Intelligent sensor-based roadbed settlement monitoring, analysis and early warning system

Abstract

The application discloses a pavement and roadbed settlement monitoring, analyzing, early warning system based on intelligent sensors, relates to the technical field of monitoring and analysis, and comprises an initial point screening unit, a point set optimization unit, a virtual model generation unit, a virtual model updating unit, an influence factor analyzing unit, a settlement effect analyzing unit and an abnormal settlement early warning unit. According to the prospective numerical simulation of dynamic updating geological parameters and load working conditions, the application can accurately identify the most sensitive potential risk area of the mechanical response before the implementation of the key construction process, and adaptively optimize the space-time layout of the monitoring resources accordingly, so that the monitoring is changed from passive response to active early warning, and a decision window is provided for the early discovery and early intervention of the core risk.

Description

Technical Field

[0001] This invention relates to the field of monitoring and analysis technology, specifically to a road and subgrade settlement monitoring, analysis and early warning system based on intelligent sensors. Background Technology

[0002] A road and subgrade settlement monitoring, analysis, and early warning system based on intelligent sensors is disclosed in patent application number 202210585785.8. This system predicts monitoring points based on geological parameters and numerical simulations, adaptively collects data, and uses digital twins and causal analysis to diagnose and warn of settlement mechanisms. The system includes a monitoring area division module, a monitoring area structural layer division module, a monitoring terminal deployment module, a monitoring area construction quality analysis module, a monitoring area pavement bearing capacity analysis module, a groundwater monitoring and analysis module, an information storage module, a subgrade settlement hazard analysis module, a subgrade settlement early warning analysis module, and a settlement early warning prompt execution module. By automatically collecting and analyzing parameters required for road and subgrade settlement analysis through monitoring terminals, weight sensors, and intelligent high-definition cameras, it not only avoids the waste of significant human and material resources and improves the accuracy of measurement data, but also greatly enhances the reliability and scientific basis of the analysis results. This compensates for the one-sidedness of current road and subgrade analysis results, providing effective data support for the analysis results and improving their rationality and reference value.

[0003] The aforementioned existing technologies have solved problems such as the inability to know the settlement of road surface and subgrade in a timely manner. However, during operation, the monitoring deployment relies on experience, which can easily miss real risk points, making it difficult to issue timely warnings. Furthermore, the settlement attribution relies on human experience, the mechanism is unclear, and the treatment measures are not targeted, often leading to repeated treatment and increased costs. At the same time, the allocation of monitoring resources cannot be dynamically adjusted with the migration of risks, resulting in insufficient monitoring in key areas. Summary of the Invention

[0004] The purpose of this invention is to provide a road and subgrade settlement monitoring, analysis and early warning system based on intelligent sensors to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a road and subgrade settlement monitoring, analysis, and early warning system based on intelligent sensors, comprising:

[0006] The initial point screening unit is used to read the geological model parameter set and numerical simulation load case set of the current construction stage, generate the initial sensitive monitoring point set, and construct the objective function and constraints.

[0007] The point set optimization unit is used to select the optimal sensor deployment scheme using the initial sensitive monitoring point set, objective function and constraints, and to correct the corresponding geological model parameters based on the actual monitoring data within the current construction phase, so as to update the initial sensitive monitoring point set for the next construction phase.

[0008] The virtual model generation unit is used to read multi-source observation data and construct a virtual roadbed model using geological models and multi-source observation data;

[0009] The virtual model update unit is used to receive multi-source observation data in the latest time period, analyze the optimal estimated state, the internal stress field, strain field and displacement field at the next moment using the observation data, and integrate them to form a new virtual roadbed model.

[0010] The impact factor analysis unit is used to directly calculate the settlement observation value obtained from the monitoring data, and simultaneously extract the impact factor observation values ​​at the same location, and calculate the contribution coefficient of each impact factor relative to the settlement value.

[0011] The settlement effect analysis unit is used to calculate the settlement change value within each construction procedure and determine the average deviation of the change value.

[0012] The abnormal settlement early warning unit is used to acquire real-time monitoring data streams, determine the location and time of abnormal points, analyze them, obtain the root cause pattern probability distribution of the current abnormal points, use virtual models to perform inferences, analyze the actual root causes according to the inference results, and generate early warning prompts.

[0013] Preferably, the initial point screening unit includes a control equation construction module, an initial point set generation module, a weight coefficient calculation module, and an objective function determination module;

[0014] The control equation construction module reads the geological model parameter set and numerical simulation load case set of the current construction stage, establishes control equations based on the geological model parameter set and numerical simulation load case set, and calculates global strain and stress.

[0015] The specific process of constructing the geological model parameter set and the numerical simulation load case set is as follows:

[0016] S101. After obtaining the original survey data of the engineering site, the geological modeling engine is used to process the survey data to form a three-dimensional geological body object set and attribute field set. The two are then parameterized to determine the geological BIM model parameter set.

[0017] S102. After obtaining engineering design specifications, construction organization plans and environmental monitoring data, the load analysis engine is used to process them to form a construction sequence load set, an operational dynamic load set and a specification combination load set. Through parametric and structured integration, the numerical simulation load case set is determined.

