A foundation settlement detection method, system, device and medium for engineering detection

Abstract

This application provides a method, system, equipment, and medium for foundation settlement detection in engineering testing, relating to the field of engineering safety testing technology. The method includes acquiring multi-source monitoring data collected in the area to be tested and adjacent sensitive structures; fusing the multi-source monitoring data to obtain a target detection dataset; inputting the target detection dataset into a rapid screening model to obtain settlement risk assessment values, forming a risk assessment point set; if the settlement risk assessment value at any point in the risk assessment point set is greater than a first warning threshold, generating a preliminary warning signal, and determining the area of ​​interest based on the spatial distribution of the risk assessment point set; responding to the preliminary warning signal, acquiring a fused dataset for in-depth analysis; performing in-depth analysis of the area of ​​interest based on multiphysics coupled numerical simulation using the fused dataset to obtain a settlement risk level; and generating graded warning information based on the settlement risk level. This application improves the accuracy of detection.

Description

Technical Field

[0001] This application relates to the field of engineering safety testing technology, and in particular to a method, system, equipment and medium for foundation settlement testing in engineering testing. Background Technology

[0002] In the field of engineering construction, foundation settlement is a key hidden danger affecting the safety and long-term stability of engineering structures, especially in densely populated urban areas and transportation hubs. Uneven foundation settlement can easily lead to building cracks, pipeline damage, and even threaten the normal use of nearby sensitive structures, causing major safety accidents and economic losses. Current foundation settlement detection technologies are insufficient to fully characterize the evolution of settlement; at the same time, early warning responses are delayed, relying heavily on manual analysis and judgment, which cannot quickly screen risk points and accurately locate areas of concern. Furthermore, the lack of in-depth coupled analysis of settlement mechanisms results in a crude early warning classification, which is insufficient to meet the needs of refined and intelligent settlement risk prevention and control in complex engineering scenarios.

[0003] Therefore, there is an urgent need for a foundation settlement detection method for engineering testing. Summary of the Invention

[0004] To address the aforementioned technical problems, this application provides a method, system, equipment, and medium for detecting foundation settlement in engineering testing.

[0005] Firstly, this application provides a method for detecting foundation settlement in engineering testing, comprising:

[0006] Acquire multi-source monitoring data collected in the project area to be inspected and adjacent sensitive structures; the multi-source monitoring data includes: time-series monitoring data and deformation monitoring images;

[0007] The multi-source monitoring data are fused to obtain a target detection dataset;

[0008] The target detection dataset is input into a preset rapid screening model to calculate the settlement risk assessment value, which constitutes a risk assessment point set.

[0009] If the settlement risk assessment value of any point in the risk assessment point set is greater than the first warning threshold, a preliminary warning signal is generated, and the corresponding area of ​​concern is determined based on the spatial distribution of the risk assessment point set.

[0010] In response to the initial warning signal, a fused dataset for in-depth analysis is acquired; the fused dataset for in-depth analysis includes the target detection dataset that triggered the initial warning signal and the newly acquired monitoring data after the initial warning signal was generated;

[0011] Based on the fused dataset, a deep analysis based on multiphysics coupling numerical simulation is performed on the region of interest to obtain the subsidence risk level of the region of interest.

[0012] Based on the aforementioned settlement risk level, corresponding graded early warning information is generated.

[0013] Secondly, this application provides a foundation settlement detection system for engineering testing, used to perform the aforementioned foundation settlement detection method for engineering testing, the system comprising:

[0014] The first data module is used to acquire multi-source monitoring data collected in the project area to be inspected and the adjacent sensitive structure area; the multi-source monitoring data includes: time-series monitoring data and deformation monitoring images;

[0015] The data processing module is used to fuse the multi-source monitoring data to obtain a target detection dataset;

[0016] The model calculation module is used to input the target detection dataset into a preset rapid screening model to calculate the settlement risk assessment value and form a risk assessment point set;

[0017] The first early warning module is used to generate a preliminary early warning signal if the settlement risk assessment value of any point in the risk assessment point set is greater than the first early warning threshold, and to determine the corresponding area of ​​concern based on the spatial distribution of the risk assessment point set.

[0018] The second data module is used to acquire a fusion dataset for in-depth analysis in response to the initial warning signal; the fusion dataset for in-depth analysis includes the target detection dataset that triggered the initial warning signal and the monitoring data newly acquired after the initial warning signal was generated;

[0019] The risk determination module is used to perform in-depth analysis of the region of interest based on multiphysics coupling numerical simulation based on the fused dataset to obtain the settlement risk level of the region of interest.

[0020] The second early warning module is used to generate corresponding graded early warning information based on the settlement risk level.

[0021] Thirdly, this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement the steps of the above-described foundation settlement detection method for engineering testing.

[0022] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described foundation settlement detection method for engineering testing.

[0023] The beneficial effects of the foundation settlement detection method, system, equipment, and medium provided in this application are as follows: This application constructs a target dataset by integrating time-series monitoring data and deformation monitoring images, breaking through the information limitations of single monitoring methods and improving data integrity and reliability; it achieves efficient identification of settlement risks and accurate positioning of areas of concern through a rapid screening model, significantly shortening risk response time and avoiding the problem of delayed early warning; it conducts in-depth analysis based on multi-physics coupled numerical simulation to explore the settlement evolution mechanism, making the risk level assessment more consistent with actual settlement patterns and improving the scientific nature of the assessment; the graded early warning mechanism outputs early warning information according to the risk level, meeting the differentiated prevention and control needs of different risk scenarios; at the same time, the analysis mode integrating newly added monitoring data realizes dynamic tracking of settlement risks, effectively ensuring the safety of engineering structures and adjacent sensitive structures, and reducing the incidence of safety accidents and economic losses. Attached Figure Description

[0024] Figure 1 A schematic flowchart of the foundation settlement detection method for engineering testing provided in this application embodiment;

[0025] Figure 2 A structural block diagram of the foundation settlement detection system for engineering testing provided in this application embodiment;

[0026] Figure 3 A schematic block diagram of an electronic device provided in an embodiment of this application.

[0027] The attached diagram is labeled as follows:

[0028] 20. Foundation Settlement Detection System for Engineering Testing; 21. First Data Module; 22. Data Processing Module; 23. Model Calculation Module; 24. First Early Warning Module; 25. Second Data Module; 26. Risk Determination Module; 27. Second Early Warning Module;

[0029] 300. Electronic device; 301. Processor; 302. Input device; 303. Output device; 304. Memory; 305. Communication bus. Detailed Implementation

[0030] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0031] To make the purpose, technical solution, and advantages of this application clearer, the following will be described in conjunction with the appendix. Figure 1-3 The following is an explanation using specific examples.

[0032] Please refer to Figure 1 , Figure 1 This is a flowchart illustrating the foundation settlement detection method for engineering testing provided in this application embodiment. The method includes:

[0033] S101: Acquire multi-source monitoring data collected in the project area to be inspected and adjacent sensitive structures; the multi-source monitoring data includes: time-series monitoring data and deformation monitoring images.

[0034] In this embodiment, multi-source monitoring data refers to various types of monitoring data collected in the area to be monitored and adjacent sensitive structures, including monitoring data that can characterize time-series changes and image data that can capture spatial deformation characteristics. Time-series monitoring data can be acquired through a deployed sensor network, including but not limited to Global Navigation Satellite System (GNSS) receivers, hydrostatic levels, inclinometers, strain gauges, etc., automatically collected and transmitted at a preset frequency (e.g., hourly or daily) to measure parameters such as settlement, displacement, and tilt angle. Deformation monitoring images can be acquired through periodic aerial photography by UAVs, fixed monitoring cameras, or synthetic aperture radar to extract surface deformation information. All data are timestamped and associated with their spatial coordinate information.

[0035] S102: The multi-source monitoring data is fused to obtain the target detection dataset.

[0036] In this embodiment, fusion processing refers to the integration, calibration, denoising, and interpolation of multi-source monitoring data from different sources, formats, and time scales to eliminate data heterogeneity, improve data quality and consistency, and thus form a unified dataset that can be used for subsequent analysis. The target detection dataset refers to the multi-source monitoring data set after fusion processing, including comprehensive monitoring information of the area to be detected and sensitive structure areas, which can be directly used for risk assessment using rapid screening models.

[0037] This embodiment employs a spatiotemporal alignment and multi-source data fusion method, specifically including: performing time series interpolation on time-series monitoring data to unify it to the same time interval; performing geometric correction, radiometric correction, and image registration on deformation monitoring images, and extracting the deformation field using digital image correlation (DIC) or interferometric synthetic aperture radar (InSAR) technology; and based on a geographic information system (GIS) platform, spatially overlaying and weighted fusion of time-series data and deformation field data to generate a target detection dataset with a unified coordinate system (such as CGCS2000) and time reference. The data format can be unified as a structured table or raster data with timestamps, spatial coordinates, and multiple attribute fields.

[0038] S103: Input the target detection dataset into the preset rapid screening model to calculate the settlement risk assessment value and form a risk assessment point set.

[0039] In this embodiment, the rapid screening model refers to a pre-established computational model used for rapid analysis and evaluation of target detection datasets. This model aims to efficiently identify potential settlement risk points and is built based on methods such as machine learning, statistical analysis, or empirical formulas, capable of quickly outputting settlement risk assessment values. The settlement risk assessment value is a numerical value calculated by the rapid screening model that quantifies the degree of foundation settlement risk. This value characterizes the likelihood and potential impact of settlement disasters occurring at a specific location or area; a higher value indicates a higher risk. The risk assessment point set refers to the spatial set of points within the area to be detected, consisting of all settlement risk assessment values ​​obtained after evaluation by the rapid screening model. Each point is associated with its corresponding spatial location and settlement risk assessment value.

[0040] As one implementation method, a machine learning model based on gradient boosting decision tree (GBDT) can be used. The model is trained with historical monitoring data and its corresponding risk labels. The input of the model is multi-dimensional features (settlement rate, cumulative settlement, deformation gradient, etc.) in the target detection dataset, and the output is a settlement risk assessment value (probability value between 0 and 1) for each monitoring point.

