A hierarchical prediction method for large tunnel deformation and related products

Abstract

The present invention relates to the field of civil engineering technology, and in particular to a hierarchical prediction method for large deformation of a tunnel and related products, comprising steps of data collection and preprocessing, feature selection and index analysis, model selection and training, model optimization and adjustment, and deformation prediction. The present invention filters out noise in data in the preprocessing stage, extracts key information at different scales, and ensures the high quality of input data. The most representative features are selected through mutual information and mRMR algorithms, which reduces the complexity of the model while improving the generalization ability of the model. The SVM model is combined with a combined parameter optimization algorithm to quickly search for optimal hyperparameters, thereby improving the prediction performance of the model under different environments. Finally, through real-time deformation data input, the present invention can realize accurate hierarchical prediction of tunnel deformation, provide an important reference for engineering management and decision-making, and improve the safety and reliability of tunnel structures.

Description

Technical Field

[0001] The invention relates to the technical field of civil engineering, and in particular to a hierarchical prediction method for large deformation of a tunnel and related products. Background Art

[0002] Swelling mudstone has significant expansion and contraction characteristics. It is easy to expand in volume when it encounters water, which in turn causes deformation and stress concentration of the surrounding rock, posing a threat to the structural safety of the tunnel. During the construction and operation of the swelling mudstone tunnel, the deformation of the tunnel is a complex dynamic process, which is affected by many factors, such as the expansion rate, water content, tunnel span and surrounding rock strength of the surrounding rock. Therefore, accurately predicting the deformation of the swelling mudstone tunnel is of great significance to ensure the safety of the tunnel and extend its service life.

[0003] In the prior art, the prediction methods for large deformation of tunnels mainly rely on empirical models or numerical simulations, which have many limitations. First, empirical models are usually based on historical data, but due to the complexity of geological conditions, it is difficult to fully adapt to different engineering scenarios and the prediction accuracy is not high. Secondly, although the numerical simulation method can consider more influencing factors, the parameter setting of the model is complex, the calculation cost is high, and the results are easily affected by the model assumptions. In addition, the existing methods generally lack effective feature extraction and penalty coefficient optimization methods, resulting in limited generalization ability of the model and difficulty in dealing with complex situations in actual engineering.

[0004] In order to meet the above challenges, in recent years, machine learning methods have gradually been applied to the field of tunnel deformation prediction, especially support vector machine (SVM) and other algorithms, which have received widespread attention due to their good nonlinear mapping capabilities and high prediction accuracy. However, the existing SVM-based prediction methods still face some problems, such as the lack of systematic guidance in feature selection and penalty coefficient optimization, resulting in unsatisfactory model prediction results. In addition, how to more effectively deal with noise, outliers and multi-scale information in tunnel deformation data is also a difficulty in current technical research.

[0005] In summary, it is a technical problem to be solved urgently to develop a hierarchical prediction method that can accurately predict large deformation of tunnels and improve the generalization ability and prediction accuracy of the model through reasonable data processing and feature selection techniques. Summary of the invention

[0006] The technical problem to be solved by the present invention is how to improve the accuracy and generalization ability of tunnel large deformation prediction, overcome the problems of unreasonable feature selection, insufficient optimization of penalty coefficient and improper data noise processing in the existing methods, and aim to provide a hierarchical prediction method and related products for tunnel large deformation, realize accurate hierarchical prediction based on support vector machine, improve the robustness of the model by optimizing feature selection and penalty coefficient tuning, and improve data quality by using effective data preprocessing methods, so as to ensure the applicability and stability of the model under complex engineering conditions.

[0007] The present invention is achieved through the following technical solutions:

[0008] A hierarchical prediction method for large deformation of a tunnel, comprising:

[0009] Data collection and preprocessing: collect deformation data from historical tunnel projects and preprocess the collected deformation data to obtain sample data;

[0010] Feature selection and index analysis: analyze sample data and select key features that characterize deformation;

[0011] Model selection and training: Train the support vector machine through key features to obtain the prediction model;

[0012] Model optimization and adjustment: verify the prediction model and optimize the model parameters according to the verification results to obtain the final hierarchical prediction model;

[0013] Deformation prediction, input the real-time deformation data of the tunnel, and the graded prediction model outputs the predicted deformation grade.

[0014] Optionally, the method for preprocessing the deformation data includes:

[0015] The noise of the deformation data was filtered by the three-step moving average method;

[0016] Apply Z-score normalization to the noise-filtered deformation data;

[0017] The missing values ​​of the deformed data after data standardization were processed by linear interpolation;

[0018] The deformation data with missing values ​​processed are analyzed at multiple scales by wavelet transform method, and the features at different scales are extracted as sample data.

