Tunnel water inrush disaster prediction method, device and storage medium

Abstract

The present invention relates to the technical field of tunnel water inrush safety assessment, and specifically to a tunnel water inrush disaster prediction method, device and storage medium, comprising S1: constructing a data set based on tunnel water inrush disaster cases, analyzing and selecting disaster-causing factors, and clarifying disaster classification standards; S2: dividing the data set into a training set and a test set according to a predetermined proportional relationship; S3: using the disaster-causing factors as characteristic variables, and performing data preprocessing on the training set and the test set in the data set respectively; compared with the existing technology, the beneficial effects of the present invention are: using machine learning methods to perform intelligent prediction of tunnel water inrush disasters can avoid the problems of strong subjectivity and poor applicability of traditional prediction methods, while avoiding the misjudgment and omission of disasters caused by technical personnel, thereby improving model performance and prediction accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of tunnel water inrush safety assessment, and in particular to a tunnel water inrush disaster prediction method, device and storage medium. Background Art

[0002] Traditional methods for predicting tunnel water inrush hazards include the Analytic Hierarchy Process (AHP), comprehensive evaluation methods, weighted average methods, and expert judgment methods. These methods are highly subjective and have poor applicability. Furthermore, the complex hydrogeological conditions during tunnel construction make it difficult to fully explore them. Inadequate technical expertise among frontline construction personnel can make it difficult to accurately predict tunnel water inrush hazards, potentially causing severe economic losses and casualties. Machine learning can be used to process complex data. In recent years, with the development of machine learning predictive models, an increasing number of researchers have applied them to data analysis and disaster prediction.

[0003] In machine learning, prediction primarily involves three steps: data preprocessing, model training, and prediction. Before machine learning tasks begin, raw data must be processed and converted into data suitable for machine learning to improve accuracy. However, there are numerous machine learning models, and their learning capabilities and adaptability vary widely. Therefore, when establishing an intelligent prediction model for tunnel water inrush disaster risk, the choice must be tailored to the specific situation. Factors such as the number of features and sample size should be considered, and the model should be evaluated to ensure a simple, easy-to-train model that avoids overfitting or underfitting.

[0004] The prediction accuracy of a model is affected by a variety of factors, including the size of the dataset, the distribution of its features, how the dataset is partitioned, and the complexity of the model. For a given model and dataset, the larger the training set, the higher the model's accuracy. However, a larger training set ratio places greater demands on the model and computing power. Excessively large training set ratios can also lead to overfitting. Therefore, it's important to choose an appropriate training set ratio based on the prediction type.

[0005] At the same time, previous researchers have mostly used relatively simple traditional machine learning models or linear fusions of simple models when predicting tunnel water inrush hazards. This has resulted in suboptimal overall prediction accuracy, which in turn has affected the determination of tunnel water inrush hazard levels and preventive measures. The present invention aims to address the technical challenges of combining machine learning with intelligent prediction of tunnel water inrush, such as the limited model size and suboptimal prediction accuracy of machine learning models.

[0006] Therefore, the development of a method, device and storage medium for predicting tunnel water inrush disasters not only has urgent research value, but also has good economic benefits and industrial application potential. This is the driving force and basis for the completion of the present invention. Summary of the Invention

[0007] In order to overcome the above-mentioned defects of the prior art, the inventors conducted in-depth research and completed the present invention after putting in a lot of creative work.

[0008] Specifically, the present invention proposes an intelligent prediction method for tunnel water inrush based on algorithm optimization to quickly and accurately predict tunnel water inrush disasters, so as to solve the technical problems of insufficient prediction accuracy and single model.

[0009] To achieve the above object, the present invention provides the following technical solutions:

[0010] A method for predicting sudden water inrush disasters in a tunnel, comprising:

[0011] S1: Construct a dataset based on tunnel water inrush disaster cases, analyze and select disaster-causing factors, and clarify disaster classification standards;

[0012] S2: Divide the dataset into training set and test set according to a predetermined ratio;

[0013] S3: Using the disaster-causing factor as the characteristic variable, perform data preprocessing on the training set and test set in the dataset respectively;

[0014] S4: Select precision, recall, and F1 score as evaluation indicators of the learning model, train the machine learning model with preprocessed data, optimize the model hyperparameters, and repeat the training to select the suitable machine learning model;

[0015] S5: Improve machine learning models using the stacking ensemble method, using a single adapted model as the first-layer base learner and linear regression and ridge regression as the second-layer meta-learners. Perform hyperparameter tuning on the single model and test and compare the prediction performance of multiple ensemble models using different combinations of base learners and meta-learners.

