A tunnel state monitoring method based on structural and apparent data

Abstract

The application discloses a tunnel state monitoring method based on structure and apparent data, and relates to the technical field of tunnel engineering health monitoring.The application significantly improves the accuracy of tunnel health assessment and the timeliness of early warning by integrating structure and apparent data.The method effectively solves the information isolation problem, enhances the comprehensiveness of the evaluation, and improves the forward-looking and reliability of the evaluation by deeply mining data features, constructing a comprehensive feature matrix, applying multi-modal data fusion technology, and combining a dynamic health assessment model with historical and real-time data to accurately predict tunnel state changes using time series analysis.In addition, the method realizes the precision of risk management and the data-driven maintenance decision by using a risk linkage network and optimized state prediction, thereby providing strong technical support for the safe operation and management of tunnels.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of tunnel engineering health monitoring, and particularly relates to a tunnel state monitoring method based on structural and apparent data. BACKGROUND

[0002] As an important field of infrastructure safety management, tunnel state monitoring is directly related to the safety and stability of traffic operation, and its research and application have irreplaceable value in protecting public safety and prolonging the service life of structures. However, the current methods still have significant deficiencies in practical application, mainly in the insufficient depth of data integration and analysis, which makes it difficult to fully reflect the true health status of the tunnel, resulting in limitations on the accuracy of the monitoring results and the timeliness of the early warning.

[0003] Under this background, tunnel state monitoring faces multiple challenges. First, due to the large differences in the sources and characteristics of structural data and apparent data, existing technologies often fail to effectively integrate the two types of data, resulting in information isolation and the inability to form a comprehensive judgment of the tunnel state. This dilemma of data isolation further leads to one-sidedness in feature extraction, making it difficult to associate the dynamic changes of structural data with the spatial distribution of apparent defects, and thus lacking comprehensiveness in health assessment, which easily overlooks the linkage effects of potential risks. This lack of comprehensiveness directly affects the accuracy of state prediction, especially when facing sudden trends in complex environments, lacking attention and prediction ability for key abnormal points. SUMMARY

[0004] The present application provides a tunnel state monitoring method based on structural and apparent data to solve the above-mentioned technical problems.

[0005] The technical solution of the present application is as follows:

[0006] A tunnel state monitoring method based on structural and apparent data, comprising the following steps:

[0007] Through a multi-source data acquisition system, structural data and apparent data within the tunnel are obtained, and the different characteristics of the two types of data are standardized to generate a unified initial data set. Then, a hierarchical feature extraction method is used to deeply mine the dynamic changes of structural data and the spatial distribution of apparent data, and output structural feature sets and apparent feature sets;

[0008] For the structural feature sets and apparent feature sets, a multi-modal data fusion technology is applied to correlate and map the two types of features, construct a comprehensive feature matrix, and determine a comprehensive representation of the tunnel state;

[0009] According to the comprehensive feature matrix, a dynamic health assessment model is constructed, combined with historical data and real-time input, to analyze the potential risk linkage effects of the tunnel state, and output the health assessment results;

[0010] According to the health assessment results, a time series analysis method is introduced to dynamically model the state change trend, generate a state prediction curve, capture possible mutation trends, and obtain prediction key points;

[0011] When a mutation trend is detected in the state prediction curve, an anomaly point detection algorithm is used to focus on the key abnormal attention area, extract abnormal feature details, and judge the risk level of the abnormal point;

[0012] According to the abnormal feature details combined with the complex environment adaptation parameters, the dynamic modeling capability is adjusted, the weight distribution of the prediction model is updated, and the optimized state prediction result is output;

[0013] Through the optimized state prediction result, a future trend report of the tunnel health state is generated, and combined with potential risk linkage information, a priority monitoring scheme for high-risk areas is determined.

[0014] Further, the structural feature set and the apparent feature set are output, specifically including:

[0015] The structural data and the apparent data are obtained from the tunnel environment through a multi-source data acquisition system, wherein the structural data is acquired in real time by deploying various sensors at key parts of the tunnel, and the apparent data is acquired by a mobile inspection robot, a track inspection vehicle, or a fixed high-definition camera system to collect images of the tunnel inner wall surface;

[0016] According to the different characteristics of the structural data and the apparent data, a preset classification rule is used to preliminarily sort the data, and a format conversion tool is used to adjust to a unified code to obtain a classified data set, and then a standardization processing method is used to normalize the data range and unit to form a standardized data set;

[0017] Based on the standardized data, the integrity and consistency of the data are detected, and when data is missing or there are abnormal values, interpolation is used to fill in the missing data, and a unified initial data set framework is constructed, and different sources of data are aligned in the field by means of data integration tools to generate an integrated data set;

[0018] According to the integrated data set, the logical consistency between the data is verified, and when there is a contradiction between the fields, the data is corrected according to the preset logical verification rule to obtain a consistent data set;

[0019] Support vector machine algorithm is used to preliminarily extract features from the consistent data to obtain a feature data set, and then a hierarchical feature extraction method is used to deeply mine the dynamic changes of the structural data and the spatial distribution of the apparent data, and output the structural feature set and the apparent feature set.

