A deep foundation pit monitoring data acquisition method and system

Abstract

This application relates to the field of data processing technology and discloses a method and system for acquiring monitoring data in deep foundation pits. The method includes: deploying sensors at three depth levels—the top, walls, and bottom of the foundation pit—to obtain grouped data; calculating the foundation pit deformation gradient, support structure stress, soil stability, and groundwater seepage parameters, and extracting spatial weight data; using a spatially layered Gaussian mixture clustering algorithm to select representative monitoring sensors; inputting the representative sensor data into a long short-term memory network (LSTM) to predict foundation pit deformation; and dynamically adjusting the monitoring data acquisition frequency through game theory weight allocation and cloud model evaluation to generate data acquisition control commands. This application solves the technical problems of low data acquisition efficiency, lack of scientific basis for representative sensor selection, and susceptibility to misjudgment in single-index early warning systems in deep foundation pit monitoring, thereby improving the intelligence level and prediction accuracy of deep foundation pit monitoring data acquisition.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a method and system for acquiring monitoring data in deep foundation pits. Background Technology

[0002] Deep foundation pit engineering is an important part of urban construction. Traditional methods for monitoring and collecting data on deep foundation pits mainly rely on manual periodic readings or simple automated acquisition systems. These systems monitor parameters such as pit deformation, support structure stress, and soil stability by deploying sensors like displacement gauges, inclinometers, and strain gauges around the pit. Current monitoring methods typically use uniform or empirically distributed sensor locations, collect data at fixed time intervals, and issue safety warnings based on single-index thresholds. Data processing primarily relies on statistical analysis and simple mathematical models to determine deformation trends.

[0003] However, existing technologies suffer from low data acquisition efficiency and omission of key information. Traditional uniform data point distribution methods fail to fully consider the three-dimensional spatial structure characteristics of deep foundation pits ("pit top-pit wall-pit bottom") and the differences in monitoring focus at different depth levels, resulting in unreasonable allocation of monitoring resources. Furthermore, fixed-frequency data acquisition modes cannot be dynamically adjusted according to the safety status of the foundation pit, leading to data redundancy during safe periods and insufficient monitoring during dangerous periods. In addition, early warning methods based on single indicators are prone to misjudgment and cannot accurately reflect the comprehensive safety status of the foundation pit. Traditional data processing methods lack the ability to intelligently analyze massive amounts of monitoring data, making it difficult to extract key features and predict deformation trends from complex multidimensional data.

[0004] Based on the above analysis, the fundamental problem with existing technologies lies in the lack of intelligent monitoring data acquisition methods for the spatial characteristics of deep foundation pits. Specifically, this manifests in the following ways: how to intelligently select representative monitoring sensors based on the spatial stratification characteristics and soil layer distribution patterns of deep foundation pits to reduce data redundancy while ensuring monitoring coverage; how to establish a multi-parameter coupled prediction model to learn the deformation patterns of foundation pits from historical monitoring data and accurately predict future deformation trends; and how to construct a comprehensive safety evaluation system that integrates multiple monitoring indicators to achieve dynamic adjustment of data acquisition frequency based on comprehensive safety status, thereby resolving the contradiction between monitoring accuracy and efficiency in traditional methods. Summary of the Invention

[0005] This application provides a method and system for acquiring monitoring data in deep foundation pits, which solves the technical problems of low data acquisition efficiency, lack of scientific basis for the selection of representative sensors, and easy misjudgment of single-index early warning in deep foundation pit monitoring, thereby improving the intelligence level and prediction accuracy of deep foundation pit monitoring data acquisition.

[0006] Firstly, this application provides a method for acquiring monitoring data in deep foundation pits. The method includes: spatially stratifying monitoring sensors according to three depth levels—the pit top, pit wall, and pit bottom—to obtain sensor group data; extracting feature values ​​from the sensor group data by calculating the pit deformation gradient, support structure stress, soil stability, and groundwater seepage parameters to obtain spatial weight data; filtering the spatial weight data using a deep foundation pit spatial stratification Gaussian mixture clustering algorithm to obtain representative monitoring sensors; inputting the monitoring data from the representative monitoring sensors into a deep foundation pit long short-term memory network for prediction processing to obtain foundation pit deformation prediction results; and dynamically adjusting the monitoring data acquisition frequency based on the foundation pit deformation prediction results through game theory weight allocation and cloud model evaluation to obtain data acquisition control commands.

[0007] Secondly, this application provides a deep foundation pit monitoring data acquisition system, the deep foundation pit monitoring data acquisition system comprising:

[0008] The deployment module is used to spatially deploy monitoring sensors according to the three depth levels of the deep foundation pit: the top of the pit, the pit wall, and the bottom of the pit, to obtain sensor group data.

[0009] The extraction module is used to extract feature values ​​from the sensor group data by calculating the foundation pit deformation gradient, support structure stress, soil stability and groundwater seepage parameters to obtain spatial weight data.

[0010] The filtering module is used to filter the spatial weight data using a deep foundation pit spatial hierarchical Gaussian mixture clustering algorithm to obtain representative monitoring sensors;

[0011] The prediction module is used to input the monitoring data of the representative monitoring sensors into the long short-term memory network of the deep foundation pit for prediction processing, and obtain the foundation pit deformation prediction result.

[0012] The adjustment module is used to dynamically adjust the monitoring data acquisition frequency based on the foundation pit deformation prediction results through game theory weight allocation and cloud model evaluation, and obtain data acquisition control commands.

[0013] Thirdly, a deep foundation pit monitoring data acquisition device is provided, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor invokes the instructions in the memory to cause the deep foundation pit monitoring data acquisition device to execute the aforementioned deep foundation pit monitoring data acquisition method.

[0014] Fourthly, a computer-readable storage medium is provided, wherein instructions are stored therein, which, when executed on a computer, cause the computer to perform the aforementioned deep foundation pit monitoring data acquisition method.

[0015] The technical solution provided in this application utilizes a spatially layered deployment of monitoring sensors based on three depth levels: the top, wall, and bottom of the deep foundation pit. This approach fully considers the spatial structural characteristics of deep foundation pit projects and the differences in monitoring focus at different depth levels. Compared to the traditional uniform deployment method, this approach is more scientific and reasonable, effectively avoiding waste of monitoring resources and monitoring blind spots in key areas. The technical feature of extracting feature values ​​from sensor group data by calculating the foundation pit deformation gradient, support structure stress, soil stability, and groundwater seepage parameters transforms complex multidimensional monitoring data into feature parameters with clear physical meaning. This lays a solid foundation for subsequent data processing and analysis, solving the problem of diverse but poorly integrated monitoring data in traditional methods. The application of the spatially layered Gaussian mixture clustering algorithm for deep foundation pits fully demonstrates the significant contribution of the algorithm to the solution. By incorporating the spatial layering features of the deep foundation pit to filter spatial weight data, this algorithm intelligently identifies sensor combinations with similar monitoring features, thereby scientifically selecting representative monitoring sensors. This ensures monitoring coverage while significantly reducing data redundancy, offering higher objectivity and accuracy compared to traditional empirical selection methods.

[0016] The predictive processing technology of long short-term memory networks for deep foundation pits plays a unique role in the field of deep foundation pit monitoring data acquisition. This network, by incorporating historical deformation data, soil pressure changes, and excavation depth—parameters specific to deep foundation pits—establishes a deformation prediction model tailored to the characteristics of deep foundation pit engineering. Compared to general neural network models, it exhibits stronger domain adaptability and prediction accuracy, effectively solving the problem that traditional monitoring methods cannot predict foundation pit deformation trends in advance. The technology of dynamically adjusting the monitoring data acquisition frequency through game theory weight allocation and cloud model evaluation, by constructing a five-dimensional monitoring index system and using a game theory Nash equilibrium algorithm to optimize weight allocation, achieves comprehensive evaluation of multiple indicators and dynamic weight adjustment. This overcomes the limitations of traditional single-indicator early warning systems, which are prone to misjudgment. Simultaneously, the introduction of cloud models effectively addresses the fuzziness and uncertainty issues in safety evaluation, enabling intelligent adjustment of the data acquisition frequency based on the comprehensive safety status of the foundation pit, significantly improving data acquisition efficiency while ensuring safe monitoring. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1This is a schematic diagram of one embodiment of the deep foundation pit monitoring data acquisition method in this application.