[0016] S6: Compare the optimal single model and the optimal ensemble model using evaluation indicators and confusion matrix to obtain the evaluation accuracy of the optimal ensemble model.

[0017] In the present invention, as an improvement, the disaster-causing factors are analyzed and selected, and the disaster classification standards are clarified to include:

[0018] Collect the frequency of occurrence of influencing factors in the statistical data set, and use high-frequency influencing factors as disaster-causing factors for prediction and evaluation;

[0019] The impact levels of disaster factors are divided according to their influence on tunnel water inrush.

[0020] In the present invention, as an improvement, data preprocessing of the data set includes:

[0021] Disaster factors are divided into qualitative variables and quantitative variables, and are quantified separately. Among them, quantitative variables and qualitative variables are assigned values ​​according to the corresponding disaster factor impact levels.

[0022] The box plot method was used to detect outliers on the characteristic variables;

[0023] Outlier replacement and missing value supplementation are performed on the data. For the same variable at different risk levels, the mode of the data at the current risk level is used as the replacement value for qualitative variables; the mean of the data at the current risk level is used as the replacement value for quantitative variables. For missing data supplementation in the dataset, the mode of the data at the impact level is used as the replacement value for qualitative variables; the mean of the data at the impact level is used as the replacement value for quantitative variables.

[0024] The data set after outlier processing is normalized.

[0025] In the present invention, as an improvement, the machine learning model is trained with preprocessed data, and after optimizing the model hyperparameters, repeated training is performed to screen and adapt the machine learning model, including:

[0026] Select multiple machine learning models as training models, and input the preprocessed training set and test set into a single machine learning model;

[0027] pass K The fold cross validation is combined with the grid search parameter tuning method to tune the hyperparameters of the machine learning model, obtain the optimal hyperparameter values ​​of the model, input the model hyperparameters, retrain the model, compare the evaluation indicators of each single model before and after hyperparameter optimization, and select the best n adaptive machine learning single models.

[0028] In the present invention, as an improvement, the linear regression and ridge regression in the second-layer meta-learner adopt a hyperparameter tuning method to obtain the optimal hyperparameters of the two meta-learners.

[0029] In the present invention, as an improvement, stacking integration is used to improve the learning model, specifically including:

[0030] The ensemble learning model is composed of n base learners and two meta-learners to form at least m sets of ensemble model combinations;

[0031] Using the training set and test set data to train m sets of integrated models, the prediction results of tunnel water inrush disasters and the evaluation indicators of each model were obtained;

[0032] For the ensemble model combinations whose precision, recall, and F1 scores are all above 0.80, re-enter the optimal hyperparameters for model training;

[0033] The evaluation indicators of the integrated model combination before and after hyperparameter tuning are compared to obtain the optimal Stacking integrated model.

[0034] In the present invention, as an improvement, the integrated model combination method is an integrated model composed of one or more base learners and any one of the meta learners.

[0035] A computer device comprising a memory, a processor and a communication component;

[0036] wherein the memory is configured to store one or more instructions;

[0037] The processor is loaded by the communication component and executes the memory instructions to complete the prediction method described in any one of claims 1-6.

[0038] A computer-readable storage medium stores executable instructions, wherein the executable instructions are loaded and executed by a processor to implement the prediction method according to any one of claims 1 to 6.

[0039] Compared with the prior art, the present invention has the following beneficial effects:

[0040] (1) The use of machine learning methods for intelligent prediction of tunnel water inrush disasters can avoid the problems of strong subjectivity and poor applicability of traditional prediction methods, and at the same time avoid the misjudgment and omission of disasters caused by technical personnel. Secondly, the present invention improves the model and optimizes the algorithm through the method of integrated model, which can quickly and accurately predict tunnel water inrush disasters and improve model performance and prediction accuracy.

