A pile foundation health state analysis early warning method and system

Abstract

The application relates to the technical field of building monitoring, and provides a pile foundation health state analysis early warning method and system. The method comprises the following steps: a monitoring platform receives and processes various monitoring data from multiple source sensors in real time, combines a target building structure with a graph neural network and a lightweight multi-head attention time sequence network, performs data fusion and dynamic trend analysis on the various monitoring data, and obtains an initial pile foundation evaluation result; the initial pile foundation evaluation result of the target building in each geographical region and the various monitoring data are uploaded to a cloud data processing center; the cloud data processing center stores the initial pile foundation evaluation result of the target building in each geographical region, comprehensively evaluates the correlation trend between the target buildings in each geographical region, and obtains a correction parameter; the correction parameter is sent to the monitoring platform, the monitoring platform dynamically combines the correction parameter to update the pile foundation health state of the target building in each geographical region, and a target pile foundation evaluation result of the target building is obtained.

Description

Technical Field

[0001] This application relates to the field of building monitoring technology, and more specifically, to an early warning method and system for analyzing the health status of pile foundations. Background Technology

[0002] With the continuous advancement of urban construction and the deepening of urbanization in my country, the dependence of building structures on the seismic performance and bearing stability of pile foundations has increased. As the core component of building foundations, the health status of pile foundations directly determines the safety and reliability of buildings throughout their entire life cycle.

[0003] Currently, the demolition and renovation of old buildings has become the mainstream direction of urban renewal, requiring precise understanding of core information such as the construction time, monitoring location, dimensional parameters, and life-cycle stress changes of pile foundations. However, existing pile foundation monitoring is mostly concentrated in the production and construction phases, with the core objective of ensuring the integrity of the pile body during construction (such as detection of broken or inclined piles) and short-term bearing capacity. After the construction unit completes the handover, pile foundation monitoring is interrupted, lacking a long-term monitoring mechanism. During the building's service life (e.g., over 50 years), data on pile foundation stress, settlement, and deformation are not continuously recorded. When the building faces renovation or expansion or reaches its service life, it is impossible to trace its long-term health trends, resulting in a lack of data support during the demolition and renovation of old buildings, which can easily lead to construction risks and resource waste. In addition, during the building's use, the stress and deformation risks of the pile foundation are constantly present due to the influence of surrounding new construction projects, underground pipeline construction, and changes in environmental loads, which also urgently require continuous monitoring to avoid structural safety hazards.

[0004] In summary, there is an urgent need to design a novel technical solution to assess the health status of pile foundations and help improve the reliability of building management and building safety. Summary of the Invention

[0005] In this context, the embodiments of this application aim to provide an early warning method and system for pile foundation health status analysis, which can realize continuous intelligent monitoring and early warning of pile foundation from the construction stage to the entire service life, provide reliable data for building renovation and expansion, safety assessment and pile foundation status analysis, and improve building management efficiency, reliability and building construction safety.

[0006] In a first aspect of this application, an early warning method for pile foundation health status analysis is provided, applied to an early warning system for pile foundation health status analysis. The system includes multi-source sensors embedded in the pile foundation, a monitoring platform, and a cloud data processing center. The method includes:

[0007] The monitoring platform integrates a communication module to receive and process various monitoring data from multiple sources of sensors in real time. These monitoring data include at least one of the following: settlement, inclination, stress, and deformation data of the pile foundation. By combining graph neural networks and lightweight multi-head attention time-series networks with the building structure of the target building, the platform performs data fusion and dynamic trend analysis on the various monitoring data to obtain the initial pile foundation assessment results. The initial pile foundation assessment results are used to represent the health status of the pile foundation of the target building.

[0008] The initial pile foundation assessment results of target buildings in each geographical area, as well as various monitoring data, are uploaded to the cloud data processing center through the monitoring platform.

[0009] The cloud data processing center stores the initial pile foundation assessment results of target buildings in each geographical region and comprehensively evaluates the correlation trends between target buildings in each geographical region to obtain the corresponding correction parameters for each geographical region. The correlation trends include: the structural correlation between the pile foundation of the target building and the surrounding buildings, the correlation between the pile foundation of the target building and the surrounding construction activities, and the correlation between the pile foundation of the target building and the surrounding geological activities.

[0010] The correction parameters are sent to the monitoring platform, which then dynamically updates the pile foundation health status of the target building in each geographical area based on the correction parameters, thereby obtaining the target pile foundation assessment results of the target building.

[0011] In a second aspect of the embodiments of this application, an early warning system for pile foundation health status analysis is provided. The system includes multi-source sensors embedded in the pile foundation, a monitoring platform, and a cloud data processing center.

[0012] The monitoring platform has a built-in integrated communication module for receiving and processing various monitoring data from multiple sources of sensors in real time. These monitoring data include at least one of the following: settlement, inclination, stress, and deformation data of the pile foundation. By combining graph neural networks and lightweight multi-head attention temporal networks with the building structure of the target building, the platform performs data fusion and dynamic trend analysis on the various monitoring data to obtain initial pile foundation assessment results. The initial pile foundation assessment results are used to represent the health status of the pile foundation of the target building.