[0041] This embodiment is based on a fast screening model using gradient boosting decision trees. The construction and training process specifically includes:

[0042] Multi-source monitoring datasets from multiple monitoring periods in historical engineering cases were collected as training samples. Each sample corresponds to a target detection dataset (i.e., input features) of a monitoring point at a certain time point, and is evaluated by experts based on subsequent actual settlement events; multi-level risk level labels are assigned according to the severity of settlement.

[0043] The input features in the training samples are standardized to eliminate the influence of units. The preprocessed training sample set is then divided into a training set and a validation set according to a predetermined ratio (e.g., 7:3). The training set is used to train the fast screening model based on gradient boosting decision trees. The loss function used during training is cross-entropy loss, and the optimization algorithm is gradient descent. The performance of the model is monitored using the validation set to prevent overfitting.

[0044] The core performance metrics for the rapid screening model include accuracy, recall, precision, and F1 score. Model hyperparameters (such as maximum tree depth, learning rate, and subsampling ratio) are fine-tuned using grid search or random search methods. Finally, the optimal combination of model parameters that performs best on the validation set is selected and solidified as the pre-defined rapid screening model.

[0045] S104: If the settlement risk assessment value of any point in the risk assessment point set is greater than the first warning threshold, a preliminary warning signal is generated, and the corresponding area of ​​concern is determined based on the spatial distribution of the risk assessment point set.

[0046] In this embodiment, the first warning threshold refers to a preset critical value used to determine whether the settlement risk has reached the early warning level. When the settlement risk assessment value is greater than the first warning threshold, it indicates that there is a potential settlement risk that requires further attention. The preliminary warning signal is an early risk warning message generated when the settlement risk assessment value of any point in the risk assessment point set is greater than the first warning threshold. This preliminary warning signal triggers the subsequent in-depth analysis process. The area of ​​concern refers to the geographical range that needs to be focused on monitoring and in-depth analysis, determined by spatial analysis methods based on the spatial distribution characteristics of the risk assessment point set after the preliminary warning signal is generated. As one implementation method, when the settlement risk assessment value of a point is greater than the first warning threshold, the system generates a preliminary warning signal and immediately performs spatial cluster analysis on all risk assessment points to determine the area where the risk is concentrated. The first warning threshold can be dynamically set based on the statistical characteristics of historical monitoring data and engineering specifications, rather than a fixed value.

[0047] S105: In response to the initial warning signal, acquire a fused dataset for deep analysis; the fused dataset for deep analysis includes the target detection dataset that triggered the initial warning signal and the newly acquired monitoring data after the initial warning signal was generated.

[0048] In this embodiment, the fusion dataset used for in-depth analysis refers to a dataset specifically prepared in response to the initial warning signal; it not only includes the initial target detection dataset that triggered the initial warning signal, but also further integrates new, more timely, and more detailed monitoring data acquired after the early warning signal is generated, in order to support more accurate in-depth analysis.

[0049] As one approach, after generating an initial warning signal, the data acquisition frequency of sensors in and around the area of ​​interest is increased (e.g., from once a day to once an hour), and new time-series monitoring data and image data are received in real time through a data interface. These data are then integrated with the initial dataset that triggered the warning based on a spatiotemporal reference to form a fusion dataset for in-depth analysis.

[0050] S106: Based on the fused dataset, a deep analysis of the region of interest is performed using multiphysics coupled numerical simulation to obtain the subsidence risk level of the region of interest.

[0051] In this embodiment, multiphysics coupled numerical simulation refers to a numerical analysis method used to simulate the behavior of a foundation under the interaction of multiple physical fields. This method establishes a mathematical model that considers the mutual influence of factors such as soil deformation, groundwater flow, and stress changes to predict the settlement trend and mechanism of the foundation. The settlement risk level refers to the detailed classification of foundation settlement risk in the area of ​​interest after in-depth analysis. This level is divided into multiple layers to guide subsequent risk management and decision-making. As one implementation method, a multiphysics model considering soil-water-mechanical coupling can be constructed using the finite element method (FEM) or the finite difference method (FDM). Parameter inversion and model calibration are performed using a fused dataset to predict future settlement trends. Combined with a pre-set comprehensive risk assessment model (fuzzy comprehensive evaluation or risk matrix), low, medium, and high settlement risk levels are output.

[0052] A multiphysics coupled numerical model was constructed based on the engineering geological conditions and settlement characteristic parameters of the region of interest. The specific implementation is as follows: The multiphysics coupled numerical model is based on Biot's consolidation theory, coupling the soil skeleton deformation field and the pore water seepage field. The governing equations include equilibrium equations (characterizing stress and external force balance) and continuity equations (characterizing water flow mass conservation), which are coupled through the effective stress principle and Darcy's law. Based on the geological survey profiles and topographic data of the region of interest, a two-dimensional or three-dimensional geometric model was established in finite element analysis software. Different material partitions were set according to the soil layer distribution. Tetrahedral elements were used to mesh the geometric model, ensuring mesh refinement in areas of expected severe deformation (near the load application point, soil layer interfaces). Model boundary conditions were set according to actual conditions: lateral boundaries typically applied normal displacement constraints or roll constraints; bottom boundaries applied fixed constraints; the surface was a free boundary. Seepage boundaries were set as constant head boundaries or impermeable boundaries based on the groundwater level. Initial conditions included the initial stress field (considering self-weight stress) and the initial pore water pressure field. The stress-strain relationship of the soil is described using an elastoplastic constitutive model, such as the Modified Cam-Clay model. The basic parameters required for the model (cohesion, internal friction angle) are provided by the geological survey report.

[0053] S107: Generate corresponding graded early warning information based on the settlement risk level.

[0054] In this embodiment, tiered early warning information refers to early warning notices with different levels and response requirements generated and issued based on the classification of settlement risk levels. Tiered early warning information aims to provide relevant parties with clear risk alerts and action suggestions to achieve differentiated risk management.

[0055] One approach is to use basic text notifications. For example, when the settlement risk level is high, a text message could be automatically generated: "High-risk warning, please take immediate action!"; when it's medium-risk, a message could be generated: "Medium-risk warning, please pay close attention!". Another approach is to pre-set a template library containing warning messages for different risk levels. Once the settlement risk level is determined, the corresponding warning template can be selected from the library, the relevant information filled in, and then sent via email.

[0056] This application overcomes the limitations of traditional methods due to the reliance on single-source data by integrating and fusing multi-source monitoring data. It utilizes a rapid screening model to efficiently identify potential risk points and locate areas of concern based on spatial distribution, effectively solving the problems of delayed early warning response and low screening efficiency. Furthermore, through in-depth analysis using multiphysics coupled numerical simulation, the settlement mechanism is revealed, resulting in more accurate settlement risk level assessments and the generation of detailed, tiered early warning information. This meets the needs for detailed and intelligent settlement risk prevention and control in complex engineering environments.

[0057] As can be seen from the above, this application constructs a target dataset by integrating time-series monitoring data and deformation monitoring images, overcoming the information limitations of a single monitoring method and improving data integrity and reliability; it achieves efficient identification of settlement risks and precise location of areas of concern through a rapid screening model, significantly shortening risk response time and avoiding the problem of delayed early warning; it conducts in-depth analysis based on multi-physics coupled numerical simulation to explore the settlement evolution mechanism, making the risk level assessment more consistent with actual settlement patterns and improving the scientific nature of the assessment; the graded early warning mechanism outputs early warning information according to the risk level, meeting the differentiated prevention and control needs of different risk scenarios; at the same time, the analysis mode integrating newly added monitoring data realizes dynamic tracking of settlement risks, effectively ensuring the safety of engineering structures and adjacent sensitive structures, and reducing the incidence of safety accidents and economic losses.

[0058] In one embodiment of this application, the corresponding area of ​​interest is determined based on the spatial distribution of the settlement risk assessment point set, including:

[0059] The risk assessment point set is used as data points and input into a spatial clustering algorithm to identify at least one risk cluster; wherein, the neighborhood search radius parameter of the spatial clustering algorithm is adaptively adjusted according to the local spatial gradient of the settlement risk assessment value of the data points;

[0060] For each risk cluster, calculate the area of ​​the risk cluster and the maximum value of the settlement risk assessment value of all data points within the risk cluster, and calculate the corresponding risk cluster's radius of concern using a preset weighting formula;

[0061] A circular buffer is generated with the geometric center of each risk cluster as the center and the corresponding radius of concern as the radius;

[0062] All circular buffer zones are spatially overlaid with the project area to be inspected and the adjacent sensitive structure areas, and the overlaid area is extracted as the area of ​​interest.

[0063] In this embodiment, the risk assessment point set is input as data points into a spatial clustering algorithm to identify at least one risk cluster. The neighborhood search radius parameter of the spatial clustering algorithm is adaptively adjusted based on the local spatial gradient of the settlement risk assessment values ​​of the data points. This step aims to effectively aggregate discrete monitoring points with settlement risk into spatially continuous risk regions, i.e., risk clusters. Clustering allows for a more macroscopic identification of areas with concentrated settlement risk, rather than focusing solely on individual high-risk points. The spatial clustering algorithm can employ density-based clustering algorithms, such as DBSCAN (Density-Based Spatial Clustering of Applications with Noise), which can discover clusters of arbitrary shapes and effectively handle noisy points. To improve the accuracy and adaptability of the clustering results, this application further proposes an adaptive adjustment of the neighborhood search radius parameter of the clustering algorithm. This adjustment mechanism adjusts the clustering parameters according to the drastic spatial changes in the settlement risk assessment values ​​(i.e., the local spatial gradient), avoiding overfitting or underfitting problems caused by fixed parameters. For example, in areas with a large settlement risk gradient, the neighborhood search radius can be reduced to identify more refined risk boundaries; while in areas with a small gradient, the radius can be increased to capture a wider range of risk trends. This adaptive adjustment is achieved by calculating the gradient mean of the settlement risk assessment values ​​of each data point and other data points within a preset neighborhood, then inputting this gradient mean into a preset mapping function to obtain the neighborhood radius adjustment coefficient corresponding to the data point, and finally multiplying the preset base neighborhood search radius by the neighborhood radius adjustment coefficient to obtain the neighborhood search radius.

