Tunnel operation period convergence deformation prediction method and system, electronic equipment and storage medium

Abstract

The invention discloses a tunnel operation period convergence deformation prediction method and system, electronic equipment and a storage medium, and the method comprises the steps: S1, obtaining the overall historical convergence deformation data and environment monitoring data of a tunnel, and carrying out the data preprocessing; s2, constructing a data sample, and performing data sample enhancement by adopting an ADASYN (adaptive synthetic sampling) method; s3, selecting a plurality of basic models as a primary learner; s4, calculating a prediction result evaluation index of the primary learner, and calculating a Pearson's correlation coefficient between the models according to the evaluation index; and S5, selecting at least two models with relatively large differences as final primary learners, selecting one model with the best evaluation result as a secondary learner, and constructing a stacking Stacking model. According to the invention, effective and high-precision convergence deformation prediction can be carried out on most sample points of the tunnel.

Description

Technical Field

[0001] The present invention belongs to the technical field of artificial intelligence, and relates to a deformation prediction method, in particular to a convergence deformation prediction method, system, electronic device and storage medium for tunnel operation period. Background Art

[0002] Deformation is a common phenomenon in nature, which refers to the change of the shape, size and position of a deformable body in the time-space domain under various loads and external forces. During the long-term operation of a shield tunnel after completion, due to the combined effects of soil disturbance, geological environment change, construction in adjacent areas and its own structural deterioration, etc., structural deformation will occur. As the internal and external forces of the tunnel accumulate day by day, once the degree of structural deformation exceeds a certain limit, it will lead to the occurrence of diseases and safety accidents, such as water seepage, segment cracks, segment damage, segment offset, wall surface corrosion, etc. When the structural deformation is serious, even large problems such as segment shedding and overall damage of the lining structure will occur, and safety accidents such as tunnel collapse, underground pipeline damage, ground subsidence and damage to ground buildings will occur, endangering the lives of personnel and causing serious consequences and social and economic losses.

[0003] As an important manifestation of tunnel structural deformation, lateral convergence deformation has received extensive attention from researchers. Different from the longitudinal deformation of the tunnel, lateral convergence deformation refers to the movement of the soil layer around the tunnel towards the center of the tunnel, causing the shield segments to be subjected to horizontal forces, resulting in a shorter transverse diameter of the tunnel cross-section and presenting a "horizontal oval" deformation mode. If it encounters soft and unstable soil or rock layers, or the tunnel crosses an aquifer or rivers, lakes and seas during construction, it is more likely to cause lateral convergence deformation.

[0004] Compared with the construction stage, the deformation of the tunnel during the operation period may bring more serious and long-term impacts. During the construction period, the deformation is usually temporary and can be controlled by appropriate construction techniques and support measures. However, during the operation period, the long-term use of the tunnel and the continuous action of geological conditions may lead to the aggravation or continuous development of deformation, causing irreversible damage to the tunnel structure. At the same time, during the operation period, large-scale activities cannot be carried out inside the tunnel. If the deformation causes structural damage or damage, it will take more time and resources to repair and maintain the tunnel, which undoubtedly brings huge challenges to maintenance management. Therefore, regular, continuous and comprehensive monitoring of the lateral convergence deformation of the shield tunnel during the operation period, predicting its future convergence deformation trend, judging and predicting possible risks in advance, and taking corresponding repair and maintenance measures in time can effectively avoid and prevent a series of disease risks caused by the imbalance of the tunnel structure, improve the operation and maintenance efficiency of the shield tunnel, and ensure the safety of the shield tunnel during the operation period.

[0005] At present, the research on the structural deformation of shield tunnels at home and abroad mainly focuses on the construction period, and there are still relatively few studies on the deformation during the operation period after the tunnel construction is completed. Since deformation monitoring projects often need to be carried out continuously for a long time, they require a large amount of manpower and material resources, high project costs, and great monitoring difficulties. The construction history of many shield tunnels is still relatively short, and the actual operation time is not very long, resulting in insufficient measured data and data volume available for research, which poses a certain obstacle to further research on structural deformation and the selection of prediction methods. In addition, the problem of tunnel structural deformation is caused by the combined influence of many factors, and it is very difficult to accurately and comprehensively identify these factors to construct a prediction model. Moreover, the overall deformation of the tunnel during the operation period is very slow, resulting in small numerical changes, and the measured monitoring data collected often has many missing values for various reasons. The accuracy and universality of using the existing data to predict the tunnel trend are often not high. All these situations put forward higher requirements for the reliability, accuracy, and practicality of shield tunnel deformation monitoring and prediction.

[0006] The existing technologies have the following three difficulties:

[0007] (1) The monitoring data used in the current prediction methods are mostly small-sample data sets, which may not be able to learn enough information for model training, making it difficult to ensure the accuracy rate of the model prediction results;

[0008] (2) Most of the current prediction methods use the time-series data of individual measuring points of the tunnel for prediction, ignoring the time-series nature of the overall measuring points of the tunnel and the relevance of the internal and external environments;

[0009] (3) The models used in the current prediction methods are relatively traditional and can only predict a part of the trends contained in the data, which is likely to cause the accumulation of single-model prediction errors and lead to the uncertainty of model prediction. The method needs to be optimized.

[0010] In view of this, there is an urgent need to design a new tunnel deformation prediction method to overcome at least some of the above defects existing in the existing tunnel deformation prediction methods. Summary of the Invention

[0011] The present invention provides a method, a system, an electronic device, and a storage medium for predicting the convergence deformation during the operation period of a tunnel, which can effectively and accurately predict the convergence deformation of most sample points of the tunnel.

[0012] To solve the above technical problems, according to one aspect of the present invention, the following technical solution is adopted:

[0013] A method for predicting the convergence deformation during the operation period of a tunnel, the method for predicting the convergence deformation during the operation period of a tunnel includes:

[0014] Step S1: Obtain the overall historical convergence deformation data and environmental monitoring data of the tunnel, and perform data preprocessing;

[0015] Step S2: Construct data samples, and use the Adaptive Synthetic Sampling (ADASYN) method to enhance the data samples;

[0016] Step S3: Select several basic models as primary learners;

[0017] Step S4: Calculate the evaluation indicators of the prediction results of the primary learners, and calculate the Pearson correlation coefficient between the models according to the evaluation indicators;

[0018] Step S5: Select at least two models with large differences as the final primary learners, select the model with the best evaluation result as the secondary learner, and construct a Stacking model.

