Intelligent grouting decision-making method for settlement control of urban underground space pipe gallery

Abstract

The invention discloses an intelligent grouting decision-making method for settlement control of an urban underground space pipe gallery, and relates to the related field of underground structure health monitoring, and the method comprises the steps: carrying out the real-time data collection and fusion processing of a multi-source sensor array on a key structure node of a target pipe gallery, and obtaining a multi-dimensional perception data set of the health of the pipe gallery; performing settlement risk analysis processing based on time sequence prediction and causal reasoning on the pipe gallery health multi-dimensional perception data set to obtain decision support information; grouting parameter dynamic optimization processing is conducted on the decision support information and the pipe gallery health multi-dimensional perception data set, and an optimal grouting strategy is obtained; and the optimal grouting strategy is sent to a grouting equipment control system, and grouting equipment is driven to execute automatic grouting operation. The technical problem that a grouting strategy cannot be accurately adjusted in real time according to actual conditions in an existing grouting decision for urban underground space pipe gallery settlement control is solved, and the technical effects that the accuracy and timeliness of grouting operation are improved, and pipe gallery settlement is effectively controlled are achieved.

Description

Technical Field

[0001] This application relates to the field of underground structure health monitoring, and in particular to an intelligent grouting decision-making method for settlement control of urban underground utility tunnels. Background Technology

[0002] As a crucial component of urban infrastructure, the safe and stable operation of underground utility tunnels directly impacts the normal functioning of the city and the quality of life for residents. Settlement of these tunnels severely threatens their structural safety, consequently affecting the normal operation of the various pipelines they support. Therefore, effectively controlling tunnel settlement is paramount. Currently, the main method for controlling settlement in urban underground utility tunnels is regular grouting maintenance based on preset grouting parameters, reinforcing the surrounding soil to prevent or mitigate settlement. However, this method, relying primarily on fixed grouting parameters and regular maintenance, struggles to adjust grouting strategies in real-time based on actual settlement conditions and dynamic changes in the surrounding environment. This results in a lack of precision and timeliness in grouting operations, potentially leading to either excessive grouting wasting resources or insufficient grouting failing to effectively control settlement.

[0003] At present, the grouting decision-making for settlement control of urban underground utility tunnels faces the technical problem of being unable to adjust the grouting strategy accurately in real time according to the actual situation. Summary of the Invention

[0004] This application provides an intelligent grouting decision-making method for settlement control of urban underground utility tunnels. It employs real-time acquisition and fusion of data from multi-source sensor arrays on key structural nodes of the target utility tunnel to obtain a health multi-dimensional perception dataset. This dataset is then used for time-series prediction and causal inference settlement risk analysis to obtain decision support information. Combining the decision support information and the dataset, grouting parameters are dynamically optimized to derive the optimal grouting strategy. This strategy is then sent to the grouting equipment control system to drive automatic grouting operations. This approach solves the technical problem of existing grouting decision-making methods for urban underground utility tunnel settlement control, which cannot accurately adjust the grouting strategy in real time according to actual conditions. This achieves the technical effect of improving the accuracy and timeliness of grouting operations and effectively controlling utility tunnel settlement.

[0005] This application provides an intelligent grouting decision-making method for settlement control of urban underground utility tunnels, comprising: real-time data acquisition and fusion processing of multi-source sensor arrays on key structural nodes of the target utility tunnel to obtain a multi-dimensional health perception dataset of the utility tunnel; performing settlement risk analysis processing based on time-series prediction and causal reasoning on the multi-dimensional health perception dataset of the utility tunnel to obtain decision support information; performing dynamic optimization processing of grouting parameters on the decision support information and the multi-dimensional health perception dataset of the utility tunnel to obtain an optimal grouting strategy; and sending the optimal grouting strategy to the grouting equipment control system to drive the grouting equipment to perform automatic grouting operations.

[0006] In a possible implementation, the key structural nodes of the target utility tunnel are identified, and the following processing is performed: Engineering design drawings and geological survey reports of the target utility tunnel are obtained, and a finite element model of the tunnel-soil interaction is constructed; the self-weight load, design traffic load, and surrounding additional loads of the utility tunnel-soil interaction finite element model are applied, and static settlement simulation calculations are performed to obtain a first displacement cloud map of the utility tunnel structure; based on the spatial coordinates of the weak soil areas identified in the geological survey report, the soil mechanical parameters at the corresponding locations in the finite element model of the tunnel-soil interaction are reduced and corrected, and settlement simulation calculations are performed again based on the first displacement cloud map to obtain a second displacement cloud map; the first and second displacement cloud maps are merged, and the risk areas of the utility tunnel with displacements greater than a first preset threshold or displacement gradients greater than a second preset threshold are extracted, and the center point and boundary point of the risk areas of the utility tunnel are determined as the key structural nodes.

[0007] In a possible implementation, the multi-dimensional health sensing dataset of the utility tunnel is subjected to settlement risk analysis based on time-series prediction and causal inference to obtain decision support information. The following processes are performed: the multi-dimensional health sensing dataset of the utility tunnel is preprocessed to construct a multivariate time series matrix of the utility tunnel; settlement of the utility tunnel is predicted based on the multivariate time series matrix of the utility tunnel to obtain the settlement prediction result; causal inference is performed based on the multivariate time series matrix of the utility tunnel to obtain a set of key settlement-causing factors; the settlement prediction result and the set of key settlement-causing factors are fused to generate decision support information.

[0008] In a possible implementation, the settlement of the utility tunnel is predicted based on the multivariate time series matrix of the utility tunnel, and the following processing is performed to obtain the settlement prediction result: each sensor node in the multivariate time series matrix of the utility tunnel is used as a graph node, and a sensor graph adjacency matrix is ​​constructed based on the physical connection relationship of the utility tunnel structure; a spatiotemporal graph convolutional network is used to predict the settlement of the utility tunnel, wherein: the encoder captures spatial dependence and temporal dynamics through a spatiotemporal convolution module based on the sensor graph adjacency matrix and the multivariate time series matrix of the utility tunnel, and generates a spatiotemporal feature vector of the utility tunnel; the decoder performs autoregressive prediction through a temporal convolutional layer based on the spatiotemporal feature vector of the utility tunnel, and outputs the predicted settlement value and confidence interval for multiple future time steps as the settlement prediction result.