[0064] Neighborhood search radius We can use the local spatial gradient at point i. Dynamic adjustments are made; the specific formula is as follows:

[0065]

[0066] in, The basic search radius is preset based on the average monitoring point density of the region. Let i be the gradient mean of the evaluation values ​​of point i and its K nearest neighbors; The global gradient mean; The global gradient standard deviation; This is an adjustment factor (which can be set to 0.5). When When it is large, The value is reduced accordingly to achieve fine clustering; conversely, it is increased to identify broad trends.

[0067] For each risk cluster, the area of ​​the risk cluster and the maximum settlement risk assessment value of all data points within the risk cluster are calculated. The radius of concern for the corresponding risk cluster is then calculated using a pre-defined weighting formula. The purpose of this step is to quantify the size and severity of each risk cluster, thereby generating a reasonable and representative buffer zone. The area of ​​a risk cluster can characterize the extent of the risk impact, for example, by calculating the area of ​​the convex hull of all data points constituting the risk cluster. The maximum settlement risk assessment value of all data points within the risk cluster directly characterizes the most severe settlement risk level within that cluster. The pre-defined weighting formula comprehensively considers the area and the maximum risk value to determine the radius of concern. For example, this weighting formula is a linear combination:

[0068]

[0069] in, Let A be the radius of interest, and A be the area of ​​the risk cluster. This represents the maximum risk assessment value within the cluster. and Preset weighting coefficients =0.7, =0.3.

[0070] A circular buffer is generated with the geometric center of each risk cluster as its center and the corresponding radius of interest as its radius. This step aims to create a defined spatial influence range for each identified risk cluster. The geometric center of a risk cluster can be determined in several ways; for example, it can be the average coordinates of all data points within the cluster as its centroid, or the point within the cluster closest to all other points can be found as its center. Alternatively, if the cluster has a relatively regular shape, the center of its minimum bounding rectangle or minimum bounding circle can be calculated. The circular buffer is a simple and effective spatial representation method that expands a uniform area of ​​interest outward from the risk cluster as its core. The generation of circular buffers can be achieved using standard Geographic Information System (GIS) buffer analysis functions, which automatically generate circular polygons based on a specified center point and radius.

[0071] All circular buffer zones are spatially overlaid with the project area to be inspected and adjacent sensitive structure areas, and the overlaid area is extracted as the area of ​​interest. The purpose of this step is to correlate the initially generated risk buffer zones with the actual project and sensitive areas, thereby defining the areas requiring focused attention. Spatial overlay is a common geographic information processing operation aimed at identifying the spatial overlap between two or more geographic features. For example, the intersection operation in GIS can be used to overlay the circular buffer zones with the boundary polygons of the project area to be inspected and the boundary polygons of adjacent sensitive structure areas. This overlay effectively excludes portions of the circular buffer zones that do not belong to the project area or sensitive structure area, ensuring that the area of ​​interest is the area actually requiring in-depth analysis.

[0072] Through the above technical solution, this application effectively solves the problems of inaccurate identification of areas of interest, inability to adapt to changes in spatial gradients, and susceptibility to subjective human judgment in traditional methods. Specifically, the risk assessment point set is input into a spatial clustering algorithm, and the neighborhood search radius parameter is adaptively adjusted according to the local spatial gradient of the settlement risk assessment value. This allows the clustering process to dynamically adapt to the risk distribution characteristics of different areas, thereby identifying risk clusters that are more accurate and consistent with the actual settlement trend. This adaptive mechanism avoids the limitations of fixed parameters and improves the precision of risk area identification. Furthermore, by calculating the area and maximum risk value of the risk cluster and combining it with a preset weighting formula to determine the radius of interest, it ensures that the generated circular buffer can reasonably represent the scale and severity of the risk cluster, avoiding the problem of buffers being too large or too small. Finally, these circular buffers are spatially superimposed with the area to be detected and the adjacent sensitive structure areas to accurately extract the areas of interest that actually require in-depth analysis, eliminating irrelevant interference. This allows subsequent in-depth analysis resources to be concentrated on the most critical areas, improving the timeliness of early warning response and the accuracy of area positioning.

[0073] In one embodiment of this application, all circular buffer zones are spatially overlaid with the area to be inspected and the adjacent sensitive structure area, and the overlaid area is extracted as the area of ​​interest, including:

[0074] The circular buffers generated for each risk cluster are spatially merged to generate a total buffer polygon.

[0075] The total buffer polygon is spatially intersected with the boundary polygons of the sensitive structure area and the boundary polygon of the project area to be detected, respectively, to obtain two intersecting sub-regions;

[0076] The two intersecting sub-regions are spatially merged to obtain the final region of interest.

[0077] In this embodiment, the circular buffers generated by each risk cluster are spatially merged to generate a total buffer polygon. The aim is to integrate multiple circular buffers generated by different risk clusters into one or a few continuous, non-overlapping geometric regions, simplifying subsequent spatial analysis and eliminating fragmentation problems caused by direct overlay. For example, the merge or union operation in Geographic Information Systems (GIS) can be used. For a set of circular buffers, by traversing all buffers, they are geometrically merged. If there are overlapping parts, they are merged into a single polygon; if there are gaps, they remain independent polygons, but overall, they form a total polygon set.

[0078] Spatial intersection operations are performed between the total buffer polygon and the boundary polygons of the sensitive structure area and the project area to be inspected, respectively, to obtain two intersecting sub-regions. The purpose is to accurately identify the portions of the total buffer polygon that overlap with the actual sensitive structure area and the project area to be inspected. Through spatial intersection operations, parts of the total buffer that are irrelevant to these key areas can be effectively clipped, thus making the area of ​​interest more focused and accurate. Specifically, the Intersect function provided by GIS software can be used. The total buffer polygon is used as an input feature, and intersection operations are performed with the boundary polygons of the sensitive structure area and the project area to be inspected, respectively. The result will be two new polygons, representing the overlapping portions of the total buffer and the sensitive structure area, and the overlapping portions of the total buffer and the project area to be inspected, respectively.

[0079] The spatial merging of two intersecting sub-regions yields the final region of interest. This step merges the two cropped sub-regions obtained in the previous step (i.e., the portions overlapping with the sensitive structure area and the portion overlapping with the project area to be inspected) again to form a complete and continuous final region of interest. This final region encompasses all risk-related geographic areas, providing a unified analytical boundary for subsequent in-depth analysis. For example, the Merge or Union operation in GIS can be used to merge these two intersecting sub-regions as input; if they are adjacent or overlap, a single, continuous polygon is formed.

[0080] Through the above technical solutions, this application eliminates potential overlaps and gaps between multiple buffer zones by merging operations, forming a unified and simplified polygonal region, reducing data processing complexity and improving subsequent computational efficiency. Spatial intersection operations are used to trim the portions of the total buffer zone that actually overlap with sensitive structure areas and engineering areas, ensuring that the region of interest only covers relevant key areas, avoiding interference from irrelevant areas, and enhancing the focus of the analysis. The merging operation integrates the trimmed sub-regions into a continuous whole region, facilitating unified deep settlement risk analysis and improving the accuracy and operability of subsequent steps. This application, through spatial processing, ensures that the region of interest derived from these buffer zones is accurate, complete, and efficient, thus providing high-quality input for subsequent deep analysis based on multiphysics coupled numerical simulation, improving the accuracy and reliability of the entire settlement detection method.

[0081] In one embodiment of this application, the neighborhood search radius parameter of the clustering algorithm is adjusted according to the local spatial gradient of the settlement risk assessment value, including:

[0082] Calculate the gradient mean of the settlement risk assessment value of each data point and other data points in the surrounding preset neighborhood;

[0083] The gradient mean is input into a preset mapping function to obtain the neighborhood radius adjustment coefficient corresponding to the data point;

[0084] The neighborhood search radius of a data point is obtained by multiplying the preset basic neighborhood search radius by the neighborhood radius adjustment coefficient.

[0085] In this embodiment, the gradient mean of the settlement risk assessment values ​​of each data point and other data points within a preset neighborhood is calculated. This aims to quantify the degree of local spatial variation in the settlement risk assessment values ​​around each data point. This gradient mean characterizes the spatial severity of the risk value, providing a direct basis for subsequent adaptive adjustment of the neighborhood radius. Specifically, a circular or square neighborhood with a fixed radius can be defined for each data point. Then, the average of the absolute values ​​of the differences between the settlement risk assessment values ​​of all data points within this neighborhood (including the center point itself) and the settlement risk assessment value of the center point can be calculated. Alternatively, a K-nearest neighbor (K-NN) method can be used to find the K nearest neighbors for each data point, then calculate the differences between the settlement risk assessment values ​​of these neighbors and the settlement risk assessment value of the center point, and take their average or weighted average as the gradient mean.

[0086] The mean gradient is input into a predefined mapping function to obtain the neighborhood radius adjustment coefficient corresponding to the data point. The purpose is to transform the quantized local spatial gradient information into a dimensionless adjustment factor. This mapping function defines how the neighborhood radius adjustment responds to gradient changes, ensuring more precise adjustments can be made in regions where risk values ​​change drastically.

[0087] For example, the mapping function can be a simple linear function, where the adjustment coefficient = a × mean gradient + b, and a and b are preset constants used to control the magnitude and offset of the adjustment. Alternatively, the mapping function can be nonlinear, such as an exponential function, a logarithmic function, or a sigmoid function, to better simulate the nonlinear response to gradient changes in real-world applications. This mapping function can be calibrated based on historical data or expert experience.