[0019] Figure 2 This is a schematic diagram of one embodiment of the deep foundation pit monitoring data acquisition system in this application.

[0020] Figure 3 This is a schematic block diagram of the deep foundation pit monitoring data acquisition device in an embodiment of the present invention. Detailed Implementation

[0021] This application provides a method and system for acquiring monitoring data in deep foundation pits. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or device 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 units not explicitly listed or inherent to such processes, methods, products, or devices.

[0022] For ease of understanding, the specific process of the embodiments of this application is described below. Please refer to [link / reference]. Figure 1 One embodiment of the deep foundation pit monitoring data acquisition method in this application includes:

[0023] Step S101: Based on the three depth levels of the deep foundation pit—the top, wall, and bottom—the monitoring sensors are spatially layered and deployed to obtain sensor group data.

[0024] Step S102: The sensor group data is processed by feature value extraction by calculating the foundation pit deformation gradient, support structure stress, soil stability and groundwater seepage parameters to obtain spatial weight data.

[0025] Step S103: The spatial weight data is filtered and processed using the deep foundation pit spatial hierarchical Gaussian mixture clustering algorithm to obtain representative monitoring sensors;

[0026] Step S104: Input the monitoring data of the representative monitoring sensor into the deep foundation pit long short-term memory network for prediction processing to obtain the foundation pit deformation prediction result;

[0027] Step S105: Based on the predicted foundation pit deformation, the monitoring data acquisition frequency is dynamically adjusted using game theory weight allocation and cloud model evaluation to obtain data acquisition control instructions.

[0028] It is understood that the executing entity of this application can be a deep foundation pit monitoring data acquisition system, or it can be a terminal or a server; the specific implementation is not limited here. This application's embodiment uses a server as an example for illustration.

[0029] Specifically, when deploying sensors in a spatially layered manner according to the three depth levels of the deep foundation pit—the top, walls, and bottom—surface settlement monitoring points and building tilt monitoring devices are installed at 15-20 meter intervals in the stable layer at the top of the pit. These devices primarily monitor the impact of the foundation pit excavation on the surface and surrounding buildings, obtaining the precise spatial coordinates of each monitoring point through GPS positioning technology. Horizontal displacement gauges and inclinometers are deployed along the pit walls at 3-5 meter vertical depths to form a deformation monitoring grid. The horizontal displacement gauges measure the horizontal deformation of the foundation pit walls, and the inclinometers measure the changes in the tilt angle of the foundation pit walls. These devices are precisely positioned based on the depth of the foundation pit and changes in soil layers. Earth pressure gauges and pore water pressure gauges are strategically deployed in the stress concentration layer at the bottom of the pit. The earth pressure gauges measure the lateral pressure of the soil on the support structure, and the pore water pressure gauges monitor changes in groundwater pressure. All sensors are assigned a unique spatial coding system based on depth level, group number within the same layer, and sensor serial number within the group.

[0030] The deformation gradient coefficient of the foundation pit is calculated by the ratio of the displacement difference between adjacent monitoring points to their distance. When the displacements of two adjacent monitoring points are 5 mm and 8 mm respectively, and the distance is 3 meters, the deformation gradient coefficient is 1 mm per meter. The stress concentration coefficient of the support structure is calculated by the ratio of the measured stress value of the support structure to the theoretical design stress value. When the measured stress is 120 kPa and the design stress is 100 kPa, the stress concentration coefficient is 1.2. The soil stability index is calculated based on the ratio of the soil shear strength to the actual shear stress. When the shear strength is 80 kPa and the actual shear stress is 60 kPa, the stability index is 1.33. The groundwater seepage influence factor is calculated by multiplying the groundwater level change rate and the permeability coefficient. When the water level change rate is 2 mm per hour and the permeability coefficient is 0.001 cm per second, the seepage influence factor has a specific value. These four feature values ​​are linearly combined according to weight coefficients to form spatial weight data. The spatial weight data is filtered using a deep foundation pit spatial hierarchical Gaussian mixture clustering algorithm. Gaussian mixture models are probabilistic models that assume data is a mixture of multiple Gaussian distributions, each representing a cluster. The deep foundation pit spatial stratification Gaussian mixture clustering algorithm incorporates the spatial stratification features of deep foundation pits into the traditional Gaussian mixture model. It initializes the parameters of the four-dimensional feature vector combination through depth-level weights, cluster centers within the same level, and the covariance matrix. The expectation-maximization algorithm calculates the posterior probability of each sensor belonging to each cluster through the expectation step, and updates the cluster centers and covariance matrix parameters through the maximization step, iterating repeatedly until convergence. After convergence, sensor clusters with similar monitoring characteristics are obtained. From each cluster, the sensor closest to the cluster center is selected as the representative monitoring sensor, thus ensuring monitoring coverage while reducing data acquisition redundancy.

[0031] Monitoring data from representative sensors are input into a Long Short-Term Memory (LSTM) network for predictive processing in deep foundation pits. LTM is a special type of recurrent neural network capable of learning long-term dependencies. The deep foundation pit LTM network incorporates specific parameters for deep foundation pits into the standard network, including cumulative deformation, soil pressure, and excavation depth data, combined in a time-series fashion to form the network input vector. The forget gate uses a sigmoid activation function to determine which information is discarded from the cell state, the input gate determines which new information is stored in the cell state, and the output gate controls which parts of the cell state are output to the hidden state. The attention mechanism calculates the weights of the hidden states at different times, highlighting important time-step information. This attention-weighted feature is input into the output layer to calculate the deformation increment of the foundation pit at the next time step, and then added to the current cumulative deformation to obtain the predicted foundation pit deformation.

[0032] Based on the predicted foundation pit deformation, dynamic adjustments are made using game theory weight allocation and cloud model evaluation. Game theory weight allocation uses the Nash equilibrium algorithm to find the optimal combination of monitoring parameter weights, ensuring that none of the participants can gain better benefits by unilaterally changing their strategies. The cloud model is an uncertainty transformation model between qualitative concepts and quantitative values, describing the randomness and fuzziness of the concept through three numerical characteristics: expected value, entropy value, and hyperentropy. Expected value represents the mathematical expectation of the concept, entropy value represents the randomness of the concept, and hyperentropy represents the uncertainty of entropy. Cloud model parameters are constructed based on soil layer characteristics, and membership functions are generated to evaluate the safety assessment indicators of deep foundation pits. The optimal acquisition weight allocation and the foundation pit safety level are combined using a comprehensive evaluation function calculated according to the foundation pit depth correction coefficient. Based on the comprehensive safety level, a threshold for the acquisition frequency is set, and representative monitoring sensors are adjusted in stages: the acquisition frequency is reduced when the safety level is high and increased when the safety level is low. Finally, data acquisition control commands are generated to guide on-site monitoring work.

[0033] In one specific embodiment, the process of performing step S101 may specifically include the following steps:

[0034] Surface settlement monitoring points and building tilt monitoring equipment were set up at intervals of 15 to 20 meters in the stable layer at the top of the pit to obtain the coordinates of the top monitoring points.

[0035] Horizontal displacement gauges and inclinometers are laid out along the pit wall at vertical depths of 3 to 5 meters to form a deformation monitoring grid, thus obtaining the spatial distribution of the monitoring equipment in the middle.