[0020] Further, a comprehensive tunnel state representation is determined, specifically including:

[0021] The time sequence feature set is obtained from the structural data by using a hierarchical extraction method, and the dynamic change of the structural data is characterized by a time sequence analysis algorithm, and the key features reflecting the change of the structural state are extracted, and the spatial feature set is extracted from the apparent data by using a pattern mining algorithm;

[0022] The image segmentation algorithm is used to extract the defect area of the tunnel surface, and the defect features are quantified, and the image enhancement processing is performed on the apparent data, and the image defect area is matched with the structural monitoring point;

[0023] The structural feature set and the apparent feature set are standardized, and when the dimension of the feature set exceeds the preset threshold, the principal component analysis algorithm is used for dimension reduction to obtain the reduced feature set, and the feature fusion model is constructed, and the structural features and the apparent features are integrated by using the weighted fusion method to obtain the fusion feature set;

[0024] Through the multi-modal data fusion technology, the multi-modal fusion model based on deep neural network is used to further correlate and map the fusion feature set, to identify the correlation between different types of data, and to comprehensively judge the health state of the tunnel;

[0025] By constructing a comprehensive feature matrix, a comprehensive tunnel state representation is determined, and the distribution characteristics of the fusion feature set are obtained, and when the distribution of the fusion feature set meets the preset normality condition, the Gaussian mixture model is used for feature optimization to obtain a comprehensive tunnel state representation.

[0026] Further, the potential risk linkage effect of the tunnel state is analyzed, and the health assessment result is output, which specifically includes:

[0027] Based on the comprehensive feature matrix, a fusion neural network model is constructed, which receives the fusion features of the structural data and the image data as input, and evaluates the health state of the tunnel by using a deep learning algorithm;

[0028] The fusion neural network model is combined with historical data and real-time input, the historical data includes the health state records of the tunnel at different time points, and the real-time input reflects the current monitoring data, according to the fused feature structure, the potential risk of the tunnel state is analyzed;

[0029] The support vector machine algorithm is used to classify and quantify the risk factors, to judge the risk level distribution, and when the risk level distribution exceeds the preset threshold, the linkage effect is further excavated, the correlation path between the potential risks is obtained, and the risk linkage network is obtained;

[0030] By constructing a risk linkage network, a dynamic assessment logic is built. Combined with the parameter adjustment of the health model, a health assessment result is generated to determine the current health status of the tunnel. Then, combined with the needs of subsequent prediction, key features in the assessment result are extracted to generate a predictive assessment result.

[0031] Furthermore, by adjusting the parameters of the health model, a health assessment result is generated to determine the current health status of the tunnel, specifically including:

[0032] The tunnel is divided into several logical segments, each of which outputs a health score independently. The overall health index of the tunnel is calculated. Based on the health assessment score of each segment and the preset health level standards, the health status of each segment is classified. Then, the health scores of each segment are combined using a weighted average method to calculate the overall health index of the tunnel.

[0033] Furthermore, the overall health index of the tunnel is calculated by combining the health score and corresponding weight of each logical segment using a weighted average method, and is expressed as follows: ,in No. The health score of each logical segment is calculated using a fusion neural network model based on the segment's structure and appearance data. No. The weight of each logical segment, Indicates the total number of logical segments into which the tunnel is divided; This indicates the overall health index of the tunnel.

[0034] Furthermore, the key points for prediction are obtained, specifically including:

[0035] Time series data of health assessment results are obtained, and combined with historical data of each detection area, a time series model is built for each area. Then, the ARIMA model is used to model the clean dataset in time series to generate a state change trend model and obtain fitting parameters.

[0036] Based on the fitted parameters, predict the future state change trend, generate a prediction curve, analyze the prediction curve using the sliding window method, extract mutation features, and identify mutation points.

[0037] When a mutation point exceeds a preset threshold, it is marked as a key prediction point, resulting in a key point set. Then, a clustering analysis method is used to group the state change patterns, determine the pattern classification results, and generate a dynamic update model of the state change trend based on the classification results to obtain the final prediction result.

[0038] Furthermore, determining the risk level of anomalies specifically includes:

[0039] When a mutation trend is detected in the state prediction curve, the Isolation Forest algorithm is used to analyze the key prediction points, locate the key abnormal area, obtain the abnormal point distribution, identify the abnormal area by detecting outliers in the data, extract the abnormal feature change amplitude, and generate the abnormal feature set;

[0040] The abnormal feature set is analyzed by the risk assessment model to determine the risk level of the abnormal points, and the frequency of abnormal occurrence is counted to generate frequency distribution data. The frequency distribution data reflects the frequency of abnormal points in different time periods, which helps to identify potential high-risk areas;

[0041] According to the frequency distribution data, the sliding window method is used to process the data stream. By sliding a fixed size window on the time series, the data in the window is analyzed and processed, and the state prediction curve is updated in real time. Through the updated state prediction curve, the abnormal point detection algorithm is iteratively optimized.

[0042] Further, the optimized state prediction result is output, specifically including:

[0043] Abnormal feature details related to the health status of the tunnel are extracted from the monitoring data. Environmental sensors are used to obtain dynamic parameters of the environment in which the tunnel is located, which reflect the complexity and dynamic changes of the environment in which the tunnel is located.

[0044] The abnormal feature details are combined with the complex environment adaptation parameters, and a machine learning algorithm is used to analyze the correlation between the two. According to the results of the correlation analysis, a weight adjustment strategy is developed;

[0045] According to the weight adjustment strategy, the modeling ability of the prediction model is dynamically adjusted. According to the results of the correlation analysis, the weight distribution of the prediction model is updated, and the model is retrained to adapt to the new weight distribution. The updated prediction model is used to generate an optimized state prediction result.

[0046] Further, a priority monitoring scheme for high-risk areas is determined, specifically including:

[0047] The optimized state prediction result is obtained, and a support vector machine algorithm is used for state prediction to obtain the tunnel health change trend. Time series features are extracted from the tunnel health change trend, and a random forest algorithm is used to analyze the trend stability to obtain a health status score;

[0048] When the health status score is lower than the preset threshold, risk linkage analysis is performed in combination with historical failure data to obtain a risk probability distribution. The K-means clustering algorithm is used to divide the high-risk area based on the risk probability distribution to obtain a high-risk area list;

[0049] According to the high-risk area list, the real-time monitoring data frequency and the abnormal point distribution are analyzed, the priority monitoring area is determined, the sensor data is extracted from the priority monitoring area, the high-risk area dynamic monitoring scheme is generated, the monitoring resource allocation strategy is obtained, the sensor data acquisition frequency is adjusted, and the optimized monitoring execution plan is obtained.

