Intelligent monitoring and early warning method and system for tunnel surrounding rock fracture water damage

Abstract

The invention discloses an intelligent monitoring and early warning method and system for tunnel surrounding rock fracture water disasters, and belongs to the field of intelligent monitoring, and the method comprises the following steps: obtaining a fusion multi-dimensional sensing signal based on a multi-dimensional sensing device; obtaining dynamic characteristic data based on the fused multi-dimensional sensing signal, and determining a space-time evolution rule of the micro crack and the stress change based on the dynamic characteristic data; constructing a feature matrix based on the space-time evolution rule, and performing dimension reduction processing on the feature matrix to obtain a low-dimensional feature vector; processing the low-dimensional feature vector based on a deep learning network to obtain a short-term change trend prediction value of crack propagation and stress change; and inputting the short-term change trend prediction values of the crack propagation and the stress change into an early warning model, and outputting a space-time probability distribution diagram of the crack water damage in combination with real-time monitoring data of a multi-dimensional sensing device.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent monitoring, and particularly relates to an intelligent monitoring and early warning method and system for fissure water disasters in tunnel surrounding rocks. Background Technique

[0002] As an important part of modern infrastructure construction, the safety and stability of tunnel projects are directly related to the guarantee of transportation and public life and property. With the continuous expansion of the scale of tunnel construction, fissure water disasters have become one of the major hidden dangers threatening the operation safety of tunnels. Fissure water disasters not only lead to structural instability but also trigger catastrophic accidents such as water inrush and mud inrush. Especially under complex geological conditions, the timeliness and accuracy of monitoring and early warning are crucial. At present, certain progress has been made in the research on monitoring and early warning of tunnel fissure water disasters, but there are still many challenges and more intelligent and efficient solutions are needed.

[0003] Existing monitoring methods mostly rely on traditional sensors or single data acquisition means, such as pressure gauges or water level gauges. These technologies have obvious deficiencies in terms of coverage, real-time performance, and data analysis depth. Especially for areas with stress concentration and complex geological conditions such as tunnel joints, traditional methods often have difficulty capturing the dynamic evolution of microcracks and cannot accurately reflect the relationship between stress changes and water disaster risks. In addition, data processing mostly stays in the post-analysis stage and lacks the ability to identify abnormal trends in advance, resulting in delayed early warning and being difficult to meet the actual engineering requirements. Therefore, the present invention proposes an intelligent monitoring and early warning method and system for fissure water disasters in tunnel surrounding rocks. Summary of the Invention

[0004] To solve the above technical problems, the present invention proposes an intelligent monitoring and early warning method and system for fissure water disasters in tunnel surrounding rocks to solve the problems existing in the above prior art.

[0005] To achieve the above object, the present invention provides an intelligent monitoring and early warning method for fissure water disasters in tunnel surrounding rocks, including:

[0006] Obtaining fused multi-dimensional perception signals based on multi-dimensional perception devices;

[0007] Obtaining dynamic characteristic data based on the fused multi-dimensional perception signals, and determining the spatio-temporal evolution laws of microcracks and stress changes based on the dynamic characteristic data;

[0008] Constructing a feature matrix based on the spatio-temporal evolution laws, and performing dimensionality reduction processing on the feature matrix to obtain low-dimensional feature vectors;

[0009] Processing the low-dimensional feature vectors based on a deep learning network to obtain predicted values of short-term change trends of crack expansion and stress changes;

[0010] Input the predicted values of the short-term change trends of the crack propagation and stress change into the early warning model, and output the spatio-temporal probability distribution map of the fissure water disaster in combination with the real-time monitoring data of the multi-dimensional perception device.

[0011] Optionally, the process of obtaining the fused multi-dimensional perception signal based on the multi-dimensional perception device includes:

[0012] Collect the initial distribution data of the micro-cracks through the sensor array of the multi-dimensional perception device to obtain the crack position distribution;

[0013] Adopt acoustic wave scanning technology to conduct depth detection on the crack position distribution to obtain the dynamic data of crack propagation;

[0014] Analyze the dynamic data through strain measurement technology to obtain the real-time trend of stress change;

[0015] Determine the crack risk level based on the real-time trend of the stress change;

[0016] Adjust the detection frequency of the multi-dimensional perception device based on the crack risk level to obtain the fused multi-dimensional perception signal.

[0017] Optionally, the process of determining the spatio-temporal evolution law of the micro-cracks and stress change includes:

[0018] Construct a spatio-temporal evolution model; use the crack position distribution and the dynamic data of crack propagation as the spatial feature variables in the spatio-temporal evolution model, and use the real-time trend of stress change as the time feature variable in the spatio-temporal evolution model;

[0019] Conduct correlation analysis on the spatial feature variables and time feature variables through the spatio-temporal evolution model to determine the evolution law of the micro-cracks and stress change at different spatial positions and time stages.

[0020] Optionally, the expression of the spatio-temporal evolution model is:

[0021]

[0022] In the formula, represents the Laplace operator, which is the diffusion effect in space, and respectively represent the change rates of the crack position distribution and crack length over time, represents the change rate of stress change over time, D is the diffusion coefficient, which represents the rate of crack diffusion, and α is the stress influence coefficient, which represents the influence degree of stress change on crack propagation.

[0023] Optionally, the process of obtaining the predicted values of the short-term change trends of the crack propagation and stress change by processing the low-dimensional feature vector based on the deep learning network includes:

[0024] Divide the low-dimensional feature vector into time series data;

[0025] Construct a deep learning network, which includes several LSTM layers;

[0026] Input the time series data into the LSTM layer, and perform forgetting gate calculation and input gate calculation in sequence to obtain the hidden state at each time step;

[0027] Based on the fully connected layer, map the hidden state to the predicted values of the short-term change trends of crack propagation and stress change.