[0019] As an implementation manner of the present invention, the data preprocessing in Step S1 includes:

[0020] Step S11: Collect the original data, including the cross-sectional transverse diameter data for calculating the convergence deformation, and at least one of the settlement monitoring data, water level data, position data, vehicle data, disease data, sensor stress data, road data, and special event data in the environmental factors; delete the data with obvious monitoring errors or anomalies in the data;

[0021] Step S12: Fill all missing values by the adjacent interpolation method to ensure the integrity of the input data;

[0022] Step S13: Align the historical convergence deformation data and the environmental monitoring data according to a unified time scale. Since the frequency of the convergence deformation data is relatively higher than that of some environmental monitoring data, the environmental monitoring data with a larger time statistical scale is aligned and complemented according to the time scale of the convergence monitoring.

[0023] As an implementation manner of the present invention, the construction of data samples in Step S2 includes the following steps:

[0024] Step S211: Define the data sample labels; the labels are the targets that the model needs to learn and predict; here, the labels adopt the evaluation indicators for the severity of convergence deformation in the "Technical Specification for the Maintenance and Operation Evaluation of Road Tunnels", that is, the relative deformation μ of the cross-sectional transverse diameter, which is calculated according to the following formula:

[0025]

[0026] In the formula, μ i represents the relative convergence deformation of the i-th measuring point section, ΔD iLet $\Delta D_i$ denote the change in the transverse diameter of the $i$-th measurement section, and $D$ denote the initial length of the transverse diameter of the $i$-th measurement section; this index is measured in per mille (‰).

[0027] Step S212: Define the feature set of the data samples; here, the feature set is the settlement monitoring data, water level data, location data, vehicle data, disease data, sensor stress data, road data, and special event data in the environmental factors.

[0028] Step S213: Encode or transform the feature set; perform numerical conversion on some features described purely in text to make them numerical variables or categorical variables; then define the encoding categories for all categorical variables and convert them into numerical variables.

[0029] Step S214: After all features are converted into numerical variables, in order to eliminate the influence of the dimensional and scale differences between features, it is necessary to perform normalization processing on the features; here, the MinMaxScaler method is used for numerical normalization, and the scaling formula is:

[0030]

[0031] where $x$ is the value of the original feature, $min_x$ is the minimum value of the feature, $max_x$ is the maximum value of the feature, and $x'$ is the scaled feature value. The numerical range of the scaled feature is $(0, 1)$.

[0032] As an implementation manner of the present invention, the ADASYN processing in step S2 includes:

[0033] Step S221: Define the majority class and the minority class; the majority class refers to the monitoring points in the tunnel section being more than the average value; the minority class refers to the monitoring points in the tunnel section being less than the average value, which are the samples that need to be focused on by the model.

[0034] Step S222: Calculate the imbalance degree between the majority class and the minority class samples. Denote the minority class samples as $m$ s , and the majority class as $m$ l , then the imbalance degree is $d = m$ s / $m$ l , $d\in[0, 1]$;

[0035] Step S223: Calculate the number of samples $G$ that need to be synthesized, $G=(m$ l - $m$ s ) * $b$, $b\in[0, 1]$. When $b = 1$, $G$ is the difference between the minority class and the majority class, and at this time, the number of majority class samples after synthesizing the data is exactly balanced with the number of minority class data.

[0036] Step S224: Calculate k neighbors for each sample belonging to the minority class using the Euclidean distance. Let Δ be the number of samples belonging to the majority class among the k neighbors, and denote the proportion of the majority class as r = Δ / k, where r ∈ [0, 1].

[0037] Step S225: For each minority class sample obtained in Step S224 i , for r i Normalize:

[0038] Step S226: Calculate the number of new samples to be generated for each minority class sample according to the sample weights.

[0039] Step S227: Calculate the number to be generated for each minority sample according to g i : s i = x i + (x zi - x i ) × λ, where s i is the synthetic sample, x i is the i-th sample in the minority class, and x zi is a randomly selected minority class sample among the k-nearest neighbors of x i . Repeat the synthesis until the required number of syntheses is met.

[0040] As an implementation manner of the present invention, in Step S3, several models among LSTM, SVM, RF, KNN, and LR are selected;

[0041] The model construction process of the neural network LSTM in Step S3 is as follows:

[0042] Step S3a1: Data preparation: Use the dataset processed by ADASYN in Step S2;

[0043] Step S3a2: Data division: Divide the dataset into a training set and a test set, with a ratio of 8:2;

[0044] Step S3a3: Model setting: Select the number of model layers, the number of neurons in each layer, select the activation function and the optimizer;

[0045] Step S3a4: Model training: Use the training set to train the LSTM model, and adjust the parameters according to the training situation, such as the learning rate, batch size, etc.;

[0046] Step S3a5: Model evaluation and verification: Use the trained model to make predictions on the test set, calculate the error between the prediction result and the actual value, and use the mean absolute error (MAE) as the evaluation index.

[0047] The model construction process of the support vector machine (SVM) in step S3 is as follows:

[0048] Step S3b1, data preparation: Use the dataset processed by ADASYN in step S2;

[0049] Step S3b2, data division: Divide the dataset into a training set and a test set, with a ratio of 8:2;

[0050] Step S3b3, model setting: Select the Gaussian kernel function (RBF), and define hyperparameters such as the penalty coefficient C of the model and the kernel function parameters;

[0051] Step S3b4, model training: Use the training set to train the SVM model, and adjust the parameters according to the training situation;

[0052] Step S3b5, model evaluation and verification: Use the trained model to make predictions on the test set, calculate the error between the prediction result and the actual value, and adopt F1-score as the evaluation index.