[0050] Compared with the prior art, the present application has the following beneficial effects:

[0051] The tunnel internal structure and apparent data obtained by the multi-source data acquisition system are processed by standardization and hierarchical feature extraction method, the data features are deeply mined, the problem of separation between structure and apparent data is effectively solved, and the collaborative analysis of data is realized; this fusion not only improves the usability of data, but also realizes the effective integration of information by constructing a comprehensive feature matrix and applying multi-modal data fusion technology, which provides more detailed and reliable basis for comprehensive evaluation of tunnel state, thereby significantly improving the comprehensiveness and accuracy of health evaluation;

[0052] By constructing a dynamic health evaluation model, combining historical and real-time data, and analyzing the potential risk linkage effect of the tunnel state, the tunnel health state can be more accurately evaluated; using time series analysis method to dynamically model the state change trend, capturing the mutation trend, obtaining the prediction key point, which not only improves the accuracy of state prediction, but also enhances the prediction ability of mutation trend in complex environment, making the health evaluation result more forward-looking and reliable;

[0053] By risk linkage network and dynamic evaluation logic, combined with health model parameter adjustment, the health evaluation result is generated, the current health status of the tunnel state is determined, and scientific basis is provided for subsequent prediction; the optimized state prediction result makes the future trend report of the tunnel health state more accurate, combined with the potential risk linkage information, the priority monitoring scheme of high-risk area is determined, this data-driven method provides more scientific decision support for the maintenance and management of the tunnel, not only effectively prolongs the structure life, but also significantly improves the public safety guarantee level. BRIEF DESCRIPTION OF DRAWINGS

[0054] Fig. 1 A flowchart of a tunnel state monitoring method based on structure and apparent data provided by the present application is provided;

[0055] Fig. 2 A flowchart of a tunnel state monitoring method based on structure and apparent data provided by the present application is provided;

[0056] Fig. 3 A flowchart of a tunnel state monitoring method based on structure and apparent data provided by the present application is provided. DETAILED DESCRIPTION

[0057] In order to make the objectives, characteristics and advantages of the present application more obvious and easy to understand, the technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application. Obviously, the embodiments described below are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present application.

[0058] As shown in Figs. 1-3 The present embodiment provides a tunnel state monitoring method based on structural and apparent data, comprising the following steps:

[0059] S1, acquiring structural data and apparent data in the tunnel through a multi-source data acquisition system, standardizing the two types of data according to their different characteristics, generating a unified initial data set, so as to obtain consistency basis for subsequent fusion operation, and then using a hierarchical feature extraction method to respectively perform deep mining on the dynamic changes of the structural data and the spatial distribution of the apparent data, and output structural feature sets and apparent feature sets, providing detailed basis for fusion;

[0060] Further, the output structural feature set and the apparent feature set specifically include:

[0061] The structural data and the apparent data are acquired from the tunnel environment by a multi-source data acquisition system, wherein the structural data is acquired in real time by arranging various sensors (such as strain gauges, crack gauges, displacement gauges, temperature and humidity sensors, accelerometers, etc.) at key parts of the tunnel (such as arch crown, arch waist, arch foot, side wall, etc.), including stress and strain, crack expansion value, displacement change, etc. Parameters; the apparent data is acquired by mobile inspection robots, track inspection cars or fixed high-definition camera systems to collect images of the tunnel inner wall surface, covering crack, water seepage, spalling, bulging, falling off, etc. Defect images;

[0062] According to the different characteristics of the structural data and the apparent data, a preset classification rule is used to preliminarily sort the data, and a format conversion tool is used to adjust to a unified code to obtain a classified data set, and then a standardization processing method is used to normalize the data range and unit, forming a standardized data set;

[0063] Based on the standardized data, the integrity and consistency of the data are detected, and once the data is missing or there is an abnormal value, the interpolation method is used for filling, a unified initial data set framework is constructed, and different sources of data are aligned in the field by means of data integration tools to generate an integrated data set;

[0064] According to the integrated data set, the logical consistency between the data is verified, and when contradictions are found between the fields, the preset logical checking rules are modified to obtain a consistent data set;

[0065] Support vector machine algorithm is used for preliminary feature extraction of consistent data to obtain a characteristic data set. Then, a hierarchical feature extraction method is used for deep mining of the dynamic changes of structural data and the spatial distribution of apparent data, respectively, to output structural feature sets and apparent feature sets, providing detailed basis for subsequent data fusion.

[0066] For example, in the health monitoring process of a subway tunnel, three typical detection sections (A section, B section and C section) in the tunnel are selected for health evaluation. Through a multi-source data acquisition system, structural data and apparent data in the tunnel are obtained. In terms of structural data acquisition, strain gauges, acceleration sensors, displacement meters, temperature and humidity sensors, etc. are arranged in A section, B section and C section to collect key parameters such as tunnel vault displacement, crack width, stress and strain distribution, vibration data, etc. in real time. The data acquisition frequency is once every 10 minutes, and the data is transmitted to the central processing system in real time after each data collection. In terms of apparent image data acquisition, high-definition cameras are installed in the same area, and track inspection cars are used for periodic inspection. The cameras shoot high-definition images of the tunnel walls, and the YOLO algorithm is used to detect apparent defects such as cracks, water seepage and peeling on the images. After each inspection, the detected defect information (such as crack length, width and depth) and the corresponding image data are uploaded to the cloud platform. The collected structural data and image data are preprocessed, standardized according to the different characteristics of the two types of data, and a unified initial data set is generated to facilitate subsequent fusion operations to obtain consistency. Then, a hierarchical feature extraction method is used for deep mining of the dynamic changes of structural data and the spatial distribution of apparent data, respectively, to output structural feature sets and apparent feature sets, providing detailed basis for fusion.