[0019] Optionally, methods for screening key features include:

[0020] Determine the mutual information between features and target variables in turn ,in, Features With the target variable The mutual information between for and The joint probability distribution between for The marginal probability distribution of for The marginal probability distribution of is the preprocessed deformation data, the target variable For deformation;

[0021] Select the feature with the largest mutual information ; and Add to selected feature set ;

[0022] Calculate the feature Redundancy between other features and target variables ,in, Features With the target variable The mutual information between It is a feature With selected feature set Medium Features The mutual information between

[0023] Select the feature with the smallest redundancy value ; and Add to selected feature set ;

[0024] Calculate Divide and Features The redundancy value between the remaining features and the target variable is calculated, and the feature with the smallest redundancy value is selected and added to the selected feature set. ;

[0025] Iterate until the stopping condition is reached to obtain the final selected feature set; and use the features in the selected feature set as key features; the stopping condition is that the number of features in the selected feature set reaches a preset value or the calculated redundancy value is lower than the set value.

[0026] Optionally, it further includes filtering out some redundant or low-contribution features in the selected feature set, the method comprising:

[0027] Construct a feature screening model based on the SVM model and set the selected feature set As input;

[0028] By selected feature set Iteratively train the feature screening model. In the first iteration, the coefficient of each feature is obtained through the feature screening model, and the decision function of the feature screening model is ,in, For the The feature in The coefficients in the iterations, For the Features, For the The selected feature set for iteration, For the At iteration The number of features in is bias;

[0029] according to Sort by size, remove the smallest feature, and obtain a new feature set ;

[0030] Repeat the iteration until the number of features in the new feature set reaches the preset number or the model performance meets the preset requirements;

[0031] Get the final feature set , and the features in the final feature set are taken as key features.

[0032] Specifically, the support vector machine is a nonlinear support vector machine, and the key features are used as input variables and the deformation level is used as the output label;

[0033] Construct a feature set, which includes input variables and output tags ,in, The final feature set The number of features in is the number of levels of deformation;

[0034] Determine the decision function and kernel function of the nonlinear support vector machine;

[0035] The penalty coefficient is optimized by combining the parameters and kernel parameters Optimize, select the optimal hyperparameters, and build a prediction model with the optimal hyperparameters.

[0036] Optionally, the combined parameter optimization algorithm includes:

[0037] Random generation includes A population of individuals, each individual in the population is ;

[0038] Evaluate the fitness of each individual and sort each individual according to the size of the fitness. Take the first multiple individuals as parents, perform crossover operations on the selected parents, and generate offspring individuals. ,in, , , and For the father generation, is the randomly generated crossover coefficient;

[0039] Randomly change the parameters of some individuals to generate new individuals;

[0040] Repeat the iteration until the maximum number of iterations is reached or the fitness converges, sort all parent individuals and child individuals according to their fitness, and select the top Individuals are used as primary hyperparameters;

[0041] Use the primary hyperparameters as the initial individuals in the initial population;

[0042] Search in the neighborhood of each initial individual to generate a new initial individual ,in, To control the random step size of the search, is another initial individual selected randomly;

[0043] Calculate the fitness of the new initial individual. If the new initial individual The fitness is greater than the current initial individual The fitness of ;otherwise, constant;

[0044] Calculate the fitness of all initial individuals and determine the search probability ,in, For the The fitness value of the initial individuals;

[0045] Sort the primary hyperparameters by the search probability and select the top A primary hyperparameter is used to perform a local search in its field to generate a new initial individual. If the fitness of the new initial individual is greater than that of the current initial individual, the current initial individual is updated; otherwise, the current initial individual remains unchanged.

[0046] In repeated iterative search, if the fitness of an initial individual remains unchanged in multiple iterations, a new initial individual is randomly generated and the new initial individual is used to replace the current initial individual;

[0047] Repeat the iteration until the preset number of iterations is reached or the fitness value of the population converges;

[0048] Sort all initial individuals according to their fitness, and filter them out The initial individuals are used as secondary hyperparameters;

[0049] Select the secondary hyperparameter with the largest fitness as the optimal hyperparameter .