[0028] Optionally, the process of inputting the predicted values of the short-term change trends of crack propagation and stress change into the early warning model and outputting the spatio-temporal probability distribution map of fissure water disasters in combination with the real-time monitoring data of the multi-dimensional perception device includes:

[0029] Input the predicted values of the short-term change trends of crack propagation and stress change and the real-time monitoring data into the early warning model to obtain the preliminary distribution characteristics of water disaster risks;

[0030] Adopt the Bayesian probability update method to process the preliminary distribution characteristics and obtain the time series change sequence of the dynamic threshold;

[0031] Judge the triggering conditions of the abnormal state through the time series change sequence of the dynamic threshold to obtain the identification data of the abnormal state;

[0032] If the identification data of the abnormal state exceeds the preset threshold, fuse the monitoring data and historical records to obtain the extended distribution of the abnormal state;

[0033] According to the extended distribution of the abnormal state, determine the division interval of the state level to obtain the classification result of the state level;

[0034] Generate a dynamic early warning signal based on the classification result of the state level;

[0035] Obtain the distribution characteristics of fissure water disasters based on the dynamic early warning signal to obtain the water disaster distribution set;

[0036] Generate the spatio-temporal probability map of fissure water disasters by using the interpolation method based on the water disaster distribution set.

[0037] The present invention also provides an intelligent monitoring and early warning system for tunnel surrounding rock fissure water disasters, which is used to implement an intelligent monitoring and early warning method for tunnel surrounding rock fissure water disasters. The system includes:

[0038] A multi-dimensional perception module, which is used to obtain the initial distribution data of micro-cracks and the real-time signal of stress change through a sensor array, and fuse the acoustic wave scanning and strain measurement technologies to obtain a fused multi-dimensional perception signal;

[0039] A feature extraction module, which is used to extract dynamic feature data from the fused multi-dimensional perception signals and determine the spatio-temporal evolution laws of micro-cracks and stress changes;

[0040] A dimensionality reduction processing module, which is used to perform dimensionality reduction processing on the dynamic feature data to obtain low-dimensional feature vectors;

[0041] A deep learning module, which is used to process the low-dimensional feature vectors and judge the potential occurrence probability of abnormal trends;

[0042] A time series analysis module, which is used to analyze the time series correlation by using a long short-term memory network for the probability results of abnormal trends and obtain the predicted values of short-term change trends of crack propagation and stress changes;

[0043] An early warning calculation module, which is used to calculate the abnormal state level at the current moment by using the Bayesian probability update method for the predicted values of short-term change trends of crack propagation and stress changes;

[0044] A data update module, which is used to adjust the sampling frequency based on the abnormal state level and update the real-time monitoring data set;

[0045] A partition processing module, which is used to perform partition processing on the crack distribution and stress concentration areas based on the updated real-time monitoring data set and judge the evolution direction and intensity distribution of abnormal trends in high-risk areas;

[0046] An early warning generation module, which is used to extract key risk features from the intensity distribution of partition processing and output the spatio-temporal probability distribution map of fissure water disasters.

[0047] Compared with the prior art, the present invention has the following advantages and technical effects:

[0048] The present invention discloses a method for early warning of water disaster risks at tunnel joints. The method obtains real-time data of micro-crack distribution and stress changes by deploying multi-dimensional perception devices at tunnel joints. Feature extraction and dimensionality reduction processing are performed on the data by using algorithms such as time-frequency analysis and principal component analysis, and then abnormal trend judgment and short-term change prediction are carried out through a deep learning network and a long short-term memory network. The water disaster risk threshold is dynamically calculated based on the Bayesian probability update method, and the data acquisition accuracy is improved when the abnormal state exceeds the threshold. Finally, clustering analysis is performed on high-risk areas to generate the spatio-temporal probability distribution map of fissure water disasters. The present invention realizes the real-time monitoring of micro-cracks and stress changes at tunnel joints, early warning of abnormal trends, and identification of risk areas, providing effective technical support for tunnel safety management. Description of the Drawings

[0049] The accompanying drawings, which form a part of this application, are used to provide a further understanding of this application. The schematic embodiments of this application and their descriptions are used to explain this application and do not constitute an improper limitation of this application. In the drawings:

[0050] Figure 1 It is a flow chart of the intelligent monitoring and early warning method for fissure water damage of tunnel surrounding rock in an embodiment of the present invention;

[0051] Figure 2 It is a structural diagram of the intelligent monitoring and early warning system for fissure water damage of tunnel surrounding rock in an embodiment of the present invention. Detailed implementation manners

[0052] It should be noted that, without conflict, the embodiments in this application and the features in the embodiments are combined with each other. The following will describe this application in detail with reference to the accompanying drawings and in combination with the embodiments.

[0053] It should be noted that the steps shown in the flowchart of the accompanying drawings are executed in a computer system such as a set of computer-executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described are executed in a different order than here.

[0054] Embodiment 1

[0055] As Figure 1 shown, in this embodiment, an intelligent monitoring and early warning method for fissure water damage of tunnel surrounding rock is provided, including the following steps:

[0056] Step S101, deploy multi-dimensional sensing devices at the tunnel joint, obtain the initial distribution data of micro-cracks and the real-time signals of stress changes through the sensor array, and obtain the fused multi-dimensional sensing signals by fusing acoustic wave scanning and strain measurement technologies.

[0057] Collect the initial distribution data of micro-cracks through the sensor array, perform preliminary processing on the fused real-time signals to obtain the crack position distribution. Use acoustic wave scanning technology to conduct depth detection on the crack position distribution to obtain the dynamic data of crack expansion. Analyze the dynamic data through strain measurement technology to obtain the real-time trend of stress changes. If the real-time trend of stress changes exceeds the preset threshold, optimize the high-sensitivity detection results through the fusion algorithm to determine the crack risk level. According to the crack risk level, adjust the detection frequency of the multi-dimensional sensing device to obtain updated real-time signals. Through the updated real-time signals, combine acoustic wave scanning and strain measurement technologies to obtain the fused multi-dimensional sensing signals.

[0058] Furthermore, as a specific implementation manner of this embodiment, a sensor array is arranged on the load-bearing beam of a concrete bridge, and a sensor is set every 10 cm. The initial data collected is the distribution of cracks with a width of 0.1 mm and a length of 5 cm.