[0041] (2) In the present invention, a classification prediction method is adopted with a Stacking integrated model for water inrush prediction. Different from the previous regression prediction, the data is numerically processed and the matched evaluation indicators are used to optimize the hyperparameters in a model integration combination mode to improve the model prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following briefly describes the drawings required for the specific embodiments or the description of the prior art. Similar elements or parts are generally identified by similar reference numerals throughout the drawings. Elements or parts in the drawings are not necessarily drawn to scale.

[0043] Figure 1 Schematic diagram of the process of the prediction method of the present invention;

[0044] Figure 2 This is a schematic diagram of the integrated model training process of the present invention;

[0045] Figure 3 are the evaluation index parameters of the four models in Example 1 of the present invention;

[0046] Figure 4 This is a schematic diagram showing the comparison of the accuracy values ​​before and after optimization of the hyperparameters of a single model in Example 1 of the present invention;

[0047] Figure 5 Schematic diagram of comparison of recall values ​​before and after single model hyperparameter optimization in Example 1 of the present invention;

[0048] Figure 6 This is a schematic diagram showing the comparison of F1 score values ​​before and after single model hyperparameter optimization in Example 1 of the present invention;

[0049] Figure 7 Schematic diagram of integrated model evaluation indicators in Example 1 of the present invention;

[0050] Figure 8 This is a schematic diagram showing the comparison of the accuracy values ​​before and after the optimization of the hyperparameters of the integrated model in Example 1 of the present invention;

[0051] Figure 9 Schematic diagram of the comparison of recall values ​​before and after optimization of hyperparameters of the integrated model in Example 1 of the present invention;

[0052] Figure 10 Schematic diagram of the comparison of F1 score values ​​before and after hyperparameter optimization of the integrated model in Example 1 of the present invention;

[0053] Figure 11 Schematic diagram of the comparison of evaluation indicators of the optimal single model and the optimal integrated model in Example 1 of the present invention. DETAILED DESCRIPTION

[0054] The following embodiments of the technical solution of the present invention will be described in detail with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention and are therefore only examples and are not intended to limit the scope of protection of the present invention.

[0055] A method for predicting sudden water inrush disasters in a tunnel comprises the following steps:

[0056] S1. Build a database of tunnel water inrush disaster cases through on-site investigations and statistical collection of disaster cases. Conduct statistical analysis of the data in the database, screen multiple influencing factors with high frequency of occurrence as disaster-causing factors for evaluation, and clarify disaster classification standards.

[0057] S2. Divide the data set into a training set and a test set according to a predetermined ratio, and perform data preprocessing on the divided test set and training set data separately;

[0058] S3. Using the disaster-causing factors as characteristic variables, preprocess the data set;

[0059] S4. Select precision, recall, and F1 score as evaluation indicators of the machine learning model, train the learning model with preprocessed data, optimize the model hyperparameters, and then repeat the training to select the suitable machine learning model;

[0060] S5. Improve the machine learning model using the stacking ensemble learning method, using the adapted single model as the first-layer base learner and linear regression and ridge regression as the second-layer meta-learners. Perform hyperparameter tuning on the single model and test and compare the prediction effects of multiple ensemble models with different combinations of base learners and meta-learners.

[0061] S6. Compare the optimal single model and the optimal integrated model using the evaluation index and confusion matrix to obtain the evaluation accuracy of the optimal integrated model.

[0062] In step 1, through statistical analysis of the data in the database, it was found that six factors, namely, stratum lithology, unfavorable geology, rock layer inclination, negative terrain area ratio, surrounding rock grade, and hydrodynamic zoning, are the main factors affecting tunnel water inrush disasters. These six influencing factors are used as disaster-causing factors to evaluate the hazard level of tunnel water inrush disasters.

[0063] By studying the tunnel water inrush hazard classification standards in previous literature and combining them with railway tunnel risk assessment and management methods, a tunnel water inrush hazard hazard classification system is established: Level I, Level II, Level III, and Level IV. Level I is the lowest hazard level, Level II is the medium hazard level, Level III is the high hazard level, and Level IV is the highest hazard level.