[0018] The specific process for calculating global strain and stress is as follows:

[0019] S201. Discretize the geological model into a set of finite elements. Each element contains nodal coordinates, material parameters, shape functions, and a mapping relationship between element node numbers and global node numbers. Perform weighted residual discretization on the governing equations to determine the stiffness matrix and load vector of each element. Assemble them into a global stiffness matrix and load vector.

[0020] S202. After solving for the global nodal displacement vector using the global stiffness matrix and load vector, statistically analyze the element strain. and stress ,according to and Calculate global strain and stress ,in , Indicates the first The shape function of each unit, Represents the node coordinates within the cell. Indicates the first strain per unit element Indicates the total number of units. , Indicates the first Unit stress, Indicates the serial number;

[0021] The initial point set generation module obtains the sensitive indicators of each node in the geological model through numerical simulation calculation and extrapolation. The sensitive indicators include the maximum principal stress, strain, safety factor and plastic state. Nodes exceeding the standard are screened according to preset thresholds, and the initial sensitive monitoring point set is formed after spatial clustering and priority sorting.

[0022] The weighting coefficient calculation module reads the initial sensitive monitoring point set. Current construction phase sequence Sensor type set and sensor performance parameters, among which , , Indicates the first One initial sensitive monitoring point, This indicates the initial total number of sensitive monitoring points. Indicates the first The start time of each construction phase Indicates the first The duration of each construction phase was calculated using real-time numerical simulations by applying the current construction load conditions to the geological model, and the weighting coefficients of each sensitive monitoring point were analyzed. ,in , Indicates the first Sensitive monitoring points during each construction phase The Normalized values ​​of several sensitive indicators Indicates the first The weight of each sensitive indicator in the current construction phase sequence. Indicates the serial number;

[0023] The objective function determination module constructs the objective function according to the weight coefficients of each sensitive monitoring point and sets the constraints.

[0024] The objective function is specifically:

[0025]

[0026]

[0027]

[0028] Among them, Co Indicates the time sequence during the construction phase. Total effective coverage Indicates sensitive monitoring points Construction phase sequence The weighting coefficients in the text Indicates sensitive monitoring points Are there sensors deployed in itself and its surrounding area? This indicates a sensitive monitoring point. If there are sensors deployed within itself and its neighborhood, This indicates a sensitive monitoring point. There are no sensors deployed in itself or its surrounding area. Indicates sensitive monitoring points The weighting coefficients, Represents the sensor deployment matrix. Indicates the time sequence during the construction phase. The overall redundancy of the internal deployment scheme, Indicates the serial number. This indicates the total number of sensitive monitoring points within the deployment plan. Indicates the first A sensitive point, Indicates the first All sensitive monitoring points within the spatial neighborhood of each sensitive monitoring point Point Whether it is covered by sensors, if , then represents a point If covered by sensors, , then represents a point Not covered by sensors Indicates the time sequence during the construction phase. Total cost of internal sensor deployment This indicates that one unit of type is deployed. The sensor cost value, Denotes the decision variable, if This indicates that at point Deployment type is The sensor, This indicates the number of sensor types.

[0029] Preferably, the point set optimization unit includes a scheme output module and an initial point set update module;

[0030] The scheme output module encodes the sensor deployment matrix into a binary chromosome according to the initial sensitive monitoring point set and sensor type set, randomly generates a set of deployment schemes that meet the constraints, and calculates the deployment scheme in each round of the loop. Influence ,in , This indicates the total effective coverage of the current deployment plan. This indicates the overall redundancy of the deployment plan. This represents the total cost of the deployment plan. This represents the control coefficient. These represent the preset maximum coverage, maximum redundancy, and maximum cost, respectively. The selection, crossover, and mutation operations are performed sequentially on all deployment schemes, and schemes that violate the constraints are repaired. When the number of iterations exceeds the maximum number of iterations, the optimal scheme is output.

[0031] The initial point set update module collects actual monitoring data within the current construction phase, corrects geological parameters using the Bayesian update formula, and performs numerical simulation for the next phase based on the updated geological model to identify newly added sensitive monitoring point sets. With low-sensitivity rejection point set Update the initial sensitive point set ,in , This represents the initial set of sensitive monitoring points within the current construction phase, and the weights of each monitoring point are recalculated using the new simulation results.

[0032] Preferably, the virtual model generation unit includes a virtual model construction module, a coordinate transformation module, an influence domain division module, a spatial field determination module, and a knowledge graph construction module;

[0033] The virtual model construction module establishes a unified spatiotemporal coordinate system, uses absolute time as the time reference, and maps it to the construction stage sequence and process number sequence. It uses the geological model to construct the static layer of the virtual roadbed model and integrates multi-source observation data into the dynamic layer of the virtual roadbed model. The multi-source observation data specifically includes construction activity flow, material flow, environmental flow, and monitoring data flow.

[0034] The coordinate transformation module treats the discrete data with three-dimensional coordinates in the multi-source observation data as point data, and uses the direct coordinate transformation method to uniformly convert the coordinate values ​​of all point data into coordinate values ​​under engineering coordinates.

[0035] The influence domain division module statistically analyzes all vector data with spatial range in the multi-source observation data, reads the coordinate sequence of its spatial range boundary, transforms the boundary points in the sequence into a unified engineering coordinate system, connects the boundary points through constrained Delaunay triangulation, generates a polygonal triangular mesh covering the spatial range, and calls it the influence domain.

