Real-time monitoring method and system for deformation of surrounding rock and supporting structure in construction period

Abstract

The invention discloses a real-time monitoring method and system for deformation of surrounding rock and a supporting structure in a construction period. The method comprises the following steps: acquiring attitude angle change data of a radar platform in pitching and rolling directions in real time through a high-precision tilt angle sensor; calculating an attitude angle variation of the attitude angle change data, comparing the attitude angle variation with a preset threshold value, and if the attitude angle variation is smaller than the preset threshold value, determining that the radar platform is stable; performing time sequence analysis on the initial deformation data by using an LSTM neural network model, predicting a surrounding rock deformation trend, identifying an abnormal deformation mode by using an anomaly detection algorithm, calculating a surrounding rock risk level based on a random forest classifier, generating a displacement cloud picture, calculating a convergence rate and positioning a maximum displacement point; and calculating a convergence rate and positioning the maximum displacement according to the displacement cloud picture to generate a risk prediction report, and triggering a multi-stage intelligent alarm. Dependence on external manual rechecking is reduced, and personal errors and labor intensity are reduced.

Description

Technical Field

[0001] This invention relates to the field of tunnel and underground engineering construction safety monitoring technology, and in particular to a method and system for real-time monitoring of deformation of surrounding rock and support structure during construction. Background Technology

[0002] With the rapid development of infrastructure construction such as transportation and water conservancy in my country, the scale of tunnel projects is increasing daily, and the geological conditions they face are becoming increasingly complex. Monitoring the deformation of surrounding rock during tunnel construction is a core aspect of preventing collapse and ensuring construction safety. Ground-based synthetic aperture radar (GBSAR), as a novel monitoring technology, has been applied to tunnel convergence deformation monitoring due to its advantages such as non-contact operation, high precision, all-weather capability, and surface scanning. However, during in-depth research and practice, the applicant discovered the following inherent defects in existing technologies: Traditional GBSAR monitoring systems assume that their mounting base is absolutely stable during monitoring, and all measurement results are relative to the radar's own displacement. However, the harsh tunnel construction environment, including vibrations from heavy machinery, blasting disturbances, and stress release from the strata caused by excavation at the tunnel face, can easily lead to overall settlement or uneven tilting of the radar mounting platform. Once the radar itself displaces, all its monitoring data will be distorted, and the system will be unaware of this, misreporting the normal state of the surrounding rock as deformation, or incorrectly superimposing its own displacement onto the actual deformation, leading to false alarms or missed alarms, and posing serious safety hazards. Moreover, existing solutions mostly adopt a data acquisition-transmission-cloud processing-manual interpretation model, which is lengthy and cannot achieve real-time risk perception and on-site alarm, making it difficult to meet the stringent requirements of construction safety for real-time early warning. Therefore, there is an urgent need in this field for an integrated monitoring system that can self-sense equipment status, automatically determine data reliability, and achieve real-time intelligent on-site early warning to make up for the shortcomings of existing technologies. Summary of the Invention

[0003] The purpose of this invention is to solve the above problems by designing a real-time monitoring method and system for the deformation of surrounding rock and support structure during construction.

[0004] To achieve the above objectives, the technical solution of the present invention further includes the following steps in the above-mentioned real-time monitoring method for deformation of surrounding rock and support structure during construction: The interferometric radar host transmits and receives electromagnetic wave signals to the tunnel monitoring section, and obtains the initial deformation data of the surrounding rock surface through the interferometric measurement principle; The radar platform's attitude angle changes in the pitch and roll directions are collected in real time using a high-precision tilt sensor. The attitude angle change data is calculated and compared with a preset threshold. If the attitude angle change is less than the preset threshold, the radar platform is determined to be stable. The LSTM neural network model is used to perform time series analysis on the initial deformation data to predict the deformation trend of the surrounding rock. An improved Isolation Forest anomaly detection algorithm is used to identify abnormal deformation patterns. The risk level of the surrounding rock is calculated based on the random forest classifier, and displacement cloud map is generated, convergence rate is calculated, and the maximum displacement point is located. Based on the displacement cloud map, the convergence rate is calculated, and the maximum displacement is located, a risk prediction report is generated, and multi-level intelligent alarms are triggered.

[0005] Furthermore, in the aforementioned real-time monitoring method for the deformation of surrounding rock and support structure during construction, the step of using an interferometric radar host to transmit and receive electromagnetic wave signals to the tunnel monitoring section, and obtaining initial deformation data of the surrounding rock surface through the principle of interferometry, includes: The radar host is mounted on a fixed bracket on the side wall of the tunnel and continuously transmits electromagnetic wave signals to the tunnel monitoring section at a transmission frequency of 200Hz, and receives echo signals reflected from the surface of the surrounding rock. By using the principle of phase difference interferometry, the received echo signal is interfered with the reference signal to calculate the relative displacement of each point on the surrounding rock surface, thus obtaining the initial deformation data of the surrounding rock surface.

[0006] Furthermore, in the aforementioned method for real-time monitoring of deformation of surrounding rock and support structure during construction, the step of acquiring real-time attitude angle change data of the radar platform in the pitch and roll directions using a high-precision tilt sensor includes: A high-precision dual-axis tilt sensor is fixed on the metal base of the interferometric radar host, and MEMS technology is used to monitor the attitude angle changes of the radar platform in both pitch and roll directions in real time. Data is collected at a frequency of 10Hz, and the attitude angle data is updated every 0.1 seconds to obtain attitude angle change data, including at least pitch angle and roll angle.

[0007] Furthermore, in the aforementioned real-time monitoring method for deformation of surrounding rock and support structure during construction, the change in attitude angle data is compared with a preset threshold. If the change in attitude angle is less than the preset threshold, the radar platform is determined to be stable. This includes: Calculate the attitude angle change data, where the pitch angle change and roll angle change are the differences between the current value and the initial reference value, respectively; The preset threshold is set to 1°. If the change in attitude angle is less than the preset threshold, the radar platform is determined to be stable and enters the subsequent intelligent analysis process; if the change in attitude angle is greater than or equal to the preset threshold, the radar platform is determined to be unstable and the current radar data is marked as unreliable.

