System for accurately monitoring linear deformation of loess tunnel portal side upward slope

Abstract

The invention discloses an accurate monitoring system for linear deformation of a loess tunnel portal side upward slope, and relates to the technical field of geotechnical engineering and surveying and mapping. The system comprises a sensing network layer, a data acquisition and transmission layer, a cloud platform and data analysis layer and an application display layer, displacement data of discrete monitoring points are reconstructed into continuous full-field displacement and strain distribution through Kriging interpolation, then a displacement vector field is regarded as a whole to be analyzed, the maximum shear strain field of the displacement vector field is calculated, an abnormal area with the deformation gradient abrupt change is accurately positioned, and the deformation gradient abrupt change accuracy is improved. The abnormal point sets are merged into independent clusters according to spatial proximity and deformation consistency, and the geometric trajectory of a potential slip crack surface or crack is accurately outlined through curve fitting, so that the technical path fundamentally changes the limitation of point-mode monitoring, and can automatically identify a hidden linear failure trend which is difficult to find by naked eyes in a non-contact manner, so as to improve the detection accuracy of the hidden linear failure trend. And refined data support is provided for slope stability evaluation.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of geotechnical engineering and surveying technology, in particular to a linear deformation precision monitoring system for the side slope of a loess tunnel portal. BACKGROUND

[0002] Tunnels and underground works are key nodes in the construction of transportation, water conservancy and other infrastructure, and the stability of their portal sections directly determines the safety of the entire project. For loess strata, due to their large pores, vertical joint development, strong water sensitivity and other special engineering geological characteristics, under the action of tunnel excavation unloading, rainfall infiltration and other factors, the side slope of the portal is prone to instability phenomena such as traction collapse and cracking, which poses a serious threat to construction safety, operational order and even life and property safety. Therefore, real-time, accurate and intelligent monitoring of the deformation of the side slope of the loess tunnel portal is a long-term and urgent major requirement in the field of geotechnical engineering.

[0003] Currently, the monitoring technology in this field mainly relies on traditional geodetic surveying, contact-type sensor methods and emerging remote sensing measurement techniques. However, these technologies have significant limitations when dealing with linear deformation monitoring of loess slopes: first, traditional point monitoring methods cannot effectively capture the continuous evolution process of the through-going slip surface or crack from initiation, expansion to penetration, making it difficult to predict sudden failures, and the monitoring results are discrete and incomplete; second, existing technologies usually analyze deformation monitoring and environmental triggers in isolation, lacking depth integration and intelligent prediction capabilities; third, optical measurement methods are easily disturbed by construction dust, rain and fog, leading to interruptions in monitoring data and making it difficult to meet the requirements of continuous monitoring. These defects make it difficult for existing technology systems to fully and reliably ensure the safety of loess tunnel portal projects.

[0004] In summary, the existing monitoring technology has the core pain point of being unable to accurately perceive linear deformation characteristics. Therefore, developing a high-reliability system that can integrate multi-source data for early warning is of great significance for breaking through industry technical bottlenecks and improving the safety level of tunnel projects. SUMMARY

[0005] The purpose of the present application is to make up for the deficiencies of the prior art, and provide a loess tunnel portal side slope linear deformation precision monitoring system, which can reconstruct the continuous full-field displacement and strain distribution of the discrete monitoring point displacement data through Kriging interpolation, and then analyze the displacement vector field as a whole, accurately locate the abnormal area of deformation gradient mutation by calculating the maximum shear strain field, merge these abnormal point sets into independent clusters according to spatial proximity and deformation consistency, and accurately outline the geometric trajectory of the potential sliding surface or crack through curve fitting, which fundamentally changes the limitations of point monitoring, can non-contact and automatically identify the hidden linear damage trend that is difficult to find with the naked eye, provides fine data support for slope stability evaluation, and makes the early warning from blind to precise.

[0006] The present application provides the following technical solutions to solve the above technical problems: a loess tunnel portal side slope linear deformation precision monitoring system, which comprises a sensing network layer, a data acquisition and transmission layer, a cloud platform and data analysis layer, and an application display layer. The sensing network layer comprises a reference subsystem arranged in a stable area and a monitoring subsystem arranged in a side slope area to be monitored, and is arranged in the side slope area to be monitored to provide a measurement reference and a monitoring object for spatial deformation. The data acquisition and transmission layer communicates with the sensing network layer, automatically acquires three-dimensional spatial data and on-site environmental parameters of the monitoring object in the sensing network layer, and transmits the data to the cloud platform and data analysis layer. The cloud platform and data analysis layer comprise a data fusion module for processing multi-source data, a linear deformation analysis module for identifying linear deformation characteristics of the side slope, and an AI intelligent early warning module for risk prediction. The application display layer is used to dynamically display linear deformation characteristic attribute information, geometric shape polynomial curves and early warning information generated by the cloud platform and data analysis layer.