[0009] In a possible implementation, causal reasoning is performed based on the multivariate time series matrix of the utility tunnel to obtain a set of key settlement factors. The following processing is then performed: conditional independence tests are conducted on all variables in the multivariate time series matrix of the utility tunnel to construct an undirected causal skeleton graph; V-structures are identified in the undirected causal skeleton graph, and the corresponding causal arrows are oriented according to the identification results to obtain a partially directed acyclic graph; based on the time priority principle and prior knowledge of the structural mechanics of the utility tunnel, the remaining undirected arrows in the partially directed acyclic graph are oriented to obtain a directed causal graph; all parent nodes that directly point to the settlement node are extracted from the directed causal graph to form a set of key settlement factors.

[0010] In a possible implementation, the decision support information and the multi-dimensional perception dataset of pipe gallery health are dynamically optimized for grouting parameters to obtain the optimal grouting strategy. The following processing is performed: the decision support information and the multi-dimensional perception dataset of pipe gallery health are defined as a state space; the grouting pressure, grouting flow rate, grout water-cement ratio, and grouting hole three-dimensional coordinates are defined as an action space; a reward function is defined based on the settlement recovery value, cumulative grouting volume, and pressure fluctuation variance; and the grouting parameters are dynamically optimized based on the state space, the action space, and the reward function to obtain the optimal grouting strategy.

[0011] In a possible implementation, the grouting parameters are dynamically optimized based on the state space, the action space, and the reward function to obtain the optimal grouting strategy. The following processing is then performed: Based on the finite element model of the pipe gallery-soil interaction, a simulation environment is constructed. A near-end strategy optimization algorithm is used to train the grouting agent according to the state space, the action space, and the reward function until training is complete. The real-time state space is input into the trained grouting agent, which outputs the optimal grouting parameter values ​​to constitute the optimal grouting strategy.

[0012] In a possible implementation, the optimal grouting strategy is sent to the grouting equipment control system to drive the execution of automatic grouting operations, and the following processes are performed: acquiring the latest multi-dimensional sensing dataset of pipe gallery health at preset time intervals; acquiring an updated optimal grouting strategy based on the latest multi-dimensional sensing dataset of pipe gallery health; and dynamically adjusting the optimal grouting strategy currently executing the automatic grouting operation based on the updated optimal grouting strategy.

[0013] The proposed intelligent grouting decision-making method for settlement control of urban underground utility tunnels involves several steps. First, real-time data acquisition and fusion processing are performed on multi-source sensor arrays at key structural nodes of the target utility tunnel to obtain a multi-dimensional health perception dataset. Next, this dataset undergoes settlement risk analysis based on time-series prediction and causal reasoning to generate decision support information. Then, the decision support information and the multi-dimensional health perception dataset are dynamically optimized to obtain the optimal grouting strategy. Finally, the optimal grouting strategy is sent to the grouting equipment control system to drive the grouting equipment to perform automatic grouting operations. This method achieves the technical effects of improving the accuracy and timeliness of grouting operations and effectively controlling utility tunnel settlement. Attached Figure Description

[0014] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings of the embodiments of the present invention will be briefly described below. Flowcharts are used in this application to illustrate the operations performed by the method according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed precisely in sequence. Instead, various steps can be processed in reverse order or simultaneously as needed. Furthermore, other operations can be added to these processes, or one or more steps can be removed from these processes.

[0015] Figure 1 This is a flowchart illustrating the intelligent grouting decision-making method for settlement control of urban underground utility tunnels provided in an embodiment of this application.

[0016] Figure 2 This is a flowchart illustrating the process of determining key structural nodes of a target utility tunnel in an intelligent grouting decision-making method for settlement control of urban underground utility tunnels provided in this application embodiment. Detailed Implementation

[0017] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application.

[0018] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description of this application will be provided in conjunction with the accompanying drawings. The described embodiments should not be considered as limitations on this application. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0019] In the following description, references to "some embodiments" describe a subset of all possible embodiments. However, it is understood that "some embodiments" may be the same or different subsets of all possible embodiments and may be combined with each other without conflict. The terms "first" and "second" are used merely to distinguish similar objects and do not represent a specific ordering of objects. The terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or modules not explicitly listed or inherent to these processes, methods, products, or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only.

[0020] This application provides an intelligent grouting decision-making method for settlement control of urban underground utility tunnels, such as... Figure 1 As shown, the method includes: Step S100: Real-time data acquisition and fusion processing are performed on the multi-source sensor array on the key structural nodes of the target utility tunnel to obtain a multi-dimensional sensing dataset of utility tunnel health.

[0021] Specifically, various types of sensors, such as strain sensors, displacement sensors, acceleration sensors, and tilt sensors, are installed on key structural nodes of the target utility tunnel. These sensors transmit the raw data they collect to the data acquisition module via wired or wireless communication. The data acquisition module acquires sensor data in real time according to a set sampling frequency, performs preliminary processing on the received multi-source raw data, and then transmits it to the data processing center. At the data processing center, data fusion algorithms are used to fuse data from different types of sensors. For example, a weighted average method is used to fuse displacement data from multiple displacement sensors at the same location, assigning different weights based on the accuracy and reliability of each sensor to calculate a more accurate displacement value. For data such as strain and acceleration, a Kalman filter algorithm is used for fusion processing to eliminate noise interference and improve the accuracy and reliability of the data. The fused data is organized according to a certain data structure to form a multi-dimensional sensing dataset for the health of the utility tunnel. This dataset contains multi-dimensional information such as displacement, strain, acceleration, and tilt of different key structural nodes of the utility tunnel.