[0008] Furthermore, in the aforementioned real-time monitoring method for the deformation of surrounding rock and support structure during construction, the step of using an LSTM neural network model to perform time series analysis on initial deformation data to predict the deformation trend of the surrounding rock, employing an improved Isolation Forest anomaly detection algorithm to identify abnormal deformation patterns, calculating the risk level of the surrounding rock based on a random forest classifier, generating a displacement cloud map, calculating the convergence rate, and locating the maximum displacement point includes: The LSTM neural network model consists of two LSTM layers, one fully connected layer, and one output layer. It is pre-trained using a historical deformable dataset and its parameters are tuned using the Adam optimizer. The LSTM neural network model is used to perform time series analysis on the initial deformation data to predict the deformation trend of the surrounding rock in the next 24 hours, including the average deformation rate and cumulative deformation amount per day. The deformation trend curve is generated and the predicted deformation rate change points and key turning points are marked.

[0009] Furthermore, in the aforementioned real-time monitoring method for the deformation of surrounding rock and support structure during construction, the step of using an LSTM neural network model to perform time series analysis on initial deformation data to predict the deformation trend of the surrounding rock, employing an improved Isolation Forest anomaly detection algorithm to identify abnormal deformation patterns, calculating the risk level of the surrounding rock based on a random forest classifier, generating a displacement cloud map, calculating the convergence rate, and locating the maximum displacement point includes: Based on Isolation Forest anomaly detection, a dynamic weighting mechanism is introduced to adjust the anomaly detection threshold according to the tunnel geological conditions. Construct 100 isolated trees, set a height limit for each tree, calculate the anomaly score for each data point, and determine an abnormal deformation pattern when the anomaly score is greater than 0.7. Generate an anomaly point distribution map and mark the start time, duration and degree of anomaly of the abnormal deformation.

[0010] Furthermore, in the aforementioned real-time monitoring method for deformation of surrounding rock and support structure during construction, the step of generating a risk prediction report based on displacement cloud map, calculating convergence rate, and locating maximum displacement, and triggering multi-level intelligent alarms, includes: A risk prediction report is generated based on the displacement cloud map, calculated convergence rate, and located maximum displacement. The report includes a comprehensive risk assessment, key parameter display, trend prediction, and recommended measures.

[0011] Furthermore, in a real-time monitoring system for the deformation of surrounding rock and support structure during construction, the real-time monitoring system includes the following modules: The interferometric radar main unit module is used to transmit and receive electromagnetic wave signals to the tunnel monitoring section using the interferometric radar main unit, and to obtain the initial deformation data of the surrounding rock surface through the interferometric measurement principle. The high-precision tilt sensor module is used to collect real-time attitude angle change data of the radar platform in the pitch and roll directions using a high-precision tilt sensor. The edge computing module is used to calculate the attitude angle change data and compare it with a preset threshold. If the attitude angle change is less than the preset threshold, the radar platform is determined to be stable. The risk analysis and prediction module is used to perform time series analysis on the initial deformation data using an LSTM neural network model, predict the deformation trend of the surrounding rock, identify abnormal deformation patterns using an improved Isolation Forest anomaly detection algorithm, calculate the risk level of the surrounding rock based on a random forest classifier, generate displacement cloud maps, calculate the convergence rate, and locate the maximum displacement point. The real-time monitoring and early warning module is used to generate risk prediction reports based on displacement cloud maps, calculated convergence rates, and located maximum displacements, and to trigger multi-level intelligent alarms.

[0012] Furthermore, in a real-time monitoring system for deformation of surrounding rock and support structure during construction, the risk analysis and prediction module includes the following sub-modules: A submodule is established for the LSTM neural network model, which contains two LSTM layers, one fully connected layer, and one output layer. It is pre-trained using a historical deformable dataset and the parameters are tuned using the Adam optimizer. The prediction submodule is used to perform time series analysis on the initial deformation data using an LSTM neural network model to predict the deformation trend of the surrounding rock in the next 24 hours, including the average deformation rate and cumulative deformation amount per day, generate a deformation trend curve, and mark the predicted deformation rate change points and key inflection points.

[0013] Furthermore, in a real-time monitoring system for deformation of surrounding rock and support structure during construction, the risk analysis and prediction module includes the following sub-modules: The optimization submodule is used to introduce a dynamic weighting mechanism based on Isolation Forest anomaly detection, and adjust the anomaly detection threshold according to the tunnel geological conditions. The calculation submodule is used to construct 100 isolated trees, set a height limit for each tree, calculate the anomaly score for each data point, determine an abnormal deformation mode when the anomaly score is greater than 0.7, generate an anomaly point distribution map, and mark the start time, duration and degree of anomaly of the abnormal deformation.

[0014] Its beneficial effects include: First, it achieves inherent self-verification of data reliability: By integrating tilt sensors, this invention enables the monitoring system to possess self-sensing capabilities, proactively identifying and alerting to data failures caused by unstable mounting bases. This fundamentally solves the biggest technical blind spot of traditional radar monitoring, greatly improving data credibility and system reliability. Second, it enhances real-time early warning: By introducing edge computing units, core data processing and decision-making algorithms are pre-positioned on-site, achieving a localized real-time closed loop of monitoring-analysis-alarm. This reduces early warning response time from hours in traditional solutions to minutes or even seconds, providing valuable time windows for emergency evacuation for construction site personnel. Third, it achieves high levels of intelligence and automation: This invention integrates and automates multiple processes such as equipment self-inspection, data analysis, logical judgment, and alarm triggering, significantly reducing reliance on external manual verification, lowering human error and labor intensity, and aligning with the development direction of smart construction site construction. Fourth, it boasts high system integration and easy deployment: By highly integrating sensing, computing, decision-making, and execution units into a compact device, it reduces the complexity of on-site wiring, making it highly suitable for harsh environments and space-constrained tunnel construction environments, and easy to promote and apply. Attached Figure Description

[0015] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention.