[0007] Further, the reference subsystem is composed of a base station array arranged at stable reference points, and a reference prism and a GNSS receiver are installed on the base station array as absolute references, which are used as absolute coordinate references and deformation reference systems of the entire monitoring system. The monitoring subsystem is composed of a plurality of intelligent targets arranged on the side slope surface, which are used to represent the deformation of the slope body.

[0008] Further, each intelligent target of the monitoring subsystem is internally integrated with: A prism: for automatic aiming of the total station; An RFID tag: with a unique ID code for target identification and rapid inventory; Bluetooth module: for short-range wireless wake-up and communication of the targets; Battery: power supply component for providing continuous power for the RFID tags and Bluetooth modules.

[0009] Further, the data acquisition and transmission layer includes: Network of measurement robots: composed of automated total stations, which communicate with the system host through wireless network bridges, cyclically measure the angle and distance of the base station array of the reference subsystem and the intelligent targets of the monitoring subsystem, and obtain their three-dimensional coordinate data; Close-range photogrammetry module: coaxially installs industrial cameras at the measurement robot station, periodically collects high-definition images of the slope, and uses the images to shoot the slope in extreme weather, and calculates the two-dimensional displacement of each intelligent target from the continuous image sequence; Environmental parameter sensing module: real-time monitoring of environmental parameters such as precipitation, soil moisture, temperature, and humidity of the slope; Intelligent edge gateway: responsible for collecting data from total stations, industrial cameras, and sensor arrays, and transmitting encrypted data to the cloud platform and data analysis layer through wireless networks.

[0010] Further, the data fusion module in the cloud platform and data analysis layer is configured to perform the following data fusion process: Data alignment and input: receives three-dimensional coordinate data from each intelligent target of the measurement robot, and two-dimensional displacement of each intelligent target from the close-range photogrammetry module, and aligns the two types of data on the timestamp; Constructing state space model: taking the state vector of the previous moment as the system state and the measurement data of the current moment as the observation value, a state space model of Kalman filter is constructed; Adaptive filtering: in the filtering process, the observation residual covariance of each measurement sensor is calculated in real time, and the observation noise matrix of the filter is adjusted accordingly; Three-dimensional estimated output: through the prediction and update cycle of Kalman filter, the three-dimensional displacement vector sequence of each intelligent target after fusion processing is output.

[0011] Further, the linear deformation analysis module is configured to perform a process of identifying and outlining continuous linear deformation features from discrete point displacement data, which takes the three-dimensional displacement vector sequence of each intelligent target point from the data fusion module as input, and outputs the geometric information of the linear feature describing potential anomalies, and the process is specifically: Full-field displacement and strain field construction based on Kriging interpolation: receives the three-dimensional displacement vector of each intelligent target point output by the data fusion engine, and for the horizontal displacement vector of the i-th target point in the three-dimensional displacement vector ​ , This represents the east-west displacement component. The north-south displacement component has the following coordinates: For any point to be determined within the slope area Its displacement components and It is obtained through the Kriging interpolation algorithm, that is ,in, It is a point The value of the function to be estimated, It is a known point The measured value, These are the weights assigned to each known point, resulting in two continuous displacement field functions for the entire computational domain: and Based on the full-field continuous displacement function and Build In-plane differential components And calculate the maximum shear strain value. ; Deformation anomaly identification based on strain field and vector field analysis: based on maximum shear strain distribution and horizontal displacement vector field Set a shear strain threshold based on historical data. , will satisfy All pixels and regions were initially identified as strain anomaly areas, constituting a candidate feature point set for potential linear deformation features. Each candidate point Its location is attached. Shear strain values ​​and divergence values; Linear feature clustering and spatial curve fitting: Using the Euclidean distance as a metric, candidate points that are spatially close are grouped into the same cluster. Each cluster Represents an independent, continuous transformation, for each cluster obtained from clustering. , and the points As data points, the least squares method is used to fit the geometric polynomial curve of the linear deformation feature. Based on the fitted curve and the original data, the attributes of the linear feature are calculated, including the geographic coordinate sequence and total length. Mean shear strain Average displacement Construct a list of linear features from all the attributes of the curve features. Each item in the list contains all the geometric and attribute information for a linear deformation feature.

[0012] Further, the AI intelligent early warning module is configured as a time series prediction and decision engine based on a long short-term memory network. The module is driven by linear deformation feature attribute information and real-time environmental data, and performs the following processes: receiving attribute information of the identified linear features, three-dimensional displacement vector sequences of the data fusion module, and environmental parameter sequences in the environmental parameter perception module; aligning and fusing the received data to construct a unified multi-dimensional input feature vector for predicting at any time ; inputting the multi-dimensional input feature vector into a long short-term memory network model, which is composed of an input layer, multiple LSTM layers, and a fully connected output layer. The input feature vector is mapped to the prediction output, i.e. the displacement amount of the monitoring point after , the displacement rate after

[0013] by the fully connected output layer. Further, the AI intelligent early warning module is preconfigured with multiple early warning thresholds, including displacement threshold , , and displacement rate threshold , . The predicted values output by the long short-term memory network model are compared with the thresholds in real time: when or , L1-level early warning is triggered; when or , L2-level early warning is triggered; Once the early warning is triggered, the system automatically generates an early warning report and pushes the early warning level, predicted value, and affected area to the application display layer through SMS, email, and the platform's built-in message center.