[0022] like Figure 2As shown, in one possible implementation, the key structural nodes of the target utility tunnel are determined. Step S100 further includes step S110, which involves obtaining the engineering design drawings and geological survey report of the target utility tunnel, and constructing a finite element model of the interaction between the utility tunnel and the soil. Specifically, the engineering design drawings of the target utility tunnel are obtained from relevant departmental or project documents. These drawings form the basis for constructing the geometric model of the utility tunnel. The drawings include information such as the geometric dimensions, structural form, and material properties of the target utility tunnel, including length, width, height, shape, and concrete strength grade. Simultaneously, a geological survey report is obtained to provide a basis for constructing the soil model and setting parameters. This report includes the stratigraphic distribution, soil type, and physical and mechanical parameters of each soil layer in the area where the target utility tunnel is located, such as density, elastic modulus, Poisson's ratio, cohesion, and internal friction angle.

[0023] Using finite element analysis software, a geometric model of the target utility tunnel was established based on the engineering design drawings. Then, based on the geological survey report, a soil model was built around the tunnel model, assigning physical and mechanical parameters of different soil layers to corresponding soil regions according to the actual distribution of the strata. Finally, contact elements were used to simulate the interaction between the tunnel and the soil, thereby defining the contact relationship between the tunnel and the soil, such as frictional contact or bonded contact. The constructed finite element model of the tunnel-soil interaction was used to simulate the interaction between the target utility tunnel and the surrounding soil, analyzing the mechanical response and deformation of the tunnel under different loads.

[0024] Step S120: Apply the self-weight load, design traffic load, and peripheral additional load to the finite element model of the utility tunnel-soil interaction, and perform static settlement simulation calculations to obtain the first displacement cloud map of the utility tunnel structure. Specifically, based on the actual situation of the target utility tunnel and relevant design specifications, calculate the self-weight load of the utility tunnel, i.e., the load generated by the weight of the utility tunnel structure itself. Calculate according to the density and volume of the utility tunnel material, and apply the self-weight load to the utility tunnel model in the form of a uniformly distributed load or a body load. The design traffic load is determined based on the type of vehicles or equipment designed to pass through the utility tunnel, referring to relevant standards to determine the load size and location, and is applied to the top of the utility tunnel or corresponding parts. Peripheral additional loads refer to the additional loads generated by environmental factors around the utility tunnel, such as buildings and surcharges. Estimate the size and range of the additional loads based on the actual situation and apply them to the corresponding locations on the soil model.

[0025] A statics analysis module was set up in the finite element analysis software, and the applied load was used as input to solve the finite element model of the pipe gallery-soil interaction. Based on the model's geometry, material properties, contact relationships, and applied loads, the stress, strain, and displacement mechanical responses of the pipe gallery and soil were calculated using numerical methods. After the calculation was completed, the displacement calculation results of the pipe gallery structure were extracted and displayed in the form of a displacement cloud map. The first displacement cloud map reflects the displacement distribution of the pipe gallery under the combined action of its own weight, design traffic load, and surrounding additional loads.

[0026] Step S130: Based on the spatial coordinates of the weak soil area identified in the geological survey report, the soil mechanical parameters at the corresponding locations in the finite element model of the pipe gallery-soil interaction are reduced and corrected. Based on the first displacement cloud map, settlement simulation calculations are performed again to obtain a second displacement cloud map. Specifically, the geological survey report is analyzed to determine the location, extent, and spatial coordinates of the weak soil area. The weak soil area refers to a region with low physical and mechanical parameters and weak bearing capacity, such as soft soil layers or silty soil layers. Failure scenarios in the weak soil area are simulated, such as groundwater erosion leading to a decrease in the bearing capacity of the soil in the weak area. The soil mechanical parameters at the corresponding locations in the finite element model of the pipe gallery-soil interaction are reduced and corrected. For example, the elastic modulus can be reduced by a certain proportion; cohesion and internal friction angle can also be appropriately reduced according to the actual situation. After completing the reduction and correction of the soil mechanical parameters, the load on the pipe gallery model remains unchanged, and static settlement simulation calculations are performed again in the finite element analysis software. Due to changes in soil mechanics parameters, the mechanical response of the utility tunnel and the soil also changes, resulting in new displacement calculation results. The displacement results of the utility tunnel structure after the second simulation calculation are extracted to generate a second displacement contour map. This second displacement contour map reflects the displacement distribution of the utility tunnel structure after considering the effects of adverse scenarios.

[0027] Step S140: Merge the first displacement cloud map and the second displacement cloud map, extract the risk areas of the utility tunnel where the displacement is greater than a first preset threshold or the displacement gradient is greater than a second preset threshold, and determine the center point and boundary point of the risk area of ​​the utility tunnel as the key structural node. Specifically, image fusion technology or data fusion method is used to merge the first displacement cloud map and the second displacement cloud map. Among them, image fusion technology can overlay the information of the two cloud maps; data fusion method can merge the displacement data corresponding to the two cloud maps.

[0028] Based on engineering experience and relevant standards, a first preset threshold and a second preset threshold are set. The first preset threshold is used to determine whether the displacement is too large; for example, it can be set to 80% of the maximum allowable settlement of the utility tunnel. The second preset threshold is used to determine whether the displacement gradient is too large. The displacement gradient reflects the drastic degree of displacement change, and an excessively large displacement gradient may cause damage such as cracks in the utility tunnel structure. By comparing the displacement and displacement gradient at each location in the fused displacement cloud map with the preset thresholds, areas where the displacement is greater than the first preset threshold or the displacement gradient is greater than the second preset threshold are extracted as risk areas for the utility tunnel.

[0029] For the extracted risk areas of the utility tunnel, their center points and boundary points are determined. The center point can be obtained by calculating the geometric center of the risk area, and the boundary points can be determined by identifying the boundary contours of the risk area. These center points and boundary points are identified as key structural nodes, which are the locations where sensors need to be installed for monitoring during the settlement and deformation process of the utility tunnel.

[0030] Step S200: Perform settlement risk analysis processing based on time-series prediction and causal reasoning on the multi-dimensional health perception dataset of the utility tunnel to obtain decision support information.