[0036] Earth pressure gauges and pore water pressure gauges were deployed in the stress concentration layer at the bottom of the pit to monitor the stress state of the soil, resulting in a bottom stress monitoring array.

[0037] A sensor spatial coordinate system is established based on the coordinates of the top monitoring point, the spatial distribution of the middle monitoring equipment, and the bottom stress monitoring array to obtain the unique spatial code of the sensor.

[0038] Based on the soil layer interface, the sensors within each depth level are grouped to obtain sensor group data containing the depth level, the group number of the same layer, and the sensor serial number within the group.

[0039] Specifically, the spatial layered deployment of deep foundation pit monitoring sensors first involves setting up surface settlement monitoring points and building tilt monitoring equipment at 15-20 meter intervals in the stable layer at the top of the pit. The surface settlement monitoring points utilize leveling benchmarks, with ground elevation changes measured using a precision level. The building tilt monitoring equipment uses tilt sensors installed on the walls of surrounding buildings to measure the building's offset angle relative to the vertical direction. Each monitoring point obtains precise three-dimensional coordinates, including longitude, latitude, and elevation information, through a GPS positioning system. These coordinate data constitute the coordinate set of the top monitoring point, with a required accuracy at the millimeter level. RTK differential positioning technology is used to eliminate GPS signal errors, ensuring the accuracy of the coordinate data. Horizontal displacement gauges and inclinometers are deployed at vertical depths of 3 to 5 meters along the pit wall to form a deformation monitoring grid. The horizontal displacement gauges measure the horizontal displacement of the pit wall by pulling steel wires or optical fibers, while the inclinometers measure the tilt angle of the pit wall relative to the vertical direction using the principle of electronic bubble or accelerometer. These devices are arranged vertically according to depth levels, with each depth level having the same horizontal elevation and the vertical spacing between different depth levels remaining consistent. The precise spatial position of each device is determined by laser rangefinders and levels, forming a three-dimensional grid structure of spatial distribution of monitoring devices in the center.

[0040] Earth pressure gauges and pore water pressure gauges are strategically deployed in the stress concentration layer at the bottom of the pit to monitor the soil stress state. The earth pressure gauges, employing vibrating wire or resistance strain gauge sensors, are installed behind the foundation pit support structure or within the soil to directly measure the lateral pressure exerted by the soil on the support structure. The pore water pressure gauges sense changes in pore water pressure within the soil through porous ceramic heads. These sensors are arranged according to the soil layer distribution and stress concentration areas, forming a bottom stress monitoring array. The installation position of each sensor is determined through drilling and measurement; depth coordinates are obtained through borehole depth measurement, and horizontal coordinates are calculated from ground reference points. A sensor spatial coordinate system is established based on the coordinates of the top monitoring points, the spatial distribution of the middle monitoring equipment, and the bottom stress monitoring array. This coordinate system uses the center point of the foundation pit as the origin, with the X-axis pointing east, the Y-axis pointing north, and the Z-axis vertically upward. Each sensor has a unique three-dimensional coordinate value in this coordinate system. The coordinate transformation is performed from the geodetic coordinate system to the engineering coordinate system to eliminate the influence of the Earth's curvature, resulting in a unique spatial code for each sensor.

[0041] Spatial coding employs a hierarchical coding structure, including depth level identifiers, equipment type identifiers, horizontal position identifiers, and vertical position identifiers. Depth level identifiers use numbers 1, 2, and 3 to represent the pit top, pit wall, and pit bottom levels, respectively. Equipment type identifiers use letters A, B, C, and D to represent settlement monitoring points, horizontal displacement gauges, inclinometers, and pressure gauges, respectively. Horizontal position identifiers use a grid numbering method to divide the monitoring area into several grid units. Vertical position identifiers indicate the layer number of the equipment in the vertical direction. Sensors within each depth level are grouped according to soil layer interfaces. Soil layer interfaces are determined through geological exploration data, including information such as soil layer thickness, soil type, and physical and mechanical parameters. Sensors within the same soil layer are grouped together, while sensors from different soil layers are grouped separately. Each group of sensors has similar soil environment and mechanical properties. The grouped data consists of three levels: depth level number, same-level group number, and sensor sequence number within the group. The depth level number indicates the vertical layer in which the sensor is located, the same-level group number indicates the grouping based on the soil interface within the same depth level, and the sensor sequence number within the group indicates the order of the sensors within the same group. This hierarchical grouping method forms the sensor grouped data structure.

[0042] In one specific embodiment, the process of performing step S102 may specifically include the following steps:

[0043] The foundation pit deformation gradient coefficient is obtained by calculating the ratio of the displacement difference to the distance between adjacent monitoring points and processing the sensor group data.

[0044] The stress concentration factor of the support structure is calculated by processing the sensor group data based on the ratio of the measured stress value of the support structure to the theoretical design stress value.

[0045] The soil stability index is calculated by processing the sensor group data based on the ratio of soil shear strength to actual shear stress.

[0046] The groundwater seepage influence factor is calculated by multiplying the groundwater level change rate and the permeability coefficient by the sensor group data.

[0047] The spatial weighted data are obtained by linearly combining the foundation pit deformation gradient coefficient, the support structure stress concentration coefficient, the soil stability index, and the groundwater seepage influence factor according to the weight coefficients.

[0048] Specifically, the calculation and processing of the foundation pit deformation gradient coefficient first extracts the displacement measurement values ​​of adjacent monitoring points from the sensor group data. These displacement measurements are obtained using a horizontal displacement gauge and a tiltmeter. The horizontal displacement gauge measures the absolute displacement of the foundation pit wall in the horizontal direction, while the tiltmeter measures the angular change of the foundation pit wall relative to the vertical direction, converting the angular change into a horizontal displacement increment. The displacement difference between adjacent monitoring points is obtained by subtracting the displacement measurement values ​​of two monitoring points. The distance is calculated from the spatial distance between the two monitoring points, which is calculated based on the three-dimensional coordinates in the sensor's spatial coordinate system. The foundation pit deformation gradient coefficient is equal to the displacement difference divided by the distance. This coefficient reflects the spatial distribution and trend of foundation pit deformation; a larger value indicates a more significant difference in deformation between adjacent areas. The calculation and processing of the support structure stress concentration coefficient is based on a comparative analysis of the measured stress values ​​and the theoretical design stress values ​​of the support structure. The measured stress values ​​are obtained directly by strain gauges or pressure sensors installed on the support structure. The strain gauge measures the strain change of the support structure, converting the strain into stress based on the material's elastic modulus. The pressure sensor directly measures the magnitude of the pressure borne by the support structure. The theoretical design stress value is determined based on the structural calculations in the foundation pit design scheme, including parameters such as the theoretical calculated value of earth pressure and the design value of structural bearing capacity. The design stress value takes into account factors such as soil layer parameters, foundation pit geometry, and the material properties of the support structure. The stress concentration factor of the support structure is equal to the measured stress value divided by the theoretical design stress value. When the measured stress value exceeds the design stress value, the factor is greater than 1, indicating that the support structure is bearing a load exceeding the design expectation. When the measured stress value is less than the design stress value, the factor is less than 1, indicating that the support structure is in a safe state.

[0049] The soil stability index is calculated based on the relationship between soil shear strength and actual shear stress to determine the degree of soil stability. Soil shear strength is determined through geotechnical tests, including direct shear tests and triaxial tests, to obtain the soil's internal friction angle and cohesion parameters. Shear strength is calculated according to Coulomb's law, i.e., shear strength equals cohesion plus the product of normal stress and the tangent of the internal friction angle. Actual shear stress is calculated through stress state analysis within the soil. The principal stress state within the soil is determined based on factors such as soil self-weight, groundwater pressure, and unloading during excavation. The maximum shear stress value is calculated using the Mohr's circle analysis method. The soil stability index equals the soil shear strength divided by the actual shear stress. This index characterizes the soil's ability to resist shear failure. An index greater than 1 indicates the soil is in a stable state, an index close to 1 indicates the soil is nearing the critical failure state, and an index less than 1 indicates the soil may experience shear failure.