[0064] The training set and test set are classified with random seeds through algorithm coding, and the selected training set and test set are processed using the same data preprocessing method.

[0065] The data preprocessing process includes: selection of input feature variables, data digitization, outlier detection and replacement, missing data supplementation and data standardization, specifically including:

[0066] S201: The disaster factor data obtained through statistical analysis is input into the data model as characteristic variables.

[0067] S202: Classify the data in the dataset into qualitative and quantitative variables, using different numerical processing standards. Qualitative variables include: stratum lithology, adverse geological and hydrodynamic zoning, and negative terrain area ratio; quantitative variables include: stratum dip and surrounding rock grade.

[0068] S203: Quantitative variables and qualitative variables are classified into different disaster risk levels according to their characteristics, and values ​​are assigned to them based on the corresponding disaster factor classification standards. The assignment can adopt the assignment standards available in the prior art.

[0069] S204: Use the boxplot method to detect and replace outliers in the digitized tunnel water inrush disaster dataset. For outlier replacement of the same variable at different hazard levels (I-IV), for qualitative variables, the mode at the current hazard level is used as the outlier replacement value; for quantitative variables, the mean at the current hazard level is used as the outlier replacement value.

[0070] S205: For missing data, use step S203 to supplement the missing data. For qualitative variables, the mode at the current hazard level is used as the supplementary value for the missing data; for quantitative variables, the average value at the current hazard level is used as the supplementary value for the missing data.

[0071] S206: The tunnel water inrush disaster dataset is normalized using the maximum and minimum normalization formula. The specific formula is as follows:

[0072]

[0073] Where, X nom is the standardized value, and its value is between [0,1]; X Get the value of a single raw data in the dataset; X max It is the maximum value of the original data under a certain characteristic variable; X min It is the minimum value of the original data under a certain characteristic variable.

[0074] When the influencing factors in the quantitative variables are discrete variables, the average value of the interval after classification is taken as the evaluation value.

[0075] Select Precision P ), Recall, R ) and F1 score (F1 score, F1) As an evaluation indicator of the machine learning model, the data obtained after preprocessing in step S2 is input into the machine learning model and training is completed to obtain an intelligent prediction result of the tunnel water inrush hazard level.

[0076] The construction and training of a single model includes:

[0077] S301: Select a single model as a machine learning training model. The single model selects at least n+1 different types, and inputs the divided and preprocessed training set and test set into the single model to obtain a prediction result of the tunnel water inrush hazard level.

[0078] S302: Passed K The fold cross-validation is combined with the grid search parameter adjustment method to tune the hyperparameters of the training model to obtain the optimal hyperparameter values. The model is retrained after inputting the machine model hyperparameters to obtain the evaluation indicators of each single model after hyperparameter optimization. The evaluation indicators of each single model before and after hyperparameter optimization are compared to select the best n models as the single model for adaptive training.

[0079] Using Stacking ensemble learning methods to improve machine learning models and optimize algorithms specifically includes:

[0080] S401: Select n adapted single models as the first-layer base learners. To prevent model overfitting, select linear regression (LR) and ridge regression (Ridge) as the second-layer meta-learners. Hyperparameter tuning methods are performed on LR and Ridge to obtain the optimal hyperparameters of the two meta-learners.

[0081] S402: constructing m groups of ensemble models by combining n base learners and two meta learners in different ways;

[0082] S403: Input the training set and test set divided from the dataset into the machine learning model, train m types of stacking ensemble models, and obtain the tunnel water inrush disaster prediction results and evaluation indicators of each model;

[0083] S404: Optimize the integrated model before P value, R Value and F After re-entering the optimal hyperparameters for the combinations with all 1 values ​​above 0.80, model training was carried out, and the evaluation indicators of the integrated model combination before and after hyperparameter tuning were compared to select the optimal Stacking integrated model in this tunnel water inrush hazard prediction model.

[0084] Compare the selected optimal Stacking ensemble model with the previous optimal prediction single model, respectively. P value, R value,F 1 value and confusion matrix to obtain the optimization degree of the Stacking ensemble model, thereby clarifying the improvement effect of the Stacking ensemble model.