[0031] Specifically, time series analysis algorithms are used to model the time-varying data such as displacement and strain in the multi-dimensional health monitoring dataset of the utility tunnel. This model is then used to predict the settlement changes of the target utility tunnel over a future period, yielding a sequence of predicted settlement values. Causal discovery algorithms are employed to analyze the causal relationships between different variables in the multi-dimensional health monitoring dataset. For example, the causal relationships between displacement changes and strain changes, and environmental factors, are analyzed. By constructing a causal graph, the causal direction and intensity between variables are analyzed to identify the key factors leading to utility tunnel settlement.

[0032] By combining time-series forecasts and causal inferences, a comprehensive assessment of the settlement risk of utility tunnels is conducted. For example, different risk level thresholds are set. When the predicted settlement value exceeds the corresponding threshold or key factors show abnormal changes, the utility tunnel is determined to have different levels of settlement risk. The risk level and related influencing factors are then compiled into decision support information.

[0033] In one possible implementation, the multi-dimensional sensing dataset of pipe gallery health is subjected to settlement risk analysis based on time-series prediction and causal inference to obtain decision support information. Step S200 further includes step S210, which preprocesses the multi-dimensional sensing dataset of pipe gallery health to construct a multivariate time series matrix of the pipe gallery. Specifically, the dataset is checked for missing values, outliers, and duplicate values. For missing values, methods such as mean imputation, median imputation, and interpolation of preceding and following data can be used. For outliers, they are identified by setting a reasonable threshold range and processed by methods such as removal, correction, or retention. For duplicate values, they are directly deleted.

[0034] Because the dimensions and numerical ranges of different variables in the multidimensional health sensing dataset of the utility tunnel vary considerably, the data is standardized or normalized to eliminate the impact of these differences on subsequent analysis. For example, the Z-score standardization method can be used to convert the data into a standard normal distribution with a mean of 0 and a standard deviation of 1, or the Min-Max normalization method can be used to linearly map the data to the [0,1] interval.

[0035] The multi-dimensional sensing dataset for utility tunnel health includes data from different sensors and at different time points. Time series alignment of this data ensures that all relevant variables at each time point are complete and correspond. Time interpolation can be used to supplement data from missing time points.

[0036] After the above processing, the data of each variable are arranged into a matrix in chronological order. The rows of the matrix represent different time points, and the columns represent different variables, thus constructing a multivariate time series matrix of the utility tunnel.

[0037] Step S220: Predict the settlement of the utility tunnel based on the multivariate time series matrix, and obtain the settlement prediction result. Specifically, based on the characteristics and data features of the utility tunnel settlement, a time series prediction model is selected. The multivariate time series matrix of the utility tunnel is divided into a training set and a test set. The prediction model is trained using the training set data, and the model parameters are adjusted so that the model can better fit the training data. The trained model is evaluated using the test set data, and the prediction accuracy of the model is measured by indicators such as mean square error, mean absolute error, and root mean square error. If the prediction accuracy of the model does not meet the requirements, the model can be optimized, such as by adjusting the model structure, increasing the amount of training data, or using ensemble learning methods. The relevant data in the multivariate time series matrix of the utility tunnel is input into the trained and optimized model to predict the settlement of the utility tunnel, and the settlement prediction result for a future period of time is obtained.

[0038] Step S230: Perform causal inference based on the multivariate time series matrix of the utility tunnel to obtain a set of key settlement-causing factors. Specifically, use causal inference methods to analyze the multivariate time series matrix of the utility tunnel. Based on the results of the causal relationship analysis, select variables that have a significant causal relationship with the settlement of the utility tunnel, and form a set of key settlement-causing factors. These factors are the main causes of the settlement of the utility tunnel.

[0039] Step S240: Integrate the settlement prediction results with the set of key settlement-causing factors to generate decision support information. Specifically, integrate the settlement prediction results and the set of key settlement-causing factors; for example, create a table containing the predicted settlement values ​​for the next week and the corresponding values ​​of key settlement-causing factors to show the relationship between the two. Based on the changing trends of the settlement prediction results and key settlement-causing factors, assess the settlement risk of the utility tunnel. By setting different risk levels, determine the current and future risk levels based on the predicted settlement and the numerical range of key settlement-causing factors, thus obtaining decision support information.

[0040] In one possible implementation, the settlement of the utility tunnel is predicted based on the multivariate time series matrix of the tunnel, and the settlement prediction result is obtained. Step S220 further includes step S221, where each sensor node in the multivariate time series matrix of the tunnel is treated as a graph node, and a sensor graph adjacency matrix is ​​constructed based on the physical connection relationship of the tunnel structure. Specifically, in the multivariate time series matrix of the tunnel, each column of data corresponds to the variable information collected by a sensor. Each sensor is regarded as a node in the graph, mapping the sensors in the actual physical world to the graph structure. The structure of the utility tunnel consists of multiple segments and nodes, and different sensors are distributed in different locations in the tunnel. Their physical connection relationship reflects the spatial correlation between the sensors. For example, sensors in adjacent segments are closer in physical structure, and their data have a stronger spatial correlation. The physical connection relationship between sensors can be determined through the design drawings of the utility tunnel, on-site surveys, etc.

[0041] Based on the physical connections between sensors, a sensor graph adjacency matrix is ​​constructed. This matrix is ​​a square matrix where rows and columns correspond to nodes in the graph, i.e., sensors. If two sensors are physically connected, the corresponding position in the adjacency matrix is ​​assigned a value of 1; otherwise, it is assigned a value of 0. The sensor graph adjacency matrix provides spatial structure information for spatiotemporal graph convolution operations.