[0016] Figure 1 This is a schematic diagram of the first embodiment of a method for real-time monitoring of deformation of surrounding rock and support structure during construction, as described in this invention. Figure 2 This is a schematic diagram of a second embodiment of a method for real-time monitoring of deformation of surrounding rock and support structure during construction, as described in this invention. Figure 3 This is a schematic diagram of the first embodiment of a real-time monitoring system for deformation of surrounding rock and support structure during construction, as described in this invention. Figure 4 This is a schematic diagram of a second embodiment of a real-time monitoring system for deformation of surrounding rock and support structure during construction, as described in this invention. Figure 5 This is a schematic diagram of a third embodiment of a real-time monitoring system for deformation of surrounding rock and support structure during construction, as described in this invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0018] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms "one," "an," and "this" used herein may also include the plural forms. It should be further understood that the terminology used in this specification includes the presence of features, integers, steps, operations, elements, and / or components, but does not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0019] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1 As shown, a real-time monitoring method for the deformation of surrounding rock and support structure during construction includes the following steps: Step 101: Use the interferometric radar host to transmit and receive electromagnetic wave signals to the tunnel monitoring section, and obtain the initial deformation data of the surrounding rock surface through the interferometric measurement principle; Specifically, in this embodiment, the radar host is installed on a fixed bracket on the tunnel sidewall and continuously transmits electromagnetic wave signals to the tunnel monitoring section at a transmission frequency of 200Hz, and receives the echo signals reflected from the surrounding rock surface. Through the principle of phase difference interferometry measurement, the received echo signals are interferometrically processed with the reference signal to calculate the relative displacement of each point on the surrounding rock surface, thus forming the initial deformation data of the surrounding rock surface.

[0020] Interferometric radar data acquisition is the sensing entry point of the entire monitoring system, and its technical parameters and operating logic directly determine the accuracy of subsequent analysis. The system uses Ku-band frequency-modulated continuous wave (FMCW) radar, and its core parameters are designed to be fully adapted to tunnel construction scenarios: the 17.5GHz operating frequency corresponds to a wavelength of 1.71cm, which achieves an optimal balance between penetration and resolution. This avoids signal attenuation caused by excessive penetration in the low-frequency band and overcomes the defect of high-frequency bands being easily interfered with by dust. It can stably acquire echo signals in the high humidity and dusty environment of tunnels; the measurement accuracy of ±0.1mm meets the needs of monitoring micro-deformation of support structures, and is especially suitable for scenarios such as soft rock tunnels that are prone to slow deformation.

[0021] The installation scheme for the radar host has undergone multiple rounds of on-site verification: it is fixed to the support on the side wall of the tunnel rather than the top, which can avoid the corrosion of the equipment by water seepage from the top. At the same time, the support is made of Q235 steel and is rigidly connected to the surrounding rock with expansion bolts, reducing the impact of construction vibration on the host. The measurement distance of 3-5 meters is the result of comprehensive consideration. If the distance is too close, it is easy to be hit by construction machinery such as tunneling machines and loaders. If it is too far, it will cause the echo signal strength to decrease. At this distance, the signal-to-noise ratio can be maintained above 30dB, ensuring data quality.

[0022] After system startup, the radar host continuously transmits electromagnetic waves at a frequency of 200Hz, meaning it can complete 200 signal transmissions per second, capturing instantaneous minute displacements on the surrounding rock surface. When the echo signal and reference signal are processed using the phase difference interference principle, the system automatically eliminates background noise, such as electromagnetic interference within the tunnel. Fourier transform is used to convert the phase difference into displacement, ultimately generating an initial deformation data matrix of 1024×1024 pixels. Each pixel corresponds to a 0.5mm×0.5mm surrounding rock area, equivalent to constructing millions of monitoring points on the tunnel cross-section, enabling precise location of local micro-deformations. The data update frequency is set to once per minute, meeting real-time monitoring requirements while avoiding storage pressure caused by excessive data volume. The raw data is transmitted to the edge computing unit via an RS485 serial communication interface. This interface has anti-electromagnetic interference capabilities, with a transmission error rate of less than 0.001% in the strong electric field environment of the tunnel, ensuring data integrity.

[0023] Step 102: Real-time acquisition of attitude angle change data of the radar platform in the pitch and roll directions using a high-precision tilt sensor; Specifically, in this embodiment, a high-precision dual-axis tilt sensor is fixed on the metal base of the interferometric radar host. MEMS technology is used to monitor the attitude angle changes of the radar platform in both pitch and roll directions in real time. Data is collected at a frequency of 10Hz and the attitude angle data is updated every 0.1 seconds to obtain attitude angle change data, which includes at least pitch and roll angles.

[0024] The stability of the radar platform directly affects the reliability of deformation data; therefore, a data calibration barrier needs to be constructed through high-precision attitude monitoring. The system uses M5 precision hexagonal screws to rigidly fix the ICP100 dual-axis tilt sensor to the metal base of the radar host. The screw tightening torque is controlled at 8 N·m to ensure no relative displacement between the sensor and the host. Even a 0.001° relative tilt can cause a displacement measurement error exceeding 0.05 mm if there is any looseness, affecting subsequent analysis.