[0067] Specifically, through multi-source data acquisition and standardization processing, deep integration of tunnel structure and apparent data is realized, and a unified and high-quality initial data set is generated. This process not only enhances the consistency and comparability of the data, but also provides more abundant and detailed data features for tunnel health evaluation through hierarchical feature extraction, significantly improving the accuracy and reliability of health monitoring, and laying a solid foundation for subsequent data analysis and state prediction.

[0068] S2, for the structural feature set and the apparent feature set, a multi-modal data fusion technology is applied to correlate and map the two types of features, construct a comprehensive feature matrix, solve the information isolation problem, and determine a comprehensive tunnel state representation;

[0069] Further, determining a comprehensive tunnel state representation specifically includes:

[0070] S21, acquire the time series feature set from the structural data using a hierarchical extraction method, and based on a time series analysis algorithm, perform feature decomposition on the dynamic changes of the structural data to extract key features reflecting the changes in the structural state, and then extract the spatial feature set from the apparent data through a pattern mining algorithm;

[0071] S22, use an image segmentation algorithm to extract the tunnel surface defect area, and quantify the defect features such as the width, length, and depth of the cracks, and perform image enhancement processing on the apparent data to improve the image quality and make the cracks and water seepage defects more obvious, and match the image defect area with the structural monitoring points;

[0072] S23, standardize the structural feature set and the apparent feature set, and when the dimension of the feature set exceeds a preset threshold, use a principal component analysis algorithm for dimension reduction to obtain a reduced dimension feature set, and by constructing a feature fusion model, the structural features and apparent features are integrated through a weighted fusion method to obtain a fused feature set;

[0073] S24, through multi-modal data fusion technology, use a multi-modal fusion model based on deep neural network (such as Transformer fusion module, feature splicing + attention mechanism, etc.) to further correlate and map the fused feature set, and the model can identify the correlation between different types of data and comprehensively judge the health status of the tunnel;

[0074] S25, by constructing a comprehensive feature matrix, the information isolation problem is solved, a comprehensive tunnel state representation is determined, and the distribution characteristics of the fused feature set are obtained, and when the distribution of the fused feature set meets the preset normality condition, a Gaussian mixture model is used for feature optimization to obtain a comprehensive tunnel state representation.

[0075] Wherein, when constructing the comprehensive feature matrix, the structural data is extracted hierarchically, the time series variation characteristics of the structural parameters are captured through the time series analysis algorithm, and the apparent data is processed through image segmentation and enhancement to quantify the geometric features of cracks, water leakage and other defects; then the processed structural features and apparent features are standardized and aligned, and the principal component analysis (PCA) is used for dimension reduction to eliminate redundancy; finally, a weighted fusion or multi-modal deep fusion model based on Transformer is used to construct a comprehensive matrix with monitoring time as row index and multi-dimensional feature vector as column, realizing the cross-modal correlation mapping of structural safety and apparent disease.

[0076] Specifically, the structural and appearance feature sets are effectively integrated through the multi-modal data fusion technology to create a comprehensive feature matrix, thereby solving the information isolation problem and determining a comprehensive tunnel state representation. This process not only improves the data utilization efficiency and the accuracy of health assessment, but also strengthens the correlation between features through a deep learning model, making the tunnel health state monitoring more accurate and reliable. In addition, through feature optimization and the application of a Gaussian mixture model, the quality of the feature set and the comprehensiveness of the tunnel state assessment are further improved, providing a more scientific basis for the maintenance and management of the tunnel.

[0077] S3, constructing a dynamic health assessment model based on the comprehensive feature matrix, combining historical data and real-time input, analyzing the potential risk linkage effect of the tunnel state, and outputting the health assessment result for subsequent prediction;

[0078] Further, analyzing the potential risk linkage effect of the tunnel state and outputting the health assessment result specifically includes:

[0079] S31, based on the comprehensive feature matrix, constructing a fusion neural network model, receiving the fusion features of structural data and image data as input, and evaluating the health state of the tunnel through a deep learning algorithm;

[0080] Wherein, the design of the model includes multiple neural network layers, which can handle complex nonlinear relationships to accurately output a health assessment score. The score range is [0, 1], or divided into different health levels (such as I-V) according to the preset grading standard;

[0081] S32, combining the fusion neural network model with historical data and real-time input, the historical data including the health state records of the tunnel at different time points, and the real-time input reflecting the current monitoring data, analyzing the potential risks of the tunnel state according to the fused feature structure;

[0082] S33, using a support vector machine algorithm to classify and quantify risk factors, determining the risk level distribution, and when the risk level distribution exceeds the preset threshold, further exploring the linkage effect to obtain the correlation path between potential risks, and obtaining a risk linkage network;

[0083] S34, constructing a dynamic assessment logic through the risk linkage network, combining parameter adjustment of the health model, generating a health assessment result, determining the current health status of the tunnel state, and extracting key features from the assessment result to generate a prediction assessment result in combination with the needs of subsequent prediction.

[0084] Further, combining parameter adjustment of the health model to generate a health assessment result and determine the current health status of the tunnel state specifically includes:

[0085] The tunnel is divided into several logical segments, each segment outputs a health score independently, and the overall health index of the whole tunnel is calculated. According to the health evaluation score of each section, combined with the preset health level standard, the health status of each section is classified (such as I level-good state, II level-mild deterioration, III level-moderate disease, etc.), and then the health scores of each section are integrated by weighted average method to calculate the overall health index of the whole tunnel, so as to comprehensively reflect the overall health status of the tunnel. This method not only can provide detailed health status of each section, but also can provide scientific basis for the overall maintenance and management of the tunnel through the comprehensive health index.

