at a specific position of a single tunnel. The specific training data setting method can be set according to the actual situation.

[0082] Step S102, iteratively training a plurality of different types of benchmark models according to the training data set, and obtaining a first performance evaluation result corresponding to each trained benchmark model, wherein the benchmark model is used to predict soil bearing capacity.

[0083] You can select multiple different types of machine learning models as benchmark models based on actual needs, such as linear regression models, extreme gradient boosting models, etc. These different types of machine learning models have their own advantages. For example, the linear regression model is fast and easy to explain; the extreme gradient boosting has the advantages of processing complex relationships between features and preventing overfitting.

[0084] During iterative training, similar to the conventional model training process, for each sample data in the training set, its characteristic parameters are input into the machine learning model, and the output results are compared with the corresponding soil bearing capacity labels to perform optimization iterations of the model parameters. Specifically, the optimization iterations can be performed through the gradient descent algorithm.

[0085] For each benchmark model, the corresponding training stop condition or accuracy requirement can be set according to the actual training needs. When the corresponding condition or requirement is met, the training can be stopped. At this time, each benchmark model can be hyperparameter optimized to determine the model's hyperparameters and obtain the trained benchmark model. At the same time, a certain amount of test data can be allocated in the training data in advance to evaluate the performance of each benchmark model and obtain the corresponding performance evaluation results of each benchmark model, so as to select the model with better performance from these benchmark models as the target benchmark model.

[0086] Step S103: determining a plurality of target benchmark models from a plurality of trained benchmark models based on the first performance evaluation result.

[0087] After obtaining the performance evaluation results corresponding to each benchmark model, that is, the prediction accuracy, several benchmark models with better prediction effects can be selected from these benchmark models as target benchmark models. For example, the top three benchmark models in the performance evaluation are selected as target benchmark models.

[0088] The benchmark model at this time already has a certain ability to predict soil bearing capacity, that is, it can predict the corresponding soil bearing capacity based on the specific engineering parameters of a tunnel project. However, different benchmark models have certain shortcomings, resulting in insufficient prediction accuracy. Further fusion through the meta-learner is needed in the future to ensure the final prediction effect.

[0089] Step S104, based on the multiple target benchmark models and the training data sets, iteratively train multiple different types of meta-learners respectively, and obtain a second performance evaluation result of each trained meta-learner.

[0090] In the process of training the meta-learner, for each round of training, the characteristic parameters in the training data are input into each target benchmark model to obtain the soil bearing capacity prediction value of each benchmark model relative to each sample data, and the prediction value is used as the input of the meta-learner. The fused soil bearing capacity prediction value is obtained through the meta-learner, and it is compared with the label of the corresponding sample data to optimize the model parameters inside the meta-learning, and so on, to complete the iterative training of each meta-learning. After the training is completed, the output result of the meta-learner can integrate the prediction characteristics of multiple target models, so as to have higher accuracy.

[0091] Multiple different types of meta-learners can be pre-set according to actual needs. For each meta-learner, training is performed according to the above training process until the training end condition is reached. The training end condition can be the training round or the prediction accuracy, which is set according to actual needs.

[0092] Next, referring to the above performance evaluation of the benchmark model, a part of the data can also be pre-selected in the training set as a test set, and multiple target benchmark models can be combined with different meta-learners respectively, and the prediction performance of different combinations can be evaluated according to the test set to obtain the performance evaluation results corresponding to different meta-learners, that is, the second performance evaluation results.

[0093] Step S105: determining a target meta-learner from the plurality of trained meta-learners based on the second performance evaluation result.

[0094] According to the evaluation results of each learner, the meta-learner with the best effect is selected as the target learner.

[0095] Step S106, constructing a fusion model based on the target meta-learner and multiple target benchmark models, and the fusion model is used to predict the soil bearing capacity according to the characteristic parameters corresponding to the target tunnel project.

[0096] The fusion model is jointly constructed based on the target meta-learner and the target benchmark model. It can be understood that the target benchmark model is used as the first-level prediction of the fusion model, and the first-level prediction results are input into the meta-learning for the second-level prediction. The results obtained by the second-level prediction can take into account the prediction results of each target benchmark model in the first layer to obtain a more accurate soil bearing capacity prediction result.

[0097] The training method of the tunnel bearing capacity prediction model provided in this embodiment uses a finite element limit analysis model to calculate a large number of data sets that can be used for model training, thereby training different types of benchmark models, and after the benchmark model training is completed, multiple target benchmark models with better effects are determined according to the performance evaluation results, so as to further train the meta-learner according to the multiple target benchmark models and the training data, and determine the best target meta-learner, and construct a fusion model in combination with the target benchmark model, so that the fusion model can replace the finite element limit analysis model to predict the tunnel bearing capacity. On the basis of taking into account the prediction accuracy, the advantages of the machine learning model with fast calculation speed and small application restrictions are brought into play to improve the efficiency of tunnel bearing capacity prediction.

[0098] According to an embodiment of the present invention, another training method embodiment of a tunnel bearing capacity prediction model is provided, which can train the machine learning model for soil bearing capacity prediction mentioned above. Figure 2is a flow chart of another method for training a tunnel bearing capacity prediction model according to an embodiment of the present invention, such as Figure 2 As shown, the process includes the following steps:

[0099] Step S201, obtaining a training data set based on a finite element limit analysis model, wherein sample data in the training data set includes: characteristic parameters corresponding to the tunnel project and soil bearing capacity labels.

[0100] Specifically, in the above step S201, obtaining a training data set based on a finite element limit analysis model includes:

[0101] Step S201-1, calling the constructed finite element limit analysis model.

[0102] Refer to the above Figure 1 From the explanation of the finite element limit analysis model in step S101 of the illustrated embodiment, it can be understood that the finite element limit analysis model can obtain the corresponding bearing capacity prediction value through numerical calculation according to the specific parameters in the tunnel project, and the pre-constructed finite element limit analysis model can be used to generate subsequent training data.

[0103] Step S201 - 2 , calculating multiple tunnel engineering parameter groups based on the finite element limit analysis model to obtain soil bearing capacity calculation values ​​corresponding to each tunnel engineering parameter group.

[0104] For the finite element limit analysis model, it can calculate the soil bearing capacity of the entire tunnel according to the engineering parameters corresponding to a certain tunnel project through numerical calculation, that is, the soil bearing capacity at each specific position in the tunnel. The engineering parameters are the tunnel depth, tunnel diameter, tunnel material, rock soil strength and density at the corresponding position of each specific tunnel section of the entire tunnel, and other physical parameters. In order to generate a large amount of training data, the characteristic parameters for calculation can be manually set by the staff based on experience, or randomly generated according to a preset program, so as to pre-generate a large amount of training data through the finite element analysis method.

[0105] Step S201-3, determining a plurality of characteristic parameters in the tunnel engineering parameter group, and constructing a training data set based on the parameter values ​​corresponding to each characteristic parameter in each tunnel engineering parameter group and the calculated value of the soil bearing capacity.

[0106] In the tunnel engineering parameter group, due to the actual complexity of the project, there are many types of parameters involved. Among these parameter types, some parameters with high importance and great influence on soil bearing capacity are determined as characteristic parameters, such as tunnel depth, tunnel diameter, rock mass parameters, etc. For details, please refer to Figure 3, is an example diagram of characteristic parameters at a specific tunnel interface according to an embodiment of the present invention. It can be seen from the figure that the characteristic parameters include: tunnel parameters: tunnel depth H, tunnel interval S, tunnel diameter D; soil parameters: geological strength index GSI, yield criterion parameter mi, gravity density γ, uniaxial compressive strength σci. The soil bearing capacity predicted by the subsequent machine learning model is the uniformly distributed load on the way. Specifically, the selection of parameters can be adjusted in combination with actual experience, and the above content is only an illustrative example.

[0107] After determining the characteristic parameters, the predicted value of soil bearing capacity corresponding to each set of engineering parameters and the parameter value corresponding to the characteristic parameters are taken as a piece of sample data, so that a large amount of sample data is constructed into a training data set.

[0108] Step S202, iteratively training a plurality of different types of benchmark models according to the training data set, and obtaining a first performance evaluation result corresponding to each trained benchmark model, wherein the benchmark model is used to predict soil bearing capacity.

[0109] Specifically, the training data set includes: a test set and a validation set. In step S202, before obtaining the first performance evaluation results corresponding to each benchmark model after training, the method further includes:

[0110] Based on the validation set, the optimal hyperparameters for each benchmark model are determined through grid search and random search.

[0111] The trained benchmark model is determined based on the model parameters and optimal hyperparameters determined through multiple rounds of iterative training.

[0112] It can be understood that after multiple rounds of iterative training of each benchmark model and determining the model parameters within the model, such as the configuration of weights, the hyperparameters of each trained benchmark model can be optimized through the validation set in the training data. Specifically, the hyperparameter combination corresponding to each benchmark model can be determined by combining grid search and random search, and the model parameters determined by the previous multiple rounds of iterations and the optimal hyperparameter combination can be used to jointly determine the trained benchmark model. At this time, the benchmark model must have a relatively good prediction effect.

[0113] Specifically, in step S202, the first performance evaluation results of each benchmark model after training are obtained, including:

[0114] The initial soil bearing capacity is predicted for the sample data of the test set through each trained benchmark model;

[0115] The mean square error corresponding to each benchmark model is determined based on the initial soil bearing capacity prediction value corresponding to each sample data, and the mean square error is used for performance evaluation.

[0116] It can be understood that when the performance of the trained benchmark model is evaluated through the test set in the training data set, the error between the predicted value and the corresponding label of each sample data of each benchmark model under the test set can be determined, and then the corresponding mean square error can be obtained through mathematical methods to determine the corresponding prediction performance of each benchmark model. The mean square error can reflect the prediction effect of the benchmark model relative to the overall test set and has good performance characterization capabilities.

[0117] Specifically, in step S202, before iteratively training a plurality of different types of benchmark models according to the training data set, the method further includes:

[0118] Corresponding benchmark models are constructed based on linear regression algorithm, support vector regression algorithm, extreme gradient boosting algorithm, random forest algorithm and multilayer perceptron neural network.

[0119] It can be understood that the corresponding machine model, i.e., the benchmark model, can be constructed based on five types of machine learning principles: linear regression algorithm, support vector regression algorithm, extreme gradient boosting algorithm, random forest algorithm, and multilayer perceptron neural network. These five types of machine learning models have their own advantages and disadvantages.

[0120] Among them, the linear regression model is suitable for the prediction of simple linear relationships. The model training speed is fast and easy to explain, but the ability to capture nonlinear relationships is limited. Support vector regression performs well in processing small samples, nonlinear and high-dimensional data by finding the best hyperplane in high-dimensional space. Extreme gradient boosting is an integrated algorithm based on gradient boosting. It has the advantages of processing complex relationships between features and preventing overfitting, and performs outstandingly in various regression and classification tasks. Random forest is an integrated model composed of multiple decision trees. It has good generalization and anti-overfitting capabilities, can process high-dimensional data and capture nonlinear features in the data. Multilayer perceptron neural network is a feedforward neural network that can capture complex nonlinear relationships through multiple hidden layers. It is suitable for large-scale data sets and complex prediction tasks. The results obtained by predicting through these five types of benchmark models can enable the subsequent meta-learner to obtain more comprehensive and accurate soil bearing capacity prediction results when performing the second-layer fusion prediction.

[0121] Step S203: Determine multiple target benchmark models from multiple trained benchmark models based on the first performance evaluation result. Figure 1 Step S103 of the illustrated embodiment will not be described in detail here.

[0122] Step S204: iteratively train multiple meta-learners of different types based on multiple target benchmark models and training data sets, and obtain second performance evaluation results of each meta-learner after training.

[0123] Specifically, in step S204, based on multiple target benchmark models and training data sets, multiple different types of meta-learners are iteratively trained respectively, including:

[0124] Step S204-1-1, obtaining the initial soil bearing capacity prediction value calculated by each target benchmark model based on the sample data in the training data set.

[0125] It can be understood that after completing the training of the benchmark model and determining the target benchmark model, when training the meta-learner, the predicted value of the soil bearing capacity of each target benchmark model relative to each data in the training data set is obtained.

[0126] Step S204-1-2: for each sample data, the initial soil bearing capacity prediction values ​​of each target benchmark model corresponding to the sample data are processed by each meta-learner to obtain the fused soil bearing capacity prediction values ​​corresponding to each meta-learner.

[0127] In the process of training each meta-learner, for each sample data in the training data set, the predicted values ​​of soil bearing capacity corresponding to each target benchmark model are used as multiple input features of each meta-learner. Through the internal calculation of the meta-learner, each meta-learner gives a fused predicted value of soil bearing capacity.

[0128] Step S204-1-3, based on the comparison result between the fused soil bearing capacity prediction value and the soil bearing capacity label corresponding to the sample data, update the internal parameters of the corresponding meta-learner.

[0129] The fused soil bearing capacity prediction value output by the meta-learner is compared with the label of the corresponding sample data, and the internal parameters of the meta-learner are updated according to the comparison result to complete one training. In this way, multiple rounds of iterative training are performed on each meta-learner, which will not be repeated here.

[0130] Step S204-2, obtaining the second performance evaluation results of each meta-learner after training.

[0131] Specifically, in step S204-2, the second performance evaluation results of each meta-learner after training are obtained, including:

[0132] Determine the predicted value of fused soil bearing capacity corresponding to each sample data in the test set through each trained meta-learner and the target benchmark model;

[0133] The mean square error corresponding to each meta-learner is determined based on the fused soil bearing capacity prediction value corresponding to each sample data.

[0134] It can be understood that when evaluating the performance of meta-learning, the sample data in the test set of the training data can be input into the fusion models constructed by multiple target benchmark models and each meta-learner respectively. Specifically, for each combined fusion model, each sample data is first input into each target benchmark model, and then the corresponding soil bearing capacity prediction value is input into the subsequent meta-learner, and finally the corresponding final prediction result is obtained through the meta-learner, that is, the fused soil bearing capacity prediction value, and the prediction result is compared with the label corresponding to the sample data. Similarly, according to the comparison between the prediction result and the label of the entire test set, the mean square error corresponding to each meta-learner under the test set is determined to characterize the performance evaluation results corresponding to each meta-learner, so as to determine the target meta-learner with the best fusion effect.

[0135] Specifically, in the above step S204, before iteratively training multiple different types of meta-learners based on multiple target benchmark models and training data sets, the method further includes:

[0136] Construct the corresponding meta-learner based on the extreme gradient boosting algorithm and linear regression algorithm.

[0137] It can be understood that the extreme gradient boosting algorithm XGBoost and the linear regression algorithm are used as the basis for meta-learning in advance to construct corresponding different types of meta-learners, among which XGBoost has the ability to handle complex nonlinear relationships, while linear regression has good interpretability and computational efficiency. Specifically, when selecting the type of meta-learner, it can be determined in combination with the actual situation. Here is only an exemplary preferred embodiment.

[0138] Furthermore, for the training of the benchmark model and the meta-learner, since the two parts have different purposes, the former is to train the prediction ability of each benchmark model, and the latter is to train the fusion effect of the prediction results of each benchmark model. Therefore, during training, the same training data can be used to complete the training. However, when there is sufficient training data, different training data can be used to train the benchmark model and the meta-learner respectively.

[0139] Specifically, the training data set includes: a first training data set and a second training data set;

[0140] Iteratively train multiple different types of benchmark models based on the training data set, including:

[0141] Iterative training is performed on multiple different types of benchmark models according to the first training data set.

[0142] Based on multiple target benchmark models and training datasets, multiple different types of meta-learners are iteratively trained, including:

[0143] Based on multiple target benchmark models and a second training dataset, multiple meta-learners of different types are iteratively trained respectively.

[0144] It can be understood that, in the training data determined in the above step S201, training data for baseline model training and meta-learner can be divided into a first training data set and a second training data set.

[0145] In the process of training the benchmark model, the first training data set is used, and in the process of training the meta-learning, the second training data set is used.

[0146] Step S205: Based on the second performance evaluation result, determine a target meta-learner from the plurality of trained meta-learners. Figure 1 Step S105 of the illustrated embodiment will not be described in detail here.

[0147] Step S206: construct a fusion model based on the target meta-learner and multiple target benchmark models, and the fusion model is used to predict the soil bearing capacity according to the characteristic parameters corresponding to the target tunnel project. Figure 1 Step S106 of the illustrated embodiment will not be described in detail here.

[0148] The training method of the tunnel bearing capacity prediction model provided by the embodiment of the present invention calculates a large number of data sets that can be used for model training through a finite element limit analysis model, thereby training different types of benchmark models, and after the benchmark model training is completed, multiple target benchmark models with better effects are determined according to the performance evaluation results, so as to further train the meta-learner according to the multiple target benchmark models and the training data, and determine the best target meta-learner, and construct a fusion model in combination with the target benchmark model, so that the fusion model can replace the finite element limit analysis model to predict the tunnel bearing capacity. On the basis of taking into account the prediction accuracy, the advantages of the machine learning model with fast calculation speed and small application restrictions are brought into play, thereby improving the efficiency of tunnel bearing capacity prediction.

[0149] This embodiment also provides a tunnel bearing capacity prediction method, which can be used to predict the soil bearing capacity in tunnel engineering. Figure 4 is a flow chart of a method for predicting tunnel bearing capacity according to an embodiment of the present invention. Figure 4 As shown, the process includes the following steps:

[0150] Step S301, calling a fusion model constructed by any of the above-mentioned training methods for the tunnel bearing capacity prediction model.

[0151] It can be understood that after the fusion model that can be used for predicting the bearing capacity of tunnel soil is obtained through training according to the embodiment corresponding to the above-mentioned model training method, when predicting the bearing capacity of soil in actual engineering projects, the model needs to be called for subsequent predictions.

[0152] Step S302, obtaining target characteristic parameters corresponding to the target tunnel project, inputting the target characteristic parameters into the fusion model, and obtaining first prediction results corresponding to each target benchmark model in the fusion model.

[0153] It can be understood that the fusion model is equivalent to an integrated model, which has two layers of prediction models. The first layer is multiple target benchmark models. After the characteristic parameters corresponding to the actual tunnel project to be predicted are input into the fusion model, it is necessary to input multiple target benchmark models separately to obtain the corresponding first prediction results. At this time, the first layer of prediction is completed.

[0154] Step 303: input the first prediction results corresponding to each target benchmark model into the target meta-learner of the fusion model to obtain the soil bearing capacity prediction result of the target tunnel project.

[0155] It can be understood that after obtaining the first prediction results corresponding to each target benchmark model, these multiple prediction results need to be input into the target meta-learner and fused to take into account the prediction results of each type of target benchmark model, so that the final output prediction results can be more accurate.

[0156] The embodiment of the present invention fuses multiple different types of benchmark models and meta-learners to obtain a fusion model that can have a more accurate prediction effect when predicting the bearing capacity of tunnel soil. At the same time, compared with the finite element limit analysis model under the traditional numerical calculation method, it can have a faster calculation effect, and the environment construction and model use thresholds are greatly reduced, which can greatly improve the use efficiency in the actual engineering environment.

[0157] In order to facilitate understanding of the above method embodiment, according to an embodiment of the present invention, a model training process example diagram is provided, such as Figure 5 shown.

[0158] First, it is necessary to obtain training data, which includes specific characteristic parameters and corresponding mechanical response information. The mechanical response can be understood as the soil bearing capacity calculated by the finite element limit analysis model.

[0159] The training data needs to be preprocessed, such as data cleaning, feature dimension reduction, linear scaling normalization, etc. The general process of data preprocessing is as follows:

[0160] First, the idea of ​​parameterized input and output is adopted to abstract the tunnel structure into a parameterized form for representation. For the tunnel structure, its input parameters can be extracted according to the finite element model. According to the input data file of the finite element, the input parameters of the finite element can be extracted. For the high-dimensional input data of the finite element, the PCA method based on the kernel function can be combined with the artificial extraction of features to perform feature extraction and dimensionality reduction, and extract the input features required by the machine learning model. Design a general data cleaning algorithm for the extracted features. Use a method based on the linear interpolation model to handle missing data values, and remove data with more features. Use the prediction method based on XGBoost to monitor and remove outliers. Use the unique hot encoding form to encode non-numerical data (such as whether the soil is damaged under a certain bearing capacity). Design a general data normalization method. Design different normalization methods for data in different ranges. For numerical data, use the reference value scaling method to limit the data to a roughly similar range, as shown in the following formula.

[0161] X is any mechanical variable, X ref For reference

[0162] For label type data, the one-hot encoding method is used to convert the label into 0-1 label data.

[0163] Next, different types of benchmark models need to be trained, that is, general machine learning models need to be trained.

[0164] The complexity of the tunnel soil bearing capacity prediction problem varies with different types and scenarios, so it is difficult to directly determine a universal intelligent prediction model. In order to cope with various changes in complexity and ensure accurate predictions for various situations, a variety of machine learning models of different complexities were used for training, and the model with the best effect was finally selected. Specifically, five benchmark models were selected for training and verification, including Linear Regression (LR), Support Vector Machine (SVR), eXtreme Gradient Boosting (XGBoost), Random Forest (RF), and Multi-layer Perceptron (MLP) neural network.

[0165] The training process of these models is similar to general machine learning methods. First, the gradient descent algorithm is used on the training set for optimization iterations, and then tested on the validation set and test set. When the predetermined accuracy requirement or training stop condition is reached, the training process is terminated.

[0166] Specifically, for the training data, the k-fold cross-validation method is used to divide the data to ensure the reasonable distribution of the training set, validation set, and test set. In the process of data division, it is necessary to classify the labels of the damaged and undamaged soil data to ensure the uniformity of their distribution in order to improve the training effect and generalization ability of the model.

[0167] After training different types of benchmark models with training data, each benchmark model needs to be optimized and automatically compared for hyperparameters. There are many hyperparameters in machine learning models that need to be manually determined. These hyperparameters may significantly affect the predictive performance of the training model. Therefore, it is necessary to optimize the key hyperparameters and find the optimal hyperparameter combination. In the hyperparameter optimization process, the advantages of random search method and grid search method are combined to balance efficiency and accuracy. The random search method can quickly explore the parameter space by randomly selecting parameter combinations, while the grid search method ensures that the optimal parameter configuration is found by systematically traversing parameter combinations. By combining these two hyperparameter search methods, the key hyperparameters of the model can be efficiently optimized in a limited time.

[0168] In the process of automatic model selection, hyperparameter optimization is performed for the above five models respectively, and the key hyperparameters are selected by combining random search and grid search.

[0169] The k-fold cross-validation method is used to evaluate the performance of each model on the validation set, with mean square error (MSE) as the evaluation criterion for model prediction effect. The automated system sorts the models according to the MSE value and selects the top three models with the smallest mean square error for further model fusion learning. This automated comparison process not only improves the efficiency of model selection, but also ensures the optimality of the selected model in terms of prediction accuracy. Through automated model comparison and hyperparameter optimization methods, high accuracy and high reliability of tunnel soil bearing capacity prediction can be achieved.

[0170] After evaluating the performance of multiple different types of machine learning models, select several best machine learning models for model fusion and output based on the specific evaluation results. For details about model fusion, please refer to Figure 6 , which is an example diagram of a model fusion process according to an embodiment of the present invention.

[0171] First, multiple models are optimized for hyperparameters and automatically selected. For details, please refer to the above related content. After completing model comparison and hyperparameter optimization, the top 3 models and their best hyperparameter combinations obtained in the above steps are further model fusion learning to improve prediction performance and stability. That is, the prediction results of multiple selected models are input into the meta-learner for fusion to obtain the final prediction results that can take into account the prediction results of different types of machine learning models.

[0172] In the training set, the first three models that have been optimized and trained in the third step are used to predict the data. The prediction results generated by each model will be used as input features for the subsequent fusion process. According to the results of the preliminary prediction, different meta-learners are selected for further prediction. Specifically, XGBoost and linear regression are selected as meta-learners. XGBoost has the ability to handle complex nonlinear relationships, while linear regression has good interpretability and computational efficiency. The prediction results of the first three models are used as new feature inputs, and these features are trained using XGBoost and linear regression respectively. In this way, the meta-learner can capture complex patterns and relationships that are not captured in the initial model, further improving the prediction accuracy.

[0173] In order to evaluate the effect of model fusion, the k-fold cross-validation method is used to evaluate the performance of different meta-learners on the validation set. By calculating the MSE on the validation set, the performance of different meta-learners can be compared. According to the cross-validation results, the meta-learner with the smallest MSE on the validation set is selected as the optimal fusion model. This fusion model not only combines the advantages of multiple base models, but also further optimizes the prediction results through the meta-learner.

[0174] Finally, after training to determine the final optimal baseline model and the optimal meta-learner, the final integrated model can be obtained, which can be applied to actual tunnel engineering business.

[0175] After completing model training and optimization, the best trained model is deployed to the production environment for real-time or batch prediction of soil bearing capacity. A complete feedback mechanism is established to compare the prediction results with the actual situation, analyze errors and deviations, and adjust and optimize the model strategy in a timely manner. Through user feedback and continuous updating of field data, the intelligence and adaptability of the model are gradually improved.

[0176] In a production environment, the model can receive new data in real time and predict soil bearing capacity. This is especially important for engineering scenarios that require real-time monitoring and rapid response. By integrating with sensors and data acquisition systems, the model can obtain the latest soil data in real time and quickly output prediction results to assist engineering decision-making.

[0177] For scenarios that require processing large amounts of historical data or regular forecasting, the model can also support batch forecasting. Large-scale data sets are input into the model, forecast results are obtained in batches, and corresponding data analysis and report generation are performed to meet the needs of engineering management and maintenance.

[0178] Compared with the finite element calculation process, the application of trained machine learning models for proxy model calculation has excellent accuracy and extremely high calculation speed. The algorithm can quickly complete the multi-parameter sensitivity analysis of tunnel bearing capacity and then optimize the design of tunnel parameters.

[0179] Compared with general machine learning models, this automatic machine learning link will automatically complete complex feature engineering, model selection, hyperparameter optimization, and model fusion learning processes, quickly train to obtain a highly accurate proxy model, and lower the application threshold of AI.

[0180] This technology can be integrated with traditional finite element analysis methods, using finite element analysis methods to provide sufficient computational data as support, and then learning and migrating the data, thereby solving the problem of insufficient data in traditional machine learning in civil engineering.

[0181] In actual model application, you can refer to Figure 7 The example flow chart shown is an example flow chart of an application of a soil bearing capacity prediction model according to an embodiment of the present invention.

[0182] The input actual engineering data is preprocessed, the parameters corresponding to the characteristic parameters are extracted and input into the above-mentioned model, and then the preliminary predictions are made by multiple machine learning models respectively. The respective prediction results are fused through the meta-learner to obtain the final soil bearing capacity prediction result.

[0183] In the present embodiment, a training device for a tunnel bearing capacity prediction model is also provided, and the device is used to implement the above-mentioned embodiments and preferred embodiments, and the descriptions that have been made will not be repeated. As used below, the term "module" can implement a combination of software and / or hardware of a predetermined function. Although the device described in the following embodiments is preferably implemented in software, the implementation of hardware, or a combination of software and hardware is also possible and conceived.

[0184] This embodiment provides a training device for a tunnel bearing capacity prediction model. Figure 8 As shown, including:

[0185] The data set acquisition module 501 is used to acquire a training data set based on a finite element limit analysis model. The sample data in the training data set includes: characteristic parameters corresponding to the tunnel engineering and soil bearing capacity labels.

[0186] The benchmark model training module 502 is used to iteratively train multiple different types of benchmark models according to the training data set, and obtain the first performance evaluation result corresponding to each trained benchmark model. The benchmark model is used to predict the bearing capacity of soil.

[0187] The target model selection module 503 is used to determine multiple target benchmark models from multiple trained benchmark models based on the first performance evaluation result.

[0188] The meta-learner training module 504 is used to iteratively train multiple meta-learners of different types based on multiple target benchmark models and training data sets, and obtain second performance evaluation results of each meta-learner after training.

[0189] The target meta-learner selection module 505 is used to determine a target meta-learner from a plurality of trained meta-learners based on the second performance evaluation result.

[0190] The fusion model construction module 506 is used to construct a fusion model based on the target meta-learner and multiple target benchmark models. The fusion model is used to predict the soil bearing capacity according to the characteristic parameters corresponding to the target tunnel project.

[0191] In some optional implementations, the meta-learner training module 504, when iteratively training multiple meta-learners of different types based on multiple target benchmark models and training data sets, includes:

[0192] Obtaining the initial soil bearing capacity prediction value calculated by each target benchmark model based on the sample data in the training data set;

[0193] For each sample data, each meta-learner processes the initial soil bearing capacity prediction value of each target benchmark model corresponding to the sample data to obtain the fused soil bearing capacity prediction value corresponding to each meta-learner;

[0194] According to the comparison results between the fused soil bearing capacity prediction value and the soil bearing capacity label corresponding to the sample data, the internal parameters of the corresponding meta-learner are updated.

[0195] In some optional implementations, the data set acquisition module 501, when acquiring a training data set based on a finite element limit analysis model, includes:

[0196] Call the constructed finite element limit analysis model;

[0197] Based on the finite element limit analysis model, multiple tunnel engineering parameter groups are calculated to obtain the soil bearing capacity calculation value corresponding to each tunnel engineering parameter group;

[0198] A plurality of characteristic parameters are determined in the tunnel engineering parameter group, and a training data set is constructed based on parameter values ​​corresponding to each characteristic parameter in each tunnel engineering parameter group and a calculated value of soil bearing capacity.

[0199] In some optional embodiments, the training data set includes: a first training data set and a second training data set;

[0200] The benchmark model training module 502, when iteratively training a plurality of different types of benchmark models according to the training data set, includes:

[0201] Iteratively training a plurality of different types of benchmark models respectively according to the first training data set;

[0202] The meta-learner training module, when iteratively training multiple meta-learners of different types based on multiple target benchmark models and training data sets, includes:

[0203] Based on multiple target benchmark models and a second training dataset, multiple meta-learners of different types are iteratively trained respectively.

[0204] In some optional implementations, the training data set includes: a test set and a validation set;

[0205] The benchmark model training module 502 is further used to:

[0206] Based on the validation set, the optimal hyperparameters for each benchmark model are determined through grid search and random search.

[0207] The trained benchmark model is determined based on the model parameters and optimal hyperparameters determined through multiple rounds of iterative training.

[0208] The benchmark model training module 502, when obtaining the first performance evaluation results of each benchmark model after training, includes:

[0209] The initial soil bearing capacity is predicted for the sample data of the test set through each trained benchmark model;

[0210] The mean square error corresponding to each benchmark model is determined based on the initial soil bearing capacity prediction value corresponding to each sample data, and the mean square error is used for performance evaluation.

[0211] The meta-learner training module 504, when obtaining the second performance evaluation results of each meta-learner after training, includes:

[0212] Determine the predicted value of fused soil bearing capacity corresponding to each sample data in the test set through each trained meta-learner and the target benchmark model;

[0213] The mean square error corresponding to each meta-learner is determined based on the fused soil bearing capacity prediction value corresponding to each sample data.

[0214] In some optional implementations, the benchmark model training module 502 is also used to construct corresponding benchmark models based on a linear regression algorithm, a support vector regression algorithm, an extreme gradient boosting algorithm, a random forest algorithm, and a multilayer perceptron neural network.

[0215] The meta-learner training module 504 is also used to construct a corresponding meta-learner based on the extreme gradient boosting algorithm and the linear regression algorithm.

[0216] This embodiment provides a tunnel bearing capacity prediction device, such as Fig. 9 As shown, including:

[0217] The model calling module 601 is used to call the fusion model constructed by any of the above-mentioned training methods of the tunnel bearing capacity prediction model.

[0218] The preliminary prediction module 602 obtains target characteristic parameters corresponding to the target tunnel project, inputs the target characteristic parameters into the fusion model, and obtains the first prediction results corresponding to each target benchmark model in the fusion model.

[0219] The prediction fusion module 603 inputs the first prediction results corresponding to each target benchmark model into the target meta-learner of the fusion model to obtain the soil bearing capacity prediction result of the target tunnel project.

[0220] The further functional description of each of the above modules and units is the same as that of the above corresponding embodiments and will not be repeated here.

[0221] The training device of the tunnel bearing capacity prediction model and the tunnel bearing capacity prediction device in this embodiment are both presented in the form of functional units, where the units refer to ASIC (Application Specific Integrated Circuit) circuits, processors and memories that execute one or more software or fixed programs, and / or other devices that can provide the above functions.

[0222] The embodiment of the present invention also provides a computer device having the above Figure 8 or Fig. 9 The device shown.

[0223] See also Fig.10 , Fig.10 is a schematic diagram of the structure of a computer device provided by an optional embodiment of the present invention, such as Fig.10As shown, the computer device includes: one or more processors 10, a memory 20, and interfaces for connecting various components, including high-speed interfaces and low-speed interfaces. Various components are connected to each other using different buses for communication, and can be installed on a common mainboard or installed in other ways as needed. The processor can process the instructions executed in the computer device, including instructions stored in or on the memory to display the graphical information of the GUI on an external input / output device (such as, a display device coupled to the interface). In some optional embodiments, if necessary, multiple processors and / or multiple buses can be used together with multiple memories and multiple memories. Similarly, multiple computer devices can be connected, and each device provides some necessary operations (for example, as a server array, a group of blade servers, or a multi-processor system). Fig.10 A processor 10 is taken as an example.

[0224] The processor 10 may be a central processing unit, a network processor or a combination thereof. The processor 10 may further include a hardware chip. The hardware chip may be a dedicated integrated circuit, a programmable logic device or a combination thereof. The programmable logic device may be a complex programmable logic device, a field programmable gate array, a general purpose array logic or any combination thereof.

[0225] The memory 20 stores instructions executable by at least one processor 10, so that at least one processor 10 executes the method shown in the above embodiment.

[0226] The memory 20 may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data created according to the use of the computer device, etc. In addition, the memory 20 may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some optional embodiments, the memory 20 may optionally include a memory remotely arranged relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0227] The memory 20 may include a volatile memory, such as a random access memory; the memory may also include a non-volatile memory, such as a flash memory, a hard disk or a solid state drive; the memory 20 may also include a combination of the above types of memory.

[0228] The computer device also includes an input device 30 and an output device 40. The processor 10, the memory 20, the input device 30 and the output device 40 may be connected via a bus or other means. Fig.10 The example of connecting through bus is taken in the following.

[0229] The input device 30 can receive input digital or character information, and generate key signal input related to the user settings and function control of the computer device, such as a touch screen, a keypad, a mouse, a track pad, a touch pad, an indicator bar, one or more mouse buttons, a trackball, a joystick, etc. The output device 40 may include a display device, an auxiliary lighting device (e.g., an LED) and a tactile feedback device (e.g., a vibration motor), etc. The above-mentioned display device includes but is not limited to a liquid crystal display, a light emitting diode, a display and a plasma display. In some optional embodiments, the display device can be a touch screen.

[0230] The embodiment of the present invention also provides a computer-readable storage medium. The method according to the embodiment of the present invention can be implemented in hardware, firmware, or can be implemented as a computer code that can be recorded in a storage medium, or can be implemented as a computer code that is originally stored in a remote storage medium or a non-temporary machine-readable storage medium and will be stored in a local storage medium through a network download, so that the method described herein can be stored in such software processing on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only storage memory, a random access memory, a flash memory, a hard disk or a solid-state hard disk, etc.; further, the storage medium can also include a combination of the above types of memories. It can be understood that a computer, a processor, a microprocessor controller, or programmable hardware includes a storage component that can store or receive software or computer code. When the software or computer code is accessed and executed by a computer, a processor, or hardware, the method shown in the above embodiment is implemented.

[0231] Although the embodiments of the present invention have been described in conjunction with the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present invention, and such modifications and variations are all within the scope defined by the appended claims.

Claims

1. A training method for a tunnel bearing capacity prediction model, characterized in that: The method comprises: Acquire a training data set based on a finite element limit analysis model, wherein the sample data in the training data set includes: characteristic parameters corresponding to the tunnel project and soil bearing capacity labels; Iteratively training a plurality of different types of benchmark models according to the training data set, and obtaining first performance evaluation results corresponding to each trained benchmark model, wherein the benchmark model is used to predict soil bearing capacity; Determining a plurality of target benchmark models from a plurality of trained benchmark models based on the first performance evaluation result; Iteratively training a plurality of meta-learners of different types based on the plurality of target benchmark models and the training data set, and obtaining a second performance evaluation result of each meta-learner after training; Based on the second performance evaluation result, determining a target meta-learner from a plurality of trained meta-learners; A fusion model is constructed based on the target meta-learner and the multiple target benchmark models, and the fusion model is used to predict the soil bearing capacity according to the characteristic parameters corresponding to the target tunnel project.

2. The method according to claim 1, characterized in that The iteratively training a plurality of different types of meta-learners based on the plurality of target benchmark models and the training data set comprises: Obtaining initial soil bearing capacity prediction values ​​calculated by each target benchmark model based on sample data in the training data set; For each sample data, the initial soil bearing capacity prediction values ​​of each target benchmark model corresponding to the sample data are processed by each meta-learner to obtain the fused soil bearing capacity prediction values ​​corresponding to each meta-learner; According to the comparison result between the fused soil bearing capacity prediction value and the soil bearing capacity label corresponding to the sample data, the internal parameters of the corresponding meta-learner are updated.

3. The method according to claim 1, characterized in that The step of acquiring a training data set based on a finite element limit analysis model includes: Call the constructed finite element limit analysis model; Calculating multiple tunnel engineering parameter groups based on the finite element limit analysis model to obtain soil bearing capacity calculation values ​​corresponding to each tunnel engineering parameter group; A plurality of characteristic parameters are determined in the tunnel engineering parameter group, and a training data set is constructed based on parameter values ​​corresponding to each characteristic parameter in each tunnel engineering parameter group and the calculated value of the soil bearing capacity.

4. The method according to claim 1, characterized in that: The training data set includes: a first training data set and a second training data set; The iterative training of multiple different types of benchmark models according to the training data set includes: Iteratively training a plurality of different types of benchmark models according to the first training data set; The iteratively training a plurality of different types of meta-learners based on the plurality of target benchmark models and the training data set comprises: Based on the multiple target benchmark models and the second training data set, multiple meta-learners of different types are iteratively trained respectively.

5. The method according to claim 1, characterized in that The training data set includes: a test set and a validation set; Before obtaining the first performance evaluation results corresponding to each trained benchmark model, the method further includes: Based on the validation set, determine the optimal hyperparameters corresponding to each benchmark model through grid search and random search; Determine a trained benchmark model based on the model parameters determined by multiple rounds of iterative training and the optimal hyperparameters; The obtaining of the first performance evaluation results of each benchmark model after training includes: Predicting the initial soil bearing capacity of the sample data of the test set by using each trained benchmark model; Determine the mean square error corresponding to each benchmark model based on the initial soil bearing capacity prediction value corresponding to each sample data, wherein the mean square error is used for performance evaluation; The obtaining of the second performance evaluation results of each meta-learner after training includes: Determine the fused soil bearing capacity prediction value corresponding to each sample data in the test set through each trained meta-learner and the target benchmark model; The mean square error corresponding to each meta-learner is determined based on the fused soil bearing capacity prediction value corresponding to each sample data.

6. The method according to claim 1, characterized in that The method further comprises: Corresponding benchmark models are constructed based on linear regression algorithm, support vector regression algorithm, extreme gradient boosting algorithm, random forest algorithm and multi-layer perceptron neural network respectively; Construct the corresponding meta-learner based on the extreme gradient boosting algorithm and linear regression algorithm.

7. A method for predicting tunnel bearing capacity, characterized in that: The method comprises: A fusion model constructed by calling the training method of the tunnel bearing capacity prediction model of any one of claims 1 to 6; Obtaining target characteristic parameters corresponding to the target tunnel project, inputting the target characteristic parameters into the fusion model, and obtaining first prediction results corresponding to each target benchmark model in the fusion model; The first prediction results corresponding to the respective target benchmark models are input into the target meta-learner of the fusion model to obtain the soil bearing capacity prediction result of the target tunnel project.

8. A training device for a tunnel bearing capacity prediction model, characterized in that: The device comprises: A data set acquisition module is used to acquire a training data set based on a finite element limit analysis model, wherein the sample data in the training data set includes: characteristic parameters corresponding to the tunnel engineering and soil bearing capacity labels; A benchmark model training module, used for iteratively training a plurality of different types of benchmark models according to the training data set, and obtaining a first performance evaluation result corresponding to each trained benchmark model, wherein the benchmark model is used to predict soil bearing capacity; a target model selection module, configured to determine a plurality of target benchmark models from a plurality of trained benchmark models based on the first performance evaluation result; A meta-learner training module, configured to iteratively train a plurality of meta-learners of different types based on the plurality of target benchmark models and the training data set, and obtain a second performance evaluation result of each meta-learner after training; a target meta-learner selection module, configured to determine a target meta-learner from among a plurality of trained meta-learners based on the second performance evaluation result; A fusion model construction module is used to construct a fusion model based on the target meta-learner and the multiple target benchmark models, and the fusion model is used to predict the soil bearing capacity according to the characteristic parameters corresponding to the target tunnel project.

9. A computer device, characterized in that: include: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the training method of the tunnel bearing capacity prediction model described in any one of claims 1 to 6 or the tunnel bearing capacity prediction method described in claim 7 by executing the computer instructions.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the training method of the tunnel bearing capacity prediction model described in any one of claims 1 to 6 or the tunnel bearing capacity prediction method described in claim 7.

Images