[0050] Optionally, set are the parameters of the prediction model before optimization, and include secondary hyperparameters; iterate over all parameters , the methods for optimizing the prediction model include:

[0051] Choose multiple different outer cross validation Values ​​and inner cross validation value;

[0052] For each The outer cross validation is performed and each The average accuracy corresponding to the value ,in, For the The number of true positives of the model prediction results in the outer cross-validation, For the The number of true negatives of the model prediction results in the outer cross-validation, For the The number of false positives in the model prediction results in the outer cross-validation, For the The number of false negatives in the model prediction results in the outer cross-validation;

[0053] Choose the optimal number of folds for outer cross validation ,in, The fold is The average accuracy of the outer cross validation is The fold is The variance of the average accuracy of the outer cross validation;

[0054] By optimal fold Determine the training set for inner cross-validation;

[0055] Through inner cross validation Optimize the parameters of the test model , For the Given parameters under inner layer cross validation The verification accuracy of For a given parameter The uncertainty of the model prediction results, To find the mean value, are the optimized parameters found in the inner cross validation;

[0056] The average accuracy in the inner cross validation is used as the objective function , optimize the parameters of the test model ,in, is the estimate of the expected objective function, is the balance parameter between exploration and exploitation, For a given parameter The uncertainty of the model prediction results, are the final parameters after optimization.

[0057] Specifically, , ,in, For a given parameter All The average accuracy of the inner layer cross validation; For a given parameter All The average accuracy of the within-fold cross validation.

[0058] A computer program related product includes a computer program / instruction, which, when executed by a processor, implements the above-mentioned hierarchical prediction method for large deformation of a tunnel.

[0059] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0060] The present invention provides a hierarchical prediction method for large deformation of a tunnel, which mainly includes the steps of data collection and preprocessing, feature selection and index analysis, model selection and training, model optimization and adjustment, and deformation prediction;

[0061] The present invention filters out noise in the data in the preprocessing stage and extracts key information at different scales, ensuring the high quality of input data; selects the most representative features through mutual information and mRMR algorithms, reduces the complexity of the model while improving the generalization ability of the model; combines the SVM model with the combined parameter optimization algorithm to quickly search for the optimal hyperparameters, improving the prediction performance of the model in different environments; finally, through real-time deformation data input, the present invention can achieve accurate hierarchical prediction of tunnel deformation, provides an important reference for engineering management and decision-making, and improves the safety and reliability of tunnel structures. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] The accompanying drawings illustrate exemplary embodiments of the present invention and, together with the description thereof, are used to explain the principles of the present invention. These drawings are included to provide a further understanding of the present invention, and the accompanying drawings are included in and constitute a part of this specification and do not constitute a limitation of the embodiments of the present invention.

[0063] Figure 1 It is a schematic flow chart of a hierarchical prediction method for large deformation of a tunnel according to the present invention.

[0064] Figure 2 It is a schematic diagram of the process of the key feature screening method according to the present invention. DETAILED DESCRIPTION

[0065] To make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and implementation methods. It is understood that the specific implementation methods described herein are only used to explain the relevant content, rather than to limit the present invention.

[0066] It should also be noted that, for the convenience of description, only the parts related to the present invention are shown in the drawings.

[0067] In the absence of conflict, the embodiments and features of the embodiments of the present invention may be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0068] Embodiment 1, as Figure 1 As shown, a hierarchical prediction method for large deformation of a tunnel includes:

[0069] The first step is data collection and preprocessing. The deformation data from historical tunnel projects are collected and preprocessed to obtain sample data.

[0070] Relevant deformation data are collected from historical tunnel projects and processed to ensure data quality. During tunnel construction, the deformation of expansive mudstone is complex and variable, and the data may contain noise, be incomplete, or have different scales. Therefore, data preprocessing usually includes noise filtering, data standardization, and missing value processing. The preprocessed data is converted into high-quality sample data to ensure that the data input for subsequent model training is accurate and stable.

[0071] The second step, feature selection and index analysis, analyzes the sample data and screens the key features that characterize the deformation.

[0072] Deformation data contains a lot of data, some of which may not affect the deformation level. Therefore, by analyzing the preprocessed data, the key features that can most effectively characterize tunnel deformation are screened out. In this embodiment, mutual information and minimum redundancy maximum correlation algorithms are used to reduce irrelevant or redundant information, retain the features that are most helpful for prediction results, simplify the model and improve its generalization ability.

[0073] The third step is model selection and training. The support vector machine is trained through key features to obtain the prediction model.

[0074] Support vector machine (SVM) is selected as the core prediction model. By learning key features, SVM can effectively build a prediction model.

[0075] The fourth step is model optimization and adjustment. The prediction model is verified and the model parameters are optimized according to the verification results to obtain the final hierarchical prediction model.