[0050] The calculation of the groundwater seepage impact factor quantifies the influence of groundwater on the stability of the foundation pit by multiplying the groundwater level change rate and the permeability coefficient. The groundwater level change rate is obtained through continuous monitoring data from pore water pressure gauges and groundwater level observation wells. Automatic water level recorders are installed in the observation wells, recording groundwater level data hourly. The rate of change is obtained by dividing the difference in water level data between adjacent time points by the time interval. The permeability coefficient is determined through pumping or injection tests. In pumping tests, a pump is installed in the observation well to pump water at a constant flow rate, measuring the water level drop and the stabilized water level. The soil permeability coefficient is calculated based on Darcy's law and well flow theory. In injection tests, a fixed amount of water is injected into the well, measuring the water level rise and infiltration time, and calculating the soil's permeability performance. The groundwater seepage impact factor equals the groundwater level change rate multiplied by the permeability coefficient; this factor reflects the intensity of the impact of groundwater seepage on the effective stress of the soil and the stability of the foundation pit.

[0051] Linear combination calculations sum the four eigenvalues ​​according to their weighting coefficients to obtain spatial weight data. These weighting coefficients are determined based on the degree of influence of each eigenvalue on the safety state of the foundation pit. The numerical range of each weighting coefficient is determined through engineering experience and statistical analysis methods. The foundation pit deformation gradient coefficient weight reflects the degree of influence of deformation on structural safety; the support structure stress concentration coefficient weight reflects the importance of the structure's bearing capacity; the soil stability index weight reflects the influence of soil stability on overall safety; and the groundwater seepage influence factor weight reflects the degree of influence of groundwater on foundation pit stability. The spatial weight data equals the foundation pit deformation gradient coefficient multiplied by its weighting coefficient, plus the support structure stress concentration coefficient multiplied by its weighting coefficient, plus the soil stability index multiplied by its weighting coefficient, plus the groundwater seepage influence factor multiplied by its weighting coefficient. The calculation result forms the spatial weight value corresponding to each sensor group.

[0052] For example, in the monitoring of a deep foundation pit project, two adjacent horizontal displacement gauges measured displacement values ​​of 8 mm and 12 mm respectively, with a distance of 5 meters between them. The deformation gradient coefficient of the foundation pit was calculated as the displacement difference of 4 mm divided by the distance of 5 meters, resulting in 0.8 mm / m. Strain gauges installed on the support structure measured a strain value of 200 microstrains. Based on the elastic modulus of steel (200 GPa), the measured stress value was calculated to be 40 MPa, the design stress value was 35 MPa, and the stress concentration factor of the support structure was 40 divided by 35, equaling 1.14. Geotechnical tests measured a cohesion of 25 kPa, an internal friction angle of 30 degrees, and a shear strength of 54 kPa under a normal stress of 50 kPa. The actual shear stress was calculated through stress analysis to be 45 kPa, and the soil stability index was 54 divided by 45, equaling 1.2. Groundwater level observation wells recorded a water level drop of 2 mm per hour, with a permeability coefficient of 1 x 10⁻⁶ m / s. The groundwater seepage influence factor was calculated by multiplying the water level change rate by the permeability coefficient to obtain a specific value. The spatial weight data of the sensor group is obtained by linearly combining these four feature values ​​with weighting coefficients of 0.3, 0.25, 0.3, and 0.15.

[0053] In one specific embodiment, the process of executing step S103 may specifically include the following steps:

[0054] The foundation pit deformation gradient coefficient, support structure stress concentration coefficient, soil stability index and groundwater seepage influencing factor in the spatial weight data are constructed as clustering feature vectors to obtain a four-dimensional feature vector combination.

[0055] The four-dimensional feature vector combination is initialized with Gaussian mixture model parameters based on the deep hierarchical weights, same-level cluster centers and covariance matrix to obtain the initial clustering parameters.

[0056] The initial clustering parameters are iteratively optimized using the expectation-maximization algorithm to obtain converged clustering parameters and posterior probability distribution.

[0057] The sensors are clustered and grouped according to the converged clustering parameters to obtain sensor clusters with similar monitoring characteristics;

[0058] Representative monitoring sensors are obtained by selecting the sensor closest to the cluster center from each sensor cluster.

[0059] Specifically, the data processing of the deep foundation pit spatial hierarchical Gaussian mixture clustering algorithm first constructs clustering feature vectors from the spatial weight data, including the foundation pit deformation gradient coefficient, support structure stress concentration coefficient, soil stability index, and groundwater seepage influence factor. Each sensor corresponds to a four-dimensional feature vector. The first component of the vector is the foundation pit deformation gradient coefficient, reflecting the deformation gradient characteristics at the sensor location; the second component is the support structure stress concentration coefficient, reflecting the structural stress state at that location; the third component is the soil stability index, reflecting the soil stability at that location; and the fourth component is the groundwater seepage influence factor, reflecting the intensity of groundwater influence at that location. The four-dimensional feature vectors are combined to form a feature matrix, with the number of rows equal to the total number of sensors and four columns. Each row represents a complete feature description of a sensor. Based on the depth-level weights, same-layer cluster centers, and covariance matrix, the four-dimensional feature vector combination is initialized with Gaussian mixture model parameters. The depth-level weights are determined according to the importance of the pit top, pit wall, and pit bottom layers. The pit wall layer has the highest weight due to its most sensitive deformation, while the pit top and pit bottom layers have relatively lower weights. Cluster centers within the same depth level are determined by calculating the mean of the feature vectors of all sensors within that same depth level. Each depth level corresponds to an initial cluster center, which is a four-dimensional vector representing the average feature of the sensors at that level. The covariance matrix describes the correlation and variability among the components of the feature vectors. It is obtained by calculating the covariance of the feature vectors of sensors within the same depth level. The covariance matrix is ​​a 4x4 symmetric matrix, with diagonal elements representing the variance of each feature component and off-diagonal elements representing the covariance between different feature components.

[0060] The Expectation-Maximization (EM) algorithm iteratively optimizes the initial clustering parameters. This algorithm is an iterative method for solving probabilistic model parameters containing latent variables, progressively optimizing the model parameters through alternating expectation and maximization steps. The expectation step calculates the posterior probability of each sensor belonging to each cluster. The posterior probability is calculated using Bayes' theorem; the numerator is the product of the prior probability of that cluster and the likelihood probability of that sensor within that cluster, and the denominator is the sum of the prior probabilities multiplied by the likelihood probabilities of all clusters. The likelihood probability is calculated using a multidimensional Gaussian probability density function. The input is the sensor's feature vector, and the parameters are the cluster centers and the covariance matrix. The maximization step updates the clustering parameters based on the posterior probabilities. The new cluster centers are equal to the weighted average of all sensor feature vectors according to their posterior probabilities of belonging to that cluster. The new covariance matrix is ​​equal to the weighted average of the outer products of the deviation vectors of all sensor feature vectors and the new cluster centers, calculated according to their posterior probabilities. The new prior probability is equal to the average of the posterior probabilities of all sensors belonging to that cluster. The algorithm repeatedly executes the expected step and the maximization step until the change in the clustering parameters is less than the preset threshold or the maximum number of iterations is reached, thus obtaining the converged clustering parameters and posterior probability distribution.