[0085] A computer device and a storable medium applying the above-mentioned prediction method, wherein the computer device includes a memory, a processor and a communication component, the memory and the processor are connected via the communication component, the memory stores one or more instructions, and the processor receives and executes the instructions.

[0086] Both the memory and the processor are used in the above prediction method to complete the data transmission and learning model instruction actions of the prediction method.

[0087] Example 1:

[0088] A total of 95 tunnels and 232 sets of cross-sectional data were selected to construct tunnel water inrush disaster cases and a database. The six most frequently occurring influencing factors were selected as disaster evaluation factors, including stratum lithology, adverse geology, hydrodynamic zoning, negative terrain area ratio, rock formation inclination and surrounding rock grade.

[0089] Based on the previous tunnel water inrush classification standard, the water inrush hazard level is divided into four levels, including: I-IV. According to existing research, the data in the dataset are divided into four levels according to their characteristics.

[0090] The data were randomly divided into training and test sets in a ratio of 8:2. The random division was performed by adding a random seed number to the program code to select the training set and test set separately without disrupting the order of the original data set.

[0091] Separate the data in the dataset into training set and test set for data preprocessing, and select the accuracy P , recall rate R and F 1 score is used as an evaluation indicator for machine learning models. P Indicates the accuracy of the model in predicting the tunnel water inrush hazard level, and its value is between [0,1]; R Indicates the comprehensiveness of the model in predicting tunnel water inrush samples, and its value is between [0,1]; F 1 for P and R The harmonic mean of .

[0092] Four different types of machine learning models, namely Naive Bayes (NB), Random Forest (RF), Support Vector Machine (SVM) and Adaptive Boosting (AB), were selected. The training set and test set were input to train four single machine learning models, and the intelligent prediction results of tunnel water inrush disaster level of the four single models were obtained. P 、 R and F 1 value such as Figure 3 shown.

[0093] by K fold cross validation ( K =5) and combined with grid search parameter tuning, hyperparameter tuning was performed on the other three models except Naive Bayes. The optimal hyperparameter values ​​of random forest, support vector machine and adaptive algorithm were obtained. The hyperparameters of the model were input to retrain the model. The evaluation indicators of each single model after hyperparameter optimization except Naive Bayes were obtained. The evaluation indicators of each single model before and after hyperparameter optimization were compared. Figure 4-6 As shown in the figure, three better models are selected, namely random forest, support vector machine and adaptive algorithm.

[0094] Random forest, support vector machine, and adaptive algorithm models were used as the first-layer base learners, and linear regression and ridge regression were used as the second-layer meta-learners. The meta-learners were hyperparameter-tuned, and the three base learners and two meta-learners were compared in different combinations. A total of eight ensemble model combinations were compared. The specific combinations are shown in the following table:

[0095] The training set and test set are input into data, and 8 sets of Stacking ensemble models are trained to obtain the tunnel water inrush disaster prediction results and the evaluation indicators of each model, such as Figure 7 shown.

[0096] For the pre-optimization integrated model P value, R Value and F The ensemble model combination with 1 values ​​above 0.80 is re-entered with the optimal hyperparameters for model training, and the evaluation indicators of the ensemble model combination before and after hyperparameter tuning are compared, such as Figure 8-10 As shown in the figure, the optimal Stacking integrated model in the tunnel water inrush hazard prediction model was selected. After comparison, it was found that SNL was the optimal Stacking integrated model.

[0097] The SNL integrated model is consistent with the above four single models. Pvalue, R value, F 1 value and confusion matrix were comprehensively compared. After comparative analysis of hyperparameter optimization, it was found that the optimal single model for the prediction of tunnel water inrush disaster risk was SVM, and the optimal Stacking ensemble model was SNL. By comparing the optimal single model set and the optimal ensemble model, Figure 11 As shown in the figure, the SNL integrated model is the optimal model, and the SNL integrated model is more accurate and efficient for predicting tunnel water inrush disasters.

[0098] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or make equivalent replacements for some or all of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present invention, and they should all be included in the scope of the claims and description of the present invention.