[0013] The monitoring platform is also used to upload the initial pile foundation assessment results of target buildings in various geographical areas and various monitoring data to the cloud data processing center;

[0014] The cloud data processing center is used to store the initial pile foundation evaluation results of target buildings in various geographical areas, and to comprehensively evaluate the correlation trends between target buildings in various geographical areas to obtain the corresponding correction parameters for each geographical area. The correlation trends include: the structural correlation between the pile foundation of the target building and the surrounding buildings, the correlation between the pile foundation of the target building and the surrounding construction activities, and the correlation between the pile foundation of the target building and the surrounding geological activities.

[0015] The cloud data processing center is also used to send the corrected parameters to the monitoring platform;

[0016] The monitoring platform is also used to dynamically update the health status of the pile foundations of target buildings in each geographical area by combining the correction parameters, so as to obtain the target pile foundation evaluation results of the target buildings.

[0017] In a third aspect of the embodiments of this application, a terminal device is provided, the terminal device comprising: at least one processor, a memory, and an input / output unit; wherein the memory is used to store a computer program, and the processor is used to invoke the computer program stored in the memory to execute the early warning method for pile foundation health status analysis as described in the first aspect.

[0018] In a fourth aspect of the embodiments of this application, a computer-readable storage medium is provided, which includes instructions that, when executed on a computer, cause the computer to perform the early warning method for pile foundation health status analysis as described in the first aspect.

[0019] In a fifth aspect of the embodiments of this application, a computer program product is provided, including a computer program that, when executed by a processor, implements the early warning method for pile foundation health status analysis as described in the first aspect.

[0020] According to an embodiment of this application, an early warning method and system for pile foundation health status analysis is provided. First, a monitoring platform with an integrated communication module receives and processes various monitoring data from multiple sources of sensors in real time. These monitoring data include at least one of the following: pile foundation settlement, inclination angle, stress, and deformation data. A graph neural network and a lightweight multi-head attention time-series network are used in conjunction with the building structure of the target building to perform data fusion and dynamic trend analysis on the various monitoring data, obtaining an initial pile foundation assessment result. This initial pile foundation assessment result represents the pile foundation health status of the target building. Next, the monitoring platform uploads the initial pile foundation assessment results of the target buildings in each geographical area, along with the various monitoring data, to a cloud data processing center. Then, the cloud data processing center stores the initial pile foundation assessment results of the target buildings in each geographical area and comprehensively evaluates the correlation trends between the target buildings in each geographical area to obtain the corresponding correction parameters for each geographical area. The correlation trends include: the structural correlation between the pile foundation of the target building and surrounding buildings, the correlation between the pile foundation of the target building and surrounding construction activities, and the correlation between the pile foundation of the target building and surrounding geological activities. Finally, the corrected parameters are sent to the monitoring platform, which dynamically updates the pile foundation health status of the target building in each geographical area based on the corrected parameters, thus obtaining the target pile foundation assessment results for the target building. This implementation method improves the comprehensiveness, accuracy, and dynamic adaptability of pile foundation health monitoring, effectively solving problems such as stage limitations, isolated assessments, and insufficient response to environmental interference in traditional pile foundation monitoring. Attached Figure Description

[0021] Figure 1 This is a flowchart illustrating an early warning method for pile foundation health status analysis as shown in this application;

[0022] Figure 2 This is a structural schematic diagram of an early warning system for pile foundation health status analysis, as shown in this application. Detailed Implementation

[0023] The following is for reference. Figure 1 , Figure 1 This is a flowchart illustrating an early warning method for pile foundation health status analysis provided in an embodiment of this application.

[0024] To address at least one of the aforementioned technical problems, embodiments of this application provide an early warning method and system for pile foundation health status analysis. Specifically, in the initial assessment stage, by leveraging the multi-source sensor data reception and processing capabilities integrated into the monitoring platform, combined with graph neural networks and lightweight multi-head attention time-series networks, multi-dimensional monitoring data such as pile foundation settlement, inclination angle, stress, and deformation are fully integrated, and the structural features of the target building are embedded. This not only avoids the one-sidedness of single-data monitoring but also, relying on the efficient operation characteristics of lightweight networks, enables real-time data fusion and dynamic trend analysis on resource-constrained monitoring platforms. It accurately outputs initial pile foundation assessment results reflecting the current overall state and short-term change trends of the pile foundation, laying a reliable foundation for subsequent monitoring. Secondly, by uploading monitoring data and initial assessment results of multiple target buildings within the region to the cloud data processing center through the monitoring platform, the cloud data processing center can improve the monitoring of single buildings. It comprehensively considers the structural correlation between the target building's pile foundation and surrounding buildings, its relationship with surrounding construction activities, and its correlation with surrounding geological activities. This allows for a comprehensive assessment of the correlation trends in the health status of pile foundations within the region, generating corrected parameters that align with the actual working conditions of the area. This effectively avoids assessment biases caused by neglecting the influence of the surrounding environment in traditional monitoring. Finally, the cloud data processing center distributes corrected parameters, and the monitoring platform dynamically updates the target pile foundation assessment results. This enables dynamic adjustment of the pile foundation health status, ensuring that the assessment results are always adapted to changes in the surrounding environment and the long-term working conditions of the pile foundation, rather than being limited to static assessments at fixed points in time. This covers the monitoring needs of the entire life cycle of the pile foundation, providing accurate and continuous data support for subsequent maintenance, renovation, and demolition of buildings, thereby improving the safety and reliability of building structures.