[0086] The overall health index of the whole tunnel is calculated by combining the health score of each logical segment and the corresponding weight by using the weighted average method, which is represented as: , wherein The health score of the first logical segment is calculated according to the structure and appearance data of the segment by using the fusion neural network model, and the score range is [0, 1] or divided into levels according to the grading standard. The weight of the first logical segment can be set according to the length, traffic flow, historical disease record and other factors of each segment. N represents the total number of logical segments divided by the tunnel. The overall health index of the whole tunnel is represented by, and the range is also [0, 1], and the higher the value, the better the overall health status of the tunnel. The dynamic evaluation logic refers to dynamically adjusting the parameters of the health evaluation model according to the risk linkage network and real-time data to reflect the changes of the tunnel state in real time, and generate the evaluation results of the current health status, while extracting key features for predicting future state.

[0087] According to the comprehensive feature matrix, a dynamic health evaluation model is constructed, combined with historical data and real-time input, the potential risk linkage effect of the tunnel state is analyzed, the structure-image fusion features are input into the fusion neural network model, and the health evaluation score is output, the score range is [0, 1] or divided into levels according to the grading standard (such as I level-V level), at the same time, the risk trend prediction model is supported, the future deterioration trend is predicted, and the health evaluation result is output, which is used for subsequent prediction link. Based on the fused data, the system outputs the health score of each section (such as A section score is 0.85, B section score is 0.72, C section score is 0.60). According to the preset health level standard, A section is "good state" (I level), B section is "mild deterioration" (II level), and C section is "moderate disease" (III level), at the same time, the risk trend prediction model is supported, the future deterioration trend is predicted, and the health evaluation result is output, which is used for subsequent prediction link.

[0088] According to the comprehensive feature matrix, a dynamic health evaluation model is constructed, combined with historical data and real-time input, the potential risk linkage effect of the tunnel state is analyzed, the structure-image fusion features are input into the fusion neural network model, and the health evaluation score is output, the score range is [0, 1] or divided into levels according to the grading standard (such as I level-V level), at the same time, the risk trend prediction model is supported, the future deterioration trend is predicted, and the health evaluation result is output, which is used for subsequent prediction link. Based on the fused data, the system outputs the health score of each section (such as A section score is 0.85, B section score is 0.72, C section score is 0.60). According to the preset health level standard, A section is "good state" (I level), B section is "mild deterioration" (II level), and C section is "moderate disease" (III level), at the same time, the risk trend prediction model is supported, the future deterioration trend is predicted, and the health evaluation result is output, which is used for subsequent prediction link. ​

[0089] Specifically, the dynamic health assessment model constructed by the comprehensive feature matrix can accurately analyze the potential risks of the tunnel state and output reliable health assessment results. This model combines historical and real-time data and uses deep learning algorithms to handle complex nonlinear relationships. It not only provides detailed health scores for each tunnel section but also calculates a comprehensive health index for the entire tunnel. This index comprehensively reflects the health status of the tunnel, providing a scientific basis for tunnel maintenance and management, and significantly improving the accuracy of health assessment and the ability to predict future trends.

[0090] S4、According to the health assessment results, introduce time series analysis method, dynamically model the state change trend, generate state prediction curve, capture possible mutation trend, get prediction key point;

[0091] Further, obtaining the prediction key point specifically includes:

[0092] Obtain time series data of health assessment results, combine historical data of each detection area, establish time series model for each area separately, then use ARIMA model to perform time series modeling on clean data set, generate state change trend model, obtain fitting parameters;

[0093] According to the fitting parameters, predict the future state change trend, generate the prediction curve, analyze the prediction curve by sliding window method, extract mutation features, and determine the mutation point;

[0094] When the mutation point exceeds the preset threshold, it is marked as a key prediction point, and a key point set is obtained. Then, a clustering analysis method is used to group the state change patterns, determine the pattern classification result, and generate a dynamic update model of the state change trend according to the classification result to obtain the final prediction result, providing forward-looking suggestions for tunnel maintenance and repair.

[0095] Specifically, the time series analysis method is used to dynamically model the tunnel health assessment results, and the generated state prediction curve can capture the potential mutation trend of the tunnel state, thereby identifying the key prediction point. This method enables the tunnel maintenance and repair work to obtain forward-looking suggestions, significantly improving the prediction accuracy of the future deterioration trend of the tunnel, and helping to take preventive maintenance measures in a timely manner to reduce safety risks and maintenance costs.

[0096] S5、When a mutation trend is detected in the state prediction curve, an anomaly point detection algorithm is used to focus on key abnormal attention areas, extract abnormal feature details, and determine the risk level of the anomaly point;

[0097] Further, determining the risk level of the anomaly point specifically includes:

[0098] When a mutation trend is detected in the state prediction curve, an isolation forest algorithm is used to analyze the key prediction points, locate the key abnormal area, obtain the abnormal point distribution, effectively identify the abnormal area by detecting outliers in the data, extract the abnormal feature change amplitude, generate the abnormal feature set, and the abnormal feature change amplitude includes but is not limited to the change of crack width, the increase of water seepage area, the mutation of displacement, etc.

[0099] Among them, the isolation forest algorithm can effectively identify the abnormal area by detecting outliers in the data, analyze each prediction point, calculate its abnormal score, and the point with higher abnormal score is considered as a potential abnormal point. Then, all points with abnormal score higher than the preset threshold are marked as abnormal points to form the abnormal point distribution.

[0100] The abnormal feature set is analyzed by the risk assessment model to determine the risk level of the abnormal point, and the frequency of abnormal occurrence is counted to generate frequency distribution data. The frequency distribution data reflects the occurrence frequency of the abnormal point in different time periods, which helps to identify potential high-risk areas.