[0053] The model construction process of the random forest (RF) in step S3 is as follows:

[0054] Step S3c1, data preparation: Use the dataset processed by ADASYN in step S2;

[0055] Step S3c2, data division: Divide the dataset into a training set and a test set, with a ratio of 8:2;

[0056] Step S3c3, model setting: Determine the number of decision trees in the random forest, determine the number of features randomly selected for each tree, adjust the depth of the tree and other parameters to prevent overfitting;

[0057] Step S3c4, model training: Use the training set to train the RF model, and adjust the parameters according to the training situation;

[0058] Step S3c5, model evaluation and verification: Use the trained model to make predictions on the test set, calculate the error between the prediction result and the actual value, and adopt F1-score as the evaluation index.

[0059] The model construction process of the k-nearest neighbors (KNN) in step S3 is as follows:

[0060] Step S3d1, data preparation: Use the dataset processed by ADASYN in step S2;

[0061] Step S3d2, data division: Divide the dataset into a training set and a test set, with a ratio of 8:2;

[0062] Step S3d3, Model Setting: Determine the value of K in the KNN algorithm, that is, select the number of neighbors, and calculate the distance between samples using the Euclidean distance;

[0063] Step S3d4, Model Training: Use the training set to train the KNN model and adjust the parameters according to the training situation;

[0064] Step S3d5, Model Evaluation and Validation: Use the trained model to make predictions on the test set, calculate the error between the prediction result and the actual value, and adopt F1-score as the evaluation index.

[0065] The model construction process of linear regression LR in Step S3 is as follows:

[0066] Step S3e1, Data Preparation: Use the dataset processed by ADASYN in Step S2;

[0067] Step S3e2, Data Partitioning: Partition the dataset into a training set and a test set at a ratio of 8:2;

[0068] Step S3e3, Model Setting and Model Training: Use the training set to train the LR model and adjust the parameters according to the training situation;

[0069] Step S3e4, Model Evaluation and Validation: Use the trained model to make predictions on the test set, calculate the error between the prediction result and the actual value, and adopt the mean absolute error (MAE) as the evaluation index.

[0070] As an implementation manner of the present invention, the calculation formula of the Pearson correlation coefficient in Step S4 is:

[0071]

[0072] where X and Y represent the evaluation index variables of two primary learners, and ρ X,Y measures the correlation of the prediction results of the two models. If the coefficient is close to 1, it indicates a strong positive correlation; if it is close to -1, it indicates a strong negative correlation; if it is close to 0, it indicates a weak correlation and a large difference between the two models.

[0073] As an implementation manner of the present invention, the construction process of the Stacking model in Step S5 is as follows:

[0074] Step S51, Preparation of Primary Learners: According to the prediction results and Pearson coefficients of the learners in Step S4, select three models with smaller prediction errors and smaller correlations as the final primary learners;

[0075] Step S52, data division: using the data set processed by ADASYN in step S2, the input data is divided into a training set and a test set, with a ratio of 8:2; secondly, the training set is divided into five parts: train1, train2, train3, train4, train5;

[0076] Step S53, training and testing the primary learner: using train1, train2, train3, train4, train5 as validation sets in turn, and the remaining 4 sets as training sets. After the training is completed, the validation set data is predicted to obtain pred1, pred2, pred3, pred4, pred5, and a 5-fold cross validation is completed. The three models are cross-validated in turn to obtain the cross-validation prediction results of each model;

[0077] Step S54, training and testing the secondary learner: stack the prediction results of the three primary learners as the training set input of the secondary learner, take the average of the test sets after each division as the test set of the secondary learner, select the primary learner with the best prediction effect as the secondary learner, and output the final prediction results on the test set.

[0078] According to another aspect of the present invention, the following technical solution is adopted: a tunnel operation period convergence deformation prediction system, the tunnel operation period convergence deformation prediction system comprising:

[0079] Data preprocessing module, used to obtain the overall historical convergence deformation data and environmental monitoring data of the tunnel and perform data preprocessing;

[0080] The sample construction enhancement module is used to construct data samples and use the adaptive synthetic sampling ADASYN method to enhance data samples;

[0081] The basic model selection module is used to select several basic models as primary learners;

[0082] The result evaluation index prediction module is used to calculate the prediction result evaluation index of the primary learner and calculate the Pearson correlation coefficient between models based on the evaluation index;

[0083] The stacking model construction module is used to select at least two models with large differences as the final primary learners, select the model with the best evaluation result as the secondary learner, and build a stacking model.

[0084] According to another aspect of the present invention, the following technical solution is adopted: an electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above method when executing the computer program.

[0085] According to another aspect of the present invention, the following technical solution is adopted: A storage medium stores computer program instructions thereon, and when the computer program instructions are executed by a processor, the steps of the above method are implemented.

[0086] The beneficial effect of the present invention is that the tunnel operation period convergence deformation prediction method, system, electronic device and storage medium proposed by the present invention can effectively and accurately predict the convergence deformation of most sample points of the tunnel. BRIEF DESCRIPTION OF THE DRAWINGS

[0087] Figure 1 It is a flowchart of the tunnel operation period convergence deformation prediction method in an embodiment of the present invention.

[0088] Figure 2 It is another flowchart of the tunnel operation period convergence deformation prediction method in an embodiment of the present invention.

[0089] Figure 3 It is a schematic diagram of the composition of the tunnel operation period convergence deformation prediction system in an embodiment of the present invention.

[0090] Figure 4 It is a schematic diagram of the composition of the electronic device in an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0091] The preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0092] In order to further understand the present invention, the preferred implementation schemes of the present invention will be described below in conjunction with embodiments. However, it should be understood that these descriptions are only for further explaining the features and advantages of the present invention, rather than limiting the claims of the present invention.

[0093] The description of this part only targets several typical embodiments, and the present invention is not limited to the scope described in the embodiments. The mutual replacement of some technical features in the prior art means that are the same or similar to those in the embodiments is also within the scope of description and protection of the present invention.