[0042] Step S222 involves using a spatiotemporal graph convolutional network to predict the settlement of the utility tunnel. The encoder, based on the sensor graph adjacency matrix and the utility tunnel's multivariate time series matrix, captures spatial dependence and temporal dynamics through a spatiotemporal convolution module, generating a spatiotemporal feature vector for the utility tunnel. The decoder, based on the spatiotemporal feature vector, performs autoregressive prediction through a temporal convolutional layer, outputting predicted settlement values ​​and confidence intervals for multiple future time steps, which serve as the settlement prediction result. Specifically, the overall architecture of the spatiotemporal graph convolutional network includes an encoder and a decoder. The encoder receives the sensor graph adjacency matrix and the utility tunnel's multivariate time series matrix as input. Through the spatiotemporal convolution module, the encoder can simultaneously capture the spatial dependence and temporal dynamics between sensor data. Spatial dependence reflects the mutual influence of data between different sensor nodes; for example, displacement sensor data from adjacent sections may be correlated. Temporal dynamics reflect the changing trend of each sensor data over time. After processing by the spatiotemporal convolution module, the encoder generates a spatiotemporal feature vector for the utility tunnel, which integrates important spatial and temporal information about the tunnel. The decoder takes the spatiotemporal feature vector of the utility tunnel generated by the encoder as input and performs autoregressive prediction through a temporal convolutional layer. Autoregressive prediction refers to using past time series data to predict future data. Based on the information in the spatiotemporal feature vector, the decoder predicts the settlement at multiple future time steps and simultaneously outputs the confidence interval of the predicted value, which represents the range of uncertainty of the prediction result.

[0043] The spatiotemporal convolution module is a component of the encoder, comprising spatial convolution and temporal convolution. Spatial convolution utilizes the adjacency matrix of the sensor graph to perform convolution operations on data between different sensor nodes to extract spatial features. For example, by sliding the convolution kernel across the graph structure defined by the adjacency matrix, a weighted sum of the neighboring nodes around each node is calculated, thereby capturing the spatial relationships between nodes. Temporal convolution performs convolution operations on the time-series data of each sensor node to extract temporal features. For example, by sliding a one-dimensional convolution kernel along the time dimension, a weighted sum of the time steps around each time point is calculated, thereby capturing the dynamic changes of the time series.

[0044] The temporal convolutional layer in the decoder is used for autoregressive prediction. It receives the spatiotemporal feature vector generated by the encoder as initial input and then predicts the settlement amount at multiple future time steps through iterative temporal convolution operations. In each iteration, the temporal convolutional layer predicts the settlement amount at the next time step based on the information from the currently predicted time step data and the spatiotemporal feature vector. Simultaneously, confidence intervals for the predicted values ​​are calculated using methods such as Monte Carlo simulations to reflect the uncertainty of the prediction results.

[0045] In one possible implementation, causal inference is performed based on the multivariate time series matrix of the utility tunnel to obtain a set of key sedimentation factors. Step S230 further includes step S231, which involves performing conditional independence tests on all pairs of variables in the multivariate time series matrix of the utility tunnel to construct an undirected causal skeleton graph. Specifically, the multivariate time series matrix of the utility tunnel contains different types of variables collected by multiple sensors, such as displacement, stress, temperature and humidity, and groundwater level. Conditional independence tests include tests based on mutual information and tests based on kernel methods. Taking the test based on mutual information as an example, mutual information measures the statistical dependence between two variables. For variables X and Y, given a set of conditional variables Z, the mutual information of X and Y under condition Z is calculated. If the mutual information is close to zero, X and Y are considered conditionally independent under the condition Z. By performing such tests on all pairs of variables, the conditional independence relationship between variables is determined. Based on the results of the conditional independence test, an undirected causal skeleton graph is constructed. In this graph, nodes represent variables. If two variables are not independent given other variables, an undirected edge is added between their corresponding nodes. The resulting undirected graph reflects the potential causal relationship structure between variables, but the direction of the edges is not yet determined.

[0046] Step S232: Identify V-structures in the undirected causal skeleton graph, and orient the corresponding causal arrows according to the identification results to obtain a partially directed acyclic graph. Specifically, a V-structure refers to a specific connection pattern between three nodes in an undirected causal skeleton graph, where two nodes simultaneously point to a third node, but there is no direct edge connecting these two nodes. For example, in a graph containing nodes A, B, and C, where both A and B point to C, and there is no edge between A and B, this is a V-structure. For the identified V-structure, the direction of the arrows can be determined according to the logic of causal relationships. That is, in a V-structure, there is no direct causal relationship between two nodes pointing to a common node, but the common node is influenced by both nodes. Therefore, in a V-structure, the edges pointing to the common node are oriented from the two pointing nodes to the common node. For example, for the V-structures AC and BC, the undirected edges between A and B and C are oriented as A→C and B→C, respectively. By orienting the arrows of all identified V-structures, some undirected edges in the undirected causal skeleton graph are converted into directed edges, resulting in a partially directed acyclic graph (DAG). A DAG is a graph in which there are no directed cycles, meaning that an event does not directly or indirectly cause itself.

[0047] Step S233: Based on the time priority principle and prior knowledge of the structural mechanics of the utility tunnel, the remaining undirected arrows in the partially directed acyclic graph are oriented to obtain a directed causal graph. Specifically, in causal relationships, events that occur earlier in time are considered causes, and events that occur later are considered effects. Therefore, when processing the remaining undirected arrows, the direction of the arrows can be determined according to the temporal order of the variables. For example, if the time series data of variable M changes before variable N in time, then M can be considered a cause of N, and the undirected edge between M and N can be oriented as M→N.

[0048] Prior knowledge of utility tunnel structural mechanics refers to the theoretical basis for understanding the causal relationships between variables based on professional knowledge of the structural characteristics, material properties, and stress conditions of the utility tunnel. For example, changes in groundwater level affect the stability of the utility tunnel foundation, which in turn affects the displacement and stress of the tunnel, and changes in displacement and stress lead to changes in settlement. Based on this prior knowledge, the direction of some causal relationships can be determined.

[0049] By combining the time-priority principle with prior knowledge of pipe gallery structural mechanics, the remaining undirected arrows in some directed acyclic graphs are oriented, ultimately resulting in a complete directed causal graph that reflects the causal relationships between variables.

[0050] Step S234: Extract all parent nodes that directly point to the settlement node from the directed causal graph to form a set of key settlement-causing factors. Specifically, in the directed causal graph, if a node D has a directed edge pointing to another node E, then node D is the parent node of node E. For a settlement node, the node directly pointing to it is its parent node. These parent nodes directly affect the settlement and are key factors causing changes in settlement. Extracting all parent nodes that directly point to the settlement node and forming a set constitutes the set of key settlement-causing factors.