[0025] The monitoring platform, located outside the pile foundation, has built-in integrated chips, communication modules, etc. It can receive and process monitoring data from multiple sources in real time, upload the data to the cloud data processing center, and update the pile foundation health status assessment results in combination with the correction parameters issued by the cloud data processing center.

[0026] The cloud data processing center can store the initial pile foundation assessment results of target buildings in various geographical areas, comprehensively assess the correlation trends between pile foundations and surrounding buildings, surrounding construction activities, and surrounding geological activities in the area, obtain corrected parameters, and send them to the monitoring platform.

[0027] Figure 1 The flowchart of an early warning method for pile foundation health status analysis provided in one embodiment of this application includes:

[0028] Step S101: The monitoring platform integrates communication modules to receive and process various monitoring data from multiple sources in real time. By combining graph neural networks and lightweight multi-head attention time-series networks with the building structure of the target building, the monitoring data is fused and dynamic trend analysis is performed to obtain the initial pile foundation evaluation results.

[0029] In this embodiment, the various monitoring data include at least one of the following: settlement, tilt angle, stress, and deformation data of the pile foundation. In practical applications, the settlement, tilt angle, stress, and deformation data collected by each sensor are transmitted to a monitoring platform outside the pile foundation through multiple data connection lines extending from the built-in cylinder. At the same time, the power supply connection line provides continuous power to each sensor from the monitoring platform to ensure the continuity of data acquisition.

[0030] The monitoring platform incorporates a built-in chip capable of running graph neural networks and lightweight multi-head attention temporal networks. The former constructs a correlation model by combining the structural characteristics of the target building (such as the connection method between the pile foundation and the load-bearing components of the superstructure, and the structural characteristics of the pile foundation itself), while the latter focuses on mining the temporal variation patterns of multi-dimensional monitoring data and identifying the inherent correlations between different data (such as the synchronicity of pile foundation settlement and stress changes). Working together, the two deeply fuse the transmitted monitoring data, eliminating invalid fluctuations caused by temporary interference (such as transient electromagnetic influences), and simultaneously performing dynamic trend analysis to determine whether data changes conform to the patterns under normal pile foundation use conditions, ultimately generating initial pile foundation assessment results.

[0031] Alternatively, the graph neural network, after optimization and compression, can be loaded into the corresponding integrated chip built into the monitoring platform. For example, by using the integrated chip built into the monitoring platform to fuse multi-source sensor data in real time and perform trend analysis in conjunction with the characteristics of the building structure, the health status changes of the pile foundation throughout its entire life cycle can be comprehensively reflected.

[0032] In this embodiment, the initial pile foundation assessment result is used to represent the pile foundation health status of the target building. Here, pile foundation health status refers to whether the pile foundation can maintain structural performance that meets the building's safety and stability requirements throughout its entire life cycle (covering construction, use, and pre-reconstruction / demolition phases). This judgment relies on continuous monitoring and comprehensive evaluation of key physical state parameters of the pile foundation, combined with the building's own structural characteristics and the influence of the surrounding environment, ultimately providing data support for the building's life-cycle maintenance and reconstruction / demolition. For example, in the pile foundation monitoring of a residential building, the settlement data captured by the displacement sensor remained stable over a long period, the tilt sensor did not detect significant tilting, and the data from the stress and deformation sensors remained within a reasonable range. After data fusion and trend analysis, the monitoring platform output an initial pile foundation assessment result indicating that the pile foundation's health status was good and no additional intervention was needed. However, for a building facing a street, due to surrounding construction activities, the monitoring platform analysis revealed that the settlement data of the building's pile foundation showed a slow upward trend, with stress data accompanied by slight fluctuations. Although these were not yet within the safe range, the initial pile foundation assessment result still indicated the need to increase the monitoring frequency to promptly track changes in the pile foundation's status and provide a basis for subsequent possible maintenance measures.

[0033] Specifically, during step S101, the monitoring data is first received and preliminarily processed in real time. The integrated communication module of the monitoring platform maintains a continuous and stable signal connection with the multi-source sensors embedded inside the pile foundation, acquiring real-time data on pile settlement, inclination angle, stress, and deformation collected by the sensors. While receiving data, the module removes invalid data caused by temporary poor sensor contact or external electromagnetic interference according to preset basic screening rules, ensuring that the data entering subsequent processing stages has basic validity.