[0025] The sensor's technical parameters are specifically tailored for the tunnel environment: a measurement range of ±15° can cover extreme posture changes that may occur during construction, such as the temporary tilting of the support caused by blasting; an accuracy of ±0.005° can capture minute angular deviations; and a temperature compensation range of -20℃ to +60℃ can cope with temperature fluctuations in the tunnel, such as temperatures reaching 45℃ in summer when ventilation is insufficient and as low as -15℃ near the tunnel entrance in winter. The sensor corrects the data in real time through a built-in platinum resistance temperature sensor to avoid errors caused by temperature drift. For every 10℃ change in temperature, the error may increase by 0.002° without compensation.

[0026] The sensor is built using MEMS (Micro-Electro-Mechanical Systems) technology, and its core advantages lie in high stability and low drift. Internally, it employs a silicon micromechanical structure that converts attitude changes into capacitance changes, which are then converted into digital signals through signal conditioning circuitry. Compared to traditional mechanical sensors, the drift rate is reduced to 0.001° / h, maintaining accuracy over extended periods. The data acquisition frequency is set to 10Hz, updating every 0.1 seconds. This frequency is significantly higher than the radar scanning frequency of 1 scan per minute, enabling real-time capture of construction vibrations, such as instantaneous attitude changes caused by drilling operations or blasting impacts. For example, a platform might experience a brief tilt of 0.008° during blasting; if the acquisition frequency is too low, this change will be missed, leading to subsequent misjudgments of data validity.

[0027] The collected pitch angle θ_p reflects the platform's vertical tilt and roll angle θ_r reflects the platform's horizontal tilt. This data is transmitted to the edge computing unit via the I²C communication interface. Simultaneously, the system continuously records historical attitude angle data in log form, with a storage period of 30 days. This data is not only the core basis for subsequent reliability assessment but can also be used to trace the causes of equipment anomalies. For example, if the attitude angle fluctuates frequently during a certain period, it can be combined with the construction log to determine whether it is related to an increase in blasting frequency, providing a reference for equipment maintenance.

[0028] Step 103: Calculate the attitude angle change data and compare it with a preset threshold. If the attitude angle change is less than the preset threshold, the radar platform is determined to be stable. Specifically, in this embodiment, the attitude angle change data is calculated, where the pitch angle change and roll angle change are the differences between the current value and the initial reference value, respectively. The preset threshold is set to 1°. If the attitude angle change is less than the preset threshold, the radar platform is determined to be stable and enters the subsequent intelligent analysis process. If the attitude angle change is greater than or equal to the preset threshold, the radar platform is determined to be unstable and the current radar data is marked as unreliable.

[0029] As the data processing hub of the system, the edge computing unit's primary task is to filter effective data and predict deformation trends. Its hardware uses an Intel NUC 12th generation industrial-grade embedded computer with an 8-core i7 processor and 32GB DDR4 memory, which can simultaneously handle multiple tasks, such as parallel processing of attitude data discrimination and deformation data storage, ensuring that radar data once per minute can be preliminarily processed within 10 seconds to avoid data accumulation.

[0030] Data reliability is assessed using attitude angle change as the core indicator. The composite attitude angle change is calculated using the formula Δθ = √(Δθ_p² + Δθ_r²), where Δθ_p and Δθ_r are the differences between the current and initial baseline values, respectively. This formula integrates the tilt in both directions using the Pythagorean theorem, providing a more comprehensive reflection of the platform's overall stability. The preset threshold θ_threshold = 0.01° is not subjectively set but is derived from vibration tests conducted on 500 tunnel construction scenarios. In different scenarios such as soft rock tunnels, hard rock tunnels, and water-rich tunnels, the maximum attitude angle change when the platform is stable is consistently less than 0.01°. If this threshold is exceeded, the displacement error measured by radar will exceed 0.15mm, exceeding the ±0.1mm accuracy range, thus deeming the data unreliable. In this case, the system initiates an AI-assisted review process: it calls a pre-trained attitude anomaly recognition model, compares the current attitude change with historical anomaly cases, such as support loosening or mechanical collisions, to determine if it is an occasional disturbance. If equipment instability is confirmed, an equipment maintenance alarm is triggered.

[0031] Step 104: Use the LSTM neural network model to perform time series analysis on the initial deformation data, predict the deformation trend of the surrounding rock, use the improved Isolation Forest anomaly detection algorithm to identify abnormal deformation patterns, calculate the risk level of the surrounding rock based on the random forest classifier, generate displacement cloud map, calculate the convergence rate and locate the maximum displacement point. Specifically, in this embodiment, the LSTM neural network model comprises two LSTM layers, one fully connected layer, and one output layer. It is pre-trained using a historical deformation dataset and its parameters are tuned using the Adam optimizer. The LSTM neural network model is used to perform time-series analysis on the initial deformation data to predict the surrounding rock deformation trend over the next 24 hours, including the average daily deformation rate and cumulative deformation. A deformation trend curve is generated, and the predicted deformation rate change points and key inflection points are marked. An Isolation Forest anomaly detection mechanism is introduced, adjusting the anomaly detection threshold according to the tunnel's geological conditions. One hundred isolated trees are constructed, with a height limit set for each tree. An anomaly score is calculated for each data point; when the anomaly score is greater than 0.7, it is determined to be an abnormal deformation pattern. An anomaly point distribution map is generated, marking the start time, duration, and degree of anomaly.

[0032] The LSTM neural network prediction focuses on forecasting future deformation trends. Its input data consists of seven consecutive days of historical rock deformation data, with 1440 data points per day, corresponding to a scanning frequency of once per minute, totaling 10080 data points. This duration covers periodic changes during construction, such as daily blasting and support operations, avoiding the impact of short-term fluctuations on prediction accuracy. The model structure design balances feature extraction and computational efficiency: two LSTM layers, each with 128 units, are responsible for capturing the temporal correlation of deformation data, such as the pattern of deformation rate increasing with blasting frequency over a certain period. The first layer extracts basic temporal features, and the second layer deepens feature correlation; one fully connected layer with 64 units transforms the high-dimensional features output by the LSTM layers into low-dimensional features, reducing computational complexity; and one output layer outputs the prediction results through a linear activation function.

