Intelligent tunnel portal structure deformation monitoring method and device based on deep learning

Abstract

The invention provides an intelligent tunnel portal structure deformation monitoring method and device based on deep learning, and relates to the technical field of civil engineering monitoring and intelligent constructions.The method comprises the steps that feature data are extracted and include beam displacement at representative points, vibration frequency, pore water pressure, tunnel wall soil pressure, accumulated water depth, earthquake acceleration and wind speed; preprocessing the feature data, calculating a soil deformation index, a water and soil stability index and a dynamic deformation index, combining the indexes with other feature data to form feature vectors, constructing a neural network model, training a neural network to minimize a prediction error, and verifying the classification precision of the model by using a test set. The deformation state and potential risks of the tunnel portal structure are obtained, and model output and a threshold value alarm mechanism are combined. The tunnel portal structure health is monitored in real time through the neural network, the deformation state and the potential risk are evaluated, and decision support is provided.

Description

Technical Field

[0001] The present invention relates to the technical fields of civil engineering monitoring and intelligent building, and specifically to an intelligent tunnel portal structure deformation monitoring method and device based on deep learning. Background Art

[0002] With the acceleration of the urbanization process, tunnel engineering has become increasingly important in infrastructure construction. Ensuring the safety of tunnel portal structures has become an urgent problem to be solved. Traditional monitoring methods often rely on manual inspections and simple sensor data, which are inefficient and difficult to detect potential risks in real time. Therefore, intelligent monitoring methods based on sensor technology and deep learning algorithms have emerged, which can effectively improve the accuracy and real-time performance of monitoring and ensure the safety management of tunnels.

[0003] Currently, many research institutions and enterprises have developed integrated intelligent monitoring systems that can analyze a large amount of sensor data in combination with deep learning to monitor the deformation of tunnel portals in real time. These systems improve the efficiency and safety of tunnel monitoring by automatically identifying structural hazards. However, despite certain progress in technology, challenges such as data acquisition, model generalization ability, and real-time performance still exist. Future development will focus on improving the intelligence level and data fusion ability of monitoring systems.

[0004] In the prior art, the publication number CN118094205A discloses an intelligent tunnel structure health monitoring system based on deep learning. By using three signal processing technologies, namely the AR model, discrete wavelet transform, and empirical mode decomposition, to analyze tunnel sequence signals, the time-domain, frequency-domain, local features, and non-linear features of tunnel sequence signals can be extracted, noise interference in tunnel sequence signals can be removed, and the structural features of tunnel sequence signals can be comprehensively analyzed from multiple angles; this system combines a one-dimensional convolutional neural network and a long short-term memory algorithm to construct a hybrid deep learning model, improving the performance and generalization ability of the tunnel structure health monitoring system.

[0005] Deficiencies of the Prior Art:

[0006] The existing tunnel structure health monitoring technologies mainly rely on traditional sensors and manual analysis methods, which lead to significant problems such as low data processing efficiency and poor real-time performance. Since monitoring work often relies on manual inspections, there is usually a time delay in data collection and analysis. This delay may not only prevent potential risks from being identified in a timely manner, thus affecting the safety of tunnels, but also lead to an exacerbation of structural damage. In addition, many traditional systems can only evaluate limited safety parameters, often unable to comprehensively capture the dynamic changes of structures, lacking the ability to effectively process complex signal data, which directly affects the accuracy and reliability of safety level prediction;

[0007] Meanwhile, existing early warning systems usually operate based on fixed thresholds and lack flexibility for different tunnel and environmental conditions. Such a simple early warning mechanism may lead to false alarms or missed alarms, thus affecting the effectiveness of emergency responses and the accuracy of decision-making. In addition, many traditional monitoring schemes fail to fully utilize modern information technologies, resulting in relatively weak capabilities for comprehensive data analysis and decision support. These limitations make it difficult for traditional monitoring technologies to cope with rapidly changing environments and diverse structural requirements. Therefore, it is urgent to improve them with advanced technologies such as deep learning to enhance the intelligence, automation level, and early warning ability of tunnel monitoring, so as to better ensure the safe and stable operation of tunnels.

[0008] Therefore, it is necessary to provide an intelligent tunnel portal structure deformation monitoring method and device based on deep learning to solve the above problems.

[0009] The above information disclosed in the background art section is only used to strengthen the understanding of the background of the present disclosure. Therefore, it may include information that does not constitute the prior art known to those of ordinary skill in the art. Summary of the Invention

[0010] The purpose of the present invention is to provide an intelligent tunnel portal structure deformation monitoring method and system based on deep learning to solve the problems raised in the above background art.

[0011] To achieve the above purpose, the present invention provides the following technical solutions:

[0012] An intelligent tunnel portal structure deformation monitoring method based on deep learning, the specific steps include:

[0013] Step 1: Collect physical data related to the tunnel portal structure deformation at equal time intervals, and extract characteristic data with time series distribution according to the obtained related physical data, including beam displacement, vibration frequency, pore water pressure, tunnel wall soil pressure, water accumulation depth, seismic acceleration, and wind speed at representative points;

[0014] Step 2: After preprocessing the calculated characteristic data, combine the beam displacement, vibration frequency, pore water pressure, and tunnel wall soil pressure at representative points to calculate the soil deformation index, water-soil stability index, and dynamic deformation index, and combine the three indexes with the water accumulation depth, seismic acceleration, and wind speed to generate a characteristic vector of the tunnel portal structure deformation. The preprocessing includes data cleaning and normalization;

[0015] Step 3: Build a neural network model, including an input layer, a hidden layer, and an output layer. The input layer receives the characteristic vector, the hidden layer is responsible for extracting the time series characteristics in the characteristic vector, and the output layer is used to identify the deformation type of the tunnel portal structure;

[0016] Step 4: Train the neural network model based on the historical deformation data of the tunnel portal structure to minimize the prediction error, optimize the model parameters, and use the test set to verify the classification accuracy of the trained model to obtain the deformation types of the tunnel portal structure;

[0017] Step 5: According to the deformation prediction results output by the model, combine the decision threshold warning mechanism to improve the precision and recall rate of the neural network model.