[0036] The spatial field determination module reconstructs the converted point monitoring data into continuous spatial field data covering the entire virtual roadbed model;

[0037] The knowledge graph building module timestamps each data point and maps it to the engineering timeline, then constructs a construction knowledge graph based on the correlation between multi-source observation data.

[0038] Preferably, the virtual model update unit includes an observation data classification module, a direct adjustment module, an indirect adjustment module, and a state optimization module;

[0039] The observation data classification module receives multi-source observation data within the latest time period and classifies it into directly observed variables and indirectly observed variables.

[0040] For directly observed variables, the direct adjustment module uses an ensemble Kalman filter algorithm to compare the directly observed variables with each predicted observation value generated by the virtual roadbed model. By calculating the Kalman gain matrix, each predicted observation value is corrected to obtain an analysis state set. The mean of all analysis states in the set is taken as the optimal estimated state. After determining the covariance matrix within the set, the error statistic corresponding to the current optimal estimated state is calculated based on the covariance matrix.

[0041] For indirectly observed variables, the indirect adjustment module uses the optimal estimated state updated by the Kalman filter algorithm and the current external load as initial and boundary conditions, and calls the control equations updated by Bayes parameter estimation to solve them, thereby deduce the internal stress field, strain field and displacement field at the next moment.

[0042] The state optimization module integrates the corrected next-moment optimal estimated state with the internal stress field, strain field and displacement field to form a new virtual roadbed model state.

[0043] Preferably, the impact factor analysis unit includes a time delay parameter analysis module and a contribution coefficient calculation module;

[0044] The time delay parameter analysis module reads the current time. Spatial location The settlement observation values ​​obtained directly from the monitoring data stream Simultaneously extract the observation values ​​of influence factors at the same location. ,in This indicates the number of influencing factors at the current location. Indicates the current time Spatial location Upper Each influencing factor observation was analyzed using a cross-correlation function to determine the factor. and Correlation coefficient sequences under different time delays ,according to Determine Relative to settlement value Optimal time delay parameters ,in , Indicates time delay, Indicates the serial number;

[0045] The contribution coefficient calculation module constructs a multiple regression model of the linear relationship between settlement and each lagged factor using the optimal time delay parameters. After determining the regression model parameter vector, the regression coefficients are solved by the least squares method, and the contribution coefficient of each influencing factor is calculated based on the regression coefficients.

[0046] Preferably, the settlement effect analysis unit includes a change value analysis module and an offset calculation module;

[0047] The change value analysis module selects any completed construction procedure and extracts the operation time of that procedure from the construction activity flow. Spatial range ,exist Internal calculation process end time With the start time Each spatial location Upper settlement change value , Indicates the end time of the process. Settlement observations Indicates the start time of the process. Settlement observations;

[0048] The offset calculation module determines the settlement control value for this type of process. By calculation The actual internal settlement change value relative to average offset ,in , It indicates the spatial range, and the settlement effect is measured according to the average offset.

[0049] Preferably, the abnormal settlement early warning unit includes a probability analysis module, a model deduction module, and a report output module;

[0050] When the real-time monitoring data stream indicates that the settlement rate is greater than the safety threshold, the probability analysis module defines the search range with the location and time of the anomaly as the center and its spatial radius as the preset search radius. It also sets a backtracking time window, extracts multi-source observation data covering the backtracking time and search range, uses the multi-source observation data to generate feature vectors, and inputs them into a pre-trained causal discovery neural network model. The model then outputs probability distribution vectors of various preset typical root cause patterns.

[0051] The model deduction module selects multiple root cause patterns to be verified according to the probability distribution vector, rolls back the state of the virtual roadbed model to a specified time before the anomaly occurs, accurately reproduces the specific scenario described by the root cause pattern in the simulation environment, modifies only the parameters existing in the root cause in the model, applies subsequent construction loads that match the actual situation, and deduces the roadbed response from that time to the time of the anomaly occurrence.

[0052] After the simulation is completed, the report output module performs a detailed comparison between the results predicted by the virtual model and the actual monitored abnormal settlement to obtain the corresponding matching degree. Based on the probability value of each root cause pattern and the matching degree, the actual root cause is analyzed, an early warning is issued based on the actual root cause, and a structured diagnostic report is generated.

[0053] Compared with the prior art, the beneficial effects of the present invention are:

[0054] 1. This invention uses forward-looking numerical simulation based on dynamically updated geological parameters and load conditions to accurately identify potential risk areas with the most sensitive mechanical response before the implementation of key construction procedures, and adaptively optimizes the spatiotemporal layout of monitoring resources accordingly. This transforms monitoring from a passive response to an active early warning, providing a decision-making window for the early detection and intervention of core risks.

[0055] 2. This invention strictly aligns the construction activity flow, material flow, environmental flow and multi-source monitoring flow in time, space and logic to construct a digital twin that evolves synchronously with the physical process. By using a causal discovery intelligent algorithm, the system can go beyond traditional phenomenon descriptions and achieve quantitative decomposition of settlement contribution, root cause tracing of abnormal events and automatic mining of causal mechanisms, thereby providing profound insights into the mechanism level of settlement control.