[0088] The neighborhood search radius of each data point is determined by multiplying the preset basic neighborhood search radius by the neighborhood radius adjustment factor. This step completes the determination of the adaptive neighborhood search radius for each data point. It combines a basic radius representing a general scale with an adjustment factor characterizing local features, allowing the clustering algorithm to dynamically adjust its search range based on the local characteristics of the settlement risk data. For example, the basic neighborhood search radius can be set based on the average monitoring point density of the area to be monitored or the expected size of the risk cluster. The adjustment factor is designed to be greater than 0. When the gradient mean is high, the adjustment factor will be less than 1, resulting in a smaller final neighborhood search radius; when the gradient mean is low, the adjustment factor will be greater than 1, resulting in a larger final neighborhood search radius.

[0089] Through the above technical solution, this application can adaptively adjust the neighborhood search radius parameter of the clustering algorithm based on the local spatial gradient of the settlement risk assessment value. In regions with a large spatial gradient of the settlement risk assessment value, the adjustment mechanism reduces the neighborhood search radius, enabling the clustering algorithm to more precisely identify risk clusters and avoid erroneously merging regions with different risk levels, thereby more accurately capturing the details of risk distribution. Conversely, in regions with a small gradient, the neighborhood search radius increases, allowing the algorithm to more effectively identify larger areas with relatively uniform risk. This further improves the reliability and effectiveness of the overall settlement risk detection method.

[0090] In one embodiment of this application, the method for determining the first warning threshold includes:

[0091] Obtain historical settlement monitoring data and engineering safety specifications for the area to be inspected;

[0092] Based on historical settlement monitoring data, the probability distribution characteristics of settlement rate in the project area to be monitored were statistically obtained;

[0093] Based on the settlement control standards and probability distribution characteristics specified in the engineering safety specifications, the first warning threshold is determined by a preset confidence level.

[0094] In this embodiment, acquiring historical settlement monitoring data and engineering safety standards for the area to be inspected refers to collecting historical settlement observation records and officially published engineering safety standards related to the area. Historical settlement monitoring data can be obtained from various sources, such as reviewing geological survey reports of the project, settlement observation records during construction, sensor data from long-term operating monitoring stations (e.g., GNSS, leveling, inclinometers), or historical deformation image data obtained using Synthetic Aperture Radar Interferometry (InSAR) technology. This data includes records of settlement amount and settlement rate changes over time, characterizing the actual historical behavior of foundation settlement in the area. Engineering safety standards refer to technical standard documents promulgated by national, industry, or local governments that clearly limit and require specific measures for the foundation settlement behavior of various engineering structures (e.g., buildings, bridges, tunnels, pipelines), such as the "Code for Design of Building Foundations" and the "Standard for Settlement Control of Railway Engineering," which specify control indicators such as the maximum allowable settlement amount and maximum settlement rate.

[0095] Secondly, based on historical settlement monitoring data, the probability distribution characteristics of the settlement rate in the area to be monitored are statistically obtained. This refers to conducting statistical analysis on the acquired historical settlement monitoring data to reveal the inherent statistical regularity of the settlement rate in the area. For example, histogram analysis can be performed on the historical settlement rate data to visualize its frequency distribution, and a probability distribution model that conforms to the data characteristics, such as a normal distribution, can be further fitted.

[0096] This embodiment determines the first warning threshold based on the settlement control standards and probability distribution characteristics specified in the engineering safety specifications and through a preset confidence level. The process includes: obtaining the settlement rate control standard value specified in the engineering safety specifications; calculating the quantiles of the settlement rate probability distribution in historical settlement monitoring data, with the quantiles corresponding to the preset confidence level; and weighted summing the settlement rate control standard value and the quantiles to obtain the first warning threshold.

[0097] The method of determining the first warning threshold based on the settlement control standards and probability distribution characteristics specified in engineering safety specifications, using a preset confidence level, refers to comprehensively considering the rigid requirements of engineering safety specifications and the statistical characteristics of historical settlement data, and setting a critical value for triggering the warning at a defined confidence level. Specifically, firstly, the settlement rate control standard value explicitly specified in the engineering safety specifications is obtained. Then, based on a high quantile of the historical settlement rate probability distribution (e.g., the 95th or 99th quantile, which corresponds to the preset confidence level), a comprehensive calculation is performed using a weighted average, taking the smaller of the two values, or an empirical formula.

[0098] Through the above technical solution, this application overcomes the limitations of traditional early warning threshold settings that rely on manual experience or static standards. By acquiring historical settlement monitoring data and engineering safety specifications for the area to be monitored, a solid empirical foundation and legal basis are provided for threshold setting, avoiding subjective assumptions. Based on the probability distribution characteristics of settlement rates obtained from the statistical analysis of historical settlement monitoring data, the early warning threshold can fully characterize the inherent laws and fluctuations of foundation settlement in a specific engineering area, enhancing the regional adaptability of the threshold. Furthermore, by combining the settlement control standards specified in the engineering safety specifications with the probability distribution characteristics and determining them through a preset confidence level, it is ensured that the set first early warning threshold not only meets engineering safety requirements but also takes into account the statistical characteristics of historical settlement, thus making the triggering of the initial early warning signal more scientific and accurate.

[0099] In one embodiment of this application, a deep analysis based on multiphysics coupled numerical simulation is performed on the region of interest using a fused dataset to obtain the settlement risk level of the region of interest, including:

[0100] Feature extraction was performed on deformation monitoring images of the area of ​​interest to obtain settlement characteristic parameters;

[0101] Based on the engineering geological conditions and settlement characteristic parameters of the region of interest, a multi-physics coupled numerical model is constructed.

[0102] Based on the time-series monitoring data in the fused dataset, an optimization algorithm is used to invert and calibrate the parameters to be optimized in the multiphysics coupled numerical model, and the calibrated multiphysics coupled numerical model is obtained.

[0103] A calibrated multiphysics coupled numerical model is used to predict the settlement trend of the area of ​​interest in a future set time period, and the settlement prediction results are obtained.

[0104] The settlement prediction results and the latest actual monitoring results of the area of ​​concern are input into a pre-set comprehensive risk assessment model for comparative analysis to obtain the settlement risk level of the area of ​​concern.

[0105] In this embodiment, when extracting features from deformation monitoring images of the area of ​​interest to obtain settlement characteristic parameters, digital image correlation (DIC) technology can be used to calculate the three-dimensional deformation vector of the surface points by analyzing the pixel displacement of images acquired at different time points, and then extract parameters such as settlement amount and settlement rate. Alternatively, synthetic aperture radar interferometry (InSAR) technology can be used to process radar images at different time phases, obtain surface deformation phase information, interpret the deformation rate and cumulative deformation along the radar line of sight, and combine them with auxiliary data (such as digital elevation model DEM) to convert them into vertical settlement characteristic parameters.

[0106] When constructing a multiphysics coupled numerical model based on the engineering geological conditions and settlement characteristic parameters of the area of ​​interest, the complexity of the geological environment must be fully considered. For example, based on engineering geological data such as soil layer distribution, lithology, groundwater level, and engineering loads of the area of ​​interest, as well as the aforementioned extracted settlement characteristic parameters (settlement amount, settlement rate), a suitable constitutive model and seepage model can be selected to construct a finite element model coupling soil deformation and groundwater seepage. Alternatively, the finite difference method can be used to establish a multiphysics coupled model considering soil particle interaction, pore water pressure changes, and the influence of temperature field, based on the soil and rock parameters and settlement characteristic parameters provided in the engineering geological survey report, to simulate complex settlement processes.

[0107] To improve the prediction accuracy of the model, this application further uses optimization algorithms to invert and calibrate the parameters to be optimized in the multiphysics coupled numerical model based on the time-series monitoring data in the fused dataset, resulting in a calibrated multiphysics coupled numerical model. Specifically, the optimization algorithm can be used to compare the settlement time-series data simulated by the model with the actual monitoring data, using the minimization of the error between the two as the objective function, and iteratively adjusting the key parameters in the model (soil elastic modulus, permeability coefficient, consolidation coefficient, etc.) until the preset convergence criterion is reached.

[0108] When using a calibrated multiphysics coupled numerical model to predict the settlement trend of an area of ​​interest over a future set time period, the calibrated model can provide more reliable settlement predictions. For example, the calibrated multiphysics coupled numerical model can be used as a prediction tool, inputting prediction conditions such as external loads and groundwater level changes over a future set time period, running the model simulation, and outputting settlement prediction results such as the cumulative settlement and settlement rate at each point in the area of ​​interest during that time period.

[0109] Finally, the settlement prediction results and the latest actual monitoring results of the area of ​​concern are input into a pre-set comprehensive risk assessment model for comparative analysis to obtain the settlement risk level of the area of ​​concern. The comprehensive risk assessment model can be a fuzzy comprehensive evaluation model based on expert experience. It takes the settlement prediction results (future maximum settlement amount, maximum settlement rate) and the latest actual monitoring results (current cumulative settlement amount, current settlement rate) as input indicators. Based on the pre-set weights and fuzzy membership functions, a comprehensive risk index is calculated, and the settlement risk level is divided according to the index range. The settlement risk levels include: low risk level, medium risk level, and high risk level.

[0110] Through the above technical solutions, this application can directly extract key settlement data from deformation monitoring images, providing a real observational basis for model construction and avoiding errors caused by subjective judgment. Simultaneously, a multi-physics coupled numerical model is constructed using engineering geological conditions and extracted settlement characteristic parameters, ensuring that the model can characterize the actual physical field interactions and enhancing its adaptability to complex geological environments. Furthermore, the model parameters are dynamically optimized using time-series monitoring data, improving the model's prediction accuracy and reducing systematic biases caused by parameter mismatches. Future settlement trend predictions based on the calibrated model ensure the reliability of future trends and provide a forward-looking basis for risk warning. Finally, by comparing predictions with actual data in real time and adjusting the assessment results, the settlement risk level can comprehensively characterize current and potential risks, thereby improving the accuracy and reliability of decision-making and effectively solving the problems of insufficient model accuracy and risk assessment bias.