[0061] The sensors are clustered based on converged clustering parameters. Each sensor is assigned to the cluster with the highest probability according to its posterior probability for each cluster. Sensors belonging to the same cluster have similar monitoring characteristics, indicating that the foundation pit areas monitored by these sensors have similar deformation patterns, stress states, stability, and groundwater influence characteristics. The number of clusters is determined based on the complexity of the foundation pit and monitoring requirements, typically set to 3 to 6 clusters, with each cluster representing a typical foundation pit monitoring state mode. The sensor clustering process solves the data redundancy problem caused by equal weighting of all sensors in traditional monitoring methods. By identifying sensor combinations with similar characteristics, it lays the foundation for subsequent selection of representative sensors.

[0062] From each sensor cluster, the sensor closest to the cluster center is selected for representative screening. The distance is calculated using the Euclidean distance method, determining the distance between each sensor's four-dimensional feature vector and the cluster center. The distance is equal to the square root of the sum of the squares of the differences between the feature vector and each component of the cluster center. The sensor with the smallest distance indicates that its feature vector is closest to the average feature of the cluster and best represents the monitoring characteristics of that cluster; therefore, it is selected as the representative monitoring sensor. This selection process ensures that each cluster has a most representative sensor participating in subsequent data acquisition and predictive analysis, guaranteeing comprehensive monitoring while avoiding redundant data processing.

[0063] In one specific embodiment, the process of executing step S104 may specifically include the following steps:

[0064] The cumulative deformation, earth pressure, and excavation depth data from the representative monitoring sensors are combined and processed according to time series to obtain the network input vector;

[0065] The network input vector is updated using long short-term memory units based on forget gates, input gates, and output gates to obtain the hidden state output.

[0066] The hidden state output is weighted and fused by calculating attention weights to obtain attention-weighted features;

[0067] The attention-weighted features are input into the output layer of the neural network to calculate the deformation amount, thereby obtaining the deformation increment of the foundation pit at the next moment.

[0068] The deformation prediction result of the foundation pit is obtained by accumulating the deformation increment of the foundation pit at the next moment and the current cumulative deformation.

[0069] Specifically, the long short-term memory network prediction processing for deep foundation pits first combines the cumulative deformation, earth pressure, and excavation depth data from representative monitoring sensors according to time series. The cumulative deformation data comes from continuous monitoring records of horizontal displacement gauges and inclinometers, with data collected hourly and accumulated to calculate the total deformation from the start of excavation to the current moment. The earth pressure data is obtained by measuring earth pressure gauges installed on the back of the support structure, reflecting the magnitude of the lateral pressure exerted by the soil on the support structure. The excavation depth data is determined based on the construction progress records, representing the excavation depth of the foundation pit at the current moment. The time series combination processing arranges the three types of data from consecutive time points in chronological order to form a sequence vector. Each time step contains three numerical components, and the sequence length is set according to the prediction requirements. Typically, the data from the most recent 10 to 20 time steps are selected as the input sequence to form the network input vector matrix. The number of rows in the matrix is ​​equal to the time step length, and the number of columns is 3, with each row representing the complete monitoring status of one time step.

[0070] The Long Short-Term Memory (LSTM) unit state update process calculates the network input vector based on forget gates, input gates, and output gates. The forget gate determines which historical information is forgotten from the cell state, the input gate determines which new information is stored in the cell state, and the output gate controls which parts of the cell state are output to the hidden state. The forget gate processes the input vector at the current time step and the hidden state at the previous time step using the sigmoid activation function. The sigmoid function maps input values ​​between 0 and 1, representing the degree of retention of historical information; values ​​close to 0 indicate complete forgetting, and values ​​close to 1 indicate complete retention. The calculation result of the forget gate is multiplied element-wise with the cell state at the previous time step, achieving selective forgetting of historical information. The input gate also processes the current input and the previous hidden state using the sigmoid activation function, and simultaneously generates a candidate value vector using the tanh activation function. The tanh function maps input values ​​between -1 and 1, representing candidate states for new information. The output of the input gate is multiplied element-wise with the candidate value vector to obtain the new information to be added to the cell state. Cell state updates are obtained by adding the historical information processed by the forget gate to the new information processed by the input gate. The output gate calculates the output control signal using the sigmoid function. The hidden state output is obtained by multiplying the cell state processed by the tanh function with the output gate signal.

[0071] Attention weight calculation performs weighted fusion processing on the hidden state output. The attention mechanism is a method for calculating the importance weights of different time steps. The attention weight is determined by calculating the similarity between the hidden state at each time step and the query vector at the current time step. The similarity calculation uses the dot product attention method, performing an inner product operation between the hidden state at each time step and the query vector. The inner product result reflects the similarity between the two vectors; a larger value indicates a higher similarity. The similarity scores of all time steps are normalized using the softmax function. The softmax function transforms any real-valued vector into a probability distribution, ensuring that the sum of all weights equals 1. The weight of each time step represents its importance to the current prediction task. Attention-weighted features are obtained by weighting and summing the hidden states of all time steps according to their corresponding attention weights. This weighted summation process highlights the features of important time steps and reduces the influence of unimportant time steps, forming a feature representation that integrates all historical information but highlights key moments.

[0072] The neural network output layer calculates the deformation increment, converting the attention-weighted features into the deformation increment of the pit at the next time step. The output layer employs a fully connected structure, mapping high-dimensional features to one-dimensional deformation increment predictions through linear transformation and activation functions. The linear transformation involves matrix multiplication between the weight matrix and the attention-weighted features. The parameters of the weight matrix are learned from the training data, reflecting the mapping relationship between the features and the deformation increment. A linear activation function is used, directly outputting the result of the linear transformation, as deformation increment prediction is a regression task and does not require non-linear activation function processing. The predicted deformation increment of the pit at the next time step represents the increase in pit deformation from the current time step to the next time step. The value may be positive, indicating continued deformation increase, or negative, indicating a decrease in deformation.

[0073] The cumulative calculation process adds the deformation increment of the foundation pit at the next time step to the current cumulative deformation to obtain the predicted deformation result. The current cumulative deformation is the total deformation from the start of excavation to the current time step, calculated by accumulating historical monitoring data. The cumulative calculation then adds the current cumulative deformation to the predicted deformation increment to obtain the predicted cumulative deformation value for the next time step. This predicted value represents the total deformation state of the foundation pit at the next time step. The cumulative calculation process solves the problem of traditional monitoring methods being unable to predict the deformation trend of the foundation pit in advance. Through learning from historical data and pattern recognition, the network can predict the future deformation state of the foundation pit.

[0074] In one specific embodiment, the process of performing long short-term memory unit state update processing on the network input vector based on the forget gate, input gate, and output gate can specifically include the following steps:

[0075] The sigmoid activation function is used to perform forget gate calculation on the historical data of pit deformation and the hidden state of the previous moment in the network input vector to obtain the forget weight of pit deformation information.

[0076] Based on the sigmoid activation function, the earth pressure change data and the previous hidden state in the network input vector are processed by input gate calculation to obtain the earth pressure information input control signal.

[0077] Based on the tanh activation function, candidate values ​​are calculated for the excavation depth parameter and the hidden state at the previous moment in the network input vector to obtain candidate updated values ​​for the deep foundation pit state.

[0078] The current deep foundation pit monitoring status is obtained by multiplying the forgetting weight of the foundation pit deformation information with the cell state at the previous moment, and adding the product of the earth pressure information input control signal and the deep foundation pit state candidate update value.

[0079] The deep foundation pit output gate control signal is calculated by using the sigmoid activation function and multiplied by the tanh value of the current deep foundation pit monitoring state to perform deformation prediction state calculation processing, thereby obtaining the hidden state output.