[0056] 3. This invention significantly improves the timeliness, coverage, and economy of settlement monitoring. More importantly, through in-depth data fusion and intelligent analysis, it empowers engineers to understand complex settlement mechanisms, anticipate potential risks, and formulate precise control strategies, fundamentally ensuring the long-term safety and stability of roadbed engineering. Attached Figure Description

[0057] Figure 1 A schematic diagram of the overall system flow is provided for embodiments of the present invention. Detailed Implementation

[0058] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0059] Please see Figure 1 This invention provides a technical solution: a road and subgrade settlement monitoring, analysis, and early warning system based on intelligent sensors, comprising:

[0060] The initial point screening unit is used to read the geological model parameter set and numerical simulation load case set of the current construction stage, generate the initial sensitive monitoring point set, and construct the objective function and constraints.

[0061] The point set optimization unit is used to select the optimal sensor deployment scheme using the initial sensitive monitoring point set, objective function and constraints, and to correct the corresponding geological model parameters based on the actual monitoring data within the current construction phase, so as to update the initial sensitive monitoring point set for the next construction phase.

[0062] The virtual model generation unit is used to read multi-source observation data and construct a virtual roadbed model using geological models and multi-source observation data;

[0063] The virtual model update unit is used to receive multi-source observation data in the latest time period, analyze the optimal estimated state, the internal stress field, strain field and displacement field at the next moment using the observation data, and integrate them to form a new virtual roadbed model.

[0064] The impact factor analysis unit is used to directly calculate the settlement observation value obtained from the monitoring data, and simultaneously extract the impact factor observation values ​​at the same location, and calculate the contribution coefficient of each impact factor relative to the settlement value.

[0065] The settlement effect analysis unit is used to calculate the settlement change value within each construction procedure and determine the average deviation of the change value.

[0066] The abnormal settlement early warning unit is used to acquire real-time monitoring data streams, determine the location and time of abnormal points, analyze them, obtain the root cause pattern probability distribution of the current abnormal points, use virtual models to perform inferences, analyze the actual root causes according to the inference results, and generate early warning prompts.

[0067] The initial point selection unit includes a governing equation construction module, an initial point set generation module, a weight coefficient calculation module, and an objective function determination module;

[0068] The control equation construction module reads the geological model parameter set and numerical simulation load case set of the current construction stage, establishes control equations based on the geological model parameter set and numerical simulation load case set, and calculates global strain and stress.

[0069] The specific process of constructing the geological model parameter set and the numerical simulation load case set is as follows:

[0070] S101. After obtaining the original survey data of the engineering site, the geological modeling engine is used to process the survey data to form a three-dimensional geological body object set and attribute field set. The two are then parameterized to determine the geological BIM model parameter set.

[0071] S102. After obtaining engineering design specifications, construction organization plans and environmental monitoring data, the load analysis engine is used to process them to form a construction sequence load set, an operational dynamic load set and a specification combination load set. Through parametric and structured integration, the numerical simulation load case set is determined.

[0072] The specific process for calculating global strain and stress is as follows:

[0073] S201. Discretize the geological model into a set of finite elements. Each element contains nodal coordinates, material parameters, shape functions, and a mapping relationship between element node numbers and global node numbers. Perform weighted residual discretization on the governing equations to determine the stiffness matrix and load vector of each element. Assemble them into a global stiffness matrix and load vector.

[0074] S202. After solving for the global nodal displacement vector using the global stiffness matrix and load vector, statistically analyze the element strain. and stress ,according to and Calculate global strain and stress ,in , Indicates the first The shape function of each unit, Represents the node coordinates within the cell. Indicates the first strain per unit element Indicates the total number of units. , Indicates the first Unit stress, Indicates the serial number;

[0075] The initial point set generation module obtains the sensitive indicators of each node in the geological model through numerical simulation calculation and extrapolation. The sensitive indicators include the maximum principal stress, strain, safety factor and plastic state. Nodes exceeding the standard are screened according to the preset threshold, and the initial sensitive monitoring point set is formed after spatial clustering and priority sorting.

[0076] The weighting coefficient calculation module reads the initial sensitive monitoring point set. Current construction phase sequence Sensor type set and sensor performance parameters, among which , , Indicates the first One initial sensitive monitoring point, This indicates the initial total number of sensitive monitoring points. Indicates the first The start time of each construction phase Indicates the first The duration of each construction phase was calculated using real-time numerical simulations by applying the current construction load conditions to the geological model, and the weighting coefficients of each sensitive monitoring point were analyzed. ,in , Indicates the first Sensitive monitoring points during each construction phase The Normalized values ​​of several sensitive indicators Indicates the first The weight of each sensitive indicator in the current construction phase sequence. Indicates the serial number;

[0077] The objective function determination module constructs the objective function according to the weight coefficients of each sensitive monitoring point and sets the constraints.