[0018] Furthermore, preprocess the calculated feature data. The preprocessing includes data cleaning and normalization, and the method is as follows:

[0019] Data cleaning of the feature data includes the detection and deletion of outliers and duplicate data, and the processing of missing values. Specifically, use the statistical method to identify outliers and duplicate data in the feature data, delete outliers and duplicate data in the feature data, and use the mean, median or mode of the feature data to fill in the missing values in the feature data;

[0020] The normalization processing of the feature data is min-max normalization, which scales the data to the range [0,1] so that all feature data have the same scale. The formula is as follows:

[0021]

[0022] Among them, Y ′ is the feature data after normalization processing, Y is the original feature data, Y min is the minimum value of the same type of feature data in the dataset, Y max is the maximum value of the same type of feature data in the dataset.

[0023] Furthermore, combine the beam displacement, vibration frequency, pore water pressure, and tunnel wall soil pressure at the representative point to calculate the soil deformation index, water and soil stability index, and dynamic deformation index. The method is as follows:

[0024] Combine the beam displacement at the representative point in the selected feature data with the tunnel wall soil pressure to generate the soil deformation index. The soil deformation index comprehensively represents the relationship between the beam displacement at the representative point and the tunnel wall soil pressure, aiming to reflect the deformation degree of the tunnel structure caused by the change of soil pressure. The formula for calculating the soil deformation index is:

[0025]

[0026] Among them, SDI k represents the soil deformation index of the tunnel portal at time k, D k+1 is the beam displacement at the representative point at time k+1, D k is the beam displacement at the representative point at time k, P k+1is the soil pressure on the tunnel wall at time k+1, P k is the soil pressure on the tunnel wall at time k, where k is the index of the acquisition time;

[0027] The pore water pressure in the selected characteristic data is combined with the soil pressure on the tunnel wall to generate a water-soil stability index. The water-soil stability index reflects the relationship between the pore water pressure and the soil pressure, aiming to evaluate the effective stress and stability of the soil, especially in the case of strength changes caused by water level changes. The formula for calculating the water-soil stability index is:

[0028]

[0029] where, WSSI k represents the water-soil stability index at the tunnel entrance at time k, u is the pore water pressure at time k, P total ′ is the effective soil pressure on the tunnel wall at time k;

[0030] The beam displacement at the representative point in the selected characteristic data is combined with the vibration frequency to generate a dynamic deformation index. The dynamic deformation index synthesizes the relationship between the beam displacement and the vibration frequency at the representative point, and is used to evaluate the dynamic response of the structure under external loads or seismic effects. The formula for calculating the dynamic deformation index is:

[0031]

[0032] where, DDI k represents the dynamic deformation index at the tunnel entrance at time k, f k+1 is the vibration frequency at time k+1, f k is the vibration frequency at time k.

[0033] Furthermore, the three indices are combined with the water depth, seismic acceleration, and wind speed to generate a characteristic vector of the structural deformation at the tunnel entrance. The method is as follows:

[0034] According to the three index data SDI k 、WSSI k 、DDI k calculated, and the extracted characteristic data includes water depth, seismic acceleration, and wind speed. The characteristic data and the index data are combined into a characteristic vector. Before combining different characteristics, it is necessary to standardize each characteristic data and index data to ensure that the characteristic data and the newly generated index data are compared on the same scale. The Z-score standardization method is used. After the standardization process, the calculated index data and the extracted characteristic data are substituted into the characteristic vector.

[0035] Furthermore, a neural network model is constructed, including an input layer, a hidden layer, and an output layer. The input layer receives feature vectors, the hidden layer is responsible for extracting the temporal features in the feature vectors, and the output layer is used to identify the deformation types of the tunnel portal structure. The method is as follows:

[0036] The input layer is responsible for receiving feature vectors, including soil deformation index, water and soil stability index, dynamic deformation index, waterlogging depth, seismic acceleration, and wind speed. The feature vectors are set as (N, t, n), where N is the number of samples of the feature vectors, t is the number of time steps, and n is the feature dimension of each time step;

[0037] The hidden layer controls the flow of feature data by setting multiple gating mechanisms, including forget gates, input gates, and output gates, to extract the temporal features in the sequence data. The forget gate determines how much initial feature information to retain to reduce the impact of noise; the input gate controls the inflow of new information at the current time step and selectively introduces features helpful for future predictions by updating the state of the memory unit; the output gate determines the output of the hidden state based on the state of the current memory unit and passes it to the next time step and the final output layer. The logical formulas are as follows:

[0038] f t =σ[W f *(h t-1 ,x t )+b f

[0039] s t =σ[W s *(h t-1 ,x t )+b s

[0040] o t =σ[W o *(h t-1 ,x t )+n o

[0041]

[0042] where f t represents the output of the forget gate, indicating which information is forgotten. W f , W s , W o represent the weight matrices corresponding to their respective gates, b f , b s , b o represent the bias terms corresponding to their respective gates. σ is the Sigmoid activation function, which compresses the output between 0 and 1. h t-1 ​​​Represents the hidden state of the previous time step, containing feature data information, x t The input of the current time step, C t The state of the memory cell at the current time step, C t-1 The state of the memory cell at the previous time step;

[0043] The output layer is responsible for mapping the features extracted by the hidden layer to specific categories, classifying the deformation types of the tunnel portal structure. The method is as follows:

[0044] According to the number of deformation types of the tunnel portal structure contained in the extracted feature data, the number of nodes in the output layer is adjusted to the same number corresponding to the number of deformation types of the tunnel portal structure. The Softmax activation function is used to convert the output into probability values to represent the probability distribution of each category. The formula is as follows:

[0045]

[0046] Among them, P represents the probability occupied by each deformation type, y i Is the jth deformation type in the ith node of the output layer, X is the input feature data, output by the hidden layer, z i Is the linear combination output of the neurons in the ith node of the output layer, j is the index of the number of deformation types, j ∈ [1, N], N is the total number of deformation types Represents the exponential value of the output of the jth deformation type, which is the exponential operation of the linear combination output of the neurons, ensuring that each value is positive.

[0047] Furthermore, the cross-entropy function is used as the loss function to optimize the neural network model, which is used to evaluate the difference value between the probability distribution output by the neural network model and the actual label. The formula is as follows:

[0048]

[0049] Among them, Represents the value of the loss function for the jth deformation type in the ith node, that is, the predicted probability of the jth deformation type in the ith node, reflecting the difference between the model prediction and the true label, o i Represents the one-hot encoding of the true label. If the sample data belongs to the category of the ith node, then o i = 1, and the loss is If the sample data does not belong to the category of the ith node, then o i = 0, and the loss is 0, Is the predicted probability of the model for the category of the ith node.