[0059] Specifically, ultrasonic waves with a frequency of 50 kHz are emitted at the crack of the bridge load-bearing beam, and the received echo shows that the crack depth has expanded from 2 cm to 3 cm, indicating that the crack is growing slowly. By using strain measurement technology to analyze dynamic data, mainly using strain gauges or fiber optic sensors to measure the stress changes around the crack.

[0060] In one implementation, strain gauges are attached on both sides of the crack, and it is monitored that the stress increases from 50 MPa to 80 MPa. The real-time trend shows that the stress concentration phenomenon intensifies. The advantage of this technology is to quantify the impact of cracks on the structural strength and provide data support for risk assessment. If the stress change trend exceeds a preset threshold, such as 100 MPa, the detection results with high sensitivity are optimized through a fusion algorithm.

[0061] In one embodiment, multi-source data of acoustic wave scanning and strain measurement are input into the fusion algorithm, and through comprehensive analysis, the crack risk level is obtained as "medium".

[0062] Specifically, the fusion algorithm weights the data of different sensors to improve the reliability of the results, thereby avoiding misjudgment of a single technology. This optimization significantly improves the accuracy of risk assessment. According to the crack risk level, the detection frequency of the multi-dimensional sensing device is adjusted. For example, the detection frequency that was originally once an hour is adjusted to once every 15 minutes to obtain updated real-time signals.

[0063] Preferably, high-frequency detection can capture the subtle changes in crack propagation more timely.

[0064] For example, the updated signal shows that the crack length has increased from 5 cm to 6 cm and the depth remains unchanged. The benefit of this adjustment is to enhance the real-time monitoring and gain time for structural maintenance. By combining the updated real-time signal with acoustic wave scanning and strain measurement technologies, a multi-dimensional sensing signal with high sensitivity is obtained.

[0065] In one embodiment, acoustic wave scanning confirms that the crack depth is stable at 3 cm, while strain measurement shows that the stress further rises to 90 MPa. The fusion of multi-dimensional signals not only improves the comprehensiveness of the data but also reduces errors through cross-validation.

[0066] Step S102, extract dynamic feature data from the multi-dimensional sensing signal with high sensitivity, decompose the frequency components of crack propagation and stress fluctuation through time-frequency analysis methods, and determine the spatio-temporal evolution law of micro-cracks and stress changes.

[0067] Further, the process of determining the spatio-temporal evolution law of micro-cracks and stress changes includes: constructing a spatio-temporal evolution model; using the crack position distribution and the dynamic data of crack propagation as spatial feature variables in the spatio-temporal evolution model, and using the real-time trend of stress changes as the time feature variable in the spatio-temporal evolution model; performing correlation analysis on the spatial feature variables and time feature variables through the spatio-temporal evolution model to determine the evolution law of micro-cracks and stress changes at different spatial positions and time stages.

[0068] Furthermore, the expression of the spatio-temporal evolution model is:

[0069]

[0070] In the formula, represents the Laplace operator, which is the diffusion effect in space, and respectively represent the change rates of crack position distribution and crack length with time, represents the change rate of stress change with time, D is the diffusion coefficient, which represents the crack diffusion rate, and α is the stress influence coefficient, which represents the influence degree of stress change on crack propagation.

[0071] Step S103, construct a feature matrix according to the spatio-temporal evolution law, and use the principal component analysis algorithm to perform dimensionality reduction processing on the dynamic features of micro-cracks and stress changes to obtain low-dimensional feature vectors for subsequent trend judgment.

[0072] Construct a feature matrix through the spatio-temporal evolution law, use the principal component analysis algorithm to perform dimensionality reduction processing on the dynamic features to obtain low-dimensional feature vectors. Extract the initial basis for trend judgment from the low-dimensional feature vectors, obtain the spatial distribution data related to micro-cracks, and determine the boundary range of the distribution data. For the data within the boundary range, use a preset threshold to judge the abnormal area of stress change to obtain the identification result of the abnormal area. According to the identification result of the abnormal area, collect the fluctuation sequence of stress change through time series data acquisition, and determine the frequency distribution characteristics of the fluctuation sequence. Use the frequency distribution characteristics to adjust the acquisition range of the low-dimensional feature vectors to obtain the updated feature vector data. Through the updated feature vector data, judge the short-term evolution direction of the trend change, and determine the feature adjustment parameters of the evolution direction. According to the feature adjustment parameters, obtain the adjusted dynamic feature set, and use the principal component analysis algorithm to perform dimensionality reduction processing on the set data again to obtain the optimized low-dimensional feature vectors.

[0073] Furthermore, as a specific implementation manner of this embodiment, when monitoring the micro-cracks of a bridge structure, the stress changes in the time dimension and the crack position data in the space dimension are integrated into a multi-dimensional matrix. Each column of the matrix represents the stress values at different times, and each row corresponds to the spatial coordinates of the cracks.

[0074] As a specific implementation of this embodiment, assume that the original feature matrix contains 20 dimensions, such as stress, displacement, temperature, etc. Through principal component analysis, the data can be reduced to 3 main dimensions, retaining more than 90% of the information volume.

[0075] Specifically, the trend is inferred by observing the weight distribution of the principal components.

[0076] If the first principal component is mainly dominated by stress fluctuations, the trend points to the stress concentration area. After obtaining the spatial distribution data related to the microcracks, when determining the boundary range of the distribution data.

[0077] The clustering method is used to group the crack points, and the boundary range is set as the area containing 95% of the data points. For the data within the boundary range, a preset threshold is used to judge the abnormal area of the stress change.

[0078] As a specific implementation of this embodiment, it is set that the stress value exceeding 2 times the average value is abnormal, and the potential risk area of the bridge bearing part is identified. After obtaining the identification result of the abnormal area, the fluctuation sequence of the stress change is obtained through time series data acquisition.

[0079] For example, the stress value is recorded every 5 minutes, and continuously collected for 24 hours to form a fluctuation sequence. When determining the frequency distribution characteristics of the fluctuation sequence, the signal is decomposed by fast Fourier transform.