[0111] In one embodiment of this application, based on time-series monitoring data in a fused dataset, an optimization algorithm is used to invert and calibrate the parameters to be optimized in a multiphysics coupled numerical model, resulting in a calibrated multiphysics coupled numerical model, including:

[0112] The parameters to be optimized in the multiphysics coupled numerical model are selected, including the soil elastic modulus, permeability coefficient and Poisson's ratio.

[0113] An objective function is constructed to measure the overall difference between the settlement time-series data output by the multiphysics coupled numerical model simulation and the actual time-series monitoring data corresponding to the area of ​​interest.

[0114] The harmony search algorithm is used to iteratively optimize the parameters to minimize the objective function;

[0115] The search parameters of the harmony search algorithm are adjusted based on the latest time-series monitoring data used in the in-depth analysis phase of the region of interest.

[0116] When the objective function value is less than the preset convergence threshold or the maximum number of iterations is reached, the iteration stops, and the optimal solution of the parameters to be optimized obtained at this time is substituted into the multiphysics coupled numerical model to obtain the calibrated multiphysics coupled numerical model.

[0117] In this embodiment, the parameters to be optimized refer to physical quantities in a multiphysics coupled numerical model whose accurate values ​​need to be determined through inversion calibration. These parameters directly affect the simulation accuracy and prediction results of the model. The elastic modulus of soil characterizes the soil's ability to resist deformation, the permeability coefficient characterizes the soil's water permeability, and Poisson's ratio describes the ratio of lateral deformation to longitudinal deformation of soil under uniaxial force. In practical applications, the value ranges of these parameters can be preliminarily determined based on engineering geological survey reports and indoor geotechnical test results, serving as the search space for the optimization algorithm.

[0118] The objective function measures the overall difference between the settlement time-series data simulated by a multiphysics coupled numerical model and the actual time-series monitoring data corresponding to the area of ​​interest. The objective function is the core of the optimization algorithm; it quantifies the degree of agreement between the model simulation results and the actual observation data. By minimizing the objective function, the model parameters can be made closer to the true values. The overall difference means that the objective function considers not only the absolute error of settlement but also differences in settlement rate, settlement trend, and other aspects. For example, the root mean square error (RMSE) can be used as the objective function, which calculates the root mean square of the sum of squares of the differences between the simulated and actual settlement time-series data; or a weighted average mean square error can be used, assigning different weights to the errors at different monitoring points or time points, such as giving higher weights to recent data or errors at key monitoring points to characterize their importance.

[0119] The Harmony Search Algorithm (HSA) is a heuristic optimization algorithm based on the process of musical improvisation. It searches for the optimal solution by simulating the process of a musician continuously adjusting pitch to achieve harmony during performance. Specifically, a Harmony Memory (HM) is initialized, storing several initial solutions for the parameters to be optimized. In each iteration, new parameter combinations are selected from or randomly generated from the Harmony Memory, and the Harmony Memory is updated according to the objective function value.

[0120] Based on the latest time-series monitoring data used in the in-depth analysis phase of the region of interest, the search parameters of the harmony search algorithm are adjusted. Adjusting the search parameters refers to dynamically changing the algorithm's behavior according to the latest monitoring data to improve its adaptability and optimization efficiency. Search parameters include the probability of values ​​taken from the harmony memory, the probability of pitch adjustment, and the pitch adjustment bandwidth, which control the algorithm's exploration (global search) and development (local search) capabilities. For example, when the latest monitoring data shows that the settlement trend changes rapidly or has high dispersion, the probability of values ​​taken from the harmony memory and the pitch adjustment bandwidth can be increased to enhance the algorithm's global search capability and adapt to new data patterns more quickly; or when the latest monitoring data shows that the settlement trend tends to be stable or has low dispersion, the probability of values ​​taken from the harmony memory and the pitch adjustment bandwidth can be decreased, and the pitch adjustment probability can be increased to enhance the algorithm's local search capability, allowing for more refined parameter adjustments and improved convergence accuracy.

[0121] The stopping condition is the criterion for terminating the optimization algorithm, aiming to balance computational efficiency and optimization accuracy. A preset convergence threshold defines the minimum acceptable error level of the objective function; when the difference between the model simulation results and the actual monitoring data is sufficiently small, the parameters can be considered calibrated. The maximum number of iterations serves as a safeguard mechanism to prevent the algorithm from failing to converge and running indefinitely in certain situations. For example, the convergence threshold can be set to a small positive number, such as 0.001, indicating that when the objective function value is below this value, the model parameters are sufficiently accurate. The maximum number of iterations can be set based on computational resources and experience. In addition to the above two conditions, another condition can be added: when the improvement in the objective function value is less than a minimum value in a number of consecutive iterations, iteration also stops, indicating that the algorithm is close to convergence and further iterations will not yield significant benefits.

[0122] Through the above technical solution, this application selects key physical parameters such as soil elastic modulus, permeability coefficient, and Poisson's ratio as parameters to be optimized, ensuring that the calibration process focuses on the core factors affecting settlement behavior. An objective function is constructed to measure the comprehensive difference between the simulated output and actual monitoring data, providing a clear and quantifiable evaluation standard for the optimization process. Iterative optimization using the harmony search algorithm, with its global search capability, helps avoid getting trapped in local optima and effectively reduces computational complexity. More importantly, the search parameters of the harmony search algorithm are dynamically adjusted based on the latest time-series monitoring data used in the in-depth analysis phase, enabling the algorithm to respond in real-time to changes in settlement trends, enhancing the model's adaptability to actual engineering environments and the robustness of calibration. Iteration stops when the objective function value reaches a preset convergence threshold or the maximum number of iterations, ensuring high-precision calibration results at a reasonable computational cost. Finally, by substituting the optimal parameters into a multiphysics coupled numerical model, the calibrated model is obtained, improving the accuracy and reliability of settlement prediction, thus providing a more solid foundation for subsequent settlement risk level assessment and graded early warning, effectively improving the timeliness and reliability of early warnings.

[0123] In one embodiment of this application, the search parameters include: harmony memory library value probability, pitch adjustment probability, and pitch adjustment bandwidth;

[0124] Based on the latest time-series monitoring data used in the in-depth analysis phase of the region of interest, the search parameters of the harmony search algorithm are adjusted, including:

[0125] Based on the latest time-series monitoring data in the fused dataset, the coefficient of variation of the settlement rate at each monitoring point in the area of ​​interest and the overall average settlement acceleration are calculated.

[0126] In response to the discrepancy coefficient being greater than a preset discrepancy threshold and the average settlement acceleration being less than a preset trend threshold, the probability of taking values ​​from the harmony memory bank is increased based on the first step length, the probability of pitch adjustment is decreased based on the first step length, and the pitch adjustment bandwidth is narrowed based on the second step length.

[0127] In response to the discrepancy coefficient being less than the preset discrepancy threshold and the average settlement acceleration being greater than the preset trend threshold, the probability of taking values ​​from the harmony memory bank is reduced based on the third step length, the probability of pitch adjustment is increased based on the third step length, and the pitch adjustment bandwidth is widened based on the fourth step length.

[0128] In this embodiment, the probability of selecting a value from the harmony memory is a key parameter in the harmony search algorithm. It controls the probability of selecting a note (i.e., a parameter value) from the harmony memory, determining the degree to which the algorithm utilizes historical optimal solutions. This probability can be preset to a fixed value, such as between 0.7 and 0.95, to balance exploration and utilization; or it can be dynamically adjusted according to the number of iterations or convergence, for example, setting a lower value in the early iterations to encourage exploration and a higher value in later iterations to accelerate convergence. The pitch adjustment probability is another important parameter in the harmony search algorithm, used to control the probability of fine-tuning the note selected from the harmony memory, affecting the algorithm's local search capability. This probability can be set to a fixed value, such as between 0.1 and 0.5; or it can be adaptively adjusted according to the search stage or the rate of change of the objective function, for example, increasing the probability to escape local optima when the objective function changes slowly. The pitch adjustment bandwidth defines the step size or range of pitch adjustment, affecting the fineness of the local search. The bandwidth can be set to a fixed small value, such as 0.01 to 0.1; or it can be dynamically scaled according to the range of parameter values ​​or the fineness of the search, for example, setting a larger bandwidth in the early stage of the search for a coarse search and setting a smaller bandwidth in the later stage for a fine search.

[0129] The latest time-series monitoring data refers to the monitoring data newly acquired after the generation of the initial warning signal, used for model calibration and prediction during the in-depth analysis phase, as well as the time-series data in the target detection dataset when the initial warning signal is triggered. The coefficient of dispersion is a statistic that measures the dispersion of a set of data, defined as the ratio of the standard deviation to the mean. It characterizes the uniformity or fluctuation of the settlement rate among monitoring points within the area of ​​interest. This coefficient of dispersion can be obtained by calculating the standard deviation of the settlement rate at each monitoring point and dividing it by the mean. The overall average settlement acceleration refers to the average rate of change of the settlement rate at all monitoring points within the area of ​​interest over time, characterizing the acceleration state of the overall settlement trend in the area of ​​interest. This acceleration can be calculated by performing time-series differencing on the settlement rate at each monitoring point and then taking the average of the accelerations at all monitoring points.

[0130] The preset discrete threshold is a reference value used to determine whether the dispersion of settlement rate is high. It can be set based on historical data analysis or expert experience, or dynamically determined through statistical methods, such as outlier detection based on box plots. The preset trend threshold is a reference value used to determine whether the overall average settlement acceleration is small. It can be set according to engineering safety specifications or historical settlement trends, for example, close to zero or a small negative value; or determined through statistical analysis of the acceleration during normal settlement processes. The first step length is used to adjust the increment or decrement of the probability of eliciting values ​​from the harmony memory library and the probability of pitch adjustment. It can be a fixed small value; or dynamically adjusted based on the distance between the current parameter value and the target range. The second step length is used to reduce the pitch adjustment bandwidth. It can be a fixed small value; or decreased proportionally to the current bandwidth value. The third step length is used to adjust the increment or decrement of the probability of eliciting values ​​from the harmony memory library and the probability of pitch adjustment. It can be a fixed small value, the same as or different from the first step length; or dynamically adjusted based on the distance between the current parameter value and the target range. The fourth step is an increment used to widen the pitch adjustment bandwidth. It can be a fixed small value, the same as or different from the second step, or it can be increased proportionally to the current bandwidth value.