[0078] The objective function is as follows:

[0079]

[0080]

[0081]

[0082] Among them, Co Indicates the time sequence during the construction phase. Total effective coverage Indicates sensitive monitoring points Construction phase sequence The weighting coefficients in the text Indicates sensitive monitoring points Are there sensors deployed in itself and its surrounding area? This indicates a sensitive monitoring point. If there are sensors deployed within itself and its neighborhood, This indicates a sensitive monitoring point. There are no sensors deployed in itself or its surrounding area. Indicates sensitive monitoring points The weighting coefficients, Represents the sensor deployment matrix. Indicates the time sequence during the construction phase. The overall redundancy of the internal deployment scheme, Indicates the serial number. This indicates the total number of sensitive monitoring points within the deployment plan. Indicates the first A sensitive point, Indicates the first All sensitive monitoring points within the spatial neighborhood of each sensitive monitoring point Point Whether it is covered by sensors, if , then represents a point If covered by sensors, , then represents a point Not covered by sensors Indicates the time sequence during the construction phase. Total cost of internal sensor deployment This indicates that one unit of type is deployed. The sensor cost value, Denotes the decision variable, if This indicates that at point Deployment type is The sensor, Indicates the number of sensor types;

[0083] The point set optimization unit includes a scheme output module and an initial point set update module;

[0084] The scheme output module encodes the sensor deployment matrix into a binary chromosome according to the initial sensitive monitoring point set and sensor type set, randomly generates a set of deployment schemes that meet the constraints, and calculates the deployment scheme in each loop. Influence ,in , This indicates the total effective coverage of the current deployment plan. This indicates the overall redundancy of the deployment plan. This represents the total cost of the deployment plan. This represents the control coefficient. These represent the preset maximum coverage, maximum redundancy, and maximum cost, respectively. The selection, crossover, and mutation operations are performed sequentially on all deployment schemes, and schemes that violate the constraints are repaired. When the number of iterations exceeds the maximum number of iterations, the optimal scheme is output.

[0085] The initial point set update module collects actual monitoring data within the current construction phase, corrects geological parameters using the Bayesian update formula, and performs numerical simulations for the next phase based on the updated geological model, identifying newly added sensitive monitoring point sets. With low-sensitivity rejection point set Update the initial sensitive point set ,in , This represents the initial set of sensitive monitoring points within the current construction phase, and the weights of each monitoring point are recalculated using the new simulation results.

[0086] The virtual model generation unit includes a virtual model construction module, a coordinate transformation module, an influence domain division module, a spatial field determination module, and a knowledge graph construction module.

[0087] The virtual model construction module establishes a unified spatiotemporal coordinate system, using absolute time as the time reference, and maps it to the construction phase sequence and process number sequence. It uses the geological model to construct the static layer of the virtual roadbed model and integrates multi-source observation data into the dynamic layer of the virtual roadbed model. The multi-source observation data specifically includes construction activity flow, material flow, environmental flow, and monitoring data flow. The construction activity flow specifically includes process type, spatial range, start and end time, associated equipment operating parameters, process control parameters, and quality inspection information. The material flow specifically includes material batches, moisture content, compaction degree, and temperature. The environmental flow specifically includes rainfall, air temperature, and evaporation. The monitoring data flow specifically includes the monitoring data sequence collected by various sensors.

[0088] The coordinate transformation module treats the discrete data with three-dimensional coordinates in the multi-source observation data as point data, and uses the direct coordinate transformation method to uniformly convert the coordinate values ​​of all point data into coordinate values ​​under engineering coordinates.

[0089] The influence domain delineation module statistically analyzes all vector data with spatial extent in the multi-source observation data, reads the coordinate sequence of its spatial extent boundary, transforms the boundary points in the sequence into a unified engineering coordinate system, connects the boundary points through constrained Delaunay triangulation, generates a polygonal triangular mesh covering the spatial extent, and calls it the influence domain.

[0090] The spatial field determination module reconstructs the converted point monitoring data into continuous spatial field data covering the entire virtual roadbed model;

[0091] The specific process of reconstructing continuous spatial field data is as follows:

[0092] S301. Select any node to be estimated in the virtual roadbed model. Using the node to be estimated as the center, search for all known observation points (i.e., the actual monitoring points where sensors have been deployed) around it according to the preset spatial correlation radius. Use all the searched known points to construct the neighborhood sample set for this interpolation.

[0093] S302. Pair up the known points in the neighborhood sample set and analyze the measured values ​​and spatial distances of all points in each combination. Based on the measured values ​​and spatial distances of all combinations in the neighborhood sample set, calculate the theoretical semivariogram value between any two points.

[0094] S303. By solving the Kriging equations, a set of optimal weights that satisfy unbiasedness and minimize the estimation variance is obtained. The coefficients in the equations are composed of the theoretical semivariogram values ​​between known points. The measured values ​​of all known points in the neighborhood sample set are linearly combined using this set of weights to obtain the best linear unbiased estimate of the node to be estimated. The Kriging variance, which characterizes the interpolation uncertainty of the point, is calculated simultaneously. This process is repeated for all nodes to be estimated in the model to generate continuous spatial field data.

[0095] After the knowledge graph building module timestamps each data point and maps it to the engineering timeline, it constructs a construction knowledge graph based on the correlation between multi-source observation data.

[0096] The virtual model update unit includes an observation data classification module, a direct adjustment module, an indirect adjustment module, and a state optimization module;

[0097] The observation data classification module receives multi-source observation data from the latest time period and classifies it into directly observed variables and indirectly observed variables.