[0101] Among them, the extracted abnormal feature set is input into the risk assessment model, and the risk assessment model calculates the risk level of each abnormal point according to the feature change amplitude in the abnormal feature set. The risk level is divided into low, medium and high levels.

[0102] According to the frequency distribution data, the sliding window method is used to process the data stream. By sliding a fixed size window on the time series, the data in the window is analyzed and processed, which can update the state prediction curve in real time to ensure the timeliness and accuracy of the prediction. Through the updated state prediction curve, the abnormal point detection algorithm is iteratively optimized. The optimized algorithm can more accurately identify new mutation trends, improve the prediction ability and reliability of the system.

[0103] Suppose in the health monitoring of a certain tunnel, a mutation trend is detected in the state prediction curve. Through the analysis of the isolation forest algorithm, it is found that the abnormal score of a certain area is higher, and the key abnormal area is located. The crack width, water seepage area and displacement of the area are extracted, the change amplitude is calculated, and the abnormal feature set is generated. Through the analysis of the risk assessment model, it is determined that the risk level of the abnormal point is "high". The frequency distribution data is generated by counting the occurrence frequency of the abnormal point in different time periods. It is found that the abnormal frequency of the area is higher in the past month. The sliding window method is used to process the data stream, the state prediction curve is updated, and the timeliness and accuracy of the prediction result are ensured. Finally, through the updated state prediction curve, the abnormal point detection algorithm is iteratively optimized, and the prediction ability and reliability of the system are improved.

[0104] Specifically, when a sudden trend appears in the state prediction curve, the key abnormal area can be quickly identified by the anomaly point detection algorithm, and detailed abnormal features such as crack width, water seepage area, and displacement are extracted to judge the risk level of the abnormal point. This process not only improves the identification ability of potential health risks of the tunnel, but also significantly enhances the prediction accuracy and reliability of the system by updating the state prediction curve and iteratively optimizing the anomaly point detection algorithm in real time, providing a scientific basis for timely maintenance and risk prevention of the tunnel.

[0105] S6、According to the abnormal feature details combined with the complex environment adaptation parameters, adjust the dynamic modeling ability, update the weight distribution of the prediction model, and output the optimized state prediction result;

[0106] Further, the output of the optimized state prediction result specifically includes:

[0107] From the monitoring data, extract the abnormal feature details related to the health status of the tunnel, including the change of crack width, the increase of water seepage area, and the mutation of displacement, etc. At the same time, use environmental sensors to obtain dynamic parameters of the environment where the tunnel is located, such as temperature, humidity, and underground water level change, etc. These parameters are collectively referred to as complex environment adaptation parameters, which are used to reflect the complexity and dynamic change of the environment where the tunnel is located. These data provide basic input for subsequent model adjustment;

[0108] Combine the abnormal feature details with the complex environment adaptation parameters, and use machine learning algorithms (such as Pearson correlation coefficient analysis, random forest, etc.) to analyze the correlation between the two, for example, humidity change may have a strong correlation with crack width increase. According to the results of correlation analysis, develop a weight adjustment strategy. For environmental parameters that have a greater impact on abnormal features, increase their weight in the model accordingly, and vice versa, thereby providing clear guidance for dynamic adjustment of the model;

[0109] According to the weight adjustment strategy, dynamically adjust the modeling ability of the prediction model. This may involve adjusting the hyperparameters of the model, changing the model structure, or introducing new features, etc. For example, increasing the weight of the humidity parameter may make the model pay more attention to the impact of humidity change on the health status of the tunnel. According to the results of correlation analysis, update the weight distribution of the prediction model, and retrain the model to make it adapt to the new weight distribution, thereby improving the accuracy and adaptability of the model in predicting the health status of the tunnel;

[0110] The updated prediction model is used to generate optimized state prediction results, which comprehensively consider the influence of abnormal feature details and complex environment adaptation parameters, and thus are significantly improved in accuracy and reliability. The optimized state prediction results are outputted to provide scientific basis for the health management and maintenance of the tunnel, help the relevant management personnel better understand the current health status and future possible development trend of the tunnel, and thus timely take effective maintenance measures to ensure the safe operation of the tunnel.

[0111] Specifically, by combining abnormal feature details and complex environment adaptation parameters, the weight distribution of the prediction model is dynamically adjusted, and thus the optimized state prediction results are outputted. This process analyzes the correlation between abnormal features and environment parameters by using a machine learning algorithm, and updates the model weight accordingly, so that the model can more accurately reflect the actual health status of the tunnel. The optimized model not only improves the accuracy of tunnel health state prediction, but also enhances the adaptability of the model to environmental changes, providing more scientific and reliable data support for the safety management and maintenance decision of the tunnel.

[0112] S7, generate a future trend report of the tunnel health status by the optimized state prediction results, and determine a priority monitoring scheme for the high-risk area in combination with potential risk linkage information.

[0113] Further, the priority monitoring scheme for the high-risk area includes:

[0114] Obtain the optimized state prediction results, perform state prediction by using a support vector machine algorithm to obtain the tunnel health change trend, extract time series features from the tunnel health change trend, analyze the trend stability by using a random forest algorithm to obtain a health status score.

[0115] When the health status score is lower than a preset threshold, perform risk linkage analysis in combination with historical failure data to obtain a risk probability distribution, divide the high-risk area by using a K-means clustering algorithm based on the risk probability distribution to obtain a high-risk area list.

[0116] According to the high-risk area list, analyze the real-time monitoring data frequency and abnormal point distribution to determine a priority monitoring area, extract sensor data from the priority monitoring area, generate a high-risk area dynamic monitoring scheme, obtain a monitoring resource allocation strategy, adjust the sensor data collection frequency, and obtain an optimized monitoring execution plan.