[0080] Specifically, the sigmoid activation function is used to perform forget gate processing on the historical data of foundation pit deformation and the hidden state of the previous time step in the network input vector. The historical data of foundation pit deformation contains a sequence of cumulative deformation from the start of excavation to the current time step, reflecting the temporal evolution of foundation pit deformation. The hidden state of the previous time step is the output of the short-term memory unit of the previous time step, containing a summary of information from all previous time steps. The sigmoid activation function is a sigmoid function that maps any real number to between 0 and 1. The forget gate calculation process concatenates the historical data of foundation pit deformation with the hidden state of the previous time step to form an extended vector, which is then multiplied by the forget gate weight matrix. The parameters of the weight matrix are learned through training data. The multiplication result, plus the forget gate bias vector, is then input into the sigmoid function for activation processing. The output value of the foundation pit deformation information forgetting weight is between 0 and 1, with values ​​close to 0 indicating complete forgetting of historical deformation information and values ​​close to 1 indicating complete retention of historical deformation information. This weight vector is multiplied element-wise with the cell state of the previous time step to achieve selective retention of historical information.

[0081] The earth pressure variation data and the previous hidden state in the network input vector are processed using the sigmoid activation function. The earth pressure variation data is continuously monitored by earth pressure gauges, reflecting the changing trend of the lateral pressure exerted by the soil on the support structure over time. This data is closely related to the excavation depth and soil properties. The input gate calculation process adopts the same calculation procedure as the forget gate: the earth pressure variation data is concatenated with the previous hidden state, multiplied by the input gate weight matrix, and then activated by the sigmoid function after adding the input gate bias. The earth pressure information input control signal also outputs a value between 0 and 1, controlling the degree to which the current earth pressure information updates the cell state. The larger the value, the more important the current earth pressure information is, and the more it needs to be incorporated into the cell state. The input gate and candidate values ​​jointly determine which new information is added to the cell state. The input gate controls the importance weight of the information, while the candidate values ​​provide the specific information content.

[0082] The tanh activation function is used to calculate candidate values ​​for the excavation depth parameter and the previous hidden state in the network input vector. The excavation depth parameter records the current excavation depth of the foundation pit, directly reflecting the construction progress and changes in the geometric state of the pit. The tanh activation function is a hyperbolic tangent function that maps the input values ​​to between -1 and 1. The candidate value calculation involves concatenating the excavation depth parameter with the previous hidden state, multiplying it by the candidate value weight matrix, adding the candidate value bias vector, and then activating it using the tanh function. The output range of the deep foundation pit state candidate update value is between -1 and 1. Positive values ​​indicate enhancement of certain state components, while negative values ​​indicate weakening of certain state components. This candidate value represents the new information that should be added to the cell state based on the current excavation depth.

[0083] Cell state update is performed by multiplying the forgotten weight of the foundation pit deformation information by the cell state at the previous time step, and then adding the product of the earth pressure information input control signal and the candidate update value of the deep foundation pit state. The cell state update formula is as follows:

[0084] in This indicates the current monitoring status of the deep foundation pit. This indicates the weight of forgetting information about the foundation pit deformation. This indicates the cell state at the previous moment. This indicates that the earth pressure information is input into the control signal. This represents the candidate update value for the deep foundation pit status. This represents element-wise multiplication. The cell state at the previous time step records important information from all previous time steps. State updates are achieved through selective retention using a forgetting gate and the addition of new information. The multiplication operation uses element-wise multiplication; each element of the foundation pit deformation information forgetting weight is multiplied by the corresponding element of the cell state at the previous time step to obtain the retained historical information. The product of the earth pressure information input control signal and the candidate update value of the deep foundation pit state is also multiplied element-wise to obtain the new information to be added. The two parts are added element-wise to form the current deep foundation pit monitoring state. This state integrates the retention of historical deformation information and the incorporation of current earth pressure and excavation depth information, forming a complete description of the current foundation pit state.

[0085] The deep foundation pit output gate control signal calculation uses a sigmoid activation function to process the current input information and the previous hidden state. The output gate determines which parts of the cell state should be output to the hidden state. The output gate calculation involves concatenating the current complete input vector with the previous hidden state, multiplying it by the output gate weight matrix, adding the output gate bias, and then activating it with the sigmoid function to obtain a control signal between 0 and 1. The deformation prediction state calculation process involves activating the current deep foundation pit monitoring state using the tanh function and then multiplying it element-wise with the deep foundation pit output gate control signal. The deformation prediction state calculation formula is:

[0086] in This indicates output in a hidden state. This represents the output gate control signal for deep foundation pits. The tanh function normalizes the numerical range of the cell state to between -1 and 1. The output gate control signal determines which components of the cell state are output to the hidden state. The output of the hidden state contains important state information of the foundation pit at the current moment, preserving key features of historical deformation evolution while incorporating the latest information on current earth pressure and excavation depth, forming a comprehensive representation of the current state of the foundation pit.

[0087] In one specific embodiment, the process of executing step S105 may specifically include the following steps:

[0088] Based on the predicted foundation pit deformation results, a five-dimensional monitoring index system was constructed, including foundation pit deformation rate, support structure stress ratio, soil stability coefficient, groundwater level change rate, and surrounding environmental impact, to obtain deep foundation pit safety evaluation indexes.

[0089] The game theory Nash equilibrium algorithm is used to perform weight optimization calculation on the safety evaluation indicators of the deep foundation pit, so as to obtain the optimal collection weight allocation of each monitoring parameter.

[0090] Based on the soil layer characteristics, cloud model expectation value, entropy value and hyperentropy parameter are constructed, and membership function is generated to perform safety level assessment on the deep foundation pit safety evaluation index, so as to obtain foundation pit safety level cloud droplet;

[0091] The optimal acquisition weight allocation and the foundation pit safety level cloud droplets are calculated using a comprehensive evaluation function based on the foundation pit depth correction coefficient to obtain the comprehensive safety level of the deep foundation pit.

[0092] Based on the comprehensive safety level of the deep foundation pit, a data acquisition frequency threshold is set, and the data acquisition frequency of the representative monitoring sensors is adjusted in stages to obtain data acquisition control commands.

[0093] Specifically, a five-dimensional monitoring index system is constructed based on the predicted deformation results of the foundation pit. The foundation pit deformation rate is calculated by the ratio of the predicted deformation result to the time interval, reflecting the speed of foundation pit deformation. The support structure stress ratio is calculated by the ratio of the measured stress value of the support structure to the design allowable stress value, assessing the safety margin of the support structure. The soil stability coefficient is determined based on the ratio of the soil shear strength to the actual shear stress, characterizing the stability state of the soil. The groundwater level change rate is calculated by the time derivative of the groundwater level monitoring data, reflecting the trend of groundwater level changes. The impact on the surrounding environment is calculated by comprehensively evaluating factors such as the settlement of surrounding buildings, road deformation, and pipeline displacement. These five indicators constitute a complete system of deep foundation pit safety evaluation indicators. Each indicator has its specific physical meaning and calculation method, forming a multi-dimensional description of the safety status of the foundation pit.

[0094] The Nash equilibrium algorithm based on game theory is used to optimize the weights of safety evaluation indicators for deep foundation pits. Game theory is a mathematical theory that studies strategy choices and conflicts of interest among decision-makers. Nash equilibrium refers to the state in a multi-player game where each participant chooses the optimal strategy given the strategies of other participants. The weight optimization process treats the five monitoring indicators as players in the game, and the weight selection of each indicator as its strategy. A payoff function is constructed to describe the monitoring effect under different weight combinations, taking into account factors such as monitoring accuracy, cost-effectiveness, and safety assurance. The Nash equilibrium algorithm iteratively calculates to find the equilibrium point where all indicator weights reach their optimal values, meaning that no single indicator can gain a better payoff by unilaterally changing its weight. After convergence, the algorithm obtains the optimal collection weight allocation for each monitoring parameter, and the weight allocation results reflect the relative importance of different monitoring indicators in the foundation pit safety evaluation.