Claims

1. A method for predicting sudden water inrush disasters in tunnels, characterized in that: include: S1: Construct a dataset based on tunnel water inrush disaster cases, analyze and select disaster-causing factors, and clarify disaster classification standards; S2: Divide the dataset into training set and test set according to a predetermined ratio; S3: Using the disaster-causing factor as the characteristic variable, perform data preprocessing on the training set and test set in the dataset respectively; S4: Select precision, recall, and F1 score as evaluation indicators for the machine learning model. Train the machine learning model with preprocessed data. After optimizing the model hyperparameters, repeat the training to select the suitable machine learning model. S5: Use stacking ensemble learning to improve the model, using a single adapted model as the first-layer base learner and linear regression and ridge regression as the second-layer meta-learners. Perform hyperparameter tuning on the single model and test and compare the prediction performance of multiple ensemble models with different combinations of base learners and meta-learners. S6: Compare the optimal single model and the optimal ensemble model using evaluation indicators and confusion matrix to obtain the evaluation accuracy of the optimal ensemble model; Data preprocessing of the dataset includes: Disaster factors are divided into qualitative variables and quantitative variables, and are quantified separately. Quantitative variables and qualitative variables are assigned values ​​based on the corresponding disaster factor impact levels. The box plot method is used to detect outliers in the numerical data set; Replace outliers and fill in missing values ​​in the data; Normalize the data set after outlier processing; The data outlier replacement and missing value supplementation include: For the outlier replacement process of the same variable at different risk levels, for qualitative variables, the mode of the data at the current risk level is used as the replacement value of the outlier; for quantitative variables, the average value of the data at the current risk level is used as the replacement value of the outlier; To supplement the missing values ​​in the data set, the mode of the data at the impact level of qualitative variables is taken as the supplementary value of the missing data; the average value of the data at the impact level of quantitative variables is taken as the supplementary value of the missing data.

2. The method for predicting sudden water inrush disaster in a tunnel according to claim 1, characterized in that: Analyze and select disaster-causing factors and clarify disaster classification standards, including: Collect the frequency of occurrence of influencing factors in the statistical data set, and use high-frequency influencing factors as disaster-causing factors for prediction and evaluation; The impact levels of disaster factors are divided according to their influence on tunnel water inrush.

3. The method for predicting sudden water inrush disaster in a tunnel according to claim 1, characterized in that: Train the machine learning model with preprocessed data, optimize the model hyperparameters, and then repeat the training. The selected adaptive machine learning models include: Select multiple machine learning models as training models, and input the preprocessed training set and test set into the single machine learning model; pass K The fold cross validation is combined with the grid search parameter tuning method to tune the hyperparameters of the machine learning model, obtain the optimal hyperparameter values ​​of the model, input the model hyperparameters, retrain the model, compare the evaluation indicators of each single model before and after hyperparameter optimization, and select the best n adaptive machine learning single models.

4. The method for predicting sudden water inrush disaster in a tunnel according to claim 1, wherein: In the second-layer meta-learner, the linear regression and ridge regression use hyperparameter tuning method to obtain the optimal hyperparameters of the two meta-learners.

5. The method for predicting sudden water inrush disaster in a tunnel according to claim 1, wherein: Using Stacking ensemble learning methods to improve machine learning models specifically includes: The ensemble learning model is composed of n base learners and two meta-learners to form at least m sets of ensemble model combinations; Using the training set and test set data to train m sets of integrated models, the prediction results of tunnel water inrush disasters and the evaluation indicators of each model were obtained; For the ensemble model combinations whose precision, recall, and F1 scores are all above 0.80, re-enter the optimal hyperparameters for model training; The evaluation indicators of the integrated model combination before and after hyperparameter tuning are compared to obtain the optimal Stacking integrated model.

6. The method for predicting sudden water inrush disaster in a tunnel according to claim 4, characterized in that: The integrated model combination method is an integrated model composed of one or more base learners and any one of the meta learners.

7. A computer device, characterized in that: Includes memory, processor and communication components; wherein the memory is configured to store one or more instructions; The processor is loaded by the communication component and executes the memory instructions to complete the prediction method described in any one of claims 1-6.

8. A computer-readable storage medium, characterized in that: The readable storage medium stores executable instructions, which are loaded and executed by a processor to implement the prediction method according to any one of claims 1 to 6.

Images