[0050] Furthermore, the neural network model is trained to minimize the prediction error, optimize the model parameters, and the classification accuracy of the trained model is verified using the test set. The method is as follows:

[0051] By using the training set and the test set for the training and evaluation of the neural network model, first, the model is trained by backpropagation using the training set to minimize the prediction error. The model uses the cross-entropy loss function to measure the difference between the prediction result and the actual label. After the training is completed, the test set is used for verification, and the recall rate and precision rate of the model are calculated to evaluate its classification performance. The formulas for calculating the precision rate and recall rate are:

[0052]

[0053] where P and R represent the precision rate and recall rate of the model respectively, TP is the number of true positives, FP is the number of false positives, and FN is the number of false negatives.

[0054] Furthermore, according to the deformation prediction result output by the model, the precision rate and recall rate of the neural network model are optimized by combining the decision threshold warning mechanism. The method is as follows:

[0055] Optimize a comprehensive index F1-score to consider both the precision rate and recall rate at the same time. Set the comprehensive index as the decision threshold. If the predicted probability value output by the model exceeds the decision threshold, then adjust the threshold of the predicted probability value as the positive class to change the precision rate and recall rate, and use cross-validation to ensure the consistent performance of the model on different data sets. The formula is as follows:

[0056]

[0057] where F1 is the comprehensive index, L is the logical value for judging whether the predicted probability value exceeds the decision threshold. When L = 0, it means that the predicted probability value does not exceed the decision threshold and no adjustment is needed; when L = 1, it means that the predicted probability value exceeds the decision threshold and the threshold of the predicted probability value as the positive class needs to be adjusted to optimize the precision rate and recall rate.

[0058] The present invention also provides an intelligent tunnel portal structure deformation monitoring device based on deep learning. The deformation monitoring device is used to execute the above-mentioned intelligent tunnel portal structure deformation monitoring method based on deep learning, including:

[0059] A data acquisition and feature extraction module, which is used to collect physical data related to the deformation of the tunnel portal structure at equal time intervals, and extract feature data with time series distribution according to the obtained relevant physical data, including beam displacement, vibration frequency, pore water pressure, tunnel wall earth pressure, water accumulation depth, seismic acceleration, and wind speed at the representative points;

[0060] A data preprocessing and index calculation module, which is used to preprocess the calculated feature data, and then combine the beam displacement, vibration frequency, pore water pressure, and soil tunnel wall pressure at the representative points to calculate the soil deformation index, the water-soil stability index, and the dynamic deformation index. Combine the three indexes with the water accumulation depth, seismic acceleration, and wind speed to generate a feature vector of the deformation of the tunnel entrance structure. The preprocessing includes data cleaning and normalization;

[0061] A neural network construction module, which is used to construct a neural network model, including an input layer, a hidden layer, and an output layer. The input layer receives the feature vector, the hidden layer is responsible for extracting the time series features in the feature vector, and the output layer is used to identify the deformation type of the tunnel entrance structure;

[0062] A model training and evaluation module, which trains the neural network model based on the historical deformation data of the tunnel entrance structure to minimize the prediction error, optimize the model parameters, and uses the test set to verify the classification accuracy of the trained model to obtain the deformation type of the tunnel entrance structure;

[0063] A real-time monitoring and decision support module, which is used to improve the precision and recall rate of the neural network model according to the deformation prediction result output by the model and in combination with a decision threshold warning mechanism.

[0064] Compared with the prior art, the beneficial effects of the present invention are:

[0065] The intelligent tunnel entrance structure deformation monitoring method of the present invention solves the deficiencies of the existing monitoring technologies in terms of data processing efficiency, model accuracy, and real-time monitoring ability by introducing deep learning technology and a comprehensive data processing process. First, through the data acquisition and feature extraction module, the physical data related to the tunnel entrance structure is obtained in a timely manner, and multiple important soil deformation indexes and stability indexes are calculated, ensuring the comprehensiveness and accuracy of the monitoring data. Then, during the data preprocessing process, data cleaning and normalization are performed to eliminate the influence of outliers and missing values, improving the quality of the model input data. These steps make the generation of the feature vector more accurate, providing a solid foundation for the subsequent training of the neural network model, and helping to improve the classification accuracy of the model and the ability to respond promptly to the deformation of the tunnel structure;

[0066] In addition, by constructing a neural network model for deep learning, the present invention cleverly extracts and analyzes temporal features, achieving efficient classification and recognition of the deformation types of the tunnel entrance structure. Combining with a threshold alarm mechanism, it monitors the health status of the tunnel entrance in real time and generates a detailed monitoring report, providing a scientific basis for decision-making support. This intelligent monitoring method not only improves the flexibility of the early warning system, reduces the risks of false alarms and missed alarms, but also ensures the adaptability to complex environments and diversified structures. In summary, the present invention significantly improves the degree of intelligence and automation level of tunnel structure health monitoring, providing an important guarantee for the safe operation of the tunnel.

[0067] By introducing deep learning technology and a comprehensive data processing process, the present invention significantly improves the efficiency, accuracy, and real-time performance of tunnel entrance structure deformation monitoring, providing an important guarantee for the safe operation of the tunnel. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 It is a schematic diagram of the overall method flow of the present invention.

[0069] Figure 2 It is a schematic diagram of the system module flow of the present invention DETAILED DESCRIPTION OF THE EMBODIMENTS

[0070] To make the objectives, technical solutions, and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below in conjunction with specific embodiments.

[0071] It should be noted that unless otherwise defined, the technical terms or scientific terms used in the present invention should have the ordinary meanings understood by those of ordinary skill in the field to which the present invention belongs. The "first", "second", and similar terms used in the present invention do not denote any order, quantity, or importance, but are only used to distinguish different components. The terms such as "including" or "comprising" mean that the elements or items appearing before this term cover the elements or items listed after this term and their equivalents, without excluding other elements or items. The terms such as "connected" or "coupled" are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. The terms such as "upper", "lower", "left", "right", etc. are only used to represent relative positional relationships, and when the absolute position of the object being described changes, the relative positional relationship may also change accordingly.