[0094] The expression of the steps in each embodiment in the specification is only for convenience of description, and the implementation manner of the present application is not limited by the order of step implementation.

[0095] "Connection" in the specification includes both direct connection and indirect connection.

[0096] The present invention discloses a tunnel operation period convergence deformation prediction method. Figure 1 、 Figure 2 It is a flowchart of the tunnel operation period convergence deformation prediction method in an embodiment of the present invention; please refer to Figure 1 、 Figure 2 , the tunnel operation period convergence deformation prediction method includes:

[0097] Step S1: Obtain the overall historical convergence deformation data and environmental monitoring data of the tunnel, and perform data preprocessing;

[0098] Step S2: Construct data samples, and use the Adaptive Synthetic Sampling (ADASYN) method to enhance the data samples;

[0099] Step S3: Select several basic models as primary learners;

[0100] Step S4: Calculate the prediction result evaluation indicators of the primary learners, and calculate the Pearson correlation coefficient between the models according to the evaluation indicators;

[0101] Step S5: Select at least two models with large differences as the final primary learners, and select the model with the best evaluation result as the secondary learner. The Stacking model is constructed.

[0102] In an embodiment of the present invention, the data preprocessing in Step S1 includes:

[0103] Step S11: Collect the original data, including at least one of the cross-section transverse diameter data for calculating the convergence deformation, settlement monitoring data, water level data, position data, vehicle data, disease data, sensor stress data, road data, and special event data in the environmental factors; delete the data with obvious monitoring errors or anomalies in the data;

[0104] Step S12: Fill all missing values by the adjacent interpolation method to ensure the integrity of the input data;

[0105] Step S13: Align the historical convergence deformation data and environmental monitoring data according to a unified time scale. Since the frequency of the convergence deformation data is relatively higher than that of some environmental monitoring data, the environmental monitoring data with a larger time statistical scale is aligned and complemented according to the time scale of the convergence monitoring.

[0106] In an embodiment of the present invention, the construction of data samples in Step S2 includes the following steps:

[0107] Step S211: Define the data sample labels; the labels are the targets that the model needs to learn and predict; here, the labels adopt the evaluation indicators for the severity of convergence deformation in the "Technical Specification for Maintenance and Operation Evaluation of Road Tunnels" (of course, other specifications can also be used), that is, the relative deformation μ of the cross-section transverse diameter, which is calculated according to the following formula:

[0108]

[0109] In the formula, μ i represents the relative convergence deformation of the i-th measuring point section, and ΔD iLet $\Delta D_i$ denote the change in the transverse diameter of the $i$-th measurement section, and $D$ denote the initial length of the transverse diameter of the $i$-th measurement section; this index is measured in per mille (‰).

[0110] Step S212: Define the feature set of the data samples; here, the feature set is the settlement monitoring data, water level data, location data, vehicle data, disease data, sensor stress data, road data, and special event data in the environmental factors.

[0111] Step S213: Encode or transform the feature set; numerically transform some features described purely in text to make them numerical variables or categorical variables; then define the encoding categories for all categorical variables and convert them into numerical variables.

[0112] Step S214: After all features are converted into numerical variables, in order to eliminate the influence of the dimensional and scale differences between features, it is necessary to normalize the features again; here, the MinMaxScaler method is used for numerical normalization, and the scaling formula is:

[0113]

[0114] where $x$ is the value of the original feature, $min_x$ is the minimum value of the feature, $max_x$ is the maximum value of the feature, and $x'$ is the scaled feature value. The numerical range of the scaled feature is $(0, 1)$.

[0115] In an embodiment of the present invention, the ADASYN processing in step S2 includes:

[0116] Step S221: Define the majority class and the minority class; the majority class means that the number of monitoring points in the tunnel section is more than the average value; the minority class means that the number of monitoring points in the tunnel section is less than the average value, and the samples that need to be focused on by the model.

[0117] Step S222: Calculate the imbalance degree between the majority class and the minority class samples. Denote the minority class samples as $m$ s , and the majority class as $m$ l , then the imbalance degree is $d = m$ s / $m$ l , $d\in[0, 1]$;

[0118] Step S223: Calculate the number of samples $G$ that need to be synthesized, $G=(m$ l - $m$ s ) * $b$, $b\in[0, 1]$. When $b = 1$, $G$ is the difference between the minority class and the majority class, and at this time, the number of majority class samples after synthesizing the data is exactly balanced with the number of minority class data.

[0119] Step S224: Calculate k neighbors for each sample belonging to the minority class using the Euclidean distance. Let Δ be the number of samples belonging to the majority class among the k neighbors, and denote the proportion of the majority class as r = Δ / k, where r ∈ [0, 1].

[0120] Step S225: For each minority class sample obtained in Step S224 i , for r i Normalize:

[0121] Step S226: Calculate the number of new samples to be generated for each minority class sample according to the sample weights

[0122] Step S227: Calculate the number to be generated for each minority sample according to g i : s i = x i +(x zi -x i )×λ, where s i is the synthetic sample, x i is the i-th sample in the minority class, and x zi is a randomly selected minority class sample among the k-nearest neighbors of x i . Repeat the synthesis until the required number of syntheses is met.

[0123] In an embodiment of the present invention, in Step S3, several models among LSTM, SVM, RF, KNN, and LR are selected;

[0124] The selection of the basic model in Step S3 can follow the following principles:

[0125] (1) Diversity principle: Using models of different types and different algorithms as primary learners can strengthen the model's learning of data features and reduce the error accumulation caused by a single model;

[0126] (2) Effectiveness principle: Select a basic model that performs excellently in a specific task such as classification or regression to ensure the prediction quality of the basic model.

[0127] In an embodiment of the present invention, the model construction process of the neural network LSTM in Step S3 is as follows:

[0128] Step S3a1: Data preparation: Use the dataset processed by ADASYN in Step S2;

[0129] Step S3a2: Data division: Divide the dataset into a training set and a test set at a ratio of 8:2;

[0130] Step S3a3: Model setting: Select the number of model layers, the number of neurons in each layer, select the activation function and the optimizer;

[0131] Step S3a4, Model Training: Use the training set to train the LSTM model and adjust parameters according to the training situation, such as the learning rate, batch size, etc.;

[0132] Step S3a5, Model Evaluation and Validation: Use the trained model to make predictions on the test set, calculate the error between the prediction result and the actual value, and adopt the mean absolute error (MAE) as the evaluation index.