[0014] Compared with the prior art, the loess tunnel portal side slope linear deformation precision monitoring system has the following beneficial effects: One, the application reconstructs the displacement data of discrete monitoring points into continuous full-field displacement and strain distribution by Kriging interpolation, then analyzes the displacement vector field as a whole, locates the abnormal area of deformation gradient mutation by calculating the maximum shear strain field, merges these abnormal point sets into independent clusters according to spatial proximity and deformation consistency, and accurately outlines the geometric trajectory of the potential slip surface or crack through curve fitting, which fundamentally changes the limitations of point monitoring, can non-contact and automatically identify the hidden linear damage trend that is difficult to find with the naked eye, provides fine data support for slope stability evaluation, and makes the early warning from blind to accurate.

[0015] Other advantages, objects, and features of the application will be set forth in part in the following specification, and in part will become apparent to those skilled in the art from a consideration of the following specification, or can be learned from practice of the application. BRIEF DESCRIPTION OF DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiment or prior art description will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application, and those skilled in the art can obtain other drawings according to these drawings without creative labor.

[0017] Figure 1 The operation flowchart of the linear deformation precision monitoring system for the loess tunnel portal side slope; Figure 2 The flowchart of linear deformation feature identification in the linear deformation precision monitoring system for the loess tunnel portal side slope; Figure 3 The composition schematic diagram of the linear deformation precision monitoring system for the loess tunnel portal side slope. DETAILED DESCRIPTION

[0018] In order to further illustrate the technical means and effects adopted by the present application to achieve the predetermined application purpose, the specific embodiments, structures, features and effects according to the present application will be described in detail below in combination with the drawings and preferred embodiments.

[0019] Embodiment one This embodiment is aimed at the problem that the loess tunnel portal side slope is prone to instability due to special geological characteristics and external action, and relies on the linear deformation precision monitoring system to carry out application, such as Figure 3As shown, the system acquires basic data through the sensor network layer, completes multi-source data collection and transmission through the data collection and transmission layer, realizes data fusion, deformation identification and intelligent early warning through the cloud platform and data analysis layer, and finally presents key information through the application display layer, breaking through the limitations of traditional monitoring discretization, weak anti-interference, etc., and providing technical support for tunnel engineering safety.

[0020] (I) Sensor network layer deployment and work The sensor network layer includes a reference subsystem and a monitoring subsystem, wherein: Reference subsystem Deployment and equipment: survey the tunnel portal periphery, select a stable area to deploy a base station array, install a reference prism and a GNSS receiver at each reference point, ensure that the prism center and the reference point coordinates are aligned, and the receiver signal is stable. The reference prism provides a measurement reference for the subsequent measurement robot, i.e., an automated total station. When the measurement robot measures the coordinates of the reference prism while measuring the coordinates of the intelligent target of the monitoring subsystem, it can determine whether there is an error in the measurement process by comparing the real-time measurement coordinates of the reference prism with the initial calibrated absolute coordinates, and then correct the measurement data of the monitoring subsystem to ensure the accuracy of the measurement results. The GNSS receiver continuously receives satellite signals, calculates and updates the absolute coordinates of the reference point in real time, and can capture any small changes in the reference point coordinates caused by the surrounding environment, providing a dynamic updated absolute coordinate reference and deformation reference system for the entire monitoring system, avoiding the impact of reference point coordinate drift on the reliability of the monitoring data.

[0021] Monitoring subsystem Deployment and equipment: survey the slope surface, deploy intelligent targets, the targets have built-in prisms, RFID tags, Bluetooth modules, and batteries. The prism provides automatic aiming for the total station, and the displacement of the prism is synchronized with the deformation of the slope body, providing a target point for coordinate measurement. The RFID tag contains a unique ID and is associated with the target position information. It can be identified and inventoried through a card reader. The Bluetooth module supports short-range wake-up and parameter configuration, and daily sleep reduces power consumption. The battery powers the tag and the Bluetooth module, and is suitable for outdoor environments to ensure battery life.

[0022] (II) Data collection and transmission layer operation The data collection and transmission layer connects the sensor network and the cloud platform, including the measurement robot network, the close-range photogrammetry module, the environmental parameter sensing module, and the intelligent edge gateway, wherein: Measurement robot network Collection process: composed of an automated total station, communicates with the system host through a wireless bridge, measures the reference prism coordinates first, then measures the angles and distances of the intelligent targets in sequence, and calculates the three-dimensional coordinates of the targets using the triangulation principle combined with the total station's own coordinates.