[0072] Embodiment:

[0073] Please refer to Figure 1 , an intelligent tunnel entrance structure deformation monitoring method based on deep learning, and the specific steps include:

[0074] Step 1: Collect physical data related to the structural deformation of the tunnel entrance at equal time intervals, and extract characteristic data with time series distribution from the obtained relevant physical data, including beam displacement, vibration frequency, pore water pressure, soil pressure on the tunnel wall, water accumulation depth, seismic acceleration, and wind speed at the representative points;

[0075] Step 2: After preprocessing the calculated characteristic data, combine the beam displacement, vibration frequency, pore water pressure, and soil pressure on the tunnel wall at the representative points to calculate the soil deformation index, water and soil stability index, and dynamic deformation index, and combine the three indexes with the water accumulation depth, seismic acceleration, and wind speed to generate a characteristic vector of the structural deformation of the tunnel entrance. The preprocessing includes data cleaning and normalization;

[0076] Step 3: Build a neural network model, including an input layer, a hidden layer, and an output layer. The input layer receives the characteristic vector, the hidden layer is responsible for extracting the time series characteristics in the characteristic vector, and the output layer is used to identify the deformation type of the tunnel entrance structure;

[0077] Step 4: Train the neural network model based on the historical tunnel entrance structure deformation data to minimize the prediction error, optimize the model parameters, and use the test set to verify the classification accuracy of the trained model to obtain the deformation type of the tunnel entrance structure;

[0078] Step 5: According to the deformation prediction result output by the model, combine the decision threshold warning mechanism to improve the precision and recall rate of the neural network model.

[0079] It should be noted that the preprocessing of the characteristic data, including data cleaning and normalization, is crucial because it directly affects the training effect and prediction accuracy of the model. Through data cleaning, outliers and duplicate data can be effectively identified and deleted to ensure the quality of the input data, thereby reducing the noise interference in model learning and improving the stability and reliability of the model. At the same time, the normalization process scales all characteristic data to the same scale, avoiding the weight imbalance caused by the difference in the dimensions between features, enabling the model to converge faster during training, and strengthening the comprehensive analysis ability of different features. Therefore, this preprocessing step lays a solid foundation for the effective learning and accurate prediction of the subsequent deep learning model.

[0080] Therefore, it is necessary to preprocess the calculated characteristic data. The preprocessing includes data cleaning and normalization, and the method is as follows:

[0081] Data cleaning of the characteristic data includes the detection and deletion of outliers and duplicate data, and the processing of missing values. Specifically, statistical methods are used to identify outliers and duplicate data in the characteristic data, delete the outliers and duplicate data in the characteristic data, and use the mean, median, or mode of the characteristic data to fill in the missing values in the characteristic data;

[0082] The normalization of the characteristic data is min-max normalization, which scales the data to the range [0,1] so that all characteristic data have the same scale. The formula is as follows:

[0083]

[0084] where Y ′ is the characteristic data after normalization, Y is the original characteristic data, Y min is the minimum value of the same type of characteristic data in the dataset, and Y max is the maximum value of the same type of characteristic data in the dataset.

[0085] It should be noted that calculating the soil deformation index, the water-soil stability index, and the dynamic deformation index by combining the beam displacement, vibration frequency, pore water pressure, and tunnel wall soil pressure at the representative point is of great significance for the health monitoring of tunnel structures. These indices synthesize the relationships between different physical quantities and can comprehensively reflect the behavior and state of the soil and the structure under various external conditions. The soil deformation index, through the combination of beam displacement and soil pressure, can effectively evaluate the degree of deformation of the tunnel caused by soil pressure changes, thus providing a reference for engineering safety; the water-soil stability index evaluates the effective stress of the soil under water level changes to ensure the stability of the soil mass; while the dynamic deformation index evaluates the response ability of the tunnel under dynamic loads through the relationship between beam displacement and vibration frequency to help identify potential risks. In summary, the setting and calculation of these indices provide a scientific basis for timely discovering and handling potential problems of tunnel structures, ensuring the safety and stability of the tunnel; the selected representative point is the crown position at the tunnel entrance, and monitoring the changes in beam displacement and soil pressure here can evaluate the overall stability and deformation trend of the tunnel structure.

[0086] Therefore, it is necessary to calculate the soil deformation index, the water-soil stability index, and the dynamic deformation index by combining the beam displacement, vibration frequency, pore water pressure, and tunnel wall soil pressure at the representative point. The method is as follows:

[0087] Combining the beam displacement at the representative point in the selected characteristic data with the tunnel wall soil pressure to generate the soil deformation index, which aims to reflect the degree of deformation of the tunnel structure caused by soil pressure changes. The formula for calculating the soil deformation index is:

[0088]

[0089] where SDI k represents the soil deformation index at the tunnel entrance at time k, D k+1 is the beam displacement at time k + 1, D k is the beam displacement at time k, and P k+1is the soil pressure on the tunnel wall at time k+1, P k is the soil pressure on the tunnel wall at time k, where k is the index of the acquisition time; in the above formula, soil deformation is usually caused by external loads, and the change in soil pressure will cause beam displacement or deformation of the soil mass. Therefore, the numerator in the formula is the change in beam displacement, and the denominator is the change in soil pressure. The ratio of the two can effectively reflect the deformation sensitivity of the soil mass under stress. By calculating SDI k , the response degree of the soil mass to the change in soil pressure can be quantified. A higher SDI k value indicates that the soil mass has a larger beam displacement under the same change in soil pressure, which may mean that the soil mass has a stronger deformation ability or there is a risk of instability;

[0090] The pore water pressure in the selected characteristic data is combined with the soil pressure to generate a water-soil stability index. The water-soil stability index reflects the relationship between the pore water pressure and the soil pressure, aiming to evaluate the effective stress and stability of the soil, especially in the case of strength changes caused by water level changes. The formula for calculating the water-soil stability index is:

[0091]

[0092] where, WSSI k represents the water-soil stability index at the tunnel entrance at time k, u is the pore water pressure at time k, and P total ′ is the effective soil pressure on the tunnel wall at time k; in the above formula, when u k increases, WSSI k also increases, indicating that the stability of the soil mass decreases, which may lead to an increase in risks such as soil landslides and liquefaction; when u k decreases, WSSI k also decreases, indicating that the proportion of effective soil pressure in the total pressure is higher, the soil mass is more stable, and the ability to resist external loads is enhanced; and the increase in effective soil pressure may come from the increase in the total soil pressure F k ′ , such as an increase in soil weight or an increase in external loads, or a decrease in the pore water pressure u k . When increases, WSSI k will decrease. A low WSSI k value means that the stability of the soil mass increases and the ability to resist external loads is enhanced;