[0080] For example, it is analyzed that the low-frequency component of 0.1Hz is related to the crack propagation, while the high-frequency component of 1Hz is caused by external vibration. The acquisition range of the low-dimensional feature vector is adjusted by using the frequency distribution characteristics.

[0081] As a specific implementation of this embodiment, focus on the signal near 0.1Hz, eliminate the irrelevant frequencies, and update the feature vector. The short-term evolution direction of the trend change is judged by the updated feature vector data.

[0082] The low-dimensional vector shows that the stress concentration point moves towards the middle of the bridge, and the crack propagation direction can be inferred. When determining the feature adjustment parameters of the evolution direction.

[0083] Furthermore, the acquisition frequency can be adjusted from every 5 minutes to every 2 minutes to improve the data resolution. After obtaining the adjusted dynamic feature set according to the feature adjustment parameters, the dimensionality reduction processing is performed on the set data again through principal component analysis.

[0084] Construct a feature matrix according to the spatio-temporal evolution law.

[0085] Obtain a feature matrix according to the spatiotemporal evolution law, extract spatial distribution data for dynamic features, and determine the boundary range of the distribution data. Based on the data within the boundary range, use a preset threshold to judge the abnormal area and obtain the identification result of the abnormal area. Through the identification result of the abnormal area, obtain the time-series data of stress changes and determine the fluctuation sequence of the time-series data. Analyze the frequency distribution characteristics using the fluctuation sequence to obtain the characteristic parameters of the frequency distribution. Adjust the acquisition range of the feature matrix based on the characteristic parameters of the frequency distribution to obtain an updated feature matrix.

[0086] In step S104, after obtaining the low-dimensional feature vector, train the dynamic features through a pre-established deep learning network, fuse the crack propagation and stress anomaly samples in the historical data, and judge the potential occurrence probability of the abnormal trend.

[0087] Furthermore, the process of obtaining the short-term change trend prediction values of crack propagation and stress changes based on the deep learning network processing the low-dimensional feature vector includes: dividing the low-dimensional feature vector into time-series data; constructing a deep learning network, where the deep learning network includes several LSTM layers; inputting the time-series data into the LSTM layers, and performing forget gate calculation and input gate calculation in sequence to obtain the hidden state at each time step; mapping the hidden state to the short-term change trend prediction values of crack propagation and stress changes based on a fully connected layer.

[0088] Even further, as a specific implementation manner of this embodiment, construct a deep learning network model, where the model includes an input layer, a hidden layer, and an output layer. The input layer receives the low-dimensional feature vector, and the output layer outputs the short-term change trend prediction values of crack propagation and stress changes;

[0089] Divide the low-dimensional feature vector into time-series data, and the input at each time step is a sample of the low-dimensional feature vector;

[0090] Use the time-series data to train the deep learning network model, and optimize the parameters of the model by minimizing the loss function.

[0091] Use the trained deep learning network model to predict a new low-dimensional feature vector to obtain the short-term change trend prediction values of crack propagation and stress changes;

[0092] Among them, the deep learning network model is a long short-term memory network (LSTM), and the process of obtaining the short-term change trend prediction values of crack propagation and stress changes based on the long short-term memory network (LSTM) includes:

[0093] Construct a long short-term memory network (LSTM) model, which includes one or more LSTM layers. Each LSTM layer contains multiple LSTM units, and the structure of each LSTM unit includes an input gate, a forget gate, and an output gate;

[0094] Use the time series data to train the LSTM model, and update the parameters of the LSTM model through the backpropagation algorithm;

[0095] Use the trained LSTM model to predict new low-dimensional feature vectors, and obtain the predicted values of the short-term change trends of crack propagation and stress change.

[0096] In step S105, input the predicted value of the change trend into the early warning model, combine the full-coverage monitoring data, and use the Bayesian probability update method to calculate the dynamic threshold of the water disaster risk, and determine the abnormal state level at the current moment.

[0097] Furthermore, the process of determining the abnormal state level at the current moment includes: inputting the predicted values of the short-term change trends of crack propagation and stress change and the real-time monitoring data into the early warning model to obtain the preliminary distribution characteristics of the water disaster risk; using the Bayesian probability update method to process the preliminary distribution characteristics to obtain the time series change sequence of the dynamic threshold; judging the triggering conditions of the abnormal state through the time series change sequence of the dynamic threshold to obtain the identification data of the abnormal state; if the identification data of the abnormal state exceeds the preset threshold, fuse the monitoring data and historical records to obtain the extended distribution of the abnormal state; determine the division interval of the state level according to the extended distribution of the abnormal state to obtain the classification result of the state level.

[0098] Even further, as a specific implementation manner of this embodiment, input the predicted value of the change trend into the early warning model, and fuse the full-coverage monitoring data to obtain the preliminary distribution characteristics of the water disaster risk.

[0099] In tunnel water disaster monitoring, generate a preliminary risk distribution map through the real-time water level data collected by sensors and historical rainfall records, reflecting the potential changes in groundwater pressure.

[0100] Exemplarily, this distribution characteristic shows that the water level in a certain area has risen by 20 centimeters within 24 hours, indicating potential water disaster risks. Use the Bayesian probability update method to process the preliminary distribution characteristics to obtain the time series change sequence of the dynamic threshold.

[0101] Specifically, the Bayesian method uses prior knowledge, such as the frequency of water disaster events in the past year, and combines the current water level rising speed to update the risk probability.

[0102] In one implementation, if historical data shows that a 15-cm rise in water level is usually risk-free, and the current rise is 20 cm, the dynamic threshold is adjusted to 18 cm, reflecting a more sensitive risk judgment. Based on the time-series change sequence of the dynamic threshold, the triggering conditions for abnormal states are judged, and the identification data of abnormal states is obtained.

[0103] Preferably, when the water level exceeds 18 cm and the duration exceeds 6 hours, it can be marked as an abnormal state.