[0117] The future trend report of the tunnel health state is generated through the optimized state prediction result, the high-risk area is determined according to the potential risk linkage information, the priority monitoring scheme is determined, the health score result is displayed in a graphical form in the BIM / GIS system or the digital operation and maintenance platform, the high-risk area is highlighted by using color gradient, risk atlas, heat map and the like, and the maintenance suggestion is automatically generated according to the score level, the report can be exported and connected to the operation and maintenance platform or the maintenance plan system.

[0118] The future trend report of the tunnel health state is generated through the optimized state prediction result, the high-risk area is determined according to the potential risk linkage information, the priority monitoring scheme is determined, the health score result is displayed in a graphical form in the BIM / GIS system or the digital operation and maintenance platform, the high-risk area is highlighted by using color gradient, risk atlas, heat map and the like, and the maintenance suggestion is automatically generated according to the score level, the report can be exported and connected to the operation and maintenance platform or the maintenance plan system.

[0119] Specifically, the future trend report of the tunnel health state is generated through the optimized state prediction result, and the high-risk area needing priority monitoring is effectively determined by combining the potential risk information, which not only improves the prediction accuracy of the change trend of the tunnel health state, but also highlights the risk area by using the BIM / GIS system in a graphical form, helps the engineering personnel to quickly identify and respond to the high-risk area, and automatically generates the maintenance suggestion and the optimized monitoring execution plan, thereby ensuring the safety and operation reliability of the tunnel and providing scientific decision support for the maintenance and management of the tunnel.

[0120] The specific embodiments of the application are described in detail above, but they are only examples, and the application is not limited to the above-described specific embodiments. Those skilled in the art should understand that the above examples and descriptions in the specification are only to illustrate the principles of the application, and various changes and improvements can be made to the application without departing from the spirit and scope of the application. These changes and improvements all fall within the scope of the application claimed. The scope of protection of the application is defined by the appended claims and their equivalents.

Claims

1. A tunnel condition monitoring method based on structural and appearance data, characterized in that: Includes the following steps: The system acquires structural and apparent data within the tunnel through a multi-source data acquisition system. It standardizes the different characteristics of the two types of data to generate a unified initial dataset. Then, it uses a hierarchical feature extraction method to deeply mine the dynamic changes of structural data and the spatial distribution of apparent data, and outputs structural feature sets and apparent feature sets. For the structural feature set and the appearance feature set, multimodal data fusion technology is applied to correlate and map the two types of features, construct a comprehensive feature matrix, and determine a comprehensive tunnel state characterization; Based on the comprehensive feature matrix, a dynamic health assessment model is constructed. Combining historical data and real-time input, the potential risk linkage effect of tunnel status is analyzed, and the health assessment results are output. Based on the health assessment results, time series analysis is introduced to dynamically model the trend of state change, generate state prediction curves, capture sudden change trends, and obtain key prediction points. When a sudden change trend is detected in the state prediction curve, the anomaly detection algorithm is used to focus on key anomaly areas of interest, extract anomaly feature details, and determine the risk level of the anomaly. Based on the details of abnormal features and the parameters for adapting to complex environments, the dynamic modeling capability is adjusted, the weight distribution of the prediction model is updated, and the optimized state prediction results are output. Based on the optimized state prediction results, a future trend report on the tunnel's health status is generated. Combined with potential risk linkage information, priority monitoring schemes for high-risk areas are determined.

2. The tunnel condition monitoring method based on structural and appearance data according to claim 1, characterized in that: The output structural feature set and appearance feature set specifically include: Structural and apparent data are acquired from the tunnel environment through a multi-source data acquisition system. Structural data is acquired in real time by deploying various sensors at key parts of the tunnel, while apparent data is acquired by capturing images of the tunnel inner wall surface through mobile inspection robots, track inspection vehicles, or fixed high-definition camera systems. Based on the different characteristics of structural data and apparent data, the data is initially sorted using preset classification rules, and then converted to a unified code using a format conversion tool to obtain a classified dataset. Then, a standardization processing method is used to normalize the data range and units to form a standardized dataset. Based on standardized data, the integrity and consistency of the data are checked. When data is missing or outliers are found, they are filled in by interpolation. By building a unified initial dataset framework and using data integration tools to align fields of data from different sources, an integrated dataset is generated. Based on the integrated dataset, the logical consistency between the data is verified. When contradictions are found between fields, they are corrected according to the preset logical verification rules to obtain a consistent dataset. The support vector machine algorithm is used to perform preliminary feature extraction on the consistency data to obtain a featured dataset. Then, a hierarchical feature extraction method is used to deeply mine the dynamic changes of structural data and the spatial distribution of appearance data, respectively, and output structural feature sets and appearance feature sets.

3. The tunnel condition monitoring method based on structural and appearance data according to claim 1, characterized in that: The determination of a comprehensive tunnel condition characterization specifically includes: A hierarchical extraction method is adopted to obtain the temporal feature set from the structural data. Based on the time series analysis algorithm, the dynamic changes of the structural data are decomposed to extract the key features reflecting the changes in the structural state. Then, the spatial feature set is extracted from the apparent data through the pattern mining algorithm. Image segmentation algorithms are used to extract the defect areas on the tunnel surface and quantify the defect features. Image enhancement processing is performed on the appearance data, and the image defect areas are matched with structural monitoring points. The structural feature set and the appearance feature set are standardized. When the dimension of the feature set exceeds a preset threshold, the principal component analysis algorithm is used to reduce the dimension to obtain the dimension-reduced feature set. By constructing a feature fusion model, the structural features and appearance features are integrated through a weighted fusion method to obtain the fused feature set. By using multimodal data fusion technology, a multimodal fusion model based on deep neural networks is used to further correlate and map the fused feature set, identify the correlation between different types of data, and comprehensively judge the health status of the tunnel. By constructing a comprehensive feature matrix, a comprehensive tunnel state representation is determined. Then, the distribution characteristics of the fused feature set are obtained. When the distribution of the fused feature set satisfies the preset normality condition, a Gaussian mixture model is used for feature optimization to obtain a comprehensive tunnel state representation.