[0093] The beam displacement at the representative point in the selected characteristic data is combined with the vibration frequency to generate a dynamic deformation index. The dynamic deformation index synthesizes the relationship between the beam displacement and the vibration frequency at the representative point and is used to evaluate the dynamic response of the structure under external loads or seismic effects. The formula for calculating the dynamic deformation index is:

[0094]

[0095] Among them, DDI k represents the dynamic deformation index, and f k+1 is the vibration frequency at the (k + 1)-th moment, and f k is the vibration frequency at the k-th moment; in the above formula, the numerator is the change in beam displacement, representing the actual deformation degree of the structure under dynamic load, and the denominator is the change in vibration frequency, reflecting the dynamic characteristics and response frequency of the structure; when the structure is subjected to dynamic load, such as an earthquake or a vehicle passing by, it will cause a change in beam displacement, and a large beam displacement usually indicates a large response of the structure, which may affect the safety and stability of the structure; if the change in vibration frequency increases while the change in beam displacement remains unchanged, DDI k will decrease. This may mean that the response of the structure to dynamic load is enhanced, and the increase in frequency may indicate that the structure exhibits greater stiffness and anti-vibration ability. If the change in vibration frequency decreases, then DDI k will increase. This may indicate that under dynamic load, the intensity of the dynamic response of the structure increases, and there may be a risk of increased deformation; a high DDI value indicates that under dynamic load conditions, the change in beam displacement is relatively large compared to the change in frequency. This may mean that when the structure is subjected to dynamic load, there is a large beam displacement or deformation, resulting in potential safety hazards.

[0096] It should be noted that in the analysis of the structural deformation characteristics of the tunnel portal, the three key indices SD k , WSSI k , DDI k and other related characteristic data including water accumulation depth, seismic acceleration, and wind speed are combined into a characteristic vector. This combination method can provide a more comprehensive perspective to help us deeply understand and analyze the structural deformation behavior of the tunnel portal. By integrating data from different sources, potential interactions and influencing factors can be captured, improving the accuracy and reliability of prediction. By integrating multiple characteristics, the characteristic vector can more comprehensively reflect the deformation characteristics of the tunnel portal structure under different environments and conditions, thereby improving the depth and breadth of analysis. After comprehensively considering multiple indicators and characteristics, the model can better capture the complex relationships affecting tunnel deformation, improve the prediction accuracy of deformation behavior, and the characteristic vector can provide a unified framework for subsequent analysis and calculation, helping engineers and decision-makers more quickly identify potential risks and formulate corresponding monitoring and maintenance measures. Finally, by establishing a systematic characteristic vector, the intelligentization and automation of tunnel structure health monitoring can be promoted, providing data support for relevant decisions.

[0097] In the process of generating feature vectors, it is crucial to adopt the Z-score normalization method, as it can eliminate the differences in scales among different features, enabling effective comparison of each feature on the same dimension. This process not only improves the effectiveness of data analysis but also enhances the model's sensitivity to each feature, thus ensuring that in subsequent analysis, the contributions of features to the overall deformation of the tunnel portal structure can be reasonably evaluated. By calculating the three indices SDI k 、WSSI k 、DDI k and other feature data such as waterlogging depth, seismic acceleration, and wind speed, normalization is carried out to ensure that these data from different sources can be combined on the same scale to form feature vectors, thereby improving the accuracy and stability of the model. This normalization process lays a good foundation for the subsequent learning and prediction of the model, enabling it to better capture and reflect the internal relationships among various features when facing complex engineering problems.

[0098] In the process of generating feature vectors, adopting the Z-score normalization method is an essential step. Through normalization, the scale differences in the impacts of different features on the model can be eliminated, enabling comparison of each feature on the same dimension. This not only improves the effectiveness of data analysis but also enhances the model's sensitivity to each feature, ensuring that in subsequent analysis, the contributions of each feature to the overall deformation behavior can be reasonably evaluated.

[0099] Therefore, it is necessary to combine the three indices with the waterlogging depth, seismic acceleration, and wind speed to generate the feature vectors of the tunnel portal structure deformation, and the method is as follows:

[0100] According to the three index data SDI k 、WSSI k 、DDI k calculated, combined with the extracted feature data including waterlogging depth, seismic acceleration, and wind speed, the feature data and index data are combined into a feature vector. Before combining different features, it is necessary to normalize each feature data and index data to ensure that the feature data and the newly generated index data are compared on the same scale. Using the Z-score normalization method, after normalization, the calculated index data and the extracted feature data are substituted into the feature vector.

[0101] It should be noted that the neural network model receives multi-dimensional feature vectors through a structured input layer, covering key parameters such as soil deformation index, water and soil stability index, waterlogging depth, etc. It uses the gating mechanism in the hidden layer to effectively extract temporal features, ensuring the efficient management of information flow and the reduction of noise. In the output layer, the model adjusts the number of nodes to correspond to the deformation types and uses the Softmax activation function to convert the output of the hidden layer into a probability distribution, accurately identifying the deformation types of the tunnel portal structure. This design not only improves the analysis ability of complex temporal data but also provides a scientific basis for the health monitoring and management of tunnel structures, thus enhancing engineering safety.