[0095] The cloud model safety level assessment process constructs three numerical characteristic parameters of the cloud model based on soil layer characteristics. The expected value represents the mathematical expectation of the evaluation level, reflecting the central position of the safety level; the entropy value represents the degree of randomness and fuzziness of the evaluation level; and the hyperentropy represents the uncertainty of entropy. Soil layer characteristics include information such as soil layer thickness, soil type, and physical and mechanical parameters. Based on these characteristics, cloud model parameter values ​​for different safety levels are determined. Safety levels are typically divided into five levels: very safe, safe, moderate, dangerous, and very dangerous. The cloud model membership function is generated through a forward cloud generator, which is an algorithm that converts qualitative concepts into quantitative values. The input is the three characteristic parameters of the cloud model, and the output is randomly generated cloud droplets and their membership values. The membership function describes the degree to which the deep foundation pit safety evaluation index belongs to each safety level. The function value is between 0 and 1, with a larger value indicating a higher probability of belonging to that level. The foundation pit safety level cloud droplets are calculated through the membership function; cloud droplets are quantitative numerical points with membership degrees in the cloud model.

[0096] The comprehensive evaluation function calculates the safety level of the foundation pit by weighting the optimal data acquisition weights and the cloud droplets according to the foundation pit depth correction coefficient. The foundation pit depth correction coefficient is determined based on the actual excavation depth; the greater the depth, the larger the correction coefficient, reflecting the amplifying effect of depth on safety risk. The comprehensive evaluation function uses a weighted summation method, multiplying the cloud droplet values ​​of each safety evaluation indicator by their corresponding optimal weights, then by the foundation pit depth correction coefficient, and finally summing the results to obtain the comprehensive safety level of the deep foundation pit. The comprehensive safety level is a value between 0 and 1; a higher value indicates a safer foundation pit, while a lower value indicates a higher safety risk. This value comprehensively reflects multiple factors such as the deformation state of the foundation pit, the state of the support structure, soil stability, the influence of groundwater, and the influence of the surrounding environment.

[0097] The data acquisition frequency is adjusted in a tiered manner based on the overall safety level of the deep foundation pit. Different acquisition frequency thresholds are set, dividing the safety level into multiple intervals, each corresponding to a different data acquisition frequency. When the overall safety level is high, the foundation pit is in a safe state, and the data acquisition frequency of representative monitoring sensors is set to a lower frequency to reduce unnecessary data acquisition and processing. When the overall safety level is medium, the foundation pit is in an early warning state, and the data acquisition frequency is increased to closely monitor changes in the foundation pit's condition. When the overall safety level is low, the foundation pit is in a dangerous state, and a high-frequency continuous monitoring mode is activated to ensure timely detection and response to safety risks. The tiered adjustment process automatically determines the acquisition frequency of each representative monitoring sensor based on the preset frequency thresholds and generates corresponding data acquisition control commands. These commands are sent to each sensor node via the communication network, and the sensor nodes automatically adjust their data acquisition parameters upon receiving the commands.

[0098] The above describes the deep foundation pit monitoring data acquisition method in the embodiments of this application. The following describes the deep foundation pit monitoring data acquisition system in the embodiments of this application. Please refer to [link / reference]. Figure 2 One embodiment of the deep foundation pit monitoring data acquisition system in this application includes:

[0099] The deployment module is used to spatially deploy monitoring sensors according to the three depth levels of the deep foundation pit: the top of the pit, the pit wall, and the bottom of the pit, to obtain sensor group data.

[0100] The extraction module is used to extract feature values ​​from the sensor group data by calculating the foundation pit deformation gradient, support structure stress, soil stability and groundwater seepage parameters to obtain spatial weight data.

[0101] The filtering module is used to filter the spatial weight data using a deep foundation pit spatial hierarchical Gaussian mixture clustering algorithm to obtain representative monitoring sensors;

[0102] The prediction module is used to input the monitoring data of the representative monitoring sensors into the long short-term memory network of the deep foundation pit for prediction processing, and obtain the foundation pit deformation prediction result.

[0103] The adjustment module is used to dynamically adjust the monitoring data acquisition frequency based on the foundation pit deformation prediction results through game theory weight allocation and cloud model evaluation, and obtain data acquisition control commands.

[0104] above Figure 2 The data acquisition system for monitoring medium and deep foundation pits in this embodiment of the invention will be described in detail from the perspective of modular functional entities. The data acquisition device for monitoring deep foundation pits in this embodiment of the invention will be described in detail from the perspective of hardware processing.

[0105] Reference Figure 3 This invention also provides a deep foundation pit monitoring data acquisition device, which can be a server, and its internal structure can be as follows: Figure 3 As shown, the deep foundation pit monitoring data acquisition device includes a processor, memory, display screen, input device, network interface, and database connected via a system bus. The processor in this computer design provides computing and control capabilities. The memory of the deep foundation pit monitoring data acquisition device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database of the deep foundation pit monitoring data acquisition device is used to store the data corresponding to this embodiment. The network interface of the deep foundation pit monitoring data acquisition device is used for communication with external terminals via network connection. When the computer program is executed by the processor, it implements the above-described method.

[0106] Those skilled in the art will understand that Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present invention and does not constitute a limitation on the deep foundation pit monitoring data acquisition device to which the present invention is applied.

[0107] The present invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when the instructions are executed on a computer, cause the computer to perform the steps of the deep foundation pit monitoring data acquisition method.

[0108] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0109] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a deep foundation pit monitoring data acquisition device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0110] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for acquiring monitoring data in deep foundation pits, characterized in that, The method includes: Based on the three depth levels of the deep foundation pit—the top, wall, and bottom—the monitoring sensors are spatially layered and deployed to obtain sensor group data. The sensor group data is processed by feature value extraction to obtain spatial weight data through calculation of the foundation pit deformation gradient, support structure stress, soil stability, and groundwater seepage parameters. This includes: calculating the foundation pit deformation gradient coefficient by calculating the ratio of displacement difference to distance between adjacent monitoring points; calculating the support structure stress concentration coefficient based on the ratio of measured stress to theoretical design stress; calculating the soil stability index based on the ratio of soil shear strength to actual shear stress; calculating the groundwater seepage influence factor based on the product of groundwater level change rate and permeability coefficient; and linearly combining the foundation pit deformation gradient coefficient, support structure stress concentration coefficient, soil stability index, and groundwater seepage influence factor according to weighting coefficients to obtain spatial weight data. The spatial weight data of the deep foundation pit is filtered using a spatially layered Gaussian mixture clustering algorithm to obtain representative monitoring sensors. This process includes: constructing clustering feature vectors from the foundation pit deformation gradient coefficient, support structure stress concentration coefficient, soil stability index, and groundwater seepage influencing factor in the spatial weight data, resulting in a four-dimensional feature vector combination; initializing the four-dimensional feature vector combination with Gaussian mixture model parameters based on depth-level weights, same-level cluster centers, and covariance matrix to obtain initial clustering parameters; iteratively optimizing the initial clustering parameters using the expectation-maximization algorithm to obtain converged clustering parameters and a posterior probability distribution; clustering the sensors according to the converged clustering parameters to obtain sensor clusters with similar monitoring characteristics; and selecting the sensor closest to the cluster center from each sensor cluster for representative screening to obtain representative monitoring sensors. The monitoring data from the representative monitoring sensors are input into the long short-term memory network of the deep foundation pit for prediction processing to obtain the foundation pit deformation prediction result. Based on the predicted foundation pit deformation, the monitoring data acquisition frequency is dynamically adjusted using game theory weight allocation and cloud model evaluation to obtain data acquisition control instructions. This includes: constructing a five-dimensional monitoring index system based on the predicted foundation pit deformation rate, support structure stress ratio, soil stability coefficient, groundwater level change rate, and surrounding environmental impact, thus obtaining deep foundation pit safety evaluation indicators; performing weight optimization calculations on the deep foundation pit safety evaluation indicators using a game theory Nash equilibrium algorithm to obtain the optimal acquisition weight allocation for each monitoring parameter; constructing cloud model expectation value, entropy value, and hyperentropy parameters based on soil characteristics and generating membership functions to evaluate the safety level of the deep foundation pit safety evaluation indicators, obtaining foundation pit safety level cloud droplets; calculating a comprehensive evaluation function by combining the optimal acquisition weight allocation with the foundation pit safety level cloud droplets according to the foundation pit depth correction coefficient to obtain the comprehensive safety level of the deep foundation pit; setting an acquisition frequency threshold based on the comprehensive safety level of the deep foundation pit and performing graded adjustment processing on the acquisition frequency of the representative monitoring sensors to obtain data acquisition control instructions.