[0098] For directly observed variables, the direct adjustment module uses an ensemble Kalman filter algorithm to compare the directly observed variables with each predicted observation value generated by the virtual roadbed model. By calculating the Kalman gain matrix, each predicted observation value is corrected to obtain an analysis state set. The mean of all analysis states in the set is taken as the optimal estimated state. After determining the covariance matrix within the set, the error statistic corresponding to the current optimal estimated state is calculated based on the covariance matrix.

[0099] For indirectly observed variables, the indirect adjustment module uses the optimal estimated state updated by the Kalman filter algorithm and the current external load as initial and boundary conditions, and calls the control equations updated by Bayes parameter estimation to solve them, thereby deduce the internal stress field, strain field and displacement field at the next moment.

[0100] The state optimization module integrates and corrects the next moment's optimal estimated state with the internal stress field, strain field, and displacement field to form a new virtual roadbed model state.

[0101] The impact factor analysis unit includes a time delay parameter analysis module and a contribution coefficient calculation module;

[0102] The time delay parameter analysis module reads the current time. Spatial location The settlement observation values ​​obtained directly from the monitoring data stream Simultaneously extract the observation values ​​of influence factors at the same location. ,in This indicates the number of influencing factors at the current location. Indicates the current time Spatial location Upper Each influencing factor observation was analyzed using a cross-correlation function to determine the factor. and Correlation coefficient sequences under different time delays ,according to Determine Relative to settlement value Optimal time delay parameters ,in , Indicates time delay, Indicates the serial number;

[0103] The contribution coefficient calculation module uses the optimal time delay parameters to construct a multiple regression model of the linear relationship between settlement and each lag factor. After determining the parameter vector of the regression model, the regression coefficients are solved by the least squares method. Based on the regression coefficients, the contribution coefficient of each influencing factor is calculated. The influencing factors include material moisture content, compaction degree, temperature, construction process control parameters, environmental rainfall, air temperature and evaporation.

[0104] The settlement effect analysis unit includes a change value analysis module and an offset calculation module;

[0105] The change value analysis module selects any completed construction process and extracts the operation time of that process from the construction activity flow. Spatial range ,exist Internal calculation process end time With the start time Each spatial location Upper settlement change value , Indicates the end time of the process. Settlement observations Indicates the start time of the process. Settlement observations;

[0106] The offset calculation module determines the settlement control value for this type of process. By calculation The actual internal settlement change value relative to average offset ,in , The value represents the spatial range and measures the settlement effect according to the average offset. A value close to 1 indicates that the construction quality meets expectations, a value significantly greater than 1 indicates excessive settlement, and a value significantly less than 1 indicates insufficient settlement.

[0107] The abnormal settlement early warning unit includes a probability analysis module, a model deduction module, and a report output module;

[0108] When the real-time monitoring data stream indicates that the settlement rate is greater than the safety threshold, the probability analysis module uses the location and time of the anomaly point as the center, and its spatial radius as the preset search radius to define the search range. It also sets a backtracking time window, extracts multi-source observation data covering the backtracking time and search range, uses this multi-source observation data to generate feature vectors, and inputs them into a pre-trained causal discovery neural network model. The model then outputs probability distribution vectors of various preset typical root cause patterns.

[0109] The causal discovery neural network model consists of a feature extraction module, a causal structure learning module, and a causal effect estimation module. The feature extraction module uses two 2D convolutional layers with 3 kernels and a stride of 1, connected in series with a long short-term memory network layer with 128 memory units, to extract spatiotemporal features from the monitored data sequence. The causal structure learning module is based on a graph attention network and includes an attention layer with 8 attention heads. The output of this layer is normalized by the Softmax function to generate an adjacency matrix representing the causal strength between variables. The causal effect estimation module consists of three fully connected network layers with 256, 128 neurons, and the number of monitored variables, respectively. The activation functions are ReLU, ReLU, and a linear function, respectively.

[0110] The hierarchical connection relationship of the causal discovery neural network model is as follows: the original multivariate monitoring sequence is input into the feature extraction module, the output feature vector is input into the causal structure learning module, the generated causal adjacency matrix and the feature vector are input into the causal effect estimation module, and finally the causal effect value of each variable is output to determine the probability distribution vector of various preset typical root cause patterns.

[0111] The causal discovery neural network model is trained using aligned time-series data obtained from a virtual roadbed model as samples. Each sample contains construction activity data, material data, environmental data, and monitoring data. The training employs a composite loss function with three components: reconstruction loss uses the mean squared error function; sparsity loss applies L1 regularization to the causal adjacency matrix with a regularization coefficient of 0.01; and domain knowledge loss penalizes causal directions that violate predefined construction logic constraints with a weight of 0.05.

[0112] The Adam optimizer was used for training, with an initial learning rate of 0.001, a batch size of 32, and 200 training epochs. An early stopping strategy on the validation set was also employed.

[0113] The model deduction module selects multiple root cause patterns to be verified according to the probability distribution vector, rolls back the state of the virtual roadbed model to a specified time before the anomaly occurs, accurately reproduces the specific scenario described by the root cause pattern in the simulation environment, modifies only the parameters existing in the root cause in the model, applies subsequent construction loads that match the actual situation, and deduces the roadbed response from that time to the time of the anomaly occurrence.