[0133] The model construction process of the support vector machine SVM in Step S3 is as follows:

[0134] Step S3b1, Data Preparation: Use the dataset processed by ADASYN in Step S2;

[0135] Step S3b2, Data Partitioning: Partition the dataset into a training set and a test set with a ratio of 8:2;

[0136] Step S3b3, Model Setting: Select the Gaussian kernel function (RBF) and define hyperparameters such as the penalty coefficient C of the model and the kernel function parameters;

[0137] Step S3b4, Model Training: Use the training set to train the SVM model and adjust parameters according to the training situation;

[0138] Step S3b5, Model Evaluation and Validation: Use the trained model to make predictions on the test set, calculate the error between the prediction result and the actual value, and adopt the F1-score as the evaluation index.

[0139] The model construction process of the random forest RF in Step S3 is as follows:

[0140] Step S3c1, Data Preparation: Use the dataset processed by ADASYN in Step S2;

[0141] Step S3c2, Data Partitioning: Partition the dataset into a training set and a test set with a ratio of 8:2;

[0142] Step S3c3, Model Setting: Determine the number of decision trees in the random forest, determine the number of features randomly selected for each tree, adjust the depth of the tree and other parameters to prevent overfitting;

[0143] Step S3c4, Model Training: Use the training set to train the RF model and adjust parameters according to the training situation;

[0144] Step S3c5, Model Evaluation and Validation: Use the trained model to make predictions on the test set, calculate the error between the prediction result and the actual value, and adopt the F1-score as the evaluation index.

[0145] The model construction process of the K-nearest neighbor (KNN) in step S3 is as follows:

[0146] Step S3d1, data preparation: Use the dataset processed by ADASYN in step S2;

[0147] Step S3d2, data partitioning: Divide the dataset into a training set and a test set with a ratio of 8:2;

[0148] Step S3d3, model setting: Determine the value of K in the KNN algorithm, that is, select the number of neighbors, and use the Euclidean distance to calculate the distance between samples;

[0149] Step S3d4, model training: Use the training set to train the KNN model and adjust the parameters according to the training situation;

[0150] Step S3d5, model evaluation and verification: Use the trained model to make predictions on the test set, calculate the error between the prediction result and the actual value, and use the F1-score as the evaluation metric.

[0151] The model construction process of the linear regression (LR) in step S3 is as follows:

[0152] Step S3e1, data preparation: Use the dataset processed by ADASYN in step S2;

[0153] Step S3e2, data partitioning: Divide the dataset into a training set and a test set with a ratio of 8:2;

[0154] Step S3e3, model setting and model training: Use the training set to train the LR model and adjust the parameters according to the training situation;

[0155] Step S3e4, model evaluation and verification: Use the trained model to make predictions on the test set, calculate the error between the prediction result and the actual value, and use the mean absolute error (MAE) as the evaluation metric.

[0156] As an implementation manner of the present invention, the calculation formula of the Pearson correlation coefficient in step S4 is:

[0157]

[0158] where X and Y represent the evaluation index variables of two primary learners, and ρ X,Y measures the correlation of the prediction results of the two models. If the coefficient is close to 1, it indicates a strong positive correlation; if it is close to -1, it indicates a strong negative correlation; if it is close to 0, it indicates a weak correlation and a large difference between the two models.

[0159] In an embodiment of the present invention, the construction process of the Stacking model in step S5 is as follows:

[0160] Step S51, Preparation of primary learners: According to the prediction results and Pearson coefficients of the learners in step S4, select three models with smaller prediction errors and smaller correlations as the final primary learners;

[0161] Step S52, Data division: Use the dataset processed by ADASYN in step S2 to divide the input data into a training set and a test set, with a ratio of 8:2; Secondly, divide the training set into five parts: train1, train2, train3, train4, train5;

[0162] Step S53, Train and test the primary learners: Use train1, train2, train3, train4, train5 as the validation sets in turn, and the remaining 4 parts as the training sets. After training, predict the validation set data to obtain pred1, pred2, pred3, pred4, pred5, and complete 5-fold cross-validation. Perform cross-validation on the three models in turn to obtain the cross-validation prediction results of each model;

[0163] Step S54, Train and test the secondary learner: Stack the prediction results of the three primary learners as the input of the training set of the secondary learner. Take the average of the test sets after each division as the test set of the secondary learner. Select the primary learner with the best prediction effect as the secondary learner, and output the final prediction result on the test set.

[0164] The present invention also discloses a prediction system for convergence deformation during the operation period of a tunnel. The prediction system for convergence deformation during the operation period of the tunnel includes: a data preprocessing module 1, a sample construction and enhancement module 2, a basic model selection module 3, a result evaluation index prediction module 4, and a stacking model construction module 5.

[0165] The data preprocessing module 1 is used to obtain the overall historical convergence deformation data and environmental monitoring data of the tunnel and perform data preprocessing; The sample construction and enhancement module 2 is used to construct data samples and enhance the data samples by using the adaptive synthetic sampling ADASYN method; The basic model selection module 3 is used to select several basic models as primary learners; The result evaluation index prediction module 4 is used to calculate the prediction result evaluation indexes of the primary learners and calculate the Pearson correlation coefficients between the models according to the evaluation indexes; The stacking model construction module 5 is used to select at least two models with large differences as the final primary learners, select the model with the best evaluation result as the secondary learner, and construct a stacking Stacking model.

[0166] The specific implementation process of each module can be referred to the introduction of the method.

[0167] In a usage scenario of the present invention, for step S1, an example of the preprocessed source data is shown in Table 1, including cross-sectional diameter data for calculating convergent deformation, settlement monitoring data, water level data, position data, vehicle data, disease data, sensor stress data, road data, and special event data in environmental factors.