[0076] After obtaining the preliminary prediction model, the parameters of SVM (such as penalty coefficient and kernel parameter) are optimized and adjusted by combining multiple penalty coefficient optimization algorithms, and finally the prediction model with the best performance is obtained, which improves the accuracy and robustness of the model and enables it to perform stably on different data sets.

[0077] Application steps: deformation prediction, real-time input of tunnel deformation data, and output of predicted deformation level by the graded prediction model. The graded prediction model can dynamically output the deformation level of the tunnel based on real-time geological data, environmental data and construction data, helping engineers to take corresponding protective measures in a timely manner.

[0078] In the second embodiment, the method for preprocessing deformation data includes:

[0079] The noise of the deformed data is filtered by the triple moving average method; the triple moving average is three consecutive moving average operations on the same data point. In each operation, the data point value is replaced by the average of the data points before and after that point. The triple moving average method smoothes the short-term fluctuations in the deformed data, thereby highlighting the trends and long-term changes in the data.

[0080] Apply Z-score normalization to the deformed data after noise filtering; Z-score normalization converts the data into a standard normal distribution with a mean of 0 and a standard deviation of 1. The purpose is to eliminate the differences in magnitude and units of different feature data so that each feature has the same dimension. The normalization formula is: ,in, is the original data, is the data mean, is the standard deviation of the data.

[0081] The missing values ​​of the deformed data after data standardization are processed by linear interpolation. Linear interpolation assumes that the data changes linearly between the points before and after the missing value, and estimates the missing data by connecting the known data points before and after. The formula is: ,in, and is the time point before and after the missing value, and is the corresponding deformation value.

[0082] The wavelet transform method is used to perform multi-scale analysis on the deformation data after missing value processing, and extract features at different scales as sample data. Wavelet transform can decompose the signal into different frequency components for analysis. Unlike the traditional Fourier transform, wavelet transform can analyze in both time domain and frequency domain, thus capturing the changes of the signal at different time scales.

[0083] Embodiment 3: After processing all deformation data, the deformation data includes but is not limited to:

[0084] Horizontal displacement: the horizontal movement of the tunnel surrounding rock.

[0085] Vertical displacement: vertical deformation of the tunnel crown or bottom, such as crown settlement or tunnel floor heaving.

[0086] Surrounding rock stress: data on stress changes in surrounding rock after tunnel excavation.

[0087] Support structure stress: the stress borne by the tunnel support system.

[0088] Water content: The water content of the rock and soil layers around the tunnel. The deformation of expansive mudstone is significantly affected by water.

[0089] Groundwater level: The impact of changes in groundwater near the tunnel on rock deformation.

[0090] Temperature variations: Fluctuations in temperature inside and outside the tunnel may affect the deformation characteristics of the material.

[0091] Humidity: In a high humidity environment, the water absorption and expansion of mudstone may be more obvious.

[0092] Crack width variation: variation in crack width on tunnel support or lining structures.

[0093] Lining surface deformation: the overall deformation of the lining structure, such as extrusion deformation.

[0094] Swelling pressure: The pressure generated by the expansion of expansive mudstone when it comes into contact with water will affect the deformation of the tunnel structure.

[0095] like Figure 2 As shown, some deformation data may have little effect on the deformation situation, so it is necessary to screen key features. The methods include:

[0096] Mutual information is a nonlinear measure of the dependence between two random variables. The larger the mutual information, the stronger the correlation between the two variables.

[0097] Determine the mutual information between features and target variables in turn ,in, Features With the target variable The mutual information between for and The joint probability distribution between for The marginal probability distribution of for The marginal probability distribution of is the preprocessed deformation data, the target variable is the deformation situation; mutual information reflects the observed features Get information about the target variable The degree of information.

[0098] Select the feature with the largest mutual information ; and Add to selected feature set ; The prediction ability of the target variable is the strongest.

[0099] When selecting the next feature, two aspects need to be considered: the correlation with the target variable and the redundancy with the selected features. Redundancy between other features and target variables ,in, Features With the target variable The mutual information between It is a feature With selected feature set Medium Features The mutual information between

[0100] After calculating the redundancy values ​​of all unselected features, the feature with the smallest redundancy value is selected and added to the selected feature set. This new feature is not only highly correlated with the target variable, but also has the smallest redundancy with the selected feature set, which can maximize the predictive ability of the model.

[0101] Select the feature with the smallest redundancy value ; and Add to selected feature set ;Sure Redundancy with the selected features. Here we use the average redundancy, which is divided by the number of selected features. .