[0114] After the simulation is completed, the results predicted by the virtual model are compared with the actual monitored abnormal settlement to obtain the corresponding matching degree. The actual root cause is analyzed according to the probability value of each root cause pattern and the matching degree. Early warning prompts are issued according to the actual root cause, and a structured diagnostic report is generated.

[0115] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0116] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A road and subgrade settlement monitoring, analysis, and early warning system based on intelligent sensors, characterized in that, include: The initial point screening unit is used to read the geological model parameter set and numerical simulation load case set of the current construction stage, generate the initial sensitive monitoring point set, and construct the objective function and constraints. The point set optimization unit is used to select the optimal sensor deployment scheme using the initial sensitive monitoring point set, objective function and constraints, and to correct the corresponding geological model parameters based on the actual monitoring data within the current construction phase, so as to update the initial sensitive monitoring point set for the next construction phase. The virtual model generation unit is used to read multi-source observation data and construct a virtual roadbed model using geological models and multi-source observation data; The virtual model update unit is used to receive multi-source observation data in the latest time period, analyze the optimal estimated state, the internal stress field, strain field and displacement field at the next moment using the observation data, and integrate them to form a new virtual roadbed model. The impact factor analysis unit is used to directly calculate the settlement observation value obtained from the monitoring data, and simultaneously extract the impact factor observation values ​​at the same location, and calculate the contribution coefficient of each impact factor relative to the settlement value. The settlement effect analysis unit is used to calculate the settlement change value within each construction procedure and determine the average deviation of the change value. The abnormal settlement early warning unit is used to acquire real-time monitoring data streams, determine the location and time of abnormal points, analyze them, obtain the root cause pattern probability distribution of the current abnormal points, use virtual models to perform inferences, analyze the actual root causes according to the inference results, and generate early warning prompts.

2. The road surface and subgrade settlement monitoring, analysis, and early warning system based on intelligent sensors according to claim 1, characterized in that: The initial point screening unit includes a control equation construction module, an initial point set generation module, a weight coefficient calculation module, and an objective function determination module. The control equation construction module reads the geological model parameter set and numerical simulation load case set of the current construction stage, establishes control equations based on the geological model parameter set and numerical simulation load case set, and calculates global strain and stress. The initial point set generation module obtains the sensitive indicators of each node in the geological model through numerical simulation calculation and extrapolation. The sensitive indicators include the maximum principal stress, strain, safety factor and plastic state. Nodes exceeding the standard are screened according to preset thresholds, and the initial sensitive monitoring point set is formed after spatial clustering and priority sorting. The weighting coefficient calculation module reads the initial sensitive monitoring point set. Current construction phase sequence Sensor type set and sensor performance parameters, among which , , Indicates the first One initial sensitive monitoring point, This indicates the initial total number of sensitive monitoring points. Indicates the first The start time of each construction phase Indicates the first The duration of each construction phase was calculated using real-time numerical simulations by applying the current construction load conditions to the geological model, and the weighting coefficients of each sensitive monitoring point were analyzed. ,in , Indicates the first Sensitive monitoring points during each construction phase The Normalized values ​​of several sensitive indicators Indicates the first The weight of each sensitive indicator in the current construction phase sequence. Indicates the serial number; The objective function determination module constructs the objective function according to the weight coefficients of each sensitive monitoring point and sets the constraints.

3. The road surface and subgrade settlement monitoring, analysis, and early warning system based on intelligent sensors according to claim 1, characterized in that: The point set optimization unit includes a scheme output module and an initial point set update module; The scheme output module encodes the sensor deployment matrix into a binary chromosome according to the initial sensitive monitoring point set and sensor type set, randomly generates a set of deployment schemes that meet the constraints, and calculates the deployment scheme in each round of the loop. Influence ,in , This indicates the total effective coverage of the current deployment plan. This indicates the overall redundancy of the deployment plan. This represents the total cost of the deployment plan. This represents the control coefficient. These represent the preset maximum coverage, maximum redundancy, and maximum cost, respectively. The selection, crossover, and mutation operations are performed sequentially on all deployment schemes, and schemes that violate the constraints are repaired. When the number of iterations exceeds the maximum number of iterations, the optimal scheme is output. The initial point set update module collects actual monitoring data within the current construction phase, corrects geological parameters using the Bayesian update formula, and performs numerical simulation for the next phase based on the updated geological model to identify newly added sensitive monitoring point sets. With low-sensitivity rejection point set Update the initial sensitive point set ,in , This represents the initial set of sensitive monitoring points within the current construction phase, and the weights of each monitoring point are recalculated using the new simulation results.