[0168] Feature Name Feature Value Example Cross-sectional Transverse Diameter Data 13.5712m Settlement Monitoring Data 2.5mm Water Level Data 307cm Location Data 15 Rings Vehicle Data 105,631 Vehicles Disease Data Cracks Sensor Stress Data 21.40 mpa Road Data (RDI) 92.14 Special Event Data None

[0169] Table 1 Example table of preprocessed source data

[0170] In this embodiment, for step S2, an example of data sample construction is shown in Table 2:

[0171]

[0172] Table 2 Example table of data sample construction

[0173] In this embodiment, the majority class refers to the monitoring points in the tunnel section being more than the average value; the minority class refers to the monitoring points in the tunnel section being less than the average value, and the samples that need to be focused on by the model. The oversampling setting ratio of step S2 ADASYN is 1:1. Table 3 is divided into six data sets according to the up and down directions of the tunnel and the convergence measurement direction, showing the number of samples before and after sampling:

[0174]

[0175]

[0176] Table 3 Table of the number of samples before and after sampling

[0177] In this embodiment, for step S3, the hyperparameters set by the five models of LSTM, SVM, RF, KNN, and LR after parameter tuning are shown in Table 4:

[0178]

[0179] Table 4 Table of hyperparameters set by five models

[0180] In this embodiment, for step S3, the prediction performance of the five models on six data sets is shown in Table 5, where the evaluation indexes used by LSTM and LR are MAE, and the evaluation indexes used by RF, SVM, and KNN are F1-score.

[0181]

[0182]

[0183] Table 5 Prediction performance of five models on six datasets

[0184] In this embodiment, for step S4, the Pearson coefficient matrix between model metrics is calculated as shown in Table 6:

[0185] LSTM RF SVM KNN LR LSTM 1 0.155995 0.120584 0.183405 0.611853 RF 0.155995 1 0.86991 0.604242 0.708073 SVM 0.120584 0.86991 1 0.524756 0.292145 KNN 0.183405 0.604242 0.524756 1 0.166362 LR 0.611853 0.139494 0.292145 0.166362 1

[0186] Table 6 Pearson coefficient matrix table between model metrics

[0187] In this embodiment, for step S5, three models with relatively large differences are selected as the final primary learners, namely LSTM, RF, and SVM; one model with the best evaluation result is selected as the secondary learner, namely LSTM.

[0188] In this embodiment, for step S5, as Figure 3 shown, the final construction process of the Stacking model is explained as follows:

[0189] (1) The first stage is the training and testing process of individual primary learners, including three models: LSTM, RF, and SVM. The model input is the training set sample features, and the output is the relative convergence deformation. The k-fold cross-validation method is used for training, and in the present invention, k = 5. First, the training set is randomly divided into 5 identical subsets. Each subset is sequentially used as the validation set, and the remaining 4 subsets are used as the training set. Secondly, model training and prediction are carried out. The prediction result of each fold is pred. The prediction results pred1, pred2, pred3, pred4, and pred5 generated by the 5 folds are concatenated to form a new training set TRAIN; the average of the test sets corresponding to each fold is taken to form a new test set TEST.

[0190] (2) The second stage is the training and testing process of the secondary learner. The model training part is the training data set merged from the new training sets TRAIN1, TRAIN2, and TRAIN3 generated by LSTM, RF, and SVM, and the testing part is the new test data set merged from TEST1, TEST2, and TEST3. The secondary learner LSTM is used to predict the test set to obtain the final result.

[0191] In this embodiment, for step S5, the final prediction result obtained by using the Stacking model is shown in Table 7, and the evaluation index is MAE:

[0192] Table 7

[0193]

[0194] The present invention focuses on the problem of predicting the convergence deformation trend of shield tunnels during the operation period. Taking the relative deformation of the transverse diameter of the measurement point as the experimental object, based on the deformation source data and environmental source data, an oversampling-integrated learning model integrating ADASYN and Stacking is established to solve the problem of effectively and accurately predicting the convergence deformation of most sample points of the tunnel.

[0195] The present invention also discloses an electronic device. Figure 4 It is a schematic diagram of the composition of the electronic device in an embodiment of the present invention; please refer to Figure 4 , at the hardware level, the electronic device includes a memory, a processor, and at least one network interface; the processor can be a microprocessor, and the memory can include a memory, such as a random access memory (RAM), and can also include a non-volatile memory, etc. Of course, the electronic device can also be provided with other hardware according to needs.

[0196] The processor, network interface, and memory can be interconnected through an internal bus, and the internal bus can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect Standard) bus, or an EISA (Extended Industry Standard Architecture) bus, etc.; the bus can include an address bus, a data bus, a control bus, etc. The memory is used to store programs (which can include an operating system program and application programs); the programs can include program codes, and the program codes can include computer operation instructions. The memory can include a memory and a non-volatile memory, and provide instructions and data to the processor.

[0197] In one embodiment, the processor can read the corresponding program from the non-volatile memory into the memory and then run; the processor can execute the program stored in the memory and is specifically used to perform the following operations (as Figure 1 shown):

[0198] Step S1: Obtain the overall historical convergence deformation data and environmental monitoring data of the tunnel, and perform data preprocessing;

[0199] Step S2: Construct data samples, and use the adaptive synthetic sampling ADASYN method to enhance the data samples;

[0200] Step S3: Select several basic models as primary learners;

[0201] Step S4: Calculate the prediction result evaluation indexes of the primary learners, and calculate the Pearson correlation coefficient between the models according to the evaluation indexes;

[0202] Step S5: Select at least two models with relatively large differences as the final primary learners, and select the model with the best evaluation result as the secondary learner. The Stacking model construction is completed.