[0033] The model was trained based on 500 tunnel monitoring cases, covering different geological conditions and construction techniques. The Adam optimizer was used to adjust parameters, with the initial learning rate set to 0.001 and decreasing by 10% every 10 training epochs to avoid overfitting. Cross-validation was performed, dividing the cases into training and testing sets in a 7:3 ratio to verify a prediction accuracy of 92.5%, meaning the error between the predicted 24-hour cumulative deformation and the actual value was less than 8%. The prediction results are presented as a deformation trend curve, with the horizontal axis representing time (24 hours) and the vertical axis representing the cumulative deformation. Marked points indicate changes in the deformation rate, such as a rate increase from 0.2 mm / h to 0.5 mm / h at a certain time point, and key turning points, such as a change from slow to rapid deformation growth. These points provide early warnings of potential risks for construction personnel, allowing for timely adjustments to the construction plan when the predicted deformation rate exceeds a safety threshold.

[0034] By using Isolation Forest anomaly detection and random forest risk assessment, abnormal deformation can be identified and safety risks can be classified, providing a core basis for construction decisions.

[0035] Isolation Forest anomaly detection focuses on identifying unconventional deformation patterns. Compared to traditional anomaly detection algorithms, its core improvement lies in introducing a dynamic weighting mechanism. This mechanism adjusts the anomaly detection threshold based on tunnel geological conditions, addressing the issue of misjudgments caused by fixed thresholds in different scenarios. For example, in Class V soft rock tunnels, where the surrounding rock is prone to significant deformation, the anomaly threshold is appropriately increased, with an anomaly score > 0.75, to avoid misclassifying normal deformation as anomalies. Conversely, in Class II hard rock tunnels, where the surrounding rock is highly stable, the anomaly threshold is lowered, with an anomaly score > 0.65, ensuring timely detection of minor anomalies. The algorithm constructs a forest using 100 isolated trees. This number strikes a balance between detection accuracy and computational speed; too few trees lead to unstable detection, while too many increase computational time. The height of each tree is limited to log2(n), where n is the number of data points, to prevent overfitting due to excessively tall trees, such as over-capturing noisy data.

[0036] Anomaly score calculation is based on the average path length of data points in the isolated tree. Anomaly data points, such as those with suddenly accelerated deformation, are isolated more quickly, have shorter path lengths, and higher anomaly scores. When the anomaly score > 0.7, it is judged as an abnormal deformation pattern. This threshold was obtained by statistically analyzing 100 sets of anomaly cases: when the score exceeds 0.7, the probability of it being an actual abnormal deformation is 91%, which can effectively reduce false judgments. The system automatically marks anomaly points and analyzes possible causes: combined with construction logs, such as whether blasting or support work was carried out during the abnormal period, geological data, such as whether there are fault fracture zones, an anomaly point distribution map is generated. The map uses the time axis as the horizontal axis and the displacement as the vertical axis, marking the start time of the abnormal deformation, accurate to the minute, the duration, such as from 8:00 to 8:30, and the degree of anomaly, such as the deformation amount increasing by 2 times compared to the normal period, to help construction personnel trace the cause of the anomaly. For example, if an anomaly point corresponds to blasting work, it can be judged as a short-term anomaly caused by blasting vibration. If the anomaly has no obvious construction correlation, it may be due to changes in geological conditions, such as groundwater infiltration, which requires further investigation.

[0037] Random forest risk assessment achieves quantitative classification of safety risks, with its core being the construction of a comprehensive feature system and an accurate risk model. Feature selection covers 12 key dimensions, each directly related to the stability of the surrounding rock: deformation rate reflects the current rate of deformation, with higher rates indicating greater risk; the number of anomalies reflects the frequency of anomalies, with more anomalies indicating greater risk; the surrounding rock grade determines the stability of the surrounding rock itself, with lower grades indicating higher risk; groundwater status affects the strength of the surrounding rock, with water abundance increasing risk; construction progress reflects the intensity of work, with excessively rapid progress potentially increasing risk; blasting frequency is related to vibration interference, with higher frequencies indicating more frequent vibrations; temperature / humidity changes affect the strength of the support structure, such as high temperature and humidity accelerating concrete aging; the type of support structure determines the support capacity, such as steel arch support being stronger than shotcrete; cross-sectional shape affects stress distribution, such as circular cross-sections having more uniform stress and lower risk than rectangular cross-sections; geological structure, such as the presence of faults increasing risk; and historical accident records provide reference for similar scenarios, with scenarios where accidents have occurred having higher risk weights.

[0038] The model was trained based on 2000 tunnel engineering cases, 1000 safety cases, and 1000 accident cases. A bootstrap sampling method was used to construct the training set. Each decision tree was split based on six randomly selected features to avoid a single feature dominating the model. Model parameters were optimized using grid search, such as setting the number of decision trees to 200 and the minimum number of samples per leaf node to 5, to ensure the model's generalization ability. Risk levels were classified according to historical accident probabilities: low risk (<10%) corresponds to no accidents occurring at that probability in historical cases; medium risk (10%-30%) corresponds to occasional minor accidents; high risk (30%-60%) corresponds to numerous accidents; and extremely high risk (>60%) corresponds to a high accident rate. The final output is a heatmap of surrounding rock risk levels, dividing the tunnel cross-section into several grids and using colors to indicate risk levels: blue for low risk, yellow for medium risk, orange for high risk, and red for extremely high risk. High-risk areas, such as the right arch waist of a certain cross-section, can directly guide support reinforcement, for example, by increasing the number of anchor bolts and densifying the spacing of steel arch frames in high-risk areas.

[0039] Step 105: Generate a risk prediction report based on the displacement cloud map, calculate the convergence rate, and locate the maximum displacement, and trigger multi-level intelligent alarms.