[0102] Therefore, it is necessary to construct a neural network model, including an input layer, a hidden layer, and an output layer. The input layer receives feature vectors, the hidden layer is responsible for extracting temporal features from the feature vectors, and the output layer is used to identify the deformation types of the tunnel portal structure. The method is as follows:

[0103] The input layer is responsible for receiving feature vectors, including soil deformation index, water and soil stability index, dynamic deformation index, waterlogging depth, seismic acceleration, and wind speed. The feature vectors are set as (N, t, n), where N is the number of samples of the feature vector, t is the number of time steps, and n is the feature dimension of each time step;

[0104] The hidden layer controls the flow of feature data by setting multiple gating mechanisms, including forget gates, input gates, and output gates, to extract temporal features from sequence data. The forget gate determines how much initial feature information to retain to reduce the impact of noise; the input gate controls the inflow of new information at the current time step and selectively introduces features that are helpful for future predictions by updating the state of the memory unit; the output gate determines the output of the hidden state based on the state of the current memory unit and passes it to the next time step and the final output layer. The logical formulas are as follows:

[0105] f t =σ[W f *(h t-1 ,x t )+b f

[0106] s t =σ[W s *(h t-1 ,x t )+b s

[0107] o t =σ[W o *(h t-1 ,x t )+b o

[0108] ​​​

[0109] Among them, f t represents the output of the forget gate, indicating which information is forgotten. W f and W s and W o respectively represent the weight matrices corresponding to their respective gates. b f and b s and b o respectively represent the bias terms corresponding to their respective gates. σ is the Sigmoid activation function, which is used to compress the output between 0 and 1. h t-1 represents the hidden state at the previous time step, which contains the feature data information between them. x t is the input at the current time step. C t is the memory cell state at the current time step. C t-1 is the memory cell state at the previous time step;

[0110] The output layer is responsible for mapping the features extracted by the hidden layer to specific categories, classifying the deformation types of the tunnel portal structure. The method is as follows:

[0111] According to the number of deformation types of the tunnel portal structure contained in the extracted feature data, the number of nodes in the output layer is adjusted to the same number corresponding to the number of deformation types of the tunnel portal structure. The Softmax activation function is used to convert the output into probability values to represent the probability distribution of each category. The formula is as follows:

[0112]

[0113] Among them, P represents the probability occupied by each deformation type. y i is the jth deformation type in the ith node of the output layer. X is the input feature data, which is output by the hidden layer. z i is the linear combination output of the neurons in the ith node of the output layer. j is the index of the number of deformation types, j ∈ [1, N], and N is the total number of deformation types. represents the exponential value of the output of the jth deformation type. This value is the exponential operation of the linear combination output of the neurons, ensuring that each value is a positive number. The deformation types included in the total number of deformation types are tunnel crown settlement, tunnel wall heave, local deformation cracking, basal lateral displacement, and tunnel structure shrinkage and expansion.

[0114] It should be noted that using the cross-entropy function as the loss function to optimize the neural network model is crucial because it can effectively evaluate the difference between the probability distribution of the model output and the actual labels, thereby guiding the model to learn more accurate classification decisions. Specifically, cross-entropy not only quantifies the reliability of the model prediction but also, by converting the true labels into one-hot encoded form, clearly distinguishes the contributions of each category, causing the loss function to produce a significant penalty when the sample belongs to a specific category. And when it does not belong, there is no loss. This design ensures that the model focuses on improving the most likely error-prone categories during training, thereby accelerating convergence and improving classification accuracy, ultimately enhancing the model's ability to identify the deformation types of tunnel portal structures.

[0115] Therefore, it is necessary to use the cross-entropy function as the loss function to optimize the neural network model for evaluating the difference value between the probability distribution of the neural network model output and the actual labels. The formula is as follows:

[0116]

[0117] Where, represents the value of the loss function for the jth deformation type in the ith node, that is, the predicted probability of the jth deformation type in the ith node, reflecting the difference between the model prediction and the true label. i represents the one-hot encoding of the true label. If the sample data belongs to the category of the ith node, then i = 1, and the loss is If the sample data does not belong to the category of the ith node, then i = 0, and the loss is 0. is the predicted probability of the model for the category of the ith node.

[0118] It should be noted that training the neural network model to minimize the prediction error and optimize the model parameters is a key step in achieving efficient classification performance. By using the training set for backpropagation training and adopting the cross-entropy loss function, the difference between the model prediction result and the actual label can be effectively measured, thereby guiding the learning process of the model. After training, it is crucial to use the test set to verify the classification accuracy of the model because it can evaluate the generalization ability of the model on unseen data. Calculating the precision and recall rate can not only provide a comprehensive evaluation of the model's classification performance but also reveal the ability balance of the model in dealing with positive and negative class samples. Therefore, the setting and calculation of the precision and recall rate provide an important basis for the practicality of the model, ensuring that the deformation state of the tunnel portal structure can be accurately identified in actual applications and corresponding maintenance measures can be taken in a timely manner.

[0119] Therefore, it is necessary to train the neural network model to minimize the prediction error, optimize the model parameters, and use the test set to verify the classification accuracy of the trained model. The method is as follows:

[0120] By using the training set and the test set to train and evaluate the neural network model, first use the training set to perform backpropagation training on the model to minimize the prediction error. The model uses the cross-entropy loss function to measure the difference between the prediction result and the actual label. After the training is completed, use the test set for verification and calculate the recall rate and precision rate of the model to evaluate its classification performance. The formulas for calculating the precision rate and recall rate are as follows:

[0121]

[0122] Among them, P and R represent the precision rate and recall rate of the model respectively, TP is the number of true positives, FP is the number of false positives, and FN is the number of false negatives.

[0123] It should be noted that it is crucial to monitor the structural health status of the tunnel entrance according to the deformation prediction results output by the model and in combination with the threshold alarm mechanism. This method optimizes the comprehensive index F1-score, comprehensively considering the precision rate and recall rate, to ensure the classification performance of the model in different situations. The set decision threshold is used to judge whether the predicted probability value exceeds the established limit, thereby affecting the prediction result of the model. When the predicted probability value exceeds the threshold, adjusting the threshold for the positive class helps to balance the precision rate and recall rate and improve the practicality of the model. In this process, cross-validation can ensure the consistent performance of the model on different data sets and enhance its reliability in practical applications. Therefore, through this mechanism, monitoring reports can be generated in a timely manner and corresponding structural health management measures can be taken, thereby effectively reducing the potential risks at the tunnel entrance and ensuring the project safety.

[0124] Therefore, it is necessary to optimize the precision rate and recall rate of the neural network model according to the deformation prediction results output by the model and in combination with the decision threshold warning mechanism. The method is as follows:

[0125] Optimize a comprehensive index F1-score to consider both the precision rate and recall rate at the same time. Set the comprehensive index as the decision threshold. If the predicted probability value output by the model exceeds the decision threshold, then adjust the threshold for the predicted probability value to be the positive class to change the precision rate and recall rate, and use cross-validation to ensure the consistent performance of the model on different data sets. The formula is as follows:

[0126]

[0127] Among them, F1 is a comprehensive index, and L is a logical value for judging whether the predicted probability value exceeds the decision threshold. When L = 0, it means that the predicted probability value does not exceed the decision threshold and no adjustment is required. When L = 1, it means that the predicted probability value exceeds the decision threshold, and it is necessary to adjust the threshold of the predicted probability value for the positive class to optimize the precision and recall rate.