[0203] The present invention further discloses a storage medium, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the following steps of the method of the present invention are implemented (as Figure 1 shown):

[0204] Step S1: Obtain the overall historical convergence deformation data and environmental monitoring data of the tunnel, and perform data preprocessing;

[0205] Step S2: Construct data samples, and use the Adaptive Synthetic Sampling (ADASYN) method to enhance the data samples;

[0206] Step S3: Select several basic models as the primary learners;

[0207] Step S4: Calculate the evaluation indexes of the prediction results of the primary learners, and calculate the Pearson correlation coefficient between the models according to the evaluation indexes;

[0208] Step S5: Select at least two models with relatively large differences as the final primary learners, and select the model with the best evaluation result as the secondary learner. The Stacking model construction is completed.

[0209] In summary, the tunnel operation period convergence deformation prediction method, system, electronic device and storage medium proposed by the present invention can effectively and accurately predict the convergence deformation of most sample points of the tunnel.

[0210] It should be noted that the present application can be implemented in software and / or a combination of software and hardware; for example, it can be implemented using an Application Specific Integrated Circuit (ASIC), a general-purpose computer or any other similar hardware device. In some embodiments, the software program of the present application can be executed by a processor to implement the above steps or functions. Similarly, the software program of the present application (including related data structures) can be stored in a computer-readable recording medium; for example, a RAM memory, a magnetic or optical drive or a floppy disk and similar devices. In addition, some steps or functions of the present application can be implemented using hardware; for example, a circuit that cooperates with a processor to execute each step or function.

[0211] The technical features of the above-described embodiments can be combined arbitrarily. For the sake of brevity of description, all possible combinations of the technical features in the above-described embodiments are not described. However, as long as there is no contradiction in the combination of these technical features, it should be considered to be within the scope described in this specification.

[0212] The description and application of the present invention herein are illustrative and are not intended to limit the scope of the present invention to the above embodiments. The effects or advantages involved in the embodiments may not be reflected in the embodiments due to various factors, and the description of the effects or advantages is not used to limit the embodiments. Modifications and changes to the disclosed embodiments are possible, and various components of substitution and equivalence of the embodiments are known to those of ordinary skill in the art. Those skilled in the art should clearly understand that the present invention can be implemented in other forms, structures, arrangements, proportions, and with other components, materials, and parts without departing from the spirit or essential characteristics of the present invention. Other modifications and changes can be made to the disclosed embodiments without departing from the scope and spirit of the present invention.

Claims

1. A prediction method for convergence deformation during the operation period of a tunnel, characterized in that, The method for predicting convergence deformation during tunnel operation period includes: Step S1, obtaining the overall historical convergence deformation data and environmental monitoring data of the tunnel and performing data preprocessing; Step S2, constructing data samples, and using the adaptive synthetic sampling ADASYN method to enhance the data samples; Step S3: Select several basic models as primary learners; Step S4, calculating the prediction result evaluation index of the primary learner, and calculating the Pearson correlation coefficient between models according to the evaluation index; Step S5: Select at least two models with large differences as the final primary learners, select the model with the best evaluation result as the secondary learner, and construct a stacking model.

2. The method for predicting convergence deformation during tunnel operation according to claim 1, characterized in that: The data preprocessing in step S1 includes: Step S11, collecting raw data, including cross-sectional diameter data for calculating convergence deformation, and at least one of settlement monitoring data, water level data, location data, vehicle data, disease data, sensor stress data, road data, and special event data in environmental factors; deleting data with obvious monitoring errors or abnormalities; Step S12: Fill all missing values ​​using adjacent interpolation method to ensure the integrity of input data; Step S13: align the historical convergence deformation data and the environmental monitoring data according to a unified time scale. Since the frequency of the convergence deformation data is relatively higher than that of some environmental monitoring data, the environmental monitoring data with a larger time statistical scale is aligned and completed according to the time scale of the convergence monitoring.

3. The method for predicting convergence deformation during tunnel operation according to claim 1, characterized in that: Constructing the data sample in step S2 includes the following steps: Step S211, define data sample labels; labels are targets that the model needs to learn and predict; the labels here use the evaluation index for the severity of convergence deformation in the setting specification, that is, the cross-sectional diameter relative deformation μ, calculated according to the following formula: In the formula, μ i represents the relative convergence deformation of the i-th measuring point section, and ΔD i represents the change in the transverse diameter of the i-th measuring section, and D represents the initial length of the transverse diameter of the i-th measuring section; this index is measured in thousandths (‰). Step S212, defining a data sample feature set; the feature set here is settlement monitoring data, water level data, location data, vehicle data, disease data, sensor stress data, road data and special event data in environmental factors; Step S213, encode or convert the feature set; convert some features described in pure text into numerical values ​​to make them into numerical variables or categorical variables; then define encoding categories for all categorical variables and convert them into numerical variables; Step S214: After all features are converted into numerical variables, in order to eliminate the influence of dimension and scale differences between features, the features need to be normalized. Here, the MinMaxScaler method is used for numerical normalization, and the scaling formula is: Among them, x is the value of the original feature, minx is the minimum value of the feature, maxx is the maximum value of the feature; x' is the scaled feature value; the scaled feature value range is (0,1).

4. The method for predicting convergence deformation during tunnel operation according to claim 1, characterized in that: The ADASYN processing in step S2 includes: Step S221: Define the majority class and the minority class. The majority class refers to the monitoring points in the tunnel section that are more than the average value. The minority class refers to the samples with fewer monitoring points in the tunnel section than the average value and that need to be focused on by the model. Step S222: Calculate the imbalance degree between the majority class and the minority class samples; denote the minority class samples as \(m\). s , and the majority class as \(M\). l Then the imbalance degree is \(d = \frac{m}{M}\), s where \(d\in[0,1]\). l ​ Step S223, calculate the number of samples G to be synthesized, G = (m l - m s ) * b, b ∈ [0, 1]; when b = 1, G is the difference between the minority class and the majority class. At this time, the number of majority class data after synthesizing data is exactly balanced with the minority class data; Step S224: Calculate k neighbors for each sample belonging to the minority class using the Euclidean distance. Let Δ be the number of samples belonging to the majority class among the k neighbors, and denote the proportion of the majority class as r = Δ / k, where r ∈ [0, 1]. Step S225: Obtain the r of each minority class sample in Step S224 i , for r i Normalize: Step S226: Calculate the number of new samples to be generated for each minority-class sample according to the sample weights Step S227. Calculate according to g i the number of synthetic samples to be generated for each minority sample: s i = x i +(x zi - x i )×λ, where s i is the synthetic sample, x i is the i-th sample in the minority class, x zi is a randomly selected minority class sample among the k nearest neighbors of x i , and λ ∈ [0, 1]; repeat the synthesis until the required number of synthetic samples is satisfied.