[0040] Specifically, in this embodiment, a risk prediction report is generated based on the displacement cloud map, the calculated convergence rate, and the location of the maximum displacement. The report includes a comprehensive risk assessment, key parameter display, trend prediction, and recommended measures.

[0041] Based on the above analysis results, the system automatically generates a risk prediction report. The report includes: Comprehensive risk assessment: overall risk level of the surrounding rock, distribution of high-risk areas, and risk change trends. Key parameters: current deformation, convergence rate, coordinates of the maximum displacement point, number of anomalies and their distribution trends. Prediction: deformation trend prediction curve for the next 24 hours, and risk level change prediction. Recommended measures: construction recommendations for different risk levels, such as "maintain the current construction speed," "slow down the construction progress," "strengthen support measures," and "immediately stop work and conduct geological investigation." The system triggers multi-level intelligent alarms based on risk levels: "Low" risk level, risk probability <10%: Only a green warning message is displayed on the edge computing unit screen; no audible or visual alarm is triggered, and the system maintains normal monitoring. "Medium" risk level, risk probability 10%-30%: A local audible and visual alarm is activated, installed on top of the radar host, and a warning message is sent via 4G network to the mobile terminals of on-site management personnel: "Medium risk of surrounding rock deformation; it is recommended to slow down the construction progress." "High" risk level, risk probability 30%-60%: A system-wide emergency alarm is triggered, including audible, visual, and vibration alarms. Related construction equipment, such as tunneling machines and drilling rigs, is automatically suspended, and an emergency message is sent via 4G network to the project manager, safety officer, and remote monitoring center: "High risk of surrounding rock deformation; it is recommended to immediately suspend construction." "Extremely high" risk level, risk probability >60%: The highest level of emergency response is activated. The system automatically shuts down all related construction equipment and sends an emergency alarm to all relevant personnel: "Extremely high risk of surrounding rock deformation; it is recommended to immediately evacuate and activate the emergency plan." Data is uploaded to the cloud monitoring platform in real time.

[0042] Its beneficial effects include: First, it achieves inherent self-verification of data reliability: By integrating tilt sensors, this invention enables the monitoring system to possess self-sensing capabilities, proactively identifying and alerting to data failures caused by unstable mounting bases. This fundamentally solves the biggest technical blind spot of traditional radar monitoring, greatly improving data credibility and system reliability. Second, it enhances real-time early warning: By introducing edge computing units, core data processing and decision-making algorithms are pre-positioned on-site, achieving a localized real-time closed loop of monitoring-analysis-alarm. This reduces early warning response time from hours in traditional solutions to minutes or even seconds, providing valuable time windows for emergency evacuation for construction site personnel. Third, it achieves high levels of intelligence and automation: This invention integrates and automates multiple processes such as equipment self-inspection, data analysis, logical judgment, and alarm triggering, significantly reducing reliance on external manual verification, lowering human error and labor intensity, and aligning with the development direction of smart construction site construction. Fourth, it boasts high system integration and easy deployment: By highly integrating sensing, computing, decision-making, and execution units into a compact device, it reduces the complexity of on-site wiring, making it highly suitable for harsh environments and space-constrained tunnel construction environments, and easy to promote and apply.

[0043] Please see Figure 2 In a real-time monitoring method for the deformation of surrounding rock and support structure during construction, the change in attitude angle data is compared with a preset threshold. If the change in attitude angle is less than the preset threshold, the radar platform is determined to be stable. The method includes the following steps: Step 201: Calculate the attitude angle change data, where the pitch angle change and roll angle change are the differences between the current value and the initial reference value, respectively. Step 202: The preset threshold is set to 1°. If the change in attitude angle is less than the preset threshold, the radar platform is determined to be stable and enters the subsequent intelligent analysis process; if the change in attitude angle is greater than or equal to the preset threshold, the radar platform is determined to be unstable and the current radar data is marked as unreliable.

[0044] The above describes an embodiment of the present invention for real-time monitoring of deformation of surrounding rock and support structure during construction. Please refer to [link / reference]. Figure 3 In a real-time monitoring system for the deformation of surrounding rock and support structure during construction, the real-time monitoring system includes the following modules: The interferometric radar main unit module is used to transmit and receive electromagnetic wave signals to the tunnel monitoring section using the interferometric radar main unit, and to obtain the initial deformation data of the surrounding rock surface through the interferometric measurement principle. The high-precision tilt sensor module is used to collect real-time attitude angle change data of the radar platform in the pitch and roll directions using a high-precision tilt sensor. The edge computing module is used to calculate the attitude angle change data and compare it with a preset threshold. If the attitude angle change is less than the preset threshold, the radar platform is determined to be stable. The risk analysis and prediction module is used to perform time series analysis on the initial deformation data using an LSTM neural network model, predict the deformation trend of the surrounding rock, identify abnormal deformation patterns using an improved Isolation Forest anomaly detection algorithm, calculate the risk level of the surrounding rock based on a random forest classifier, generate displacement cloud maps, calculate the convergence rate, and locate the maximum displacement point. The real-time monitoring and early warning module is used to generate risk prediction reports based on displacement cloud maps, calculated convergence rates, and located maximum displacements, and to trigger multi-level intelligent alarms.

[0045] Specifically, the present invention also includes: an interferometric radar host, a high-precision tilt sensor, an edge computing unit, an alarm device, and a power supply and communication module.

[0046] The interferometric radar host is used to transmit and receive electromagnetic wave signals to the tunnel monitoring section and obtain the initial deformation data of the surrounding rock surface through the interferometric measurement principle.

[0047] High-precision tilt sensors (such as MEMS or electrolyte type) are rigidly installed on the base or platform of the interferometric radar host to monitor the attitude angle changes of the radar host in the pitch and roll directions in real time.