[0034] Subsequently, the data fusion and dynamic trend analysis phase begins. Considering the limited hardware resources of the integrated chip on the monitoring platform, the graph neural network needs to be specifically optimized and compressed first. By simplifying the network structure and reducing redundant parameters, the computational load of the model is reduced, making it compatible with the chip's operating capabilities. After optimization, the graph neural network is deployed collaboratively with a lightweight multi-head attention temporal network on the integrated chip. The former combines the structural characteristics of the target building, such as the depth, diameter, concrete grade of the pile foundation, and the connection method between the pile foundation and the superstructure, to construct a correlation model of the pile foundation and the overall load-bearing system of the building. The latter focuses on mining the temporal variation patterns of multi-dimensional monitoring data and learning the intrinsic correlations between different data, such as the linkage between pile foundation settlement and stress values ​​over time. Together, they deeply fuse the initially processed monitoring data and implement dynamic trend analysis to capture the hidden tendencies in pile foundation status changes behind the data.

[0035] Finally, the chip's computational processing outputs the initial pile foundation assessment results, which clearly reflect the current health status of the target building's pile foundation and demonstrate short-term trends. Taking the pile foundation monitoring of a multi-story residential building as an example, various sensors were installed inside the internal cylinder during the pile foundation pouring process. During the monitoring phase, the integrated communication module of the monitoring platform receives data from each sensor in real time. The integrated chip STM32F407 initiates a collaborative network to process the data. Combined with the structural parameters of the building's pile foundation (15 meters deep, 1 meter in diameter), the analysis reveals that the pile foundation settlement data has remained stable within a small range over a long period, with no tilt angle deviation, and stress and deformation data meeting design standards. Therefore, the initial pile foundation assessment result indicates that the building's pile foundation is currently in good health and there is no abnormal risk in the short term. If, during the pile foundation monitoring of a commercial building, graph neural network analysis reveals a slow upward trend in settlement data recently, accompanied by slight fluctuations in stress data, the initial assessment result will indicate that the pile foundation requires continuous monitoring, providing a basis for further investigation.

[0036] For example, when implementing the steps of acquiring various monitoring data and generating initial pile foundation assessment results, the data acquisition device must first be pre-deployed. During the pile foundation pouring process, an internal cylinder is pre-embedded inside the pile foundation. The cylindrical and conical parts of the internal cylinder need to be installed in reliable positions according to the distribution of the pile foundation reinforcement to ensure that the stability of the pile foundation structure is not affected. The displacement sensor, tilt sensor, stress sensor, and deformation sensor inside the internal cylinder need to be fixed in preset positions. The tilt sensor is installed at the tip of the conical part to ensure monitoring accuracy, while the stress sensor and deformation sensor are evenly distributed along the inner wall of the cylindrical part to ensure comprehensive capture of stress and deformation in different directions of the pile foundation. The displacement sensor, together with the fixed measuring point set on the ground and the fixed rod and magnetic ring extending into the pile foundation, realizes settlement data acquisition by sensing the change in the position of the magnetic ring.

[0037] After the pile foundation is poured and stabilized, the monitoring platform is activated. It establishes a connection with the data acquisition device through a built-in integrated communication module. Multiple power supply cables extending from the built-in cylinder provide continuous power to each sensor, while the data connection cables transmit the settlement, tilt, stress, and deformation data collected in real time by each sensor to the monitoring platform. The monitoring platform relies on a built-in integrated chip, running a graph neural network and a lightweight multi-head attention temporal network. The former combines the structural characteristics of the target building, such as pile depth, diameter, and concrete grade, to construct a correlation model between the pile foundation and the superstructure. The latter focuses on the temporal variation patterns of multi-dimensional monitoring data, uncovering the inherent correlations between different data, such as the linkage between settlement and stress data over time. Both work together to deeply fuse and dynamically analyze the monitoring data transmitted to the platform, eliminating abnormal fluctuations caused by temporary interference, and ultimately generating initial pile foundation assessment results. This directly reflects the current health status of the pile foundation, such as whether there are abnormal settlement, tilt deviation, stress overload, or excessive deformation.

[0038] Taking the pile foundation monitoring of a 6-story residential building in a certain residential community as an example, during the pouring of the pile foundation, internal cylinders were pre-embedded according to specifications, and each sensor was fixed in its position. After the building was put into use, the monitoring platform received sensor data in real time. The graph neural network combined with the structural parameters of the pile foundation, such as a depth of 12 meters and a diameter of 0.8 meters, and the lightweight multi-head attention time series network analyzed the monitoring data of the past month. It was found that the settlement data remained within a stable range, the tilt angle did not shift, and the stress and deformation data met the design requirements. The generated initial pile foundation assessment result determined that the pile foundation was in good health. As another example, the pile foundation monitoring of a street-facing shop showed a slight and continuous upward trend in recent pile foundation settlement data due to road construction in the surrounding area. The stress data fluctuated slightly. Although it did not exceed the safety threshold, the initial pile foundation assessment result indicated that the pile foundation needed to increase the monitoring frequency and closely monitor the impact of surrounding construction on its health status, providing a basis for subsequent maintenance.