[0131] Through the above technical solution, this application can intelligently adjust the search parameters of the harmony search algorithm according to the dynamic characteristics of settlement data, effectively solving the problem of lack of adaptability in parameter adjustment during the inversion process. Specifically, when the dispersion coefficient of the settlement rate of each monitoring point in the area of ​​interest is large but the overall average settlement acceleration is small, it indicates that although the settlement data is scattered, the trend of change is relatively stable. At this time, by increasing the probability of taking values ​​from the harmony memory, decreasing the probability of pitch adjustment, and narrowing the pitch adjustment bandwidth, the utilization of historical excellent solutions can be strengthened, unnecessary random exploration can be reduced, and the focus can be placed on local fine search, thereby improving the accuracy and convergence speed of model parameter inversion in stable areas. Conversely, when the dispersion coefficient is small but the average settlement acceleration is large, it indicates that although the settlement data is concentrated, it changes rapidly. At this time, by decreasing the probability of taking values ​​from the harmony memory, increasing the probability of pitch adjustment, and widening the pitch adjustment bandwidth, the ability to explore new solutions can be enhanced, the search diversity can be increased, and the search range can be expanded, thereby accelerating the convergence of the algorithm in rapidly changing areas and avoiding getting trapped in local optima. This data-feature-based conditional adaptive adjustment mechanism enables the harmony search algorithm to better adapt to different settlement stages and modes, improving the efficiency and accuracy of parameter inversion and calibration of multi-physics coupled numerical models.

[0132] In one embodiment of this application, a foundation settlement detection method for engineering testing further includes:

[0133] After the deep analysis step is completed, the target detection datasets within the historical period that are confirmed as valid early warnings or false early warnings, along with their final settlement risk levels, are collected to form the model optimization sample set;

[0134] Using the target detection dataset in the model optimization sample set as input features and its corresponding final settlement risk level as supervision label, the parameters of the rapid screening model are retrained and optimized.

[0135] The retrained model was tested on the validation set to calculate its early warning accuracy, false alarm rate, and average early warning time.

[0136] When the accuracy of early warning is improved by more than the first set threshold and the average early warning time is increased by more than the second set threshold, while the false alarm rate is not greater than the preset tolerance limit, the performance of the judgment model is effectively improved, and the retrained and validated model is updated to the currently used fast screening model.

[0137] In this embodiment, an effective early warning refers to a preliminary warning signal being confirmed as a real settlement risk after in-depth analysis based on multiphysics coupled numerical simulation, and a corresponding settlement risk level is obtained; while a false early warning refers to a preliminary warning signal being confirmed as having no actual settlement risk after in-depth analysis. The system continuously collects target detection datasets and their final settlement risk levels corresponding to these confirmed early warning events within a preset historical period, such as the past week, month, or quarter. The target detection dataset is data after multi-source monitoring data fusion processing, while the final settlement risk level is the output result of the in-depth analysis.

[0138] The target detection dataset in the model optimization sample set is used as input features, such as settlement rate, deformation features, and spatial distribution, and its corresponding final settlement risk level is used as the supervision label to retrain and optimize the parameters of the rapid screening model. The rapid screening model can be built based on machine learning algorithms (random forest). The parameter retraining and optimization process can be carried out using gradient descent, aiming to adjust the weights, biases, or other hyperparameters within the model to minimize the difference between the model's prediction results and the actual supervision label, thereby improving the model's ability to identify settlement risk.

[0139] After model retraining, its performance needs rigorous evaluation. Therefore, the retrained model is tested on an independent validation set. The validation set is a subset of historical data, not involved in the model training process, and is used to assess the model's generalization ability on unknown data. During testing, the system calculates the model's warning accuracy, false alarm rate, and average early warning time. Warning accuracy measures the model's ability to correctly identify actual risk events; the false alarm rate measures the frequency with which the model incorrectly issues warnings; and the average early warning time measures the timeliness of the warning signal, i.e., the average time difference between the warning time and the actual occurrence of the risk.

[0140] Finally, based on the evaluation results, it is determined whether the model performance has been effectively improved. When the early warning accuracy improves by more than a preset first threshold (e.g., 5%) compared to before retraining, and the average early warning time increases by more than a preset second threshold (e.g., 1 day), while the false alarm rate does not exceed a preset tolerance limit (e.g., 5%), the model performance is considered to have been effectively improved. At this point, the retrained and validated model is updated to the currently used rapid screening model and put into actual operation, replacing the old model.

[0141] Through the above technical solutions, this application establishes a continuous optimization mechanism for a rapid screening model, effectively solving the performance degradation problem caused by environmental changes, data distribution shifts, or long-term use.

[0142] In one embodiment of this application, after obtaining the settlement risk level of the area of ​​interest, the method further includes:

[0143] Based on the settlement risk level, the corresponding basic monitoring frequency is obtained by querying the preset risk level-monitoring frequency mapping table;

[0144] Based on the settlement prediction results, the location with the largest cumulative settlement or settlement rate during the prediction period is identified as the prediction point.

[0145] Based on the spatial distance between the monitoring equipment and the prediction point, the spatial influence factor corresponding to each monitoring equipment is calculated using a preset spatial attenuation function.

[0146] The risk adjustment coefficient is determined based on the settlement risk level. Based on the risk adjustment coefficient and the spatial impact factor, the adjustment factor for each monitoring device is calculated using a preset adjustment function.

[0147] The target monitoring frequency for each monitoring device is determined by multiplying the adjustment factor by the base monitoring frequency.

[0148] Based on the calculated target monitoring frequency of each monitoring device, a set of monitoring task instructions, including device identifier and adjusted frequency, is generated and sent to the corresponding monitoring device for execution.

[0149] In this embodiment, the risk level-monitoring frequency mapping table is pre-defined based on a large amount of historical settlement monitoring data, differences in engineering geological conditions, and relevant industry technical specifications. For example, for areas corresponding to low risk levels, the basic monitoring frequency can be set to once per week; for areas corresponding to medium risk levels, it can be set to twice per week; and for areas corresponding to high risk levels, it can be set to once per day. The design logic of the mapping table is that the higher the risk level, the higher the basic monitoring frequency, ensuring that settlement data in high-risk areas can be captured more intensively, reserving sufficient response time for risk warnings. At the same time, the mapping table supports dynamic calibration according to the actual monitoring scenario (such as along urban rail transit lines, around high-rise buildings, soft soil subgrade areas, etc.).

[0150] Based on the settlement prediction results obtained through previous numerical simulations, we further explore key settlement feature points within the prediction period. Specifically, we select the spatial locations with the largest cumulative settlement or the highest settlement rate from the predicted settlement distribution data and identify them as prediction points. The core purpose of this step is to pinpoint the areas with the most concentrated and drastic settlement risks, providing a key targeting basis for subsequent frequency adjustments of monitoring equipment.

[0151] In practice, the settlement data within the prediction period needs to be interpolated in both time and space dimensions to ensure the continuity and integrity of the data. Then, the coordinates (latitude, longitude, and elevation information) of the prediction points are located by using an extreme value solving algorithm (such as gradient descent). The prediction points are then linked and matched with the geographic information system (GIS) of the area of ​​interest to form a visualized prediction point.

[0152] Based on the actual installation coordinates of each monitoring device deployed within the area of ​​interest and the coordinates of the prediction point, the straight-line spatial distance between each monitoring device and the prediction point is first calculated using a spatial distance calculation algorithm (Euclidean distance formula). Then, a preset spatial attenuation function is used to calculate the spatial influence factor corresponding to each monitoring device. The characteristics of the spatial attenuation function are: the closer the spatial distance between the monitoring device and the prediction point, the larger the spatial influence factor, indicating higher sensitivity and stronger data representativeness of the monitoring device for the settlement data of the prediction point; the farther the distance, the smaller the spatial influence factor and the weaker the data representativeness. The spatial attenuation function can adopt an exponential attenuation model, with the attenuation coefficient pre-calibrated according to the topography and geological conditions of the monitoring area.

[0153] The risk adjustment coefficient is determined based on the settlement risk level of the area of ​​interest. The risk adjustment coefficient is set according to the principle of a positive correlation between risk level and adjustment coefficient. For example, the risk adjustment coefficient is 1.0 for low-risk areas (without additional frequency increase), 1.5 for medium-risk areas, and 2.0 for high-risk areas. Specific coefficient values ​​need to be calibrated based on historical risk management cases and settlement disaster loss assessment results. Subsequently, based on the obtained risk adjustment coefficients and the spatial influence factors of each monitoring device, the comprehensive adjustment factor for each monitoring device is calculated using a preset adjustment function. The preset adjustment function can adopt a multiplicative model, i.e., adjustment factor = risk adjustment coefficient × spatial influence factor.

[0154] Based on the baseline monitoring frequency, the target monitoring frequency for each monitoring device is obtained by multiplying it by the calculated comprehensive adjustment factor for each device. During the calculation process, the results need to be validated and standardized: First, if the calculated target monitoring frequency is lower than the device's minimum operating frequency (determined by the device's hardware performance, power consumption constraints, etc.), the minimum operating frequency is taken as the final target frequency. Second, if the target monitoring frequency is higher than the device's maximum operating frequency, the maximum operating frequency is taken, and auxiliary prompts indicating the need for additional monitoring devices are generated simultaneously. Third, the calculation results are rounded to ensure that frequency commands can be recognized and executed by the monitoring devices.