4. The tunnel condition monitoring method based on structural and appearance data according to claim 1, characterized in that: The analysis of the potential risk linkage effects of the tunnel status outputs health assessment results, specifically including: Based on the comprehensive feature matrix, a fusion neural network model is constructed. By receiving the fusion features of structural data and image data as input, the health status of the tunnel is evaluated through a deep learning algorithm. By combining a fusion neural network model with historical data and real-time input, the historical data includes records of the tunnel's health status at different points in time, while the real-time input reflects current monitoring data, and the potential risks to the tunnel's status are analyzed based on the fusion feature structure. The support vector machine algorithm is used to classify and quantify risk factors, determine the risk level distribution, and when the risk level distribution exceeds the preset threshold, the linkage effect is further explored to obtain the correlation path between potential risks and obtain the risk linkage network. By constructing a risk linkage network, a dynamic assessment logic is built. Combined with the parameter adjustment of the health model, a health assessment result is generated to determine the current health status of the tunnel. Then, combined with the needs of subsequent prediction, key features in the assessment result are extracted to generate a predictive assessment result.

5. The tunnel condition monitoring method based on structural and appearance data according to claim 4, characterized in that: The parameter adjustment combined with the health model generates a health assessment result, determining the current health status of the tunnel, specifically including: The tunnel is divided into several logical segments, each of which outputs a health score independently. The overall health index of the tunnel is calculated. Based on the health assessment score of each segment and the preset health level standards, the health status of each segment is classified. Then, the health scores of each segment are combined using a weighted average method to calculate the overall health index of the tunnel.

6. The tunnel condition monitoring method based on structural and appearance data according to claim 5, characterized in that: The overall health index of the tunnel is calculated by combining the health score and corresponding weight of each logical segment using a weighted average method, specifically as follows: ,in Indicates the first The health score of each logical segment is calculated using a fusion neural network model based on the segment's structure and appearance data. Indicates the first The weight of each logical segment, Indicates the total number of logical segments into which the tunnel is divided; This indicates the overall health index of the tunnel.

7. The tunnel condition monitoring method based on structural and appearance data according to claim 1, characterized in that: The obtained predicted key points specifically include: Time series data of health assessment results are obtained, and combined with historical data of each detection area, a time series model is built for each area. Then, the ARIMA model is used to model the clean dataset in time series to generate a state change trend model and obtain fitting parameters. Based on the fitted parameters, predict the future state change trend, generate a prediction curve, analyze the prediction curve using the sliding window method, extract mutation features, and identify mutation points. When a mutation point exceeds a preset threshold, it is marked as a key prediction point, resulting in a key point set. Then, a clustering analysis method is used to group the state change patterns, determine the pattern classification results, and generate a dynamic update model of the state change trend based on the classification results to obtain the final prediction result.

8. The tunnel condition monitoring method based on structural and appearance data according to claim 1, characterized in that: The determination of the risk level of anomalies specifically includes: When a sudden change trend is detected in the state prediction curve, the isolated forest algorithm is used to analyze the key prediction points, locate the key abnormal areas, obtain the distribution of abnormal points, identify abnormal areas by detecting outliers in the data, extract the magnitude of abnormal feature changes, and generate an abnormal feature set. By analyzing the abnormal feature set through a risk assessment model, the risk level of the abnormal points is determined, and the frequency of abnormal occurrence is statistically analyzed to generate frequency distribution data. The frequency distribution data reflects the frequency of occurrence of abnormal points in different time periods, which helps to identify potential high-risk areas. Based on frequency distribution data, a sliding window method is used to process the data stream. By sliding a fixed-size window across the time series, the data within the window is analyzed and processed, and the state prediction curve is updated in real time. The anomaly detection algorithm is then iteratively optimized using the updated state prediction curve.

9. The tunnel condition monitoring method based on structural and appearance data according to claim 1, characterized in that: The optimized state prediction results specifically include: Extract abnormal feature details related to the health status of the tunnel from the monitoring data, and use environmental sensors to obtain dynamic parameters of the environment in which the tunnel is located, so as to reflect the complexity and dynamic changes of the environment in which the tunnel is located. By combining the details of abnormal features with the parameters of adaptation to complex environments, machine learning algorithms are used to analyze the correlation between the two, and a weight adjustment strategy is formulated based on the results of the correlation analysis. Based on the weight adjustment strategy, the modeling capability of the prediction model is dynamically adjusted. Based on the results of correlation analysis, the weight distribution of the prediction model is updated, and the model is retrained to adapt to the new weight distribution. The updated prediction model is then used to generate the optimized state prediction results.

10. A tunnel condition monitoring method based on structural and appearance data according to claim 1, characterized in that: The priority monitoring scheme for identifying high-risk areas specifically includes: The optimized state prediction results are obtained, and the state prediction is performed by the support vector machine algorithm to obtain the tunnel health change trend. Time series features are extracted from the tunnel health change trend, and the trend stability is analyzed by the random forest algorithm to obtain the health status score. When the health status score is lower than the preset threshold, risk linkage analysis is performed in combination with historical fault data to obtain the risk probability distribution. Based on the risk probability distribution, the K-means clustering algorithm is used to divide high-risk areas and obtain a list of high-risk areas. Based on the list of high-risk areas, we analyze the frequency of real-time monitoring data and the distribution of anomalies to determine priority monitoring areas. We extract sensor data from these priority monitoring areas to generate a dynamic monitoring plan for high-risk areas, obtain a monitoring resource allocation strategy, adjust the sensor data acquisition frequency, and obtain an optimized monitoring execution plan.

Images