[0102] Calculate Divide and Features The redundancy value between the remaining features and the target variable is calculated, and the feature with the smallest redundancy value is selected and added to the selected feature set. That is, repeat the above steps, continue to calculate the redundancy values ​​of the remaining features and the target variable, and continue to select the features with the smallest redundancy to add to the selected feature set. The redundancy value represents the similarity or information overlap between the new feature and the selected feature. Selecting features with small redundancy values ​​means that the new feature shares less redundant information with the features in the selected feature set, and carries more independent and useful information. Features with small redundancy values ​​are preferentially selected to add to the selected feature set to minimize information redundancy between features.

[0103] Iterate until the stopping condition is reached to obtain the final selected feature set; and use the features in the selected feature set as key features; the stopping condition is that the number of features in the selected feature set reaches the preset value (that is, the number of features reaches the upper limit) or the calculated redundancy value is lower than the set value (indicating that the new feature contributes little to the model). When the redundancy value is lower than the set value, it means that the redundant information between the new feature and the selected feature is already very small, that is, continuing to add new features will not bring significant incremental information to the model.

[0104] Embodiment 4: After embodiment 3 is completed, the selected feature set may contain features with small contribution, so it is necessary to filter out some redundant or small-contribution features in the selected feature set, and the method includes:

[0105] Construct a feature screening model based on the SVM model and set the selected feature set As input; using SVM can help evaluate the specific contribution of each feature to the model performance, which helps to further optimize the feature set.

[0106] By selected feature set Iteratively train the feature screening model. In the first iteration, the coefficient of each feature is obtained through the feature screening model, and the decision function of the feature screening model is ,in, For the The feature in The coefficient in the iteration reflects the influence of the feature on the model decision function. For the Features, For the The selected feature set for iteration, For the At iteration The number of features in is bias;

[0107] according to Sort by size, remove the smallest feature, and obtain a new feature set ; A larger coefficient indicates that the feature contributes more to the model, while a smaller coefficient indicates that the feature has less impact on the model decision and does not contribute much. Removing features with small contributions can help simplify the model while retaining features that have a greater impact on the prediction results and optimize the performance of the model.

[0108] Repeat the iteration until the number of features in the new feature set reaches the preset number (the size of the feature set is reduced to the preset number of features) or the model performance (such as accuracy, F1 score, etc.) meets the preset requirements;

[0109] Get the final feature set , and the features in the final feature set are taken as key features.

[0110] In the fifth embodiment, the support vector machine is a nonlinear support vector machine, and the key features are used as input variables and the deformation level is used as the output label;

[0111] Construct a feature set, which includes input variables and output tags , represents different deformation levels of the tunnel (e.g., different degrees of deformation, risk levels, etc.), where The final feature set The number of features in is the number of levels of deformation.

[0112] The feature set contains all the key features obtained through feature screening, which may include various data such as tunnel displacement, stress, environmental monitoring, etc. The output label is the number of possible deformation levels, which can be defined as different risk levels or deformation conditions according to actual needs, such as slight, medium, severe deformation, etc.

[0113] Determine the decision function and kernel function of the nonlinear support vector machine.

[0114] The decision function is used to determine the classification boundary. For linearly inseparable data, the nonlinear decision function of SVM is as follows: ,in, is the Lagrange multiplier, Is the kernel function, used to calculate the input features and the current data point The similarity of is the bias term.

[0115] Common kernel functions include linear kernel function, polynomial kernel function and radial basis function RBF. When processing tunnel deformation prediction, tunnel deformation data may be nonlinear, so RBF kernel function is usually selected: .

[0116] The penalty coefficient is optimized by combining the parameters and kernel parameters Optimize, select the optimal hyperparameters, and build a prediction model with the optimal hyperparameters.

[0117] The penalty coefficient controls the model's tolerance to misclassification, and the kernel parameter controls the influence range of the RBF kernel function.

[0118] Among them, the combined parameter optimization algorithm includes:

[0119] Random generation includes A population of individuals, each individual in the population is that is, by randomly generating multiple parameter combinations.

[0120] Evaluate the fitness of each individual. The fitness represents the performance of the hyperparameter combination in the SVM model. The accuracy or other performance indicators of the model are usually calculated through methods such as cross-validation.

[0121] Sort each individual according to the size of the fitness, take the first multiple individuals as parents, perform crossover operation on the selected parents, and generate offspring individuals ,in, , , and For the father generation, The randomly generated crossover coefficient is used to mix the parameters of the parent individuals to generate new individuals. The crossover operation simulates the biological reproduction process, generating a new generation of individuals by mixing the parameters of different individuals, maintaining the diversity of the population, and exploring new solutions.