[0155] This embodiment achieves precise allocation of monitoring resources by adjusting frequency based on both risk level and spatial location. This avoids resource waste caused by over-monitoring in low-risk areas while ensuring high-frequency monitoring coverage in high-risk core areas, thus improving resource utilization efficiency. Secondly, it enhances the timeliness and accuracy of settlement risk early warning by optimizing monitoring frequency at key settlement prediction points, ensuring data capture density in areas with drastic risk changes, providing reliable data support for early warning, and reducing disaster losses. Finally, it optimizes operation and maintenance management efficiency through standardized command generation and issuance processes and fault feedback mechanisms, reducing manual intervention costs and ensuring stable execution of monitoring tasks. A flexible parameter calibration mechanism adapts to different geological and engineering scenarios, enhancing the process's versatility and practicality.

[0156] In one embodiment of this application, a calibrated multiphysics coupled numerical model is used to predict the settlement trend of the area of ​​interest over a future set time period. After obtaining the settlement prediction result, the method further includes:

[0157] In the future, within a set time period, new time-series monitoring data of the area of ​​interest will be continuously acquired to form a validation dataset;

[0158] The actual settlement in the validation dataset is compared with the model-predicted settlement at the same time and location, and the prediction error for each monitoring point is calculated.

[0159] When the average prediction error of a certain monitoring point exceeds the preset error tolerance within a consecutive preset number of monitoring cycles, a local recalibration process is performed on the model parameters of the local area to obtain the recalibrated model subdomain parameters. The local recalibration process uses the latest validation dataset as input and, under the constraint of keeping the parameters of the other areas of the calibrated multiphysics coupled numerical model unchanged, only the soil parameters of the model subdomain corresponding to the local area are inverted and optimized.

[0160] Using the recalibrated model subdomain parameters, the multiphysics coupled numerical model is updated, and the settlement trend for the remaining prediction period is re-predicted to generate corrected settlement prediction results.

[0161] The revised settlement prediction results are compared with the latest actual monitoring results and input into the preset comprehensive risk assessment model for analysis, so as to update the settlement risk level of the area of ​​concern.

[0162] In this embodiment, new time-series monitoring data of the area of ​​interest needs to be continuously acquired within a set future time period to form a verification dataset.

[0163] This embodiment performs precise matching of timestamps and spatial locations between the actual monitoring data and the model prediction data in the validation dataset. For example, for each monitoring point, at the end of each monitoring period, its latest actual settlement is extracted and compared with the settlement predicted by the model at the same time and location. The prediction error can be calculated in various ways, such as calculating the absolute difference between the two (predicted value minus actual value) or calculating the relative error (the ratio of the absolute difference to the actual value). Through this comparison, the prediction accuracy of the model at different monitoring points and time periods can be intuitively understood.

[0164] When the average prediction error of a monitoring point exceeds a preset error tolerance within a consecutive preset number of monitoring periods, a local recalibration process for the model parameters in that local area is triggered to obtain the recalibrated model subdomain parameters. The consecutive preset number of monitoring periods and the preset error tolerance are set as trigger conditions. For example, recalibration can be initiated if the average prediction error (e.g., root mean square error) of a monitoring point exceeds 5 mm within three consecutive monitoring periods. The local recalibration process aims to efficiently correct prediction biases in specific local areas, rather than performing a global recalculation of the entire model. This process uses the latest validation dataset as input and, while maintaining the parameters of the rest of the calibrated multiphysics coupled numerical model unchanged, only performs inversion optimization on the soil parameters of the corresponding model subdomain in that local area. For example, for areas with large errors, optimization algorithms can be used to iteratively adjust key parameters such as the soil elastic modulus, permeability coefficient, and Poisson's ratio to make the simulation results more consistent with the latest actual monitoring data.

[0165] After obtaining the recalibrated model subdomain parameters, these parameters are used to update the multiphysics coupled numerical model. Specifically, the optimized local soil parameters replace the parameters of the corresponding subdomains in the original model, while the model parameters for other error-free areas remain unchanged. Subsequently, based on the updated model, the settlement trend for the remaining prediction period is re-predicted, thereby generating corrected settlement prediction results. This step ensures that the model can reflect the latest geological conditions and settlement behavior in a timely manner, providing more accurate and realistic prediction data for subsequent risk assessment. The corrected settlement prediction results will be presented as new settlement curves or discrete prediction values, covering the entire prediction interval from the current time to the end of the future set time period.

[0166] Finally, the revised settlement prediction results are compared with the latest actual monitoring results and input into a pre-set comprehensive risk assessment model for analysis, thereby updating the settlement risk level of the area of ​​concern. This step aims to combine the latest prediction information with actual observation data to conduct a comprehensive and dynamic assessment of the settlement risk of the area of ​​concern. Through comparative analysis, the comprehensive risk assessment model can more accurately determine the degree of impact of settlement on the safety of the engineering structure in the current and future periods, and dynamically adjust and update the settlement risk level of the area of ​​concern accordingly, such as adjusting it from a medium-risk level to a high-risk level, or downgrading it from a high-risk level to a medium-risk level, thereby achieving timely response and precise management of risk warnings.

[0167] Through the above technical solution, this application introduces a dynamic verification and local recalibration mechanism, which effectively solves the problems of prediction deviation and assessment lag caused by the failure to capture dynamic changes in geological conditions or uneven settlement in local areas in a timely manner, thereby ensuring the real-time updating of settlement risk level.

[0168] In one embodiment of this application, after spatially overlaying all circular buffer zones with the area to be detected and the adjacent sensitive structure areas, and extracting the overlaid area as the area of ​​interest, the method further includes:

[0169] Obtain known engineering geological zoning maps, underground pipeline distribution maps, and historical damage records within the area of ​​interest;

[0170] The initial boundary polygon of the area of ​​interest is overlaid with the boundary of unfavorable geological bodies in the engineering geological zoning map, the location of key pipelines in the underground pipeline distribution map, and the spatial location in the historical disease records.

[0171] If the spatial distance between the initial boundary and any of the aforementioned prior elements is less than a preset association distance threshold, then the boundary or location buffer of the prior elements will be merged into the region of interest to form the final corrected region of interest.

[0172] In this embodiment, the engineering geological zoning map can be obtained from geological survey reports, regional geological maps, engineering design drawings, or relevant geological databases. It provides spatial distribution information on soil and rock types, geological structures, and adverse geological phenomena (such as weak soil layers, fault zones, and mining subsidence areas) within the region. The underground pipeline distribution map can be obtained from urban planning departments, pipeline management units, as-built drawings, or underground pipeline detection data. It details the spatial location and burial depth of critical underground pipelines such as gas, water supply and drainage, electricity, and communications, which are highly susceptible to damage during settlement. Historical damage records can be obtained from engineering archives, monitoring reports, maintenance records, accident reports, or relevant management systems. These records document the spatial location and occurrence time of past damage events such as foundation settlement, structural cracking, and pipeline damage, providing direct historical evidence for identifying potential risk points. Spatial comparison identifies potential correlations between the initial area of ​​concern and these prior risk factors. Overlay analysis can be implemented using Geographic Information System (GIS) software, allowing for the overlay display and analysis of different geographic information layers. For the boundary of an unfavorable geological body, if its distance from the initial area of ​​interest is less than a threshold, the polygon represented by the boundary of the unfavorable geological body can be spatially merged with the initial area of ​​interest (e.g., using GIS union or fusion operations), thereby expanding the area of ​​interest. For critical pipeline locations or historical defect points, since they are point or line features, a buffer zone with a preset radius can be generated centered on these features, and then this buffer zone can be spatially merged with the initial area of ​​interest. The preset association distance threshold can be set according to the project type, geological conditions, risk level, and empirical values; for example, it can be set to 5 meters, 10 meters, or 20 meters.

[0173] Through the aforementioned technical solution, this application, based on the initial identification of the area of ​​interest, further integrates known engineering geological conditions, underground pipeline distribution, and historical damage information. By performing spatial overlay analysis on the initial area of ​​interest and these prior risk factors, and dynamically correcting the boundary of the area of ​​interest according to a preset correlation distance threshold, the final area of ​​interest can more comprehensively and accurately cover all potential risk factors, including adverse geological bodies, critical underground pipelines, and historical damage points. This effectively solves the problem that relying solely on monitoring data can lead to inaccurate boundaries of the area of ​​interest and omission of potential risks.

[0174] Corresponding to the foundation settlement detection method for engineering testing in the above embodiment, Figure 2 This is a structural block diagram of a foundation settlement detection system for engineering testing provided in an embodiment of this application. For ease of explanation, only the parts relevant to the embodiments of this application are shown. References Figure 2The foundation settlement detection system 20 used for this project includes: a first data module 21, a data processing module 22, a model calculation module 23, a first early warning module 24, a second data module 25, a risk determination module 26, and a second early warning module 27.

[0175] The first data module 21 is used to acquire multi-source monitoring data collected in the area to be inspected and the adjacent sensitive structure area; the multi-source monitoring data includes: time-series monitoring data and deformation monitoring images;

[0176] Data processing module 22 is used to fuse multi-source monitoring data to obtain target detection dataset;

[0177] The model calculation module 23 is used to input the target detection dataset into the preset rapid screening model, calculate the settlement risk assessment value, and form a risk assessment point set;

[0178] The first early warning module 24 is used to generate a preliminary early warning signal if the settlement risk assessment value of any point in the risk assessment point set is greater than the first early warning threshold, and to determine the corresponding area of ​​concern based on the spatial distribution of the risk assessment point set.

[0179] The second data module 25 is used to acquire a fused dataset for in-depth analysis in response to the initial warning signal; the fused dataset for in-depth analysis includes the target detection dataset that triggered the initial warning signal and the monitoring data newly acquired after the initial warning signal was generated;

[0180] The risk determination module 26 is used to perform in-depth analysis of the area of ​​interest based on multi-physics coupled numerical simulation based on the fused dataset, and obtain the settlement risk level of the area of ​​interest.

[0181] The second early warning module 27 is used to generate corresponding graded early warning information based on the settlement risk level.

[0182] See Figure 3 , Figure 3 This is a schematic block diagram of an electronic device provided in an embodiment of this application. (For example...) Figure 3 The electronic device 300 in this embodiment may include one or more processors 301, one or more input devices 302, one or more output devices 303, and one or more memories 304. The processors 301, input devices 302, output devices 303, and memories 304 communicate with each other via a communication bus 305. The memories 304 store computer programs, including program instructions. The processors 301 execute the program instructions stored in the memories 304. Specifically, the processors 301 are configured to invoke the program instructions to perform the functions of the modules in the aforementioned device embodiments.