[0039] As an optional embodiment, in step S101, a graph neural network and a lightweight multi-head attention temporal network are used in conjunction with the building structure of the target building to perform data fusion and trend analysis on various monitoring data to obtain initial pile foundation evaluation results, including:

[0040] First, based on the classification and labeling of various monitoring data as dynamic monitoring features, the pre-input building structural parameters are classified and labeled as static structural features. Next, a lightweight multi-head attention temporal network is activated, employing a grouped query attention mechanism to fuse features from the multimodal input matrix. By pre-setting the number of groups, the attention computation task is split, reducing the model parameter scale and computational load. The network also learns the interdependencies between different dynamic monitoring features, determining the contribution weight of each monitoring data point to the pile foundation health status assessment at different time nodes. Combined with the weighted fusion of the source monitoring data contributing to the weights, the fused features of the target building's pile foundation are obtained. Then, a graph neural network is used to construct a structural association graph of the target building, with the pile foundation as the core node and the superstructure load-bearing components and foundation caps as associated nodes. The load transfer coefficient and the connection strength parameters between the pile foundation and associated nodes from the static structural features are loaded. Through the message passing mechanism of the graph neural network, the structural association relationships between corresponding components of each node are transformed into feature correction factors, which correct the fused features output by the lightweight multi-head attention temporal network. Finally, dynamic trend analysis is performed on the corrected fusion characteristics to obtain the initial pile foundation evaluation results, so as to reflect the current overall state of the pile foundation and the tendency of health changes within a future preset period through the initial pile foundation evaluation results.

[0041] Specifically, the above steps begin with feature classification and labeling. Staff retrieve structural parameters from the target building's pre-stored design data on the monitoring platform, including pile depth, diameter, concrete grade, and load transfer coefficient between the pile foundation and the superstructure. These parameters, which do not change over time, are classified and labeled as static structural features. Simultaneously, the pile settlement, inclination, stress, and deformation data collected in real-time by multi-source sensors are organized according to time series and labeled as dynamic monitoring features. Dynamic monitoring features must retain continuous temporal change trajectories to facilitate subsequent analysis of data fluctuation patterns over time, providing a clear feature input basis for subsequent network processing.

[0042] Subsequently, a lightweight multi-head attention temporal network was launched. This network is pre-adapted to the integrated chip built into the monitoring platform. Through a group query attention mechanism, the multimodal input matrix (formed by the correlation mapping between dynamic monitoring features and static structural features) is split into a preset number of task groups according to sensor type or data dimension, significantly reducing the model's computational load and parameter scale, and adapting to the chip's resource limitations. During network operation, it automatically learns the interdependencies between different dynamic monitoring features. For example, it identifies whether stress data changes synchronously when pile foundation settlement data rises. At the same time, it determines the contribution weight of each monitoring data based on the degree of influence of data at different time points on health assessment. If deformation data fluctuates significantly within a certain period, the network will assign higher weights to the deformation data of that period. These weights are then combined to perform weighted fusion of multi-source monitoring data, ultimately obtaining a fused feature that can initially reflect the pile foundation's condition.

[0043] Next, a structural relationship graph of the target building is constructed using a graph neural network. The pile foundations with embedded sensors are used as core nodes, and the beams, columns, and other load-bearing components of the superstructure, as well as the foundation caps, are used as related nodes. The graph is loaded with load transfer coefficients (reflecting the force transfer efficiency between the pile foundations and the superstructure) and connection strength parameters between the pile foundations and related nodes (reflecting the stability of the structural connection) from the static structural features. Through the message passing mechanism of the graph neural network, each node in the graph transmits its own feature parameters to the core node (pile foundations). During this process, the network converts the structural relationships between the corresponding components of each node (such as the constraint effect of the foundation caps on the pile foundations) into feature correction factors. These correction factors are used to adjust the fused features output by the lightweight multi-head attention temporal network. For example, if the connection strength between the pile foundations and the foundation caps is high, the correction factors will appropriately reduce the impact of local stress data fluctuations on the fused features, making the fused features more consistent with the actual structural logic of the pile foundation's stress and avoiding deviations caused by isolated analysis of monitoring data.

[0044] Finally, dynamic trend analysis is performed on the corrected fusion characteristics. The network compares the fusion characteristics with historical data from the same period and the standard data range under normal operating conditions to determine whether the current data is within a reasonable range. It also captures short-term trends in data change (such as whether settlement data shows a continuous upward, stable, or downward trend). Based on these analysis results, an initial pile foundation assessment is generated: if the fusion characteristics remain stable within the standard range for a long period without any abnormal trends, the assessment result will determine that the pile foundation is in good health. If the fusion characteristics fluctuate beyond the standard range or show obvious abnormal trends, the assessment result will indicate the corresponding abnormal dimensions (such as abnormal settlement or excessive stress), providing direction for further monitoring or intervention.