[0128] Please refer to Figure 2 , the present invention also provides an intelligent tunnel portal structure deformation monitoring device based on deep learning. The deformation monitoring device is used to execute the above-mentioned intelligent tunnel portal structure deformation monitoring method based on deep learning, including:

[0129] A data acquisition and feature extraction module, which is used to collect physical data related to the tunnel portal structure deformation at equal time intervals, and extract feature data with time series distribution according to the obtained relevant physical data, including beam displacement, vibration frequency, pore water pressure, tunnel wall soil pressure, water accumulation depth, seismic acceleration, and wind speed at the representative points;

[0130] A data preprocessing and index calculation module, which is used to preprocess the calculated feature data, and then combine the beam displacement, vibration frequency, pore water pressure, and soil tunnel wall pressure at the representative points to calculate the soil deformation index, water and soil stability index, and dynamic deformation index, and combine the three indexes with the water accumulation depth, seismic acceleration, and wind speed to generate a feature vector of the tunnel portal structure deformation. The preprocessing includes data cleaning and normalization;

[0131] A neural network construction module, which is used to construct a neural network model, including an input layer, a hidden layer, and an output layer. The input layer receives the feature vector, the hidden layer is responsible for extracting the time series features in the feature vector, and the output layer is used to identify the deformation type of the tunnel portal structure;

[0132] A model training and evaluation module, which trains the neural network model based on the historical tunnel portal structure deformation data to minimize the prediction error, optimize the model parameters, and uses the test set to verify the classification accuracy of the trained model to obtain the deformation type of the tunnel portal structure;

[0133] A real-time monitoring and decision support module, which is used to improve the precision and recall rate of the neural network model according to the deformation prediction result output by the model and in combination with the decision threshold warning mechanism.

[0134] The above formulas are all dimensionless and take their numerical values for calculation. The formulas are obtained by collecting a large amount of data for software simulation to get a formula closest to the real situation. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0135] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. Those skilled in the art will appreciate that the units and algorithm steps of the examples described in connection with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed by hardware or software methods depends on the specific application and design constraints of the technical solution.

[0136] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units. They may be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the objectives of the solution of this embodiment.

[0137] As described above, the specific implementation manners of the present application are only described, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed in the present application, and all of them should be covered by the protection scope of the present application.

Claims

1. An intelligent tunnel entrance structure deformation monitoring method based on deep learning, characterized in that: The specific steps include: Step 1: Collect physical data related to the deformation of the tunnel entrance structure at equal time intervals, and extract characteristic data of time series distribution based on the acquired relevant physical data, including beam displacement at representative points, vibration frequency, pore water pressure, tunnel wall soil pressure, water accumulation depth, seismic acceleration and wind speed; Step 2: After preprocessing the calculated characteristic data, the soil deformation index, soil and water stability index and dynamic deformation index are calculated by combining the beam displacement, vibration frequency, pore water pressure and tunnel wall earth pressure at the representative point. The three indexes are combined with the water accumulation depth, earthquake acceleration and wind speed to generate the characteristic vector of the tunnel entrance structure deformation. The preprocessing includes data cleaning and normalization. Step 3: Construct a neural network model, including an input layer, a hidden layer, and an output layer. The input layer receives the feature vector, the hidden layer is responsible for extracting the temporal features in the feature vector, and the output layer is used to identify the deformation type of the tunnel entrance structure. Step 4: Train the neural network model based on the historical tunnel mouth structure deformation data to minimize the prediction error, optimize the model parameters, and use the test set to verify the classification accuracy of the trained model to obtain the deformation type of the tunnel mouth structure; Step 5: Based on the deformation prediction results output by the model, the decision threshold warning mechanism is combined to optimize the precision and recall of the neural network model.

2. According to claim 1, a method for intelligent tunnel entrance structure deformation monitoring based on deep learning is characterized in that: The calculated characteristic data is preprocessed, and the preprocessing includes data cleaning and normalization, and the method is as follows: Data cleaning of feature data includes detection and deletion of outliers and duplicate data, and processing of missing values. Specifically, it uses statistical methods to identify outliers and duplicate data in feature data, deletes outliers and duplicate data in feature data, and uses the mean, median or mode of feature data to fill in missing values ​​in feature data. The normalization of feature data is minimum-maximum normalization, which scales the data to the range of [0,1] so that all feature data have the same scale. The formula is: Among them, Y′ is the normalized feature data, Y is the original feature data, and Y min is the minimum value of the same type of feature data in the data set, Y max It is the maximum value of the same type of feature data in the dataset.

3. The intelligent tunnel entrance structure deformation monitoring method based on deep learning according to claim 1 is characterized in that: The beam displacement, vibration frequency, pore water pressure, and tunnel wall earth pressure at the representative points are combined to calculate the soil deformation index, soil and water stability index, and dynamic deformation index based on the following method: The beam displacement at the representative point in the selected characteristic data is combined with the tunnel wall soil pressure to generate the soil deformation index. The soil deformation index comprehensively considers the relationship between the beam displacement at the representative point and the tunnel wall soil pressure, and aims to reflect the deformation degree of the tunnel structure caused by the change of soil pressure. The formula for calculating the soil deformation index is: Among them, SDI k represents the soil deformation index at the tunnel entrance at time k, D k+1 is the beam displacement at the representative point at time k+1, D k is the beam displacement at the representative point at time k, P k+1 is the tunnel wall earth pressure at time k+1, P k is the tunnel wall soil pressure at time k, k is the index of the acquisition time; The pore water pressure in the selected characteristic data is combined with the tunnel wall soil pressure to generate the soil and water stability index. The soil and water stability index reflects the relationship between pore water pressure and soil pressure. It aims to evaluate the effective stress and stability of the soil, especially in the case of strength changes caused by water level changes. The formula for calculating the soil and water stability index is: Q totalk ′=P k -u k Among them, WSSI k represents the soil and water stability index at the tunnel entrance at time k, u is the pore water pressure at time k, P total ′ is the effective tunnel wall earth pressure at time k; The beam displacement at the representative point in the selected characteristic data is combined with the vibration frequency to generate a dynamic deformation index. The dynamic deformation index combines the relationship between the beam displacement at the representative point and the vibration frequency, and is used to evaluate the dynamic response of the structure under external loads or earthquakes. The formula for calculating the dynamic deformation index is: Among them, DDI k represents the dynamic deformation index of the tunnel entrance at time k, f k+1 is the vibration frequency at time k+1, f k is the vibration frequency at time k.