4. The road surface and subgrade settlement monitoring, analysis, and early warning system based on intelligent sensors according to claim 1, characterized in that: The virtual model generation unit includes a virtual model construction module, a coordinate transformation module, an influence domain division module, a spatial field determination module, and a knowledge graph construction module; The virtual model construction module establishes a unified spatiotemporal coordinate system, uses absolute time as the time reference, and maps it to the construction stage sequence and process number sequence. It uses the geological model to construct the static layer of the virtual roadbed model and integrates multi-source observation data into the dynamic layer of the virtual roadbed model. The multi-source observation data specifically includes construction activity flow, material flow, environmental flow, and monitoring data flow. The coordinate transformation module treats the discrete data with three-dimensional coordinates in the multi-source observation data as point data, and uses the direct coordinate transformation method to uniformly convert the coordinate values ​​of all point data into coordinate values ​​under engineering coordinates. The influence domain division module statistically analyzes all vector data with spatial range in the multi-source observation data, reads the coordinate sequence of its spatial range boundary, transforms the boundary points in the sequence into a unified engineering coordinate system, connects the boundary points through constrained Delaunay triangulation, generates a polygonal triangular mesh covering the spatial range, and calls it the influence domain. The spatial field determination module reconstructs the converted point monitoring data into continuous spatial field data covering the entire virtual roadbed model; The knowledge graph building module timestamps each data point and maps it to the engineering timeline, then constructs a construction knowledge graph based on the correlation between multi-source observation data.

5. The road surface and subgrade settlement monitoring, analysis, and early warning system based on intelligent sensors according to claim 1, characterized in that: The virtual model update unit includes an observation data classification module, a direct adjustment module, an indirect adjustment module, and a state optimization module. The observation data classification module receives multi-source observation data within the latest time period and classifies it into directly observed variables and indirectly observed variables. For directly observed variables, the direct adjustment module uses an ensemble Kalman filter algorithm to compare the directly observed variables with each predicted observation value generated by the virtual roadbed model. By calculating the Kalman gain matrix, each predicted observation value is corrected to obtain an analysis state set. The mean of all analysis states in the set is taken as the optimal estimated state. After determining the covariance matrix within the set, the error statistic corresponding to the current optimal estimated state is calculated based on the covariance matrix. For indirectly observed variables, the indirect adjustment module uses the optimal estimated state updated by the Kalman filter algorithm and the current external load as initial and boundary conditions, and calls the control equations updated by Bayes parameter estimation to solve them, thereby deduce the internal stress field, strain field and displacement field at the next moment. The state optimization module integrates the corrected next-moment optimal estimated state with the internal stress field, strain field and displacement field to form a new virtual roadbed model state.

6. The road surface and subgrade settlement monitoring, analysis, and early warning system based on intelligent sensors according to claim 1, characterized in that: The impact factor analysis unit includes a time delay parameter analysis module and a contribution coefficient calculation module; The time delay parameter analysis module reads the current time. Spatial location The settlement observation values ​​obtained directly from the monitoring data stream Simultaneously extract the observation values ​​of influence factors at the same location. ,in This indicates the number of influencing factors at the current location. Indicates the current time Spatial location Upper Each influencing factor observation was analyzed using a cross-correlation function to determine the factor. and Correlation coefficient sequences under different time delays ,according to Determine Relative to settlement value Optimal time delay parameters ,in , Indicates time delay, Indicates the serial number; The contribution coefficient calculation module constructs a multiple regression model of the linear relationship between settlement and each lagged factor using the optimal time delay parameters. After determining the regression model parameter vector, the regression coefficients are solved by the least squares method, and the contribution coefficient of each influencing factor is calculated based on the regression coefficients.

7. The pavement and subgrade settlement monitoring, analysis, and early warning system based on intelligent sensors according to claim 1, characterized in that: The settlement effect analysis unit includes a change value analysis module and an offset calculation module; The change value analysis module selects any completed construction procedure and extracts the operation time of that procedure from the construction activity flow. Spatial range ,exist Internal calculation process end time With the start time Each spatial location Upper settlement change value , Indicates the end time of the process. Settlement observations Indicates the start time of the process. Settlement observations; The offset calculation module determines the settlement control value for this type of process. By calculation The actual internal settlement change value relative to average offset ,in , It indicates the spatial range, and the settlement effect is measured according to the average offset.

8. The road surface and subgrade settlement monitoring, analysis, and early warning system based on intelligent sensors according to claim 1, characterized in that: The abnormal settlement early warning unit includes a probability analysis module, a model deduction module, and a report output module; When the real-time monitoring data stream indicates that the settlement rate is greater than the safety threshold, the probability analysis module defines the search range with the location and time of the anomaly as the center and its spatial radius as the preset search radius. It also sets a backtracking time window, extracts multi-source observation data covering the backtracking time and search range, uses the multi-source observation data to generate feature vectors, and inputs them into a pre-trained causal discovery neural network model. The model then outputs probability distribution vectors of various preset typical root cause patterns. The model deduction module selects multiple root cause patterns to be verified according to the probability distribution vector, rolls back the state of the virtual roadbed model to a specified time before the anomaly occurs, accurately reproduces the scenario described by the root cause pattern in the simulation environment, modifies only the parameters existing in the root cause in the model, applies subsequent construction loads that match the actual situation, and deduces the roadbed response from that time to the time of the anomaly occurrence. After the simulation is completed, the report output module performs a detailed comparison between the results predicted by the virtual model and the actual monitored abnormal settlement to obtain the corresponding matching degree. Based on the probability value of each root cause pattern and the matching degree, the actual root cause is analyzed, an early warning is issued based on the actual root cause, and a structured diagnostic report is generated.

Images