[0051] Step S300: Dynamically optimize the grouting parameters of the decision support information and the multi-dimensional perception dataset of the pipe gallery health to obtain the optimal grouting strategy.

[0052] Specifically, a grouting parameter optimization model is established using risk levels and key influencing factors from decision support information, as well as relevant data from the multi-dimensional perception dataset of utility tunnel health, as input variables, and grouting parameters such as grouting pressure, grouting volume, and grouting speed as output variables. This model can employ a physical model-based approach, such as a theoretical model combining the structural mechanical properties of the utility tunnel and soil properties; or a data-driven approach, such as training a neural network model using historical grouting data and corresponding utility tunnel state change data. A dynamic optimization algorithm is then used to solve the grouting parameter optimization model. Taking a genetic algorithm as an example, an initial population is first generated, with each individual representing a set of grouting parameter combinations. Then, the fitness value of each individual is calculated according to the fitness function. The population is evolved through selection, crossover, and mutation operations, iteratively optimizing until the termination condition is met, yielding the optimal grouting parameter combination. Based on the optimized grouting parameter combination and the actual situation of the target utility tunnel, an optimal grouting strategy is formulated, including the grouting sequence, grouting parameters at each grouting point, and grouting time arrangement.

[0053] In one possible implementation, the decision support information and the multi-dimensional sensing dataset of pipe gallery health are dynamically optimized to obtain the optimal grouting strategy. Step S300 further includes step S310, defining the decision support information and the multi-dimensional sensing dataset of pipe gallery health as a state space. Specifically, the decision support information and the multi-dimensional sensing dataset of pipe gallery health are used as components of the state space, where the state space can be represented as a multi-dimensional vector, with each dimension corresponding to a specific data type or feature. In this way, the data set is transformed into a structured state space.

[0054] Step S320 defines the grouting pressure, grouting flow rate, grout water-cement ratio, and grouting hole three-dimensional coordinates as the action space. Specifically, grouting pressure, grouting flow rate, grout water-cement ratio, and grouting hole three-dimensional coordinates are key parameters affecting the grouting effect. Grouting pressure refers to the pressure exerted by the grout on the soil or structure surrounding the pipe gallery during the grouting process. It needs to be set according to the actual situation of the pipe gallery and geological conditions. Too high a pressure may damage the pipe gallery structure, while too low a pressure may not effectively fill the voids. Grouting flow rate refers to the volume of grout injected around the pipe gallery per unit time, used to ensure that the grout can be evenly injected into the area that needs to be repaired. Grout water-cement ratio refers to the ratio of the mass of water to the mass of cement in the grout, affecting the fluidity of the grout and its strength after curing. Grouting hole three-dimensional coordinates refer to the spatial coordinates of the grouting hole, which determines the location and range of the grouting.

[0055] These parameters are defined as an action space, which can be represented as a multi-dimensional vector, with each dimension corresponding to an adjustable parameter. For example, one dimension could represent the grouting pressure, and another dimension could represent the grouting flow rate. In practical applications, the range of values ​​for these parameters needs to be constrained to ensure the feasibility and effectiveness of the action space.

[0056] Step S330: Define a reward function based on the settlement recovery value, cumulative grouting volume, and pressure fluctuation variance. Specifically, the settlement recovery value refers to the difference in settlement of the pipe gallery before and after grouting, reflecting the improvement effect of grouting on pipe gallery settlement and is one of the indicators for measuring grouting effect; the cumulative grouting volume refers to the total volume of grout injected around the pipe gallery during the grouting process, which is related to grouting cost and resource consumption, and the grouting volume should be minimized while ensuring grouting effect; the pressure fluctuation variance refers to the variance of pressure data during grouting, reflecting the stability of pressure during grouting, and excessive pressure fluctuation will affect grouting quality.

[0057] The reward function is constructed based on these three parameters. The reward function can be a weighted combination function, for example: R = w1 × ΔS - w2 × Q - w3 × σ 2 Where R represents the reward value, ΔS represents the settlement recovery value, Q represents the cumulative grouting volume, and σ 2 The variance of pressure fluctuation is represented by w1, w2, and w3, which are the weighting coefficients of the three parameters. The determination of the weighting coefficients needs to be adjusted according to actual needs and objectives. For example, if more emphasis is placed on the grouting effect, the value of w1 can be increased; if more emphasis is placed on cost and resource consumption, the value of w2 can be appropriately increased.

[0058] Step S340: Dynamically optimize grouting parameters based on the state space, action space, and reward function to obtain the optimal grouting strategy. Specifically, a reinforcement learning algorithm can be used to dynamically optimize grouting parameters based on the state space, action space, and reward function. The reinforcement learning algorithm allows an agent to interact with the environment, continuously try different actions, and receive feedback based on the reward function, thereby learning the optimal action strategy. The agent perceives the current state in the state space, selects an action (i.e., a set of grouting parameters) based on the action space, executes the action, observes the environmental response (i.e., settlement recovery value, cumulative grouting volume, and pressure fluctuation variance), and calculates the reward value based on the reward function. By continuously repeating this process, the agent gradually learns to select the optimal action under different states, thus obtaining the optimal grouting strategy. When the agent's learning process reaches convergence, i.e., when the reward value no longer significantly increases, the action strategy selected by the agent at this point is the optimal grouting strategy. The optimal grouting strategy includes parameters such as the optimal grouting pressure, grouting flow rate, grout water-cement ratio, and three-dimensional coordinates of the grouting holes to be adopted under different pipe gallery health states.