[0023] Data transmission: After the measurement is completed, the total station transmits the data to the host via a wireless bridge. The host verifies the data format and integrity, and abnormal data triggers retransmission or is marked as suspicious.

[0024] Close-range photogrammetry module Installation and data acquisition: An industrial camera is coaxially mounted at the total station site and its parameters are calibrated. Under extreme weather conditions, the slope is photographed at a preset frequency to obtain a high-definition image sequence with timestamps.

[0025] Data processing and transmission: After the camera compresses and processes the image, it is transmitted to the host. The host checks the image clarity and timestamp. Invalid images are not included in subsequent processing.

[0026] Environmental parameter sensing module Deployment and data collection: Deploy sensor arrays in the slope area. Place precipitation sensors on the top of the slope where there is no obstruction. Deploy soil moisture sensors according to soil depth and area. Place temperature and humidity sensors in a well-ventilated and shaded place. Collect data according to the preset cycle and add timestamps.

[0027] Data transmission: The sensor transmits data to the host via wired or wireless means. The host verifies the validity of the data, marks abnormal data, and associates it with sensor information for maintenance.

[0028] Smart Edge Gateway Data aggregation and encryption: Receive multi-source data forwarded by the host: coordinates, images, environmental parameters, classify and store them according to type and source, check their integrity, encrypt the data and generate data packets with verification codes.

[0029] Transmission to the cloud platform: Encrypted data packets are transmitted to the cloud platform via a dedicated wireless network. The transmission status is monitored in real time, and parameters are adjusted to ensure stable and efficient data transmission. After receiving the data, the cloud platform decrypts and verifies the code, and then parses and stores the data.

[0030] (III) Cloud Platform and Data Analysis Layer Processing The cloud platform and data analysis layer are responsible for processing, analyzing, and providing intelligent early warnings of multi-source data transmitted from the data acquisition and transmission layer. It consists of a data fusion module, a linear deformation analysis module, and an AI intelligent early warning module. Its specific operation process is as follows: Data fusion module The data fusion module receives two types of data from the cloud platform storage unit: one type is the 3D coordinate data of the smart targets from the measurement robot network, containing data for each smart target. In timestamp Three-dimensional coordinates ,in East-west coordinates North-South coordinates For the height coordinate, the data directly reflects the space of the target; the second is the intelligent target two-dimensional displacement data from the close-range photogrammetry module, which is calculated by matching and analyzing the continuous image sequence to obtain the three-dimensional displacement vector of each target at the time stamp The horizontal displacement component of the lower phase relative to the initial moment , where is the east-west displacement, is the north-south displacement, this data can supplement the total station data under extreme weather, but lacks elevation information and the accuracy is affected by image resolution, and the timestamps of the measurement robot network and the close-range photogrammetry module are aligned.

[0031] Model construction and filtering: taking the "spatial state of intelligent target" as the core, the system state vector is defined as the three-dimensional displacement vector and displacement rate of the target at the previous moment , that is: , where (east-west cumulative displacement), (north-south cumulative displacement), (elevation cumulative displacement), , , , respectively, the displacement rate of the corresponding direction, the observation vector is defined as the multi-source measurement data at the current moment , that is: , where (total station calculated east-west displacement), (total station calculated north-south displacement), (total station calculated elevation displacement), , is the horizontal displacement data of close-range photogrammetry, through the state transition matrix , the displacement evolution law of the target from to moment, and the observation matrix , the mapping relationship between system state and observation data is established, and the model system equation is constructed: , where is the system noise, which is subject to Gaussian distribution with mean 0 and covariance , the observation equation: , where is the observation noise, which is subject to Gaussian distribution with mean 0 and covariance , the observation noise covariance of traditional Kalman filtering is a fixed value, which cannot adapt to the dynamic changes of the measurement environment, this module realizes self-adaptive adjustment by real-time calculation of observation residual covariance: first, calculate the observation residual , where is based on​ Prediction of state at time t; then calculate the residual error covariance matrix by sliding window method ; finally, update the observation noise covariance as the weighted value of , which ensures that the filtering process can dynamically adjust the trust degree of different data sources according to the actual measurement accuracy.

[0032] Three-dimensional estimated output Perform the "prediction-update" cycle of Kalman filtering: first, predict the state estimation value at time t by the system equation , and calculate the prediction error covariance ; then calculate the Kalman gain , which reflects the contribution degree of observation data to state correction; finally, update the state estimation value by observation data, and update the error covariance , where is the unit matrix, and the three-dimensional displacement component is extracted from the updated state vector to obtain the fused three-dimensional displacement vector of each intelligent target at time t . Arrange all displacement vectors under different time stamps by target number to form a continuous three-dimensional displacement vector sequence , which has both the three-dimensional integrity of the total station and the anti-interference of close-range photogrammetry, providing a high-precision data basis for subsequent linear deformation analysis.