[0045] Taking the pile foundation monitoring of a multi-story residential building as an example, its static structural characteristics include a pile depth of 12 meters, a pile diameter of 0.8 meters, and the load transfer coefficient of the corresponding grade of concrete. Dynamic monitoring characteristics show that settlement, tilt angle, stress, and deformation data have remained stable for three consecutive months. The lightweight multi-head attention temporal network learned that there were no abnormal correlations between the data. The graph neural network, through the correction factor generated by the structural correlation graph, further verified that the data conformed to the stress law of the building structure. Finally, the initial pile foundation assessment result determined that the pile foundation of the residential building was in good health. However, due to the construction of underground pipelines in the vicinity, the settlement data of a commercial building facing the street showed a slow upward trend in its dynamic monitoring characteristics. The lightweight network assigned higher weights to the settlement data during this period. After the graph neural network combined the connection strength parameters between the pile foundation and the upper commercial floor slab to generate a correction factor, the dynamic trend analysis found that the settlement trend exceeded the normal fluctuation range. The initial pile foundation assessment result marked the settlement trend as abnormal and indicated that the monitoring frequency needed to be increased to provide a basis for subsequent tracking of the impact of construction on the pile foundation.

[0046] Further optionally, in one embodiment of step S101, the pre-input building structure parameters are classified and labeled as static structural features based on multiple monitoring data classification and labeling as dynamic monitoring features. This includes: first, preprocessing multiple monitoring data from multiple source sensors, removing noise data and invalid values ​​based on preset data cleaning rules, retrieving the building structure parameters of the target building, including pile foundation depth, pile foundation diameter, concrete grade, and load transfer coefficient between the pile foundation and the superstructure, classifying and labeling the preprocessed monitoring data as dynamic monitoring features, and classifying and labeling the building structure parameters as static structural features. Then, a lightweight multi-head attention temporal network input layer is constructed on the integrated chip of the monitoring platform. The dynamic monitoring features are divided into continuous data segments according to the time series, and associated and mapped with the static structural features to form a multimodal input matrix. The dynamic monitoring features at least include the temporal changes of pile foundation settlement, inclination angle, stress, and deformation data. The static structural features are embedded as prior information into the input matrix.

[0047] Specifically, after multi-source sensors collect real-time data on pile foundation settlement, inclination, stress, and deformation, the raw data is transmitted to the monitoring platform. The platform then filters and processes this data according to pre-defined data cleaning rules. For example, if a sensor experiences a brief abnormal peak due to transient electromagnetic interference, or if blank data is generated due to poor contact, the platform automatically identifies and removes such noisy data and invalid values ​​to prevent them from affecting subsequent analysis results. Simultaneously, the monitoring platform retrieves the target building's structural parameters from pre-stored building archives. These parameters specifically cover pile foundation depth, pile diameter, concrete grade, and the load transfer coefficient between the pile foundation and the superstructure—key information reflecting the inherent structural characteristics of the pile foundation. After data cleaning, the platform categorizes and labels the processed settlement, inclination, stress, and deformation data as dynamic monitoring features. These features exhibit temporal changes over time, reflecting the current stress and state changes of the pile foundation in real time. The retrieved structural parameters are categorized and labeled as static structural features. These features remain relatively stable throughout the building's service life and serve as a fundamental reference for assessing the health status of the pile foundation.

[0048] Next, the construction of the input layer of the lightweight multi-head attention temporal network proceeds. The integrated chip of the monitoring platform (such as the embedded chip adapted in the document) serves as the hardware carrier for network operation, and the network's input layer structure is first built within the chip. For labeled dynamic monitoring features, the system divides them into continuous data segments at fixed time intervals, such as integrating settlement, tilt, and other data within a time period by hour, forming data units with time-series attributes. Then, the platform associates and maps these dynamic data segments with static structural features. For example, it matches stress data for a certain time period with pile diameter and concrete grade, and determines whether the stress value for that time period is within a reasonable range based on the structural characteristics of the pile. This association forms a multimodal input matrix. In this matrix, static structural features are embedded as prior information, providing structural reference for the network's subsequent analysis of dynamic data. This ensures that the network can interpret the state changes of the pile based on its inherent characteristics, rather than relying solely on dynamic data for biased judgments.