[0059] In one possible implementation, the grouting parameters are dynamically optimized based on the state space, the action space, and the reward function to obtain the optimal grouting strategy. Step S340 further includes step S341, which involves constructing a simulation environment based on the finite element model of the pipe gallery-soil interaction, and using a near-end strategy optimization algorithm to train the grouting agent according to the state space, the action space, and the reward function until training is complete. Specifically, the finite element model of the pipe gallery-soil interaction is integrated with a reinforcement learning algorithm, and an environment interface is designed to receive the actions output by the grouting agent, i.e., the grouting parameters. These parameters are input into the finite element model of the pipe gallery-soil interaction for calculation to simulate the impact of the grouting process on the pipe gallery-soil system. The calculation results, such as settlement recovery value, cumulative grouting volume, and pressure fluctuation variance, are fed back to the grouting agent as reward signals and new state information.

[0060] Initial simulation conditions are set, such as the initial settlement state of the utility tunnel and the initial stress state of the soil. In each simulation step, the grouting agent selects an action from the action space based on the current state. The environment interface transmits this action to the finite element model of the utility tunnel-soil interaction for calculation, obtaining a new state and reward value. This information is then returned to the grouting agent, which learns and updates its strategy based on this feedback. Through multiple iterative training iterations, the grouting agent gradually learns the optimal grouting strategy. Training completion is indicated by the reward value reaching a stable state or reaching a preset number of training iterations.

[0061] Among them, the proximal policy optimization algorithm is a reinforcement learning algorithm that effectively improves learning efficiency while ensuring the stability of policy updates. The proximal policy optimization algorithm avoids policy performance degradation due to excessively large updates by limiting the magnitude of policy updates. Specifically, it introduces a pruning function to limit the difference between the old and new policies, making policy updates more stable.

[0062] Step S342: The real-time state space is input into the trained grouting agent, which outputs the optimal grouting parameter values ​​to form the optimal grouting strategy. Specifically, the real-time acquired decision support information and the multi-dimensional sensing dataset of the pipe gallery health are used to construct the real-time state space. This real-time state space is input into the trained grouting agent, which, based on its internally learned optimal strategy, selects the optimal grouting parameter values ​​from the action space, including grouting pressure, grouting flow rate, grout water-cement ratio, and the three-dimensional coordinates of the grouting hole. The optimal grouting parameter values ​​output by the grouting agent are combined to form the optimal grouting strategy.

[0063] Step S400: The optimal grouting strategy is sent to the grouting equipment control system to drive the grouting equipment to perform automatic grouting operations.

[0064] Specifically, a standard industrial communication protocol is used to encode and package the optimal grouting strategy according to the protocol's specified format. The encoded data is then transmitted to the grouting equipment control system via wired Ethernet or wireless communication networks. The grouting equipment control system has a corresponding communication interface capable of receiving and parsing the data transmitted according to the communication protocol. After receiving the optimal grouting strategy, the control system parses the data and extracts parameters such as grouting pressure and flow rate. Based on these parameters, the control system automatically adjusts the operating status of the grouting equipment by controlling actuators such as valves and pumps, performing automatic grouting operations according to the optimal grouting strategy. For example, the grouting pressure is controlled by controlling a pressure regulating valve.

[0065] In one possible implementation, the optimal grouting strategy is sent to the grouting equipment control system to drive the automatic grouting operation. Step S400 further includes step S410, acquiring the latest multi-dimensional sensing dataset of the pipe gallery health at preset time intervals. Specifically, the time interval for acquiring data is set according to factors such as the importance of the target pipe gallery, geological conditions, and the characteristics of the grouting operation. For example, for areas with complex geological conditions and sensitive pipe gallery structures, the time interval can be set shorter, such as acquiring data every 5 minutes; for areas with relatively stable geological conditions and slow changes in the pipe gallery status, the time interval can be appropriately extended, such as acquiring data every 30 minutes. Similar to step S100, the latest multi-dimensional sensing dataset of the pipe gallery health is acquired through a multi-source sensor array on the key structural nodes of the target pipe gallery.

[0066] Step S420: Obtain the updated optimal grouting strategy based on the latest multi-dimensional sensing dataset of utility tunnel health. Specifically, repeat steps S200-S300, perform settlement risk analysis based on time-series prediction and causal reasoning using the latest multi-dimensional sensing dataset of utility tunnel health to obtain the latest decision support information, and dynamically optimize the grouting parameters using the latest decision support information and the latest multi-dimensional sensing dataset of utility tunnel health to obtain the updated optimal grouting strategy.

[0067] Step S430: Dynamically adjust the optimal grouting strategy currently performing automatic grouting operations according to the updated optimal grouting strategy. Specifically, after receiving the updated optimal grouting strategy, the grouting equipment control system dynamically adjusts the currently performing automatic grouting operations according to the grouting parameter adjustment instructions in the strategy. For example, if the updated strategy requires a reduction in grouting pressure, the control system will achieve this by adjusting the output pressure of the grouting pump; if it requires a change in grouting flow rate, the control system will adjust the speed of the grouting pump or the opening of the valve. After the strategy adjustment, the effect of the grouting operation and the health status of the pipe gallery are continuously monitored. If the adjusted effect is found to be unsatisfactory or the health status of the pipe gallery is found to be abnormal, the strategy is re-evaluated and adjusted, forming a closed-loop control process.

[0068] This application employs real-time acquisition and fusion of data from multi-source sensor arrays on key structural nodes of the target utility tunnel to obtain a health multi-dimensional perception dataset. This dataset is then used for time-series prediction and causal inference-based settlement risk analysis to obtain decision support information. Combined with the decision support information and the dataset, grouting parameters are dynamically optimized to derive the optimal grouting strategy. This strategy is then sent to the grouting equipment control system to drive automatic grouting operations. These technical means solve the existing technical problem of grouting decision-making for urban underground utility tunnel settlement control, which cannot accurately adjust the grouting strategy in real time according to actual conditions. This achieves the technical effect of improving the accuracy and timeliness of grouting operations and effectively controlling utility tunnel settlement.