[0033] Linear deformation analysis module The linear deformation analysis module converts the displacement data of discrete intelligent targets into a continuous slope deformation field through Kriging interpolation, strain field analysis, clustering, and curve fitting, and accurately identifies potential sliding surfaces, cracks, and other linear deformation features. The specific steps are as follows: Full-field displacement and strain field construction based on Kriging interpolation Clearly input data: three-dimensional displacement vector sequence of each intelligent target from the data fusion module, take the horizontal displacement component as the analysis object, the horizontal displacement vector of each target is , and the corresponding plane coordinates are (determined by the initial measurement of the total station, consistent with the geographical coordinates of the slope surface), use Kriging interpolation algorithm to construct the full-field continuous displacement field: for any one point to be solved in the slope area , the calculation of its horizontal displacement components (east-west direction) and (north-south direction) is , where represents ​Displacement component to be estimated for a point or ), Displacement component measurement value of a known target or ), ), is the weight assigned to each target , and satisfies By performing the interpolation on all grid points in the slope region, the continuous horizontal displacement field function of the entire slope is obtained and .

[0034] Strain field calculation based on displacement field: the strain field reflects the deformation difference inside the slope and is the key to identifying potential damage areas. In the plane (the slope surface plane), three basic strain components are calculated: the east-west direction normal strain , the north-south direction normal strain , and the shear strain ; further, the maximum shear strain is calculated: the maximum shear strain directly reflects the risk of shear failure of the slope, and the formula is , The larger the value, the more intense the shear deformation of the point, and the more likely it is to occur.

[0035] Deformation anomaly area identification based on strain field and vector field analysis Set the shear strain threshold When , it means that the shear deformation of the point exceeds the stable range, and there is a risk of damage. Traverse the strain field grid points of the entire slope, and mark all points that satisfy as "strain anomaly points". These strain anomaly points constitute a candidate feature point set of potential linear deformation features Each candidate point is accompanied by its position coordinates , shear strain value , and divergence value (the divergence value reflects the convergence or divergence of local deformation, a positive value corresponds to crack expansion; a negative value corresponds to extrusion deformation).

[0036] Linear feature clustering and spatial curve fitting Perform candidate point clustering: take "spatial proximity" as the clustering criterion, use Euclidean distance as the metric, and set the distance threshold , and merge points in the candidate point set whose Euclidean distance is less than into the same cluster Each cluster Representing an independent, continuous deformation region (such as a potential slip surface or a crack), for each cluster , with all the points it contains ( Using data points as the base, a polynomial curve is fitted using the least squares method to describe the geometric shape of the linear deformation characteristics. The polynomial degree is determined based on the shape complexity of the cluster. The fitting objective is to minimize the sum of squared residuals from all data points to the curve. Based on the fitted curve and the original data, each cluster is extracted. The corresponding linear deformation feature attributes include: geographic coordinate sequence and total length. Mean shear strain Average displacement All clusters The linear deformation feature attributes are organized into a list. Each item in the list corresponds to a linear deformation feature, which includes its geometric shape (fitting curve parameters, coordinate sequence, total length) and deformation intensity information, providing feature input for AI intelligent early warning.

[0037] AI Intelligent Early Warning Module The AI-powered intelligent early warning module uses a Long Short-Term Memory (LSTM) network as its core algorithm, integrating linear deformation feature data with environmental parameter data to accurately predict the deformation trend of slopes over a future period. It also triggers early warnings through multi-level thresholds, providing a time window for engineering decisions. The specific operation process is as follows: Data reception and filtering: The system receives three types of core input data: first, linear feature attribute data from the linear deformation analysis module, including the total length of all linear deformation features. Mean shear strain Average displacement The data includes: 1) geographic coordinate sequence; 2) three-dimensional displacement vector sequence of intelligent targets from the data fusion module, which selects displacement data of 5-10 key targets around the linear deformation feature, whose deformation can directly reflect the evolution trend of the linear feature; and 3) real-time environmental parameter sequence from the environmental parameter perception module, including precipitation, soil moisture content, atmospheric temperature, and relative humidity, which are key external factors that induce loess slope deformation.

[0038] Multidimensional feature vector construction: based on "prediction time" "as the goal" To predict the time window, the preprocessed data of various types are aligned according to the time dimension to construct the time frame. Unified multidimensional input feature vector .

[0039] LSTM Network Structure Design and Model Training: The multi-dimensional input feature vector is input into a Long Short-Term Memory (LSTM) network model. The LSTM network consists of an input layer, multiple LSTM layers, and a fully connected output layer. The input feature vector is mapped to the predicted output through the fully connected output layer. Displacement of monitoring points after time Displacement rate .