[0122] Randomly change the parameters of some individuals to generate new individuals; that is, perform mutation operations and further introduce new parameter combinations to prevent the population from falling into a local optimal solution.

[0123] Repeat the crossover and mutation operations to generate a new generation of individuals, and evaluate the fitness of the newly generated individuals. Repeat the iteration until the maximum number of iterations is reached or the fitness converges. Sort all parent individuals and child individuals according to their fitness size, and select the top individuals as primary hyperparameters.

[0124] Use the primary hyperparameters as the initial individuals in the initial population;

[0125] Search in the neighborhood of each initial individual to generate a new initial individual ,in, To control the random step size of the search, is another initial individual selected randomly;

[0126] Calculate the fitness of the new initial individual. If the new initial individual The fitness is greater than the current initial individual The fitness of ;otherwise, unchanged; local search helps to further fine-tune around the optimal parameters and find possible better parameter combinations.

[0127] Calculate the fitness of all initial individuals and determine the search probability ,in, For the The fitness value of the initial individuals; through the search probability, the search can be concentrated in the most promising area according to the size of the fitness, gradually approaching the optimal solution.

[0128] Sort the primary hyperparameters by the search probability and select the top A primary hyperparameter is used to perform a local search in its field to generate a new initial individual. If the fitness of the new initial individual is greater than that of the current initial individual, the current initial individual is updated; otherwise, the current initial individual remains unchanged.

[0129] In repeated iterative search, if the fitness of an initial individual remains unchanged in multiple iterations, a new initial individual is randomly generated and the new initial individual is used to replace the current initial individual; replacing stagnant individuals can avoid falling into a local optimal solution.

[0130] Iterations are repeated until the preset number of iterations is reached or the fitness value of the population converges (i.e., the model performance has not been significantly improved).

[0131] Sort all initial individuals according to their fitness, and filter them out The initial individuals are used as secondary hyperparameters;

[0132] Select the secondary hyperparameter with the largest fitness as the optimal hyperparameter .

[0133] The optimal hyperparameters can be further optimized.

[0134] set up are the parameters of the prediction model before optimization, and include secondary hyperparameters; that is, after the aforementioned hyperparameter optimization process, a set of secondary hyperparameters has been obtained - a set of candidate better hyperparameters.

[0135] Cross-validation is a common method to verify model performance. The dataset is divided into K folds, and K-1 folds are selected as training sets each time, and the remaining fold is used as a validation set. The model is trained and validated on each fold, and the average value is taken to evaluate the generalization ability of the model.

[0136] The outer cross-validation is used to evaluate the overall model performance, while the inner cross-validation is used to tune the model parameters.

[0137] Iterate over all parameters , the methods for optimizing the prediction model include:

[0138] Choose multiple different outer cross validation Values ​​and inner cross validation value;

[0139] For each The outer cross validation is performed and each The average accuracy corresponding to the value ,in, For the The number of true positives of the model prediction results in the outer cross-validation, For the The number of true negatives of the model prediction results in the outer cross-validation, For the The number of false positives in the model prediction results in the outer cross-validation, For the The number of false negatives in the model prediction results in the outer fold cross-validation; by calculating the average accuracy of multiple folds, the performance of the model under different data partitions can be comprehensively evaluated.

[0140] The number of folds K in the outer cross-validation will affect the performance evaluation of the model. The larger the number of folds, the less data in each fold, resulting in a more refined model evaluation, but at the same time it may bring about a larger variance (volatility). Therefore, it is necessary to find a fold that can balance accuracy and variance, that is, to choose the optimal fold for the outer cross-validation. ,in, The fold is The average accuracy of the outer cross validation is The fold is The purpose of optimizing the folds is to find the folds that are most suitable for evaluating the performance of the model, so that the model has high accuracy while maintaining stability and avoiding overfitting or underfitting.

[0141] By optimal fold Determine the training set for inner cross-validation; use the number of folds Divide the data set and use the divided data for inner cross-validation.

[0142] Inner cross validation is mainly used to tune model parameters and maximize validation accuracy. Optimize the parameters of the test model , , For the Given parameters under inner layer cross validation The verification accuracy of For a given parameter The uncertainty of the model prediction results, To find the mean value, The optimized parameters found in the inner cross-validation; in the inner cross-validation, the model parameters are continuously adjusted To find the parameters that maximize validation accuracy and minimize uncertainty .

[0143] In the inner cross-validation process, by continuously tuning the parameters of the model , multiple candidate parameters will be generated .