[0069] The specific embodiments described above do not constitute a limitation on the scope of protection of this application. Those skilled in the art should understand that various modifications, combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of this application. In some cases, the actions or steps described in this application can be performed in a different order than that shown in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

Claims

1. An intelligent grouting decision-making method for settlement control of urban underground utility tunnels, characterized in that, The method includes: Real-time data acquisition and fusion processing of multi-source sensor arrays on key structural nodes of the target utility tunnel are performed to obtain a multi-dimensional health perception dataset of the utility tunnel. The settlement risk analysis based on time-series prediction and causal reasoning is performed on the multi-dimensional health perception dataset of the utility tunnel to obtain decision support information; The decision support information and the multi-dimensional health perception dataset of the utility tunnel are subjected to dynamic optimization of grouting parameters to obtain the optimal grouting strategy; The optimal grouting strategy is sent to the grouting equipment control system to drive the grouting equipment to perform automatic grouting operations.

2. The intelligent grouting decision-making method for settlement control of urban underground utility tunnels as described in claim 1, characterized in that, Identify the key structural nodes of the target utility tunnel, including: Obtain the engineering design drawings and geological survey report of the target utility tunnel, and construct a finite element model of the interaction between the utility tunnel and the soil. Apply the self-weight load of the utility tunnel, the design traffic load, and the surrounding additional load to the finite element model of the utility tunnel-soil interaction, and perform static settlement simulation calculation to obtain the first displacement cloud map of the utility tunnel structure. Based on the spatial coordinates of the weak soil area identified in the geological survey report, the soil mechanical parameters at the corresponding locations in the finite element model of the pipe gallery-soil interaction are reduced and corrected. Based on the first displacement cloud map, settlement simulation calculation is performed again to obtain the second displacement cloud map. By integrating the first displacement cloud map and the second displacement cloud map, the risk areas of the utility tunnel with displacement greater than a first preset threshold or displacement gradient greater than a second preset threshold are extracted, and the center point and boundary point of the risk area of ​​the utility tunnel are determined as the key structural nodes.

3. The intelligent grouting decision-making method for settlement control of urban underground utility tunnels as described in claim 1, characterized in that, The settlement risk analysis based on time-series prediction and causal reasoning is performed on the multi-dimensional health perception dataset of the utility tunnel to obtain decision support information, including: The multidimensional health perception dataset of the utility tunnel is preprocessed to construct a multivariate time series matrix of the utility tunnel; The settlement of the utility tunnel is predicted based on the multivariate time series matrix of the tunnel, and the settlement prediction result is obtained. Causal inference is performed based on the multivariate time series matrix of the utility tunnel to obtain a set of key sedimentation factors. By integrating the predicted settlement amount with the set of key settlement-causing factors, decision support information is generated.

4. The intelligent grouting decision-making method for settlement control of urban underground utility tunnels as described in claim 3, characterized in that, Based on the multivariate time series matrix of the utility tunnel, the settlement of the utility tunnel is predicted, and the settlement prediction results are obtained, including: Each sensor node in the multivariate time series matrix of the utility tunnel is used as a graph node, and a sensor graph adjacency matrix is ​​constructed based on the physical connection relationship of the utility tunnel structure. The spatiotemporal graph convolutional network is used to predict the settlement of the utility tunnel, where: The encoder captures spatial dependence and temporal dynamics of the utility tunnel using a spatiotemporal convolution module based on the adjacency matrix of the sensor map and the multivariate time series matrix of the utility tunnel, and generates a spatiotemporal feature vector of the utility tunnel. The decoder performs autoregressive prediction through a time convolutional layer based on the spatiotemporal feature vector of the utility tunnel, and outputs the predicted settlement values ​​and confidence intervals for multiple future time steps as the settlement prediction result.

5. The intelligent grouting decision-making method for settlement control of urban underground utility tunnels as described in claim 3, characterized in that, Causal inference is performed based on the multivariate time series matrix of the utility tunnel to obtain a set of key sedimentation factors, including: Conditional independence tests are performed on all variables in the multivariate time series matrix of the utility tunnel, and an undirected causal skeleton graph is constructed. V-structures are identified in the undirected causal skeleton graph, and the corresponding causal arrows are oriented according to the identification results to obtain a partially directed acyclic graph. Based on the time priority principle and prior knowledge of pipe gallery structural mechanics, the remaining undirected arrows in the aforementioned directed acyclic graph are oriented to obtain a directed causal graph. Extract all parent nodes that directly point to the settlement node from the directed causal graph to form a set of key sedimentation factors.

6. The intelligent grouting decision-making method for settlement control of urban underground utility tunnels as described in claim 2, characterized in that, The decision support information and the multi-dimensional health perception dataset of the utility tunnel are subjected to dynamic optimization of grouting parameters to obtain the optimal grouting strategy, including: The decision support information and the multi-dimensional health perception dataset of the utility tunnel are defined as a state space; The grouting pressure, grouting flow rate, grout water-cement ratio, and three-dimensional coordinates of the grouting hole are defined as the action space; The reward function is defined based on the settlement recovery value, cumulative grouting volume, and pressure fluctuation variance; The optimal grouting strategy is obtained by dynamically optimizing the grouting parameters based on the state space, the action space, and the reward function.

7. The intelligent grouting decision-making method for settlement control of urban underground utility tunnels as described in claim 6, characterized in that, Dynamic optimization of grouting parameters is performed based on the state space, the action space, and the reward function to obtain the optimal grouting strategy, including: Based on the finite element model of the pipe gallery-soil interaction, a simulation environment is constructed, and a near-end strategy optimization algorithm is adopted to train the grouting agent according to the state space, the action space and the reward function until the training is completed. The real-time state space is input into the trained grouting agent, which outputs the optimal grouting parameter values ​​to form the optimal grouting strategy.

8. The intelligent grouting decision-making method for settlement control of urban underground utility tunnels as described in claim 1, characterized in that, The optimal grouting strategy is sent to the grouting equipment control system to drive the execution of automatic grouting operations, including: The latest multi-dimensional sensing dataset of utility tunnel health is acquired at preset time intervals; Based on the latest multi-dimensional sensing dataset of pipe gallery health, obtain the updated optimal grouting strategy; The optimal grouting strategy for the ongoing automatic grouting operation is dynamically adjusted based on the updated optimal grouting strategy.

Images