[0104] Exemplarily, at a monitoring point in a certain mining area, the water level rises from 15 cm to 19 cm within 3 hours. Since the triggering conditions are not fully met, it is not yet marked as abnormal for the time being, but it is prompted to continue to pay attention. If the identification data of the abnormal state exceeds the preset threshold, the monitoring data and historical records are fused to obtain the extended distribution of the abnormal state.

[0105] In one embodiment, if the water level rises to 22 cm and there has been water seepage in a similar area in the historical records, the extended distribution shows that the risk spreads from a single point to a surrounding area of 50 meters. This kind of fusion helps to more comprehensively evaluate the affected range of water disasters. According to the extended distribution of the abnormal state, the division interval of the state level is determined, and the classification result of the state level is obtained.

[0106] For example, the risk is divided into three levels: the water level of 18 - 20 cm is the first level, 20 - 25 cm is the second level, and above 25 cm is the third level.

[0107] Specifically, the current water level of 22 cm is classified as a second-level risk, indicating that drainage measures need to be taken. According to the classification result of the state level, the parameters of the early warning model are adjusted to obtain an optimized risk assessment output.

[0108] It should be noted that if second-level risks occur frequently, the model can increase the sensitivity to water level changes. For example, the threshold is adjusted from 18 cm to 17 cm.

[0109] In one embodiment, after adjustment, the model predicts that the water level in a certain area will reach 24 cm within the next 12 hours, and the accuracy of the optimized output is improved. Based on the optimized risk assessment output, the abnormal state level at the current moment is determined.

[0110] If the current water level is 22 cm and it is predicted that there will be no significant drop within 24 hours, the state level remains at the second level. However, if there is a rainfall forecast, it is upgraded to the third level. The advantage of this method is that it can give early warnings and gain time for emergency measures.

[0111] For example, in the actual monitoring of a certain tunnel, on a certain day, the water level rapidly rose from 16 cm to 23 cm. The dynamic threshold was adjusted to 17 cm, and the extended distribution showed that the risk affected the surrounding area, and it was finally rated as level two. By timely adjusting the power of the drainage pump, the water level dropped to the safe range within 8 hours, avoiding greater losses. This multi-faceted analysis ensures the comprehensiveness and practicality of risk assessment.

[0112] Step S106, if the abnormal state level exceeds the preset threshold, adjust the sampling frequency through the multi-dimensional sensing device to obtain the dynamic characteristics of cracks and stress change data with higher resolution, and obtain an updated real-time monitoring data set.

[0113] If the abnormal state exceeds the preset threshold, adjust the sampling frequency through the multi-dimensional sensing device to obtain high-resolution data on crack dynamics and stress changes, and obtain an updated version of the real-time monitoring data set. Extract the temporal change characteristics of crack dynamics from the updated data set, perform dimensionality reduction processing using the principal component analysis method, and obtain a simplified representation of crack dynamics. Calculate the fluctuation amplitude of crack dynamics based on the simplified representation, and judge the persistence of the abnormal state through a preset fluctuation range to obtain an abnormal persistence flag. If the abnormal persistence flag is positive, obtain the distribution characteristics of stress changes from the real-time monitoring data, use the clustering analysis method to divide the stress change area, and obtain the stress partition result. Match the stress partition result with the historical monitoring data to obtain a stress change pattern similar to the current crack dynamics, and obtain a pattern matching sequence. Adjust the sampling direction of the multi-dimensional sensing device according to the pattern matching sequence to obtain the joint distribution characteristics of crack dynamics and stress changes, and obtain a joint feature set. Update the storage structure of the real-time monitoring data through the joint feature set, and optimize the data query efficiency using the hash index method to obtain an optimized monitoring data set.

[0114] Specifically, adjusting the sampling frequency through the multi-dimensional sensing device is a flexible means to cope with the abnormal state exceeding the threshold.

[0115] Exemplarily, when monitoring the cracks in an underground reservoir, if it is detected that the abnormal state exceeds the preset threshold, the sampling frequency can be increased from once per hour to once every 10 minutes to capture the subtle changes in crack propagation. The high-resolution data obtained, such as the crack width increasing from 2 mm to 2.5 mm and the stress value rising from 50 Pa to 60 Pa, can more accurately reflect the real-time dynamics.

[0116] In one implementation, extract the temporal change characteristics of crack dynamics from the updated data set, and it is observed that the crack width shows periodic fluctuations within 24 hours. For example, the increase is larger during the day and tends to be stable at night. When performing dimensionality reduction processing using the principal component analysis method, multi-dimensional data such as crack width, length, and stress can be simplified into two main feature vectors, retaining more than 90% of the information for subsequent analysis.

[0117] Specifically, when calculating the fluctuation amplitude of the crack dynamics, it is assumed that the fluctuation range is preset to ±0.3 mm. If the measured amplitude reaches 0.4 mm, it is determined that the abnormal state is persistent, and the abnormal persistence flag is positive.

[0118] For example, if the crack width exceeds the preset range continuously for 3 hours, it indicates that the anomaly is not an instantaneous interference but a continuous evolution.

[0119] It should be noted that after obtaining the distribution characteristics of stress changes from the real-time monitoring data, the clustering analysis method can divide the stress values into three regions: high, medium, and low.

[0120] For example, the high-stress area is concentrated at the crack tip with a value of 70 Pa, and the medium-stress area is around it with a value of about 40 Pa. This zoning result clearly reflects the stress concentration law.

[0121] In one embodiment, similar cases can be found by matching the stress zoning results with the historical monitoring data.

[0122] For example, when the crack width in a certain area was 2.4 mm three months ago, the stress distribution was highly consistent with the current one. A pattern matching sequence was thus generated, indicating that the crack was accelerating in a certain direction. After adjusting the sampling direction of the multi-dimensional sensing device, such as focusing the sensor on the crack tip, the joint distribution characteristics of the crack dynamics and stress changes can be obtained. For example, the crack propagation speed and the stress value show a positive correlation trend.

[0123] Preferably, when the joint feature set updates the storage structure of the real-time monitoring data, the hash index method can significantly improve the query efficiency.