[0040] Multi-level warning threshold comparison: The AI ​​intelligent warning module has multiple preset warning thresholds, including displacement thresholds. , and displacement rate threshold , The predicted values ​​output by the Long Short-Term Memory network model and Real-time comparison with threshold: when or If so, an L1 level warning will be triggered; when or If so, an L2 level warning will be triggered; Once an alert is triggered, the system automatically generates an alert report and pushes the alert level, forecast value, and affected area to the application display layer via SMS, email, and the platform's built-in message center.

[0041] (iv) Application Presentation Layer The application presentation layer, serving as the system's "interactive window," visualizes monitoring data and analysis results, providing users with convenient query and management functions. This includes both web and mobile visualization platforms, which, together with the cloud platform and data analysis layer, form a "data output-visualization" linkage. Platform base: Reconstructs the slope terrain, tunnel structure, and monitoring equipment in three-dimensional form to achieve spatial positioning; Data Display: Displays device status, supports timeline backtracking, regional filtering, and early warning record query.

[0042] In summary, this embodiment, through understanding the working principle of the precise monitoring system for linear deformation of loess tunnel entrance slopes, establishes a closed-loop monitoring system encompassing "data acquisition, fusion analysis, feature recognition, and intelligent early warning." This system can not only accurately identify hidden linear deformation features of loess slopes, such as slip surfaces and cracks, but also predict deformation trends in advance, providing ample time for emergency response. It effectively improves the accuracy and intelligence level of monitoring loess tunnel entrance slopes, offering a complete technical solution for safety monitoring in similar geotechnical engineering projects.

[0043] Example 2 Based on Example 1, such as Figure 1 As shown in the figure, this embodiment provides the specific steps for the precise monitoring system of linear deformation of the slope at the entrance of a loess tunnel to monitor the linear deformation of the slope. The specific steps are as follows: (1) System initialization and deployment Establish a benchmark subsystem on the stable bedrock / foundation behind the slope: install observation piers, and set up ultra-high precision prisms or GNSS receivers on the piers.

[0044] Install a monitoring subsystem: Deploy smart targets at selected locations. Each target includes a prism, RFID tag, Bluetooth module, and power supply components.

[0045] A measuring robot was set up and calibrated at a stable location near the opening, and an industrial camera was coaxially mounted on it.

[0046] Micro-weather stations, including sensors for rainfall, soil moisture, and temperature and humidity, were installed on the slope.

[0047] Deploy intelligent edge gateways and configure them to communicate with measurement robots, industrial cameras, weather stations, and cloud platforms.

[0048] (2) Automated data acquisition The measuring robot initiates scanning: according to a preset schedule or trigger command, it automatically and cyclically aims at the prisms of the reference subsystem and all smart targets of the monitoring subsystem.

[0049] Total station data acquisition: The surveying robot records the angle and distance measurements of each prism and calculates its high-precision three-dimensional coordinates.

[0050] Close-range photography-assisted shooting: During the scanning intervals of the measurement robot or in extreme weather, the industrial camera automatically captures a high-definition image sequence of the slope, and camera calibration ensures that the image is associated with the measurement coordinate system.

[0051] Simultaneous collection of environmental parameters: The micro-weather station continuously records precipitation, soil moisture content, temperature, and humidity data in real time.

[0052] Bluetooth wake-up and identification: When necessary, wake up the target via Bluetooth for communication or quick inventory.

[0053] (3) Data transmission and edge processing Data aggregation: The intelligent edge gateway receives coordinate data from the measuring robot, image data from the industrial camera, and environmental data from the weather station in real time.

[0054] Edge preprocessing: The gateway performs preliminary filtering, compression, and timestamp alignment on the raw data.

[0055] Encrypted transmission: The gateway encrypts and transmits the pre-processed data packets to the cloud platform via wireless / wired network.

[0056] (4) Cloud-based data fusion processing Data reception and alignment: The cloud platform receives data to ensure that the total station coordinate data and image data are accurately synchronized in time.

[0057] Adaptive data fusion: The three-dimensional coordinates of discrete points obtained by the total station are fused with the two-dimensional displacements calculated from the image sequence. An adaptive filtering algorithm is used to dynamically process the data, eliminate errors, and output a more reliable and accurate three-dimensional displacement vector sequence for each smart target.

[0058] (5) Linear deformation feature recognition (e.g.) Figure 2 (As shown) Construction of full-field displacement and strain field: Based on the fused displacement data of all smart targets, a continuous displacement field distribution covering the entire slope area is generated through spatial interpolation algorithm. Based on the continuous displacement field, the strain field distribution of the entire slope, especially the maximum shear strain field, is calculated.

[0059] Deformation anomaly zone identification: Analyze the displacement vector field and the maximum shear strain field, set a strain threshold, identify the region where the shear strain exceeds the threshold, and mark it as a potential linear deformation feature candidate point.

[0060] Feature point clustering and curve fitting: Candidate points that are spatially close and have similar strain characteristics are grouped into independent clusters.