4. The intelligent tunnel entrance structure deformation monitoring method based on deep learning according to claim 1 is characterized in that: A neural network model is constructed, including an input layer, a hidden layer, and an output layer. The input layer receives the feature vector, the hidden layer is responsible for extracting the temporal features in the feature vector, and the output layer is used to identify the deformation type of the tunnel entrance structure. The method is as follows: The input layer is responsible for receiving feature vectors, including soil deformation index, soil and water stability index, dynamic deformation index, water accumulation depth, earthquake acceleration and wind speed, and setting the feature vector to (N, t, n), where N is the number of samples of the feature vector, t is the number of time steps, and n is the feature dimension of each time step; The hidden layer controls the flow of feature data by setting multiple gating mechanisms including forget gate, input gate and output gate to extract time series features from sequence data. The forget gate determines how much initial feature information to retain to reduce the impact of noise; the input gate controls the inflow of new information at the current time step and selectively introduces features that are helpful for future predictions by updating the state of the memory unit; The output gate determines the output of the hidden state based on the state of the current memory unit and passes it to the next time step and the final output layer; The output layer is responsible for mapping the features extracted by the hidden layer to specific categories and classifying the deformation types of the tunnel entrance structure based on the following method: According to the number of deformation types of the tunnel mouth structure contained in the extracted feature data, the number of nodes in the output layer is adjusted to the same number corresponding to the number of deformation types of the tunnel mouth structure, and the output is converted into a probability value using the Softmax activation function to represent the probability distribution of each category. The formula is: Among them, P represents the probability of each deformation type, y i is the jth deformation type in the i-th node in the output layer, X is the input feature data, output by the hidden layer, z i is the linear combination output of the neurons in the i-th node of the output layer, j is the index of the number of deformation types, j∈[1,N], N is the total number of deformation types, Represents the exponential value of the output of the j-th deformation type, which is the exponential operation of the output of the linear combination of neurons, ensuring that each value is a positive number.

5. The intelligent tunnel entrance structure deformation monitoring method based on deep learning according to claim 1 is characterized in that: The cross entropy function is used as the loss function to optimize the neural network model to evaluate the difference between the probability distribution output by the neural network model and the actual label. The formula is: in, represents the value of the loss function of the j-th deformation type in the ith node, that is, the predicted probability of the j-th deformation type in the ith node, reflecting the difference between the model prediction and the true label, o i Represents the unique hot encoding of the true label. If the sample data belongs to the i-node category, then o i =1, the loss is If the sample data does not belong to the i-th node category, then o i =0, the loss is 0, is the predicted probability of the model for the i-th node category.

6. The intelligent tunnel entrance structure deformation monitoring method based on deep learning according to claim 1 is characterized in that: The neural network model is trained to minimize the prediction error, optimize the model parameters, and the classification accuracy of the trained model is verified using the test set, according to the following method: The neural network model is trained and evaluated by using the training set and the test set. First, the model is trained by back propagation using the training set to minimize the prediction error. The model uses the cross entropy loss function to measure the difference between the prediction result and the actual label. After the training is completed, the test set is used for verification, and the recall and precision of the model are calculated to evaluate its classification performance. The formula for calculating precision and recall is: Among them, P and R represent the precision and recall of the model respectively, TP is the number of true positives, FP is the number of false positives, and FN is the number of false negatives.

7. The intelligent tunnel entrance structure deformation monitoring method based on deep learning according to claim 1 is characterized in that: According to the deformation prediction results output by the model, the decision threshold warning mechanism is combined to optimize the precision and recall rate of the neural network model. The method is based on: Optimize a comprehensive indicator F1-score to consider both precision and recall, set the comprehensive indicator as the decision threshold, and if the predicted probability value output by the model exceeds the decision threshold, adjust the predicted probability value to the threshold of the positive class to change the precision and recall, and use cross-validation to ensure the consistent performance of the model on different data sets. The formula is: Among them, F1 is a comprehensive indicator, L is a logical value for judging whether the predicted probability value exceeds the decision threshold. When L = 0, it means that the predicted probability value does not exceed the decision threshold and no adjustment is required; when L = 1, it means that the predicted probability value exceeds the decision threshold, and the predicted probability value is adjusted to the threshold of the positive class to optimize the precision and recall.

8. An intelligent tunnel entrance structure deformation monitoring device based on deep learning, characterized in that: The deformation monitoring device is used to execute the intelligent tunnel entrance structure deformation monitoring method based on deep learning according to any one of claims 1 to 7, comprising: A data acquisition and feature extraction module, which is used to collect physical data related to the deformation of the tunnel entrance structure at equal time intervals, and extract characteristic data of time series distribution according to the acquired relevant physical data, including beam displacement at representative points, vibration frequency, pore water pressure, tunnel wall soil pressure, water accumulation depth, seismic acceleration and wind speed; A data preprocessing and index calculation module, which is used to preprocess the calculated characteristic data, and then calculate the soil deformation index, water and soil stability index and dynamic deformation index by combining the beam displacement, vibration frequency, pore water pressure and soil tunnel wall pressure at the representative point, and combine the three indexes with the water accumulation depth, earthquake acceleration and wind speed to generate a characteristic vector of the tunnel mouth structure deformation. The preprocessing includes data cleaning and normalization; A neural network building module, which is used to build a neural network model, including an input layer, a hidden layer and an output layer, wherein the input layer receives a feature vector, the hidden layer is responsible for extracting the temporal features in the feature vector, and the output layer is used to identify the deformation type of the tunnel mouth structure; A model training and evaluation module, wherein the model training and evaluation module trains the neural network model based on historical tunnel mouth structure deformation data to minimize prediction errors, optimize model parameters, and uses a test set to verify the classification accuracy of the trained model to obtain the deformation type of the tunnel mouth structure; The real-time monitoring and decision support module is used to improve the precision and recall rate of the neural network model based on the deformation prediction results output by the model and the decision threshold early warning mechanism.

Images