[0124] For example, the time-consuming for querying the crack data in a certain time period is reduced from 5 seconds to 1 second. This optimization ensures the efficient access to data, especially when the crack dynamics change rapidly, and can provide timely decision support.

[0125] The above method forms a complete chain from anomaly detection to data optimization through multi-dimensional data acquisition, feature extraction, and analysis.

[0126] For example, the adjustment of the sampling frequency directly improves the data accuracy, the dimensionality reduction processing reduces the analysis complexity, and the pattern matching enhances the prediction ability. These links support each other, ensuring the real-time and reliability of crack monitoring, and at the same time providing a solid foundation for subsequent risk assessment.

[0127] Step S107, according to the updated real-time monitoring data set, use the clustering analysis algorithm to perform zoning processing on the crack distribution and stress concentration area, and judge the abnormal trend evolution direction and intensity distribution of the high-risk area.

[0128] An updated dataset is obtained through real-time monitoring. Cluster analysis is used to divide the regional characteristics of crack distribution, and a preliminary zoning result is obtained. Based on the preliminary zoning result, the distribution characteristics of stress concentration are extracted, and statistical methods are used to calculate the range of strength distribution, obtaining the strength distribution characteristics. If the strength distribution characteristics exceed the preset threshold, time-series data of abnormal trends are obtained from the high-risk area, resulting in a trend change sequence. The persistence of the evolution direction is judged through the trend change sequence, and the direction vector method is used to determine the main path of the evolution direction, obtaining a path distribution set. According to the path distribution set, the sampling strategy of data processing is adjusted, and the joint characteristics of crack distribution and stress concentration are obtained, resulting in an optimized feature set. The storage structure of real-time monitoring is updated through the optimized feature set, and the index method is used to improve the query efficiency, obtaining an efficient dataset. The dynamic characteristics of abnormal trends are extracted from the efficient dataset, and the matching degree between the evolution direction and strength distribution in the high-risk area is judged, obtaining a matching result.

[0129] Specifically, an updated dataset is obtained through real-time monitoring. Cluster analysis is used to divide the regional characteristics of crack distribution, and a preliminary zoning result is obtained.

[0130] For example, in the real-time monitoring of a bridge, after the sensor collects crack distribution data, the K-means clustering method can be used to distinguish between the crack-dense area and the sparse area, forming a preliminary zoning map. This zoning can intuitively reflect the spatial distribution characteristics of cracks and provide a basis for subsequent analysis. Based on the preliminary zoning result, the distribution characteristics of stress concentration are extracted, and statistical methods are used to calculate the range of strength distribution, obtaining the strength distribution characteristics.

[0131] Specifically, the stress peak data of each region are extracted from the zoning, and their mean and variance are statistically calculated.

[0132] For example, the peak value of a high-stress area is concentrated around 200 MPa, and the variance is small, indicating obvious stress concentration. This statistical method can quickly quantify the regularity of stress distribution. If the strength distribution characteristics exceed the preset threshold, time-series data of abnormal trends are obtained from the high-risk area, resulting in a trend change sequence.

[0133] In one implementation, if the threshold is set to 250 MPa and the stress in a certain area reaches 280 MPa, abnormal monitoring is triggered, and the stress changes in this area are continuously recorded for 48 hours to form a time-series sequence. This sequence can reflect whether the abnormality is short-lived or continuously intensifying. The persistence of the evolution direction is judged through the trend change sequence, and the direction vector method is used to determine the main path of the evolution direction, obtaining a path distribution set.

[0134] For example, after analyzing the time-series data, it is found that the stress increases in a certain direction over time. The change trend can be represented by a vector, and the main path is calculated to point to the bridge support point. This method can clearly locate the core area of abnormal development. Adjust the sampling strategy of data processing according to the path distribution set, obtain the joint characteristics of crack distribution and stress concentration, and obtain an optimized feature set.

[0135] Exemplarily, if the path points to a certain area, the sampling frequency of the sensors in this area can be increased, from once per minute to once per second, to collect more refined crack width and stress data. This adjustment can improve the pertinence of the data. Update the storage structure of real-time monitoring through the optimized feature set, adopt an indexing method to improve the query efficiency, and obtain an efficient data set.

[0136] In one embodiment, the joint feature set can be divided by time and area, and a hash index can be constructed to shorten the query time of data in a certain period and area from seconds to milliseconds. This optimization facilitates the rapid positioning of key information. Extract the dynamic features of abnormal trends from the efficient data set, judge the matching degree between the evolution direction of high-risk areas and the intensity distribution, and obtain a matching result.

[0137] If the evolution direction points to the area with the highest stress peak and the intensity continuously exceeds the threshold, the matching degree is high, indicating that the abnormal risk is increasing.

[0138] For example, the crack propagation direction in a bridge support area coincides with the stress concentration point, indicating that this area needs to be given priority attention. This matching analysis can provide a reliable basis for decision-making.

[0139] Preferably, through multi-faceted verification, such as comparing the current path distribution with historical data, or cross-confirming through trend sequences at different times, the credibility of the results can be enhanced.

[0140] Specifically, if similar paths in the historical records have led to serious consequences, the warning significance of the current matching result is stronger. This method ensures comprehensive and practical analysis through logical progression.

[0141] Step S108, extract key risk features from the intensity distribution processed by partitioning, generate dynamic warning signals for high-risk areas through a warning model, and output the spatio-temporal probability distribution map of fissure water hazards.

[0142] Extract key risk features from the intensity distribution, generate a feature set using the random forest algorithm, and obtain the key risk distribution. Divide the high-risk area based on the key risk distribution, use the clustering analysis method to determine the area boundary, and obtain the scope of the high-risk area. Generate dynamic warning signals for the scope of the high-risk area, process the feature set using the warning model, and obtain the warning signal sequence. Extract the time-varying features from the warning signal sequence, calculate the change trend through the spatio-temporal analysis method, and obtain the trend distribution. Adjust the area division strategy according to the trend distribution, obtain the distribution characteristics of fissure water hazards, and obtain the water hazard distribution set. Update the probability distribution through the water hazard distribution set, generate the spatio-temporal probability map using the interpolation method, and obtain the fissure water hazard prediction map. If the probability in the prediction map exceeds the preset threshold, extract the abnormal features from the high-risk area, judge the matching degree of the trend distribution, and obtain the matching result.