[0061] For each set of points within a cluster, a curve fitting algorithm is used to accurately depict the geometric shape and spatial trajectory of the linear deformation feature.

[0062] Feature attribute calculation: Calculate the attribute information of each fitted curve: geographic coordinate sequence, total length, average shear strain value, average displacement, etc., and generate a list of linear deformation features, which includes all identified features and their attributes.

[0063] (6) AI-powered intelligent prediction and early warning Data preparation: Receive linear deformation characteristic attribute information, displacement vector sequence of each target, and real-time environmental parameters.

[0064] LSTM time series prediction: The above multi-source data are fused into a unified time series input feature vector, which is then input into a trained Long Short-Term Memory (LSTM) network model. The model predicts the displacement and displacement rate of key monitoring points within a specific time period Δt in the future.

[0065] Multi-level early warning decision: The predicted displacement and displacement rate are compared with the preset multi-level early warning thresholds in real time. If the predicted value exceeds the threshold, the corresponding level of early warning is triggered.

[0066] Early warning information generation and pushing: automatically generate early warning report containing early warning level, prediction value, affected area, and suggested measures.

[0067] Push early warning information to relevant managers in real time through SMS, email, platform message center, etc.

[0068] (7) Application display and interaction Three-dimensional visualization display: based on BIM+GIS three-dimensional real scene model on Web / mobile platform.

[0069] Dynamic presentation: real-time or near real-time display of displacement nephogram and strain nephogram of slope, superimposed display of identified linear deformation feature geometric track, and eye-catching prompt of early warning information.

[0070] Interactive query: users can click to view detailed attributes, historical evolution, and associated environmental data of any linear feature.

[0071] Device state monitoring: display online state and power of each device in the sensor network layer, i.e. target, measuring robot, and gateway.

[0072] The above is only the preferred embodiment of the present application, and does not limit the present application in any form. Although the present application has been disclosed as the above preferred embodiment, it is not intended to limit the present application. Any person skilled in the art can make some changes or modifications to the above disclosed technical content without departing from the scope of the present application, and any simple modification, equivalent change and modification of the above embodiment based on the technical essence of the present application are still within the scope of the present application.

Claims

1. A loess tunnel portal side slope linear deformation precision monitoring system, characterized in that, The system comprises a sensing network layer, a data acquisition and transmission layer, a cloud platform and data analysis layer, and an application display layer. The sensing network layer comprises a reference subsystem arranged in a stable area and a monitoring subsystem arranged in a to-be-monitored high slope area, and is arranged in the to-be-monitored high slope area to provide a measurement reference and a monitoring object for spatial deformation. The data acquisition and transmission layer communicates with the sensing network layer, comprises a measurement robot network composed of an automatic total station, a close-range photogrammetry module, an environmental parameter sensing module and an intelligent edge gateway, and is used for automatically acquiring three-dimensional spatial data and on-site environmental parameters of the monitoring object in the sensing network layer and transmitting the data to the cloud platform and data analysis layer. The cloud platform and data analysis layer comprises a data fusion module for processing multi-source data, a linear deformation analysis module for identifying linear deformation characteristics of a slope, and an AI intelligent early warning module for risk prediction, the data fusion module fuses three-dimensional coordinate data and two-dimensional displacement data into a three-dimensional displacement vector sequence through Kalman adaptive filtering, the linear deformation analysis module constructs a full-field displacement and strain field based on Kriging interpolation, identifies an abnormal area through a maximum shear strain threshold, and outlines a geometric trajectory of a potential sliding surface and a crack through clustering and curve fitting, and the AI intelligent early warning module predicts displacement and rate based on a long short-term memory network and triggers multi-level early warning. The application display layer is used for dynamically displaying linear deformation characteristic attribute information, a geometric shape polynomial curve and early warning information generated by the cloud platform and data analysis layer.

2. The loess tunnel portal side slope linear deformation precision monitoring system according to claim 1, characterized in that, The reference subsystem is composed of a base station array arranged at stable reference points, and a reference prism and a GNSS receiver are installed on the base station array as an absolute reference, and are used as an absolute coordinate reference and a deformation reference system of the entire monitoring system. The monitoring subsystem is composed of a plurality of intelligent targets arranged on a slope surface of a high slope, and is used to represent deformation of the slope.

3. The loess tunnel portal side slope linear deformation precision monitoring system according to claim 2, characterized in that, Each intelligent target of the monitoring subsystem is internally integrated with: a prism for automatic aiming of a total station; an RFID tag with a unique ID code for identification and rapid inventory of the target; a Bluetooth module for short-range wireless wake-up and communication of the target; a battery as a power supply component for providing continuous power supply for the RFID tag and the Bluetooth module.