[0049] Taking the pile foundation monitoring of a multi-story residential building in a certain residential community as an example, when the building pile foundation sensors were collecting data, a brief abnormal stress peak was generated due to vibrations from temporary construction in the surrounding area. The monitoring platform identified and removed this noise data through preset cleaning rules. At the same time, it retrieved parameters such as the depth, diameter, concrete grade, and load transfer coefficient between the pile foundation and the upper floor slab. The processed settlement, tilt angle, and other time-series data were labeled as dynamic monitoring features, and the structural parameters were labeled as static features. When constructing the network input layer, the chip divided the dynamic features into data segments every two hours, and correlated the deformation data of each time period with the pile foundation diameter and concrete grade to form a multimodal input matrix. The static features serve as prior information to help the network subsequently analyze whether the deformation data conforms to the structural tolerance range of this type of pile foundation. For example, in the monitoring of the pile foundation of a street-facing shop, the sensor generated some invalid data due to a brief circuit failure. After the platform cleaned the data, the valid data was marked as dynamic features. Static parameters such as the load transfer coefficient between the shop's pile foundation and the upper load-bearing beam were retrieved. When constructing the input layer, the dynamic data was segmented by hour and associated with the static parameters to form a matrix, providing a basic input for network analysis of the state changes of the shop's pile foundation under daily operating loads.

[0050] Further optionally, in one embodiment of step S101, dynamic trend analysis is performed on the modified fusion features to obtain the initial pile foundation assessment result, including: First, calculating a multi-dimensional health index through the output layer of a lightweight multi-head attention temporal network, where the value of each dimension of the health index corresponds to the pile foundation health status level. Then, using a lightweight moving average algorithm and a linear regression model, the short-term trend curves of each dimension of the health index are calculated. Finally, the health indices of each dimension and their corresponding trend curves together constitute the initial pile foundation assessment result, reflecting the current overall state of the pile foundation and its potential health changes within a preset period.

[0051] In the above steps, dynamic trend analysis is performed on the corrected fusion features to obtain the initial pile foundation assessment results, starting with the calculation of multi-dimensional health indices. The corrected fusion features integrate the correlation information between preprocessed multi-source monitoring data and building structural parameters. The output layer of the lightweight multi-head attention temporal network first decomposes these fusion features by dimension and matches them with the core correlation elements of each health index. For example, information related to settlement in the fusion features is correlated with the monitoring data of the magnetostrictive displacement sensor and the pile foundation depth and concrete grade. Information related to tilt angle is correlated with the tilt sensor data and the pile foundation diameter and the load transfer coefficient of the superstructure. Information related to stress and deformation is also correlated with structural parameters such as concrete grade, pile foundation cross-sectional area or depth. Subsequently, the integrated chip of the monitoring platform calls the pre-stored threshold calculation model, combines industry pile foundation safety standards with the specific structural characteristics of the target building, determines the three-level threshold range of each dimension of the health index, and then converts the monitoring data of each dimension in the fusion features into values ​​that conform to the threshold range. Through normalization processing, these values ​​are mapped to indices in the range of 0-100, with each index value corresponding to a specific health status level. At the same time, the overall health index of the pile foundation is calculated by weighting the indices of each dimension according to their impact on the overall safety of the pile foundation, so as to ensure that it can reflect both the health status of a single dimension and the overall health level of the pile foundation.

[0052] Next, the calculation of the short-term trend curve is performed. First, based on the monitoring platform's regular data collection cycle, historical data of various health indices over a recent period are selected as analysis samples, forming a dataset with a one-to-one correspondence between time and health index. Then, the integrated chip runs a lightweight moving average algorithm to smooth out accidental fluctuations in the dataset caused by momentary interference. For example, if the health index experiences a brief jump due to electromagnetic interference at a certain moment, this anomaly is eliminated by averaging multiple consecutive sampling points, resulting in a smooth time-series dataset that more closely reflects the actual health changes of the pile foundation. Subsequently, a linear regression model is used to fit the smoothed dataset to generate short-term trend curves for each dimension, with time as the horizontal axis and the health index as the vertical axis. The direction of change is determined by analyzing the curve slope. If the slope is in a stable range, it indicates that the health status of that dimension has no significant fluctuations. If the slope tends to increase, the corresponding health status is improving. If the slope tends to decrease, it suggests that the health status may be deteriorating. Simultaneously, key nodes are marked on the trend curve, such as the time points when the health index changes across levels, the locations where the slope suddenly changes, and a description of the rate of trend change is added, forming a complete short-term trend curve.

[0053] Finally, the multi-dimensional health indices are integrated with the corresponding short-term trend curves to form the initial pile foundation assessment results. The multi-dimensional health indices directly reflect the current health status level of the pile foundation in each dimension and overall, while the short-term trend curves reveal the tendency for health changes within a preset period. The combination of both provides a comprehensive view of the dynamic health of the pile foundation. For example, in the pile foundation monitoring of a multi-story residential building, after processing the corrected fusion characteristics, the health indices for settlement, tilt angle, stress, and deformation were all at healthy levels, and the overall health index also met high safety standards. The short-term trend curves showed that the slopes of each dimension's indices were within a stable range, with no abnormal fluctuations. Therefore, the initial pile foundation assessment results determined that the pile foundation's current health status was good, and there was no significant health risk in the short term. However, in the pile foundation monitoring of a street-front shop, due to underground pipeline construction nearby, the settlement health index calculated by the corrected fusion characteristics was at a sub-healthy level, while other dimension indices remained within the healthy range. The short-term trend curve shows that the slope of the settlement health index is slowly decreasing, and a slight change in slope has been marked in the last two days. Although it has not yet reached the warning level, the initial pile foundation assessment results still clearly indicate that there is a potential tendency for the current settlement dimension of the pile foundation to change. In the future, it is necessary to increase the monitoring frequency of settlement data to track the impact of construction on the health status of the pile foundation.