[0143] Specifically, when extracting key risk features from the intensity distribution.

[0144] For example, in an underground tunnel, assume that the intensity distribution of a certain area shows that the maximum stress value reaches 50 MPa, while the surrounding area is only 20 MPa. This difference implies potential high-risk points. When generating the feature set using the random forest algorithm, the stress value, crack length, and distribution density can be used as input features, and the variables with the greatest contribution to the risk are screened out through the tree structure of the algorithm to obtain the key risk distribution. The advantage of this method is that it can extract the dominant factors from multi-dimensional data and improve the pertinence of subsequent analysis.

[0145] In one implementation, when dividing the high-risk area based on the key risk distribution, assume that the top 10% of the key risk values in a certain mining area are marked as the high-risk area. When using the clustering analysis method to determine the area boundary, based on the spatial coordinates and risk values, adjacent high-risk points are clustered into a group.

[0146] For example, the risk values of 5 points in a certain area are 45, 48, 50, 46, and 42 MPa respectively. After clustering, a high-risk area with a clear boundary is formed. This division can effectively demarcate the scope that needs to be focused on, facilitating centralized resource management. When generating dynamic warning signals for the scope of the high-risk area.

[0147] Exemplarily, the warning model can be trained based on historical data. For example, in the past 30 days, when the stress value in a certain area exceeds 40 MPa for 3 consecutive days, the crack propagation probability increases to 80%. After processing the feature set, the obtained warning signal sequence is the alarm level per hour, such as "low - medium - high". This serialized output helps to grasp the risk dynamics in real time. When extracting the time-varying features from the warning signal sequence.

[0148] It should be noted that the spatio-temporal analysis method combines the timestamp and the spatial position.

[0149] For example, the warning signal in a high-risk area rose from "medium" to "high" from March 15th to 17th. Through analysis, it was found that the stress value increased by 5 MPa daily, and the trend distribution pointed to the northeast direction. This trend distribution provides a directional basis for subsequent strategy adjustments. When adjusting the regional division strategy according to the trend distribution.

[0150] Specifically, if the trend shows that the fissure expands towards the northeast, the density of monitoring points in this direction can be increased, adjusted from 1 per 100 meters to 1 per 50 meters. When obtaining the distribution characteristics of fissure water hazards, assuming that the water pressure value in a certain area rises from 0.5 MPa to 1.2 MPa, the water hazard distribution set can reflect the correlation between water flow and cracks, which helps to identify the source of water hazards. When updating the probability distribution through the water hazard distribution set.

[0151] Preferably, interpolation methods can be used to fill the blank areas between monitoring points.

[0152] For example, when the water pressures at two monitoring points are 1.0 MPa and 1.5 MPa respectively, and the distance is 200 meters, the probability in the middle area can be interpolated and estimated to be 1.2 MPa. The generated spatio-temporal probability map can visually display the spatial evolution of water hazard risks, facilitating early intervention. If the probability in the prediction map exceeds the preset threshold, such as setting the threshold to 1.3 MPa, abnormal features are extracted from the high-risk area.

[0153] Embodiment 2

[0154] As Figure 2 shown, in this embodiment, an intelligent monitoring and early warning system for tunnel surrounding rock fissure water hazards is provided, including:

[0155] A multi-dimensional perception module for deploying multi-dimensional perception devices at the tunnel joint, obtaining the initial distribution data of micro-cracks and real-time signals of stress changes through a sensor array, and fusing acoustic wave scanning and strain measurement technologies to obtain highly sensitive multi-dimensional perception signals;

[0156] A feature extraction module for extracting dynamic feature data from highly sensitive multi-dimensional perception signals, decomposing the frequency components of crack expansion and stress fluctuations through time-frequency analysis methods, and determining the spatio-temporal evolution laws of micro-cracks and stress changes;

[0157] A dimensionality reduction processing module for constructing a feature matrix according to the spatio-temporal evolution laws, using the principal component analysis algorithm to perform dimensionality reduction processing on the dynamic features of micro-cracks and stress changes, and obtaining low-dimensional feature vectors for subsequent trend judgment;

[0158] A deep learning module for, after obtaining the low-dimensional feature vectors, training the dynamic features through a pre-established deep learning network, fusing crack expansion and stress anomaly samples in historical data, and judging the potential occurrence probability of abnormal trends;

[0159] The timing analysis module is used to analyze the temporal correlation between real-time monitoring data and historical data by using a long short-term memory network for the probability results of abnormal trends, and obtain the predicted values of short-term change trends of crack propagation and stress changes.

[0160] The early warning calculation module is used to input the predicted values of change trends into the early warning model, combine the full-coverage monitoring data, and calculate the dynamic threshold of water disaster risk by using the Bayesian probability update method to determine the abnormal state level at the current moment.

[0161] The data update module is used to adjust the sampling frequency through multi-dimensional sensing devices if the abnormal state level exceeds the preset threshold, obtain the dynamic characteristics of cracks and stress change data with higher resolution, and obtain the updated real-time monitoring data set.

[0162] The zoning processing module is used to perform zoning processing on the crack distribution and stress concentration areas by using a clustering analysis algorithm according to the updated real-time monitoring data set, and judge the evolution direction and intensity distribution of abnormal trends in high-risk areas.

[0163] The early warning generation module is used to extract key risk characteristics from the intensity distribution of zoning processing, generate dynamic early warning signals for high-risk areas through the early warning model, and output the spatio-temporal probability distribution map of fissure water disasters.