4. The loess tunnel portal side slope linear deformation precision monitoring system according to claim 1, characterized in that, The data acquisition and transmission layer comprises: a measurement robot network composed of an automatic total station, the total station communicates with a system host through a wireless network bridge, and cyclically measures angles and distances of a base station array of the reference subsystem and intelligent targets of the monitoring subsystem to obtain three-dimensional coordinate data thereof; a close-range photogrammetry module, an industrial camera is coaxially installed on a measurement robot station, and high-definition images of the slope are periodically acquired, and the slope is photographed in extreme weather, and two-dimensional displacement amounts of the intelligent targets are calculated from a continuous image sequence; an environmental parameter sensing module, which monitors environmental parameters such as precipitation, soil moisture, temperature and humidity of the high slope in real time through a sensor array. Intelligent edge gateway: responsible for collecting data from total station, industrial camera, sensor array, and transmitting data to cloud platform and data analysis layer through wireless network after encryption.

5. The loess tunnel portal edge batter slope linear deformation precision monitoring system according to claim 1, characterized in that, The data fusion module in the cloud platform and data analysis layer is configured to perform the following data fusion process: Data alignment and input: receive three-dimensional coordinate data from each intelligent target of the measuring robot, and two-dimensional displacement data from each intelligent target of the close-range photogrammetry module, and align the two types of data on the timestamp; Constructing state space model: taking the state vector of the previous time as the system state and the measurement data of the current time as the observation value, a state space model of Kalman filter is constructed; Adaptive filtering: in the filtering process, the observation residual covariance of each measurement sensor is calculated in real time, and the observation noise matrix of the filter is adjusted accordingly; Three-dimensional prediction output: through the prediction and update cycle of Kalman filter, the three-dimensional displacement vector sequence of each intelligent target after fusion processing is output.

6. The loess tunnel portal edge batter slope linear deformation precision monitoring system according to claim 1, characterized in that, The linear deformation analysis module is configured to perform a process of identifying and outlining continuous linear deformation features from discrete point displacement data, with the input being the three-dimensional displacement vector sequence of each intelligent target point from the data fusion module, and the output being the geometric information of the linear feature describing potential anomalies, and the process specifically being: Construction of the full-field displacement and strain field based on Kriging interpolation: Receive the three-dimensional displacement vectors of each intelligent target point output by the data fusion engine; for the third displacement vector... The horizontal displacement vector of each target point is , This represents the east-west displacement component. The north-south displacement component has the following coordinates: For any point to be determined within the slope area Its displacement components and It is obtained through the Kriging interpolation algorithm, that is ,in, It is a point The value of the function to be estimated, It is a known point The measured value, These are the weights assigned to each known point, resulting in two continuous displacement field functions for the entire computational domain: and Based on the full-field continuous displacement function and Build In-plane differential components And calculate the maximum shear strain value. ; Deformation anomaly area identification based on strain field and vector field analysis: based on maximum shear strain distribution and horizontal displacement vector field , set the shear strain threshold value based on historical data , all pixel points and regions that meet are preliminarily identified as strain anomaly areas, and constitute a candidate feature point set of potential linear deformation features Each candidate point is attached with its position , shear strain value and divergence value; Linear feature clustering and spatial curve fitting: Using the Euclidean distance as a metric, candidate points that are spatially close are grouped into the same cluster. Each cluster Represents an independent, continuous transformation, for each cluster obtained from clustering. , and the points As data points, the least squares method is used to fit the geometric polynomial curve of the linear deformation feature. Based on the fitted curve and the original data, the attributes of the linear feature are calculated, including the geographic coordinate sequence and total length. Mean shear strain Average displacement Construct a list of linear features from all the attributes of the curve features. Each item in the list contains all the geometric and attribute information for a linear deformation feature.

7. The loess tunnel portal side slope linear deformation precision monitoring system according to claim 6, characterized in that, The AI intelligent early warning module is configured as a time series prediction and decision engine based on long short-term memory network, which takes linear deformation feature attribute information and real-time environmental data as the driving force and performs the following process: Receive the attribute information of each linear feature identified, the three-dimensional displacement vector sequence of the data fusion module, and the environmental parameter sequence in the environmental parameter perception module; The received data is aligned and fused to construct a unified multi-dimensional input feature vector for predicting the time ​ The multi-dimensional input feature vector is input into a long short-term memory network model, the long short-term memory network is composed of an input layer, multiple LSTM layers and a fully connected output layer, and for the input feature vector, the fully connected output layer is mapped to a prediction output, that is The displacement amount of the monitoring point after the time , displacement rate .

8. The loess tunnel portal side slope linear deformation precision monitoring system according to claim 7, characterized in that, The AI ​​intelligent early warning module has multiple preset early warning thresholds, including displacement thresholds. , and displacement rate threshold , The predicted values ​​output by the Long Short-Term Memory network model and Real-time comparison with threshold: When or a L1 level warning is triggered; When or a L2 level warning is triggered; Once the early warning is triggered, the system automatically generates an early warning report and pushes the early warning level, predicted value, and affected area to the application display layer through SMS, email, and platform built-in message center.

Images