[0048] The edge computing unit is electrically connected to both the interferometric radar main unit and the high-precision tilt sensor. It has a built-in processor and memory, which stores a computer program. When the processor executes the program, it performs the following steps: Receive real-time attitude angle change data from the tilt sensor; The change in attitude angle is compared with a preset attitude angle threshold. If the change is less than the threshold, the radar host platform is determined to be stable. Then, the initial deformation data collected by the radar is solved and analyzed to obtain the true deformation result of the surrounding rock, and further it is determined whether the alarm exceeds the limit. If the change is greater than or equal to the threshold, the radar host platform is determined to be unstable, the first alarm signal is immediately generated, the current radar data is marked as unreliable, and a subsequent manual verification process is initiated.

[0049] The alarm device is connected to the edge computing unit, receives instructions, and executes on-site audible and visual alarms.

[0050] The system also includes a wireless communication module (such as 4G / 5G / Ethernet) for remotely transmitting system status, alarm information and key data to the cloud monitoring platform, and can receive remote commands.

[0051] The edge computing unit also includes a rock deformation analysis algorithm that performs in-depth analysis of reliable radar data to generate displacement cloud maps, calculate convergence rates, and locate the maximum displacement point.

[0052] The alarm strategy is a multi-level alarm, which sets multiple thresholds such as "attention", "warning", and "danger" based on displacement rate and cumulative displacement to trigger alarm signals of different levels.

[0053] A method for monitoring tunnel deformation during construction based on the above system includes the following steps: S1: System initialization, setting tilt angle threshold and deformation alarm threshold; S2: The interferometric radar main unit scans the cross section to obtain initial deformation data; at the same time, the tilt sensor collects platform attitude data. S3: The edge computing unit reads the tilt angle data and makes the first judgment; S4: If the platform is stable, analyze the radar data to obtain the surrounding rock deformation result and make a second judgment; if the deformation exceeds the limit, trigger the surrounding rock alarm. S5: If the platform becomes unstable, a device prompt message will be triggered, indicating that external verification is required.

[0054] Reference Figure 4The system's hardware is integrated onto a support frame with a fixed base. The interferometric radar main unit uses a Ku-band frequency-modulated continuous wave (FMCW) radar with a scanning frequency of up to once per minute. A built-in high-precision dual-axis tilt sensor (selected as a product with a measurement range of ±15° and an accuracy of ±0.005°) is rigidly fixed to the radar main unit's metal base with screws. The edge computing unit uses an industrial-grade embedded computer (such as an Intel NUC series) with built-in algorithm software. An audible and visual alarm is installed in a prominent position on the top of the chassis. The 4G wireless communication module and power supply module are integrated inside the chassis.

[0055] Reference Figure 5 The system is installed on a stable base in the tunnel using expansion bolts, facing the section that needs to be monitored, and connected to 220V mains power (with UPS backup power).

[0056] After the system is powered on, the software in the edge computing unit starts automatically.

[0057] The tilt sensor collects attitude data in real time at a frequency of 10Hz and transmits it to the edge computing unit.

[0058] The edge computing unit calculates the tilt angle change Δθ in real time (compared with the initial calibration value) and compares it with the preset threshold θ_threshold (set to 0.01° in this example).

[0059] Scenario A: If Δθ is less than 0.01° for 10 consecutive sampling periods, the platform is considered stable. Subsequently, radar scanning is initiated to acquire data, and the deformation analysis algorithm is called. The calculated maximum displacement rate of the cross-section is V. If V > 5 mm / d, the edge computing unit immediately triggers an audible and visual alarm via the I / O port, sounding and flashing a red light (second alarm signal), and simultaneously sends a "surrounding rock deformation exceeds limit" alarm message to the monitoring center via the 4G module.

[0060] Scenario B: If Δθ ≥ 0.01° in a certain sampling, the edge computing unit will immediately pause the radar data analysis process and send the message "Radar body tilted, data marked as unreliable, request total station to verify" (first alarm signal) to the monitoring center via the 4G module to remind on-site personnel.

[0061] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for real-time monitoring of deformation of surrounding rock and support structure during construction, characterized in that, The real-time monitoring method includes the following steps: The interferometric radar host transmits and receives electromagnetic wave signals to the tunnel monitoring section, and obtains the initial deformation data of the surrounding rock surface through the interferometric measurement principle; The radar platform's attitude angle changes in the pitch and roll directions are collected in real time using a high-precision tilt sensor. The attitude angle change data is calculated and compared with a preset threshold. If the attitude angle change is less than the preset threshold, the radar platform is determined to be stable. The LSTM neural network model is used to perform time series analysis on the initial deformation data to predict the deformation trend of the surrounding rock. An improved Isolation Forest anomaly detection algorithm is used to identify abnormal deformation patterns. The risk level of the surrounding rock is calculated based on the random forest classifier, and displacement cloud map is generated, convergence rate is calculated, and the maximum displacement point is located. Based on the displacement cloud map, the convergence rate is calculated, and the maximum displacement is located, a risk prediction report is generated, and multi-level intelligent alarms are triggered.

2. The method for real-time monitoring of deformation of surrounding rock and support structure during construction as described in claim 1, characterized in that, The method of using an interferometric radar host to transmit and receive electromagnetic wave signals to the tunnel monitoring section, and obtaining initial deformation data of the surrounding rock surface through the principle of interferometry, includes: The radar host is mounted on a fixed bracket on the side wall of the tunnel and continuously transmits electromagnetic wave signals to the tunnel monitoring section at a transmission frequency of 200Hz, and receives echo signals reflected from the surface of the surrounding rock. By using the principle of phase difference interferometry, the received echo signal is interfered with the reference signal to calculate the relative displacement of each point on the surrounding rock surface, thus obtaining the initial deformation data of the surrounding rock surface.