[0164] The above is only a preferred specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed in the present application should be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. An intelligent monitoring and early warning method for fissure water disasters in tunnel surrounding rocks, characterized in that, It includes the following steps: Obtain fused multi-dimensional perception signals based on multi-dimensional perception devices; Obtain dynamic feature data based on the fused multi-dimensional perception signals, and determine the spatio-temporal evolution laws of micro-cracks and stress changes based on the dynamic feature data; Construct a feature matrix based on the spatio-temporal evolution laws, and perform dimensionality reduction processing on the feature matrix to obtain low-dimensional feature vectors; Process the low-dimensional feature vectors based on a deep learning network to obtain predicted values of short-term change trends of crack propagation and stress changes; Input the predicted values of the short-term change trends of crack propagation and stress changes into an early warning model, and output the spatio-temporal probability distribution map of fissure water disasters in combination with the real-time monitoring data of multi-dimensional perception devices.

2. The intelligent monitoring and early warning method for tunnel surrounding rock fissure water disaster according to claim 1, wherein The process of obtaining fused multi-dimensional perception signals based on multi-dimensional perception devices includes: Collect the initial distribution data of micro-cracks through the sensor array of the multi-dimensional perception device to obtain the crack position distribution; Adopt acoustic wave scanning technology to conduct depth detection on the crack position distribution to obtain dynamic data of crack propagation; Analyze the dynamic data through strain measurement technology to obtain the real-time trend of stress changes; Determine the crack risk level based on the real-time trend of stress changes; Adjust the detection frequency of the multi-dimensional perception device based on the crack risk level to obtain the fused multi-dimensional perception signals.

3. The intelligent monitoring and early warning method for tunnel surrounding rock fissure water disaster according to claim 2, characterized in that The process of determining the spatio-temporal evolution laws of micro-cracks and stress changes includes: Construct a spatio-temporal evolution model; use the crack position distribution and the dynamic data of crack propagation as spatial feature variables in the spatio-temporal evolution model, and use the real-time trend of stress changes as the time feature variable in the spatio-temporal evolution model; Conduct correlation analysis on the spatial feature variables and time feature variables through the spatio-temporal evolution model to determine the evolution laws of micro-cracks and stress changes at different spatial positions and time stages.

4. The intelligent monitoring and early warning method for tunnel surrounding rock fissure water disaster according to claim 3, characterized in that The expression of the spatio-temporal evolution model is: In the formula, represents the Laplace operator, which is the diffusion effect in space, and respectively represent the distribution of crack positions and the change rate of crack length over time, represents the change rate of stress change over time, D is the diffusion coefficient, representing the rate of crack diffusion, and α is the stress influence coefficient, representing the degree of influence of stress change on crack propagation.

5. The intelligent monitoring and early warning method for tunnel surrounding rock fissure water disaster according to claim 1, characterized in that The process of processing the low-dimensional feature vectors based on a deep learning network to obtain predicted values of short-term change trends of crack propagation and stress changes includes: Divide the low-dimensional feature vectors into time series data; Construct a deep learning network, and the deep learning network includes several LSTM layers; Input the time series data into the LSTM layers, and perform forget gate calculation and input gate calculation in sequence to obtain the hidden state at each time step; Map the hidden state to the predicted values of the short-term change trends of crack propagation and stress changes based on a fully connected layer.

6. The intelligent monitoring and early warning method for tunnel surrounding rock fissure water disaster according to claim 1, characterized in that The process of inputting the predicted values of the short-term change trends of crack propagation and stress changes into an early warning model, and outputting the spatio-temporal probability distribution map of fissure water disasters in combination with the real-time monitoring data of multi-dimensional perception devices includes: Input the predicted values of the short-term change trends of crack propagation and stress changes and the real-time monitoring data into the early warning model to obtain the preliminary distribution characteristics of water disaster risks; Adopt the Bayesian probability update method to process the preliminary distribution characteristics to obtain the time series change sequence of dynamic thresholds; Judge the triggering conditions of abnormal states through the time series change sequence of dynamic thresholds to obtain the identification data of abnormal states; If the identification data of the abnormal state exceeds the preset threshold, fuse the monitoring data and historical records to obtain the extended distribution of the abnormal state. Determine the division interval of the state level according to the extended distribution of the abnormal state, and obtain the classification result of the state level; Generate a dynamic early warning signal based on the classification result of the state level; Obtain the distribution characteristics of fissure water disasters based on the dynamic early warning signal, and obtain the water disaster distribution set; Generate a spatio-temporal probability map of fissure water disasters by using the interpolation method based on the water disaster distribution set.

7. An intelligent monitoring and early warning system for fissure water disasters in tunnel surrounding rock, characterized in that, For implementing the intelligent monitoring and early warning method for tunnel surrounding rock fissure water disasters as described in any one of claims 1-6, the system includes: A multi-dimensional perception module, which is used to obtain the initial distribution data of micro-cracks and the real-time signal of stress change through a sensor array, and fuse the acoustic wave scanning and strain measurement technologies to obtain a fused multi-dimensional perception signal; A feature extraction module, which is used to extract dynamic feature data from the fused multi-dimensional perception signal and determine the spatio-temporal evolution law of micro-cracks and stress change; A dimensionality reduction processing module, which is used to perform dimensionality reduction processing on the dynamic feature data to obtain a low-dimensional feature vector; A deep learning module, which is used to process the low-dimensional feature vector and judge the potential occurrence probability of an abnormal trend; A time series analysis module, which is used to analyze the time series correlation by using a long short-term memory network for the probability result of the abnormal trend, and obtain the short-term change trend prediction values of crack propagation and stress change; An early warning calculation module, which is used to calculate the abnormal state level at the current moment by using the Bayesian probability update method for the short-term change trend prediction values of crack propagation and stress change; A data update module, which is used to adjust the sampling frequency based on the abnormal state level and update the real-time monitoring data set; A partition processing module, which is used to perform partition processing on the crack distribution and stress concentration areas based on the updated real-time monitoring data set, and judge the evolution direction and intensity distribution of the abnormal trend in the high-risk area; An early warning generation module, which is used to extract key risk features from the intensity distribution of the partition processing and output the spatio-temporal probability distribution map of fissure water disasters.

Images