[0054] Step S102: The initial pile foundation assessment results of target buildings in each geographical area and various monitoring data are uploaded to the cloud data processing center through the monitoring platform.

[0055] In step S102, the monitoring platform first systematically organizes and identifies the data to be uploaded, ensuring a precise association between the data and the target building and geographical area. The monitoring platform first collects the initial pile foundation assessment results for each target building. These initial pile foundation assessment results include multi-dimensional health indices (settlement, tilt angle, stress, and deformation dimensions) and corresponding short-term trend curves, while also summarizing various pre-processed raw monitoring data (time-series data of settlement, tilt angle, stress, and deformation). To facilitate subsequent classification, storage, and regional analysis by the cloud data processing center, the platform adds unique identification information to each set of data, including the geographical area division of the target building (e.g., a street area in a certain city), the building's unique number (e.g., Building 1 and Building 3 in a certain residential area), and the data collection time range (e.g., a certain date and time period), ensuring that the uploaded data can be clearly located to a specific area and building.

[0056] Subsequently, the monitoring platform establishes a stable data link with the cloud data processing center using its built-in communication module and initiates the data upload process. During transmission, the platform activates a data integrity verification mechanism to perform real-time verification of uploaded data blocks. If data transmission is interrupted or verification fails, the system automatically triggers a retransmission mechanism to ensure that the data arrives at the cloud center intact. Simultaneously, the monitoring platform's dual-power supply architecture (municipal power as primary and energy storage power as backup) provides continuous assurance for data upload. Even in the event of a temporary interruption of municipal power supply, the energy storage power supply will seamlessly switch power to avoid data upload interruptions due to power outages. If the power outage duration exceeds a short period, the platform also has a power outage resume function; once power is restored, it will resume uploading unfinished data from the point of interruption, ensuring the continuity of data transmission.

[0057] Taking a residential community in a certain city as an example, the community comprises five multi-story buildings, each equipped with an independent monitoring platform. At fixed times each day, the monitoring platforms of each building compile the initial pile foundation assessment results and corresponding monitoring data. For example, the settlement health index of Building 1 is at a healthy level with a stable short-term trend, while the stress health index of Building 2 shows no abnormalities. This data is labeled with the geographical area identifier "XX City, XX Community," the building numbers "Building 1" and "Building 2," and the time range for that day. Then, a connection is initiated to the cloud data processing center via a 4G communication module. During one upload process, a municipal power outage occurred in the area due to line maintenance. The monitoring platform's energy storage power supply immediately activated, ensuring uninterrupted power supply and unaffected data transmission. On another day, a brief fluctuation in network signal caused a data transmission verification failure for one building. The platform automatically triggered a retransmission, and ultimately, all data was uploaded completely to the cloud center.

[0058] Step S103: The initial pile foundation evaluation results of target buildings in each geographical area are stored through the cloud data processing center, and the correlation trends between target buildings in each geographical area are comprehensively evaluated to obtain the correction parameters corresponding to each geographical area. In this embodiment, the correlation trends include, but are not limited to: the structural correlation between the pile foundations of the target building and surrounding buildings, the correlation between the pile foundations of the target building and surrounding construction activities, and the correlation between the pile foundations of the target building and surrounding geological activities.

[0059] Specifically, in step S103, the cloud data processing center first performs structured storage on the received initial pile foundation assessment results and monitoring data. After monitoring platforms in various geographical areas upload data, the cloud center establishes dedicated databases according to geographical regions. Within each database, data is further archived hierarchically based on the unique identifier of the target building (such as building number and building name), ensuring that data is accurately bound to specific regions and buildings. During storage, the initial pile foundation assessment results (including multi-dimensional health indices and short-term trend curves) are associated with the corresponding original monitoring data (settlement, tilt angle, stress, and deformation time-series data), and metadata such as data upload timestamps and monitoring platform numbers are appended. This facilitates subsequent data tracing and supports batch retrieval and historical comparison of data within the region. For example, building data from different streets in a certain urban area may belong to different sub-databases, and data for each building within the same street may be archived separately. When analyzing a particular building, its historical assessment results and data from other buildings in the same area can be quickly retrieved.

[0060] Subsequently, the cloud data processing center initiated a comprehensive assessment process, focusing on analyzing three types of correlation trends between the target building's pile foundation and its surrounding environment. When assessing the structural correlation between the target building's pile foundation and surrounding buildings, the cloud center retrieves the pile foundation structural parameters (such as pile depth, diameter, and concrete grade) and initial pile foundation assessment results of