2. The method for acquiring monitoring data of deep foundation pits according to claim 1, characterized in that, The monitoring sensors are spatially layered and deployed according to three depth levels: the top, wall, and bottom of the deep foundation pit, to obtain sensor group data, including: Surface settlement monitoring points and building tilt monitoring equipment were set up at intervals of 15 to 20 meters in the stable layer at the top of the pit to obtain the coordinates of the top monitoring points. Horizontal displacement gauges and inclinometers are laid out along the pit wall at vertical depths of 3 to 5 meters to form a deformation monitoring grid, thus obtaining the spatial distribution of the monitoring equipment in the middle. Earth pressure gauges and pore water pressure gauges were deployed in the stress concentration layer at the bottom of the pit to monitor the stress state of the soil, resulting in a bottom stress monitoring array. A sensor spatial coordinate system is established based on the coordinates of the top monitoring point, the spatial distribution of the middle monitoring equipment, and the bottom stress monitoring array to obtain the unique spatial code of the sensor. Based on the soil layer interface, the sensors within each depth level are grouped to obtain sensor group data containing the depth level, the group number of the same layer, and the sensor serial number within the group.

3. The method for acquiring monitoring data of deep foundation pits according to claim 1, characterized in that, The step of inputting the monitoring data from the representative monitoring sensors into the deep foundation pit long short-term memory network for prediction processing to obtain the foundation pit deformation prediction result includes: The cumulative deformation, earth pressure, and excavation depth data from the representative monitoring sensors are combined and processed according to time series to obtain the network input vector; The network input vector is updated using long short-term memory units based on forget gates, input gates, and output gates to obtain the hidden state output. The hidden state output is weighted and fused by calculating attention weights to obtain attention-weighted features; The attention-weighted features are input into the output layer of the neural network to calculate the deformation amount, thereby obtaining the deformation increment of the foundation pit at the next moment. The deformation prediction result of the foundation pit is obtained by accumulating the deformation increment of the foundation pit at the next moment and the current cumulative deformation.

4. The method for acquiring monitoring data of deep foundation pits according to claim 3, characterized in that, The process of updating the long short-term memory unit state of the network input vector based on the forget gate, input gate, and output gate to obtain the hidden state output includes: The sigmoid activation function is used to perform forget gate calculation on the historical data of pit deformation and the hidden state of the previous moment in the network input vector to obtain the forget weight of pit deformation information. Based on the sigmoid activation function, the earth pressure change data and the previous hidden state in the network input vector are processed by input gate calculation to obtain the earth pressure information input control signal. Based on the tanh activation function, candidate values ​​are calculated for the excavation depth parameter and the hidden state at the previous moment in the network input vector to obtain candidate updated values ​​for the deep foundation pit state. The current deep foundation pit monitoring status is obtained by multiplying the forgetting weight of the foundation pit deformation information with the cell state at the previous moment, and adding the product of the earth pressure information input control signal and the deep foundation pit state candidate update value. The deep foundation pit output gate control signal is calculated by using the sigmoid activation function and multiplied by the tanh value of the current deep foundation pit monitoring state to perform deformation prediction state calculation processing, thereby obtaining the hidden state output.

5. A deep foundation pit monitoring data acquisition system, characterized in that, For implementing the deep foundation pit monitoring data acquisition method as described in any one of claims 1-4, the deep foundation pit monitoring data acquisition system comprises: The deployment module is used to spatially deploy monitoring sensors according to the three depth levels of the deep foundation pit: the top of the pit, the pit wall, and the bottom of the pit, to obtain sensor group data. The extraction module is used to extract feature values ​​from the sensor group data by calculating the foundation pit deformation gradient, support structure stress, soil stability, and groundwater seepage parameters to obtain spatial weight data. This includes: calculating the foundation pit deformation gradient coefficient by calculating the ratio of the displacement difference to the distance between adjacent monitoring points; calculating the support structure stress concentration coefficient based on the ratio of the measured stress value to the theoretical design stress value; calculating the soil stability index based on the ratio of the soil shear strength to the actual shear stress; calculating the groundwater seepage influence factor based on the product of the groundwater level change rate and the permeability coefficient; and linearly combining the foundation pit deformation gradient coefficient, support structure stress concentration coefficient, soil stability index, and groundwater seepage influence factor according to weighting coefficients to obtain spatial weight data. A filtering module is used to filter the spatial weight data using a deep foundation pit spatial hierarchical Gaussian mixture clustering algorithm to obtain representative monitoring sensors. This includes: constructing clustering feature vectors from the foundation pit deformation gradient coefficient, support structure stress concentration coefficient, soil stability index, and groundwater seepage influencing factor in the spatial weight data, resulting in a four-dimensional feature vector combination; initializing the four-dimensional feature vector combination with Gaussian mixture model parameters based on depth-level weights, same-level cluster centers, and covariance matrix to obtain initial clustering parameters; iteratively optimizing the initial clustering parameters using the expectation-maximization algorithm to obtain converged clustering parameters and a posterior probability distribution; clustering the sensors according to the converged clustering parameters to obtain sensor clusters with similar monitoring characteristics; and selecting the sensor closest to the cluster center from each sensor cluster for representative filtering to obtain representative monitoring sensors. The prediction module is used to input the monitoring data of the representative monitoring sensors into the long short-term memory network of the deep foundation pit for prediction processing, and obtain the foundation pit deformation prediction result. The adjustment module is used to dynamically adjust the monitoring data acquisition frequency based on the foundation pit deformation prediction results through game theory weight allocation and cloud model evaluation, and to obtain data acquisition control instructions. This includes: constructing a five-dimensional monitoring index system based on the foundation pit deformation prediction results, including foundation pit deformation rate, support structure stress ratio, soil stability coefficient, groundwater level change rate, and surrounding environmental impact, to obtain deep foundation pit safety evaluation indicators; performing weight optimization calculations on the deep foundation pit safety evaluation indicators using a game theory Nash equilibrium algorithm to obtain the optimal acquisition weight allocation for each monitoring parameter; constructing cloud model expectation value, entropy value, and hyperentropy parameters based on soil characteristics and generating a membership function to evaluate the safety level of the deep foundation pit safety evaluation indicators, obtaining foundation pit safety level cloud droplets; calculating a comprehensive evaluation function by combining the optimal acquisition weight allocation with the foundation pit safety level cloud droplets according to the foundation pit depth correction coefficient, to obtain the comprehensive safety level of the deep foundation pit; setting an acquisition frequency threshold based on the comprehensive safety level of the deep foundation pit and performing graded adjustment processing on the acquisition frequency of the representative monitoring sensors to obtain data acquisition control instructions.

6. A deep foundation pit monitoring data acquisition device, characterized in that, The method includes a memory and a processor, wherein the memory stores a computer program that can run on the processor, and the processor executes the computer program to implement the deep foundation pit monitoring data acquisition method according to any one of claims 1 to 4.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is run by the processor, it causes the processor to execute the deep foundation pit monitoring data acquisition method as described in any one of claims 1 to 4.

Images