[0144] The average accuracy in the inner cross validation is used as the objective function , optimize the parameters of the test model ,in, is the estimate of the expected objective function, is the balance parameter between exploration and exploitation, For a given parameter The uncertainty of the model prediction results, are the final parameters after optimization.

[0145] , ,in, For a given parameter All The average accuracy of the inner layer cross validation; For a given parameter All The average accuracy of the intra-fold cross-validation. Uncertainty measures the performance fluctuation of each parameter combination in different validation folds, that is, the stability of the model under different training and test set partitions.

[0146] Embodiment 6. A computer-readable storage medium stores a computer program. When the computer program is executed by a processor, the method for hierarchical prediction of large deformation of a tunnel as described above is implemented.

[0147] Without loss of generality, computer readable media may include computer storage media and communication media. Computer storage media include volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information such as computer readable instruction data structures, program modules or other data. Computer storage media include RAM, ROM, EPROM, EEPROM, flash memory or other solid-state storage technology, CD-ROM, DVD or other optical storage, cassettes, magnetic tapes, disk storage or other magnetic storage devices. Of course, those skilled in the art will appreciate that computer storage media are not limited to the above. The above-mentioned system memory and mass storage devices can be collectively referred to as memory.

[0148] A computer program product includes a computer program / instruction, which, when executed by a processor, implements the above-mentioned hierarchical prediction method for large deformation of a tunnel.

[0149] A computer program product includes a computer program or set of instructions for performing specific tasks or implementing specific functions. These programs or instructions are designed to be executed by a processor to implement a series of predefined steps or operations. The program product may be stored in various forms of computer storage media, such as memory, hard disk, solid-state drive, optical disk or other forms of digital storage devices. It may exist in the form of compiled binary code or in the form of scripts or bytecodes that can be executed by an interpreter. The program product uses carefully designed algorithms and logical instructions to enable the processor to process data in a specific order and manner to complete various functions such as data analysis, user interaction, device control, etc.

[0150] In the description of this specification, the description with reference to the terms "one embodiment / method", "some embodiments / methods", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment / method or example are included in at least one embodiment / method or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment / method or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments / methods or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments / methods or examples described in this specification and the features of the different embodiments / methods or examples, unless they are contradictory.

[0151] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In the description of this application, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0152] It should be understood by those skilled in the art that the above embodiments are only for the purpose of clearly illustrating the present invention, and are not intended to limit the scope of the present invention. For those skilled in the art, other changes or modifications may be made based on the above invention, and these changes or modifications are still within the scope of the present invention.