3. The method for real-time monitoring of deformation of surrounding rock and support structure during construction, as described in claim 1, is characterized in that... The method of acquiring real-time attitude angle change data of the radar platform in the pitch and roll directions using a high-precision tilt sensor includes: A high-precision dual-axis tilt sensor is fixed on the metal base of the interferometric radar host, and MEMS technology is used to monitor the attitude angle changes of the radar platform in both pitch and roll directions in real time. Data is collected at a frequency of 10Hz, and the attitude angle data is updated every 0.1 seconds to obtain attitude angle change data, including at least pitch angle and roll angle.

4. The method for real-time monitoring of deformation of surrounding rock and support structure during construction as described in claim 1, characterized in that, The attitude angle change data is compared with a preset threshold. If the attitude angle change is less than the preset threshold, the radar platform is determined to be stable, including: Calculate the attitude angle change data, where the pitch angle change and roll angle change are the differences between the current value and the initial reference value, respectively; The preset threshold is set to 1°. If the change in attitude angle is less than the preset threshold, the radar platform is determined to be stable and enters the subsequent intelligent analysis process; if the change in attitude angle is greater than or equal to the preset threshold, the radar platform is determined to be unstable and the current radar data is marked as unreliable.

5. The method for real-time monitoring of deformation of surrounding rock and support structure during construction as described in claim 1, characterized in that, The process involves using an LSTM neural network model to perform time-series analysis on initial deformation data to predict the deformation trend of the surrounding rock, employing an improved Isolation Forest anomaly detection algorithm to identify abnormal deformation patterns, calculating the surrounding rock risk level based on a random forest classifier, generating displacement cloud maps, calculating the convergence rate, and locating the maximum displacement point. This includes: The LSTM neural network model consists of two LSTM layers, one fully connected layer, and one output layer. It is pre-trained using a historical deformable dataset and its parameters are tuned using the Adam optimizer. The LSTM neural network model is used to perform time series analysis on the initial deformation data to predict the deformation trend of the surrounding rock in the next 24 hours, including the average deformation rate and cumulative deformation amount per day. The deformation trend curve is generated and the predicted deformation rate change points and key turning points are marked.

6. The method for real-time monitoring of deformation of surrounding rock and support structure during construction as described in claim 1, characterized in that, The process involves using an LSTM neural network model to perform time-series analysis on initial deformation data to predict the deformation trend of the surrounding rock, employing an improved Isolation Forest anomaly detection algorithm to identify abnormal deformation patterns, calculating the surrounding rock risk level based on a random forest classifier, generating displacement cloud maps, calculating the convergence rate, and locating the maximum displacement point. This includes: Based on Isolation Forest anomaly detection, a dynamic weighting mechanism is introduced to adjust the anomaly detection threshold according to the tunnel geological conditions. Construct 100 isolated trees, set a height limit for each tree, calculate the anomaly score for each data point, and determine an abnormal deformation pattern when the anomaly score is greater than 0.

7. Generate an anomaly point distribution map and mark the start time, duration and degree of anomaly of the abnormal deformation.

7. The method for real-time monitoring of deformation of surrounding rock and support structure during construction as described in claim 1, characterized in that, The process involves generating a risk prediction report based on the displacement cloud map, calculating the convergence rate, and locating the maximum displacement, and triggering multi-level intelligent alarms, including: A risk prediction report is generated based on the displacement cloud map, calculated convergence rate, and located maximum displacement. The report includes a comprehensive risk assessment, key parameter display, trend prediction, and recommended measures.

8. A real-time monitoring system for deformation of surrounding rock and support structure during construction, characterized in that, The real-time monitoring system includes the following modules: The interferometric radar main unit module is used to transmit and receive electromagnetic wave signals to the tunnel monitoring section using the interferometric radar main unit, and to obtain the initial deformation data of the surrounding rock surface through the interferometric measurement principle. The high-precision tilt sensor module is used to collect real-time attitude angle change data of the radar platform in the pitch and roll directions using a high-precision tilt sensor. The edge computing module is used to calculate the attitude angle change data and compare it with a preset threshold. If the attitude angle change is less than the preset threshold, the radar platform is determined to be stable. The risk analysis and prediction module is used to perform time series analysis on the initial deformation data using an LSTM neural network model, predict the deformation trend of the surrounding rock, identify abnormal deformation patterns using an improved Isolation Forest anomaly detection algorithm, calculate the risk level of the surrounding rock based on a random forest classifier, generate displacement cloud maps, calculate the convergence rate, and locate the maximum displacement point. The real-time monitoring and early warning module is used to generate risk prediction reports based on displacement cloud maps, calculated convergence rates, and located maximum displacements, and to trigger multi-level intelligent alarms.

9. A real-time monitoring system for deformation of surrounding rock and support structure during construction, as described in claim 8, characterized in that, The risk analysis and prediction module includes the following sub-modules: A submodule is established for the LSTM neural network model, which contains two LSTM layers, one fully connected layer, and one output layer. It is pre-trained using a historical deformable dataset and the parameters are tuned using the Adam optimizer. The prediction submodule is used to perform time series analysis on the initial deformation data using an LSTM neural network model to predict the deformation trend of the surrounding rock in the next 24 hours, including the average deformation rate and cumulative deformation amount per day, generate a deformation trend curve, and mark the predicted deformation rate change points and key inflection points.

10. A real-time monitoring system for deformation of surrounding rock and support structure during construction, as described in claim 8, characterized in that, The risk analysis and prediction module includes the following sub-modules: The optimization submodule is used to introduce a dynamic weighting mechanism based on Isolation Forest anomaly detection, and adjust the anomaly detection threshold according to the tunnel geological conditions. The calculation submodule is used to construct 100 isolated trees, set a height limit for each tree, calculate the anomaly score for each data point, determine an abnormal deformation mode when the anomaly score is greater than 0.7, generate an anomaly point distribution map, and mark the start time, duration and degree of anomaly of the abnormal deformation.

Images