[0183] It should be understood that, in the embodiments of this application, the processor 301 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), etc. A general-purpose processor may be a microprocessor or any conventional processor.

[0184] Input device 302 may include a touchpad, a fingerprint sensor (for collecting the user's fingerprint information and fingerprint orientation information), a microphone, etc., and output device 303 may include a display (LCD, etc.), a speaker, etc.

[0185] The memory 304 may include read-only memory and random access memory, and provides instructions and data to the processor 301. A portion of the memory 304 may also include non-volatile random access memory. For example, the memory 304 may also store device type information.

[0186] In specific implementations, the processor 301, input device 302, and output device 303 described in the embodiments of this application can execute the implementation methods described in any embodiment of the foundation settlement detection method for engineering testing provided in the embodiments of this application, or they can execute the implementation methods of the electronic devices described in the embodiments of this application, which will not be repeated here.

[0187] This application also provides a computer-readable storage medium storing a computer program. The computer program includes program instructions, which, when executed by a processor, implement all or part of the processes in the methods described above. The computer program can also instruct related hardware to implement these processes. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer-readable storage medium is used to store the computer program and other programs and data required by the electronic device.

[0188] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for detecting foundation settlement in engineering testing, characterized in that, include: Acquire multi-source monitoring data collected in the project area to be inspected and adjacent sensitive structures; The multi-source monitoring data includes: time-series monitoring data and deformation monitoring images; The multi-source monitoring data are fused to obtain a target detection dataset; The target detection dataset is input into a preset fast screening model to calculate the settlement risk assessment value, which constitutes a risk assessment point set. The fast screening model is a machine learning model based on gradient boosting decision tree, which uses settlement rate, cumulative settlement amount, and deformation gradient as input features to calculate the settlement risk assessment value that characterizes the degree of settlement risk. If the settlement risk assessment value of any point in the risk assessment point set is greater than the first warning threshold, a preliminary warning signal is generated, and the risk assessment point set is used as data points to input a spatial clustering algorithm to identify at least one risk cluster, and a region of interest is determined based on the risk cluster; the neighborhood search radius parameter of the spatial clustering algorithm is adaptively adjusted according to the local spatial gradient of the settlement risk assessment value of the data point; including: calculating the average gradient of the settlement risk assessment values ​​of each data point and other data points in the surrounding preset neighborhood; The gradient mean is input into a preset mapping function to obtain the neighborhood radius adjustment coefficient corresponding to the data point; The neighborhood search radius of the data point is obtained by multiplying the preset basic neighborhood search radius by the neighborhood radius adjustment coefficient. In response to the initial warning signal, a fused dataset for in-depth analysis is acquired; the fused dataset for in-depth analysis includes the target detection dataset that triggered the initial warning signal and the newly acquired monitoring data after the initial warning signal was generated; Based on the fused dataset, a deep analysis based on multiphysics coupling numerical simulation is performed on the region of interest to obtain the subsidence risk level of the region of interest. The in-depth analysis includes: after constructing a multi-physics coupled numerical model, using the time-series monitoring data in the fused dataset, employing the harmony search algorithm to invert and calibrate the model's parameters to be optimized, and dynamically adjusting the search parameters of the harmony search algorithm according to the latest time-series monitoring data used in the in-depth analysis phase for the region of interest; Based on the aforementioned settlement risk level, corresponding graded early warning information is generated.

2. The method for detecting foundation settlement for engineering testing according to claim 1, characterized in that, The step of determining the region of interest based on the risk cluster includes: For each risk cluster, calculate the area of ​​the risk cluster and the maximum value of the settlement risk assessment value of all data points within the risk cluster, and calculate the corresponding risk cluster's radius of concern using a preset weighting formula; A circular buffer is generated with the geometric center of each risk cluster as the center and the corresponding radius of concern as the radius; All circular buffer zones are spatially overlaid with the project area to be detected and the adjacent sensitive structure areas, and the overlaid area is extracted as the region of interest.

3. The method for detecting foundation settlement for engineering testing according to claim 2, characterized in that, The step of spatially overlaying all circular buffer zones with the area to be inspected and the adjacent sensitive structure areas, and extracting the overlaid area as the area of ​​interest, includes: The circular buffers generated for each risk cluster are spatially merged to generate a total buffer polygon. The total buffer polygon is spatially intersected with the boundary polygons of the sensitive structure area and the boundary polygon of the project area to be detected to obtain two intersecting sub-regions. The two intersecting sub-regions are spatially merged to obtain the final region of interest.

4. The method for detecting foundation settlement for engineering testing according to claim 1, characterized in that, The method for determining the first warning threshold includes: Obtain historical settlement monitoring data and engineering safety specifications for the project area to be inspected; Based on the historical settlement monitoring data, the probability distribution characteristics of the settlement rate in the project area to be monitored were statistically obtained. Based on the settlement control standards specified in the engineering safety specifications and the probability distribution characteristics, the first early warning threshold is determined by a preset confidence level.

5. The method for detecting foundation settlement for engineering testing according to claim 4, characterized in that, The determination of the first warning threshold based on the settlement control standards specified in the engineering safety specifications and the probability distribution characteristics, through a preset confidence level, includes: Obtain the settlement rate control standard value specified in the engineering safety specifications; Calculate the quantiles of the probability distribution of settlement rate in the historical settlement monitoring data, where the quantiles correspond to the preset confidence level; The first warning threshold is obtained by weighted summation of the settlement rate control standard value and the quantile.

6. The method for detecting foundation settlement for engineering testing according to claim 1, characterized in that, The deep analysis of the region of interest based on the fused dataset using multiphysics coupled numerical simulation, to obtain the settlement risk level of the region of interest, includes: Feature extraction is performed on the deformation monitoring images of the area of ​​interest to obtain settlement feature parameters; Based on the engineering geological conditions of the area of ​​interest and the settlement characteristic parameters, a multi-physics coupled numerical model is constructed. Based on the time-series monitoring data in the fused dataset, an optimization algorithm is used to invert and calibrate the parameters to be optimized in the multiphysics coupled numerical model, resulting in a calibrated multiphysics coupled numerical model. Using the calibrated multiphysics coupled numerical model, the settlement trend of the area of ​​interest in the future within a set time period is predicted, and the settlement prediction results are obtained. The settlement prediction results and the latest actual monitoring results of the area of ​​concern are input into a preset comprehensive risk assessment model for comparison and analysis to obtain the settlement risk level of the area of ​​concern.

7. The method for detecting foundation settlement for engineering testing according to claim 6, characterized in that, Based on the time-series monitoring data in the fused dataset, an optimization algorithm is used to invert and calibrate the parameters to be optimized in the multiphysics coupled numerical model, resulting in a calibrated multiphysics coupled numerical model, including: Select the parameters to be optimized in the multiphysics coupled numerical model, including the soil elastic modulus, permeability coefficient and Poisson's ratio; Construct an objective function to measure the overall difference between the settlement time series data simulated by the multiphysics coupled numerical model and the actual time series monitoring data corresponding to the region of interest; The harmony search algorithm is used to iteratively optimize the parameters to be optimized in order to minimize the objective function; The search parameters of the harmony search algorithm are adjusted based on the latest time-series monitoring data used in the deep analysis phase of the region of interest. When the objective function value is less than the preset convergence threshold or the maximum number of iterations is reached, the iteration stops, and the optimal solution of the parameters to be optimized obtained at this time is substituted into the multiphysics coupling numerical model to obtain the calibrated multiphysics coupling numerical model.

8. A foundation settlement detection system for engineering testing, used to perform the method according to any one of claims 1 to 7, characterized in that, include: The first data module is used to acquire multi-source monitoring data collected in the project area to be inspected and the adjacent sensitive structure area; The multi-source monitoring data includes: time-series monitoring data and deformation monitoring images; The data processing module is used to fuse the multi-source monitoring data to obtain a target detection dataset; The model calculation module is used to input the target detection dataset into a preset fast screening model to calculate the settlement risk assessment value and form a risk assessment point set. The fast screening model is a machine learning model based on gradient boosting decision tree, which uses settlement rate, cumulative settlement amount and deformation gradient as input features to calculate the settlement risk assessment value that represents the degree of settlement risk. The first early warning module is used to generate a preliminary early warning signal if the settlement risk assessment value of any point in the risk assessment point set is greater than a first early warning threshold, and input the risk assessment point set as data points into a spatial clustering algorithm to identify at least one risk cluster, and determine the area of ​​interest based on the risk cluster; the neighborhood search radius parameter of the spatial clustering algorithm is adaptively adjusted according to the local spatial gradient of the settlement risk assessment value of the data point; including: calculating the gradient mean of the settlement risk assessment values ​​of each data point and other data points in the surrounding preset neighborhood; The gradient mean is input into a preset mapping function to obtain the neighborhood radius adjustment coefficient corresponding to the data point; The neighborhood search radius of the data point is obtained by multiplying the preset basic neighborhood search radius by the neighborhood radius adjustment coefficient. The second data module is used to acquire a fusion dataset for in-depth analysis in response to the initial warning signal; the fusion dataset for in-depth analysis includes the target detection dataset that triggered the initial warning signal and the monitoring data newly acquired after the initial warning signal was generated; The risk determination module is used to perform in-depth analysis of the region of interest based on multiphysics coupling numerical simulation based on the fused dataset to obtain the settlement risk level of the region of interest. The in-depth analysis includes: after constructing a multiphysics coupling numerical model, using the harmony search algorithm to invert and calibrate the parameters to be optimized of the model based on the time-series monitoring data in the fused dataset, and dynamically adjusting the search parameters of the harmony search algorithm according to the latest time-series monitoring data used in the in-depth analysis stage of the region of interest. The second early warning module is used to generate corresponding graded early warning information based on the settlement risk level.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images