5. The tunnel operation period convergence deformation prediction method according to claim 1, characterized in that: In step S3, select several models from LSTM, SVM, RF, KNN, and LR. The model construction process of the neural network LSTM in step S3 is as follows: Step S3a1: Data preparation: Use the dataset processed by ADASYN in step S2. Step S3a2: Data division: Divide the dataset into a training set and a test set at a ratio of 8:

2. Step S3a3: Model setting: Select the number of model layers, the number of neurons in each layer, and select the activation function and optimizer. Step S3a4: Model training: Use the training set to train the LSTM model, and adjust parameters according to the training situation, such as the learning rate, batch size, etc. Step S3a5: Model evaluation and verification: Use the trained model to make predictions on the test set, calculate the error between the prediction result and the actual value, and use the mean absolute error MAE as the evaluation index. The model construction process of the support vector machine SVM in step S3 is as follows: Step S3b1: Data preparation: Use the dataset processed by ADASYN in step S2. Step S3b2: Data division: Divide the dataset into a training set and a test set at a ratio of 8:

2. Step S3b3: Model setting: Select the Gaussian kernel function RBF, and define hyperparameters such as the penalty coefficient C of the model and the kernel function parameters. Step S3b4: Model training: Use the training set to train the SVM model, and adjust parameters according to the training situation. Step S3b5: Model evaluation and verification: Use the trained model to make predictions on the test set, calculate the error between the prediction result and the actual value, and use the F1-score as the evaluation index. The model construction process of the random forest RF in step S3 is as follows: Step S3c1: Data preparation: Use the dataset processed by ADASYN in step S2. Step S3c2: Data division: Divide the dataset into a training set and a test set at a ratio of 8:

2. Step S3c3: Model setting: Determine the number of decision trees in the random forest, determine the number of features randomly selected for each tree, and adjust the depth of the tree and other parameters to prevent overfitting. Step S3c4: Model training: Use the training set to train the RF model, and adjust parameters according to the training situation. Step S3c5: Model evaluation and verification: Use the trained model to make predictions on the test set, calculate the error between the prediction result and the actual value, and use the F1-score as the evaluation index. The model construction process of the k-nearest neighbor KNN in step S3 is as follows: Step S3d1: Data preparation: Use the dataset processed by ADASYN in step S2. Step S3d2, data division: divide the data set into a training set and a test set with a ratio of 8:2; Step S3d3, model setting: determine the K value in the KNN algorithm, that is, select the number of neighbors, and use the Euclidean distance to calculate the distance between samples; Step S3d4, model training: use the training set to train the KNN model and adjust the parameters according to the training situation; Step S3d5, model evaluation and verification: Use the trained model to make predictions on the test set, calculate the error between the prediction result and the actual value, and use F1-score as the evaluation indicator; The model building process of linear regression LR in step S3 is as follows: Step S3e1, data preparation: using the data set processed by ADASYN in step S2; Step S3e2, data division: divide the data set into a training set and a test set with a ratio of 8:2; Step S3e3, model setting and model training: use the training set to train the LR model and adjust the parameters according to the training situation; Step S3e4, model evaluation and verification: Use the trained model to make predictions on the test set, calculate the error between the predicted result and the actual value, and use the mean absolute error (MAE) as the evaluation indicator.

6. The method for predicting convergence deformation during tunnel operation according to claim 1, characterized in that: The calculation formula of the Pearson correlation coefficient in step S4 is: Among them, X and Y represent the evaluation index variables of two primary learners, and ρ X,Y measures the correlation of the prediction results of the two models. If the coefficient is close to 1, it indicates a strong positive correlation; if it is close to -1, it indicates a strong negative correlation; if it is close to 0, it indicates a weak correlation and a large difference between the two models.

7. The method for predicting convergence deformation during tunnel operation according to claim 1, characterized in that: The construction process of the Stacking model in step S5 is as follows: Step S51, primary learner preparation: according to the prediction results of the learner in step S4 and the Pearson coefficient, three models with smaller prediction errors and smaller correlations are selected as the final primary learners; Step S52, data division: using the data set processed by ADASYN in step S2, the input data is divided into a training set and a test set, with a ratio of 8:2; secondly, the training set is divided into five parts: train1, train2, train3, train4, train5; Step S53, training and testing the primary learner: using train1, train2, train3, train4, train5 as validation sets in turn, and the remaining 4 sets as training sets. After the training is completed, the validation set data is predicted to obtain pred1, pred2, pred3, pred4, pred5, and a 5-fold cross validation is completed; the three models are cross-validated in turn to obtain the cross-validation prediction results of each model; Step S54, training and testing the secondary learner: stack the prediction results of the three primary learners as the training set input of the secondary learner, take the average of the test sets after each division as the test set of the secondary learner, select the primary learner with the best prediction effect as the secondary learner, and output the final prediction results on the test set.

8. A convergence deformation prediction system during the operation period of a tunnel, characterized in that, The tunnel operation period convergence deformation prediction system comprises: Data preprocessing module, used to obtain the overall historical convergence deformation data and environmental monitoring data of the tunnel and perform data preprocessing; A sample construction enhancement module for constructing data samples and enhancing the data samples by using the Adaptive Synthetic Sampling (ADASYN) method; A basic model selection module for selecting a number of basic models as primary learners; A result evaluation index prediction module for calculating the prediction result evaluation indexes of the primary learners and calculating the Pearson correlation coefficient between the models according to the evaluation indexes; A stacked model construction module for selecting at least two models with large differences as the final primary learners, selecting one model with the best evaluation result as the secondary learner, and constructing a stacked Stacking model.

9. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images