Claims

1. A hierarchical prediction method for large tunnel deformation, characterized in that: include: Data collection and preprocessing: collect deformation data from historical tunnel projects and preprocess the collected deformation data to obtain sample data; Feature selection and index analysis: analyze sample data and select key features that characterize deformation through mutual information and mRMR algorithm; Model selection and training: Train the support vector machine through key features to obtain the prediction model; Model optimization and adjustment: verify the prediction model and optimize the model parameters according to the verification results to obtain the final hierarchical prediction model; Deformation prediction: input the real-time deformation data of the tunnel, and the graded prediction model outputs the predicted deformation grade; The method for obtaining the final classification prediction model includes: The support vector machine is a nonlinear support vector machine and takes the key features as input variables and the deformation level as the output label; Construct a feature set, which includes input variables and output tags ,in, The final feature set The number of features in is the number of levels of deformation; Determine the decision function and kernel function of the nonlinear support vector machine; The penalty coefficient is optimized by combining the parameters and kernel parameters Optimize, select the optimal hyperparameters, and build a prediction model with the optimal hyperparameters; Among them, the combined parameter optimization algorithm includes: Random generation includes A population of individuals, each individual in the population is ; Evaluate the fitness of each individual and sort each individual according to the size of the fitness. Take the first multiple individuals as parents, perform crossover operations on the selected parents, and generate offspring individuals. ,in, , , and For the father generation, is the randomly generated crossover coefficient; Randomly change the parameters of some individuals to generate new individuals; Repeat the iteration until the maximum number of iterations is reached or the fitness converges, sort all parent individuals and child individuals according to their fitness, and select the top Individuals are used as primary hyperparameters; Use the primary hyperparameters as the initial individuals in the initial population; Search in the neighborhood of each initial individual to generate a new initial individual ,in, To control the random step size of the search, is another initial individual selected randomly; Calculate the fitness of the new initial individual. If the new initial individual The fitness of the initial individual is greater than that of the current individual The fitness of ;otherwise, constant; Calculate the fitness of all initial individuals and determine the search probability ,in, For the The fitness value of the initial individuals; Sort the primary hyperparameters by the search probability and select the top A primary hyperparameter is used to perform a local search in its field to generate a new initial individual. If the fitness of the new initial individual is greater than that of the current initial individual, the current initial individual is updated; otherwise, the current initial individual remains unchanged. In repeated iterative search, if the fitness of an initial individual remains unchanged in multiple iterations, a new initial individual is randomly generated and the new initial individual is used to replace the current initial individual; Repeat the iteration until the preset number of iterations is reached or the fitness value of the population converges; Sort all initial individuals according to their fitness, and filter them out The initial individuals are used as secondary hyperparameters; Select the secondary hyperparameter with the largest fitness as the optimal hyperparameter ; set up are the parameters of the prediction model before optimization, and include secondary hyperparameters; iterate over all parameters , the methods for optimizing the prediction model include: Choose multiple different outer cross validation Values ​​and inner cross validation value; For each The outer cross validation is performed and each The average accuracy corresponding to the value ,in, For the The number of true positives of the model prediction results in the outer cross-validation, For the The number of true negatives of the model prediction results in the outer cross-validation, For the The number of false positives in the model prediction results in the outer cross-validation, For the The number of false negatives in the model prediction results in the outer cross-validation; Choose the optimal number of folds for outer cross validation ,in, The fold is The average accuracy of the outer cross validation is The fold is The variance of the average accuracy of the outer cross validation; By optimal fold Determine the training set for inner cross-validation; Through inner cross validation Optimize the parameters of the test model , For the Given parameters under inner layer cross validation The verification accuracy of For a given parameter The uncertainty of the model prediction results, To find the mean value, are the optimized parameters found in the inner cross validation, For a given parameter All The average accuracy of the inner layer cross validation; The average accuracy in the inner cross validation is used as the objective function , optimize the parameters of the test model ,in, is the estimate of the expected value of the objective function, is the balance parameter between exploration and exploitation, For a given parameter The uncertainty of the model prediction results, is the final parameter after optimization, For a given parameter All The average accuracy of the within-fold cross validation.

2. A hierarchical prediction method for large deformation of a tunnel according to claim 1, characterized in that: Methods for preprocessing deformation data include: The noise of the deformation data was filtered by the three-step moving average method; Apply Z-score normalization to the noise-filtered deformation data; The missing values ​​of the deformed data after data standardization were processed by linear interpolation; The deformation data with missing values ​​processed are analyzed at multiple scales by wavelet transform method, and the features at different scales are extracted as sample data.

3. A hierarchical prediction method for large deformation of a tunnel according to claim 1, characterized in that: Methods for screening key characteristics include: Determine the mutual information between features and target variables in turn ,in, Features With the target variable The mutual information between for and The joint probability distribution between for The marginal probability distribution of for The marginal probability distribution of is the preprocessed deformation data, the target variable For deformation; Select the feature with the largest mutual information ; and Add to selected feature set ; Calculate the feature Redundancy between other features and target variables ,in, Features With the target variable The mutual information between It is a feature With selected feature set Medium Features The mutual information between Select the feature with the smallest redundancy value ; and Add to selected feature set ; Calculate Divide and Features The redundancy value between the remaining features and the target variable is calculated, and the feature with the smallest redundancy value is selected and added to the selected feature set. ; Iterate until the stopping condition is reached to obtain the final selected feature set; and use the features in the selected feature set as key features; the stopping condition is that the number of features in the selected feature set reaches a preset value or the calculated redundancy value is lower than the set value.

4. A hierarchical prediction method for large deformation of a tunnel according to claim 1, characterized in that: It also includes filtering out some redundant or low-contribution features in the selected feature set, including: Construct a feature screening model based on the SVM model and set the selected feature set As input; By selected feature set Iteratively train the feature screening model. In the first iteration, the coefficient of each feature is obtained through the feature screening model, and the decision function of the feature screening model is ,in, For the The feature in The coefficients in the iterations, For the Features, For the The selected feature set for iteration, For the At iteration The number of features in is bias; according to Sort by size, remove the smallest feature, and obtain a new feature set ; Repeat the iteration until the number of features in the new feature set reaches the preset number or the model performance meets the preset requirements; Get the final feature set , and the features in the final feature set are taken as key features.

5. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instruction is executed by a processor, a hierarchical prediction method for large deformation of a tunnel as described in any one of claims 1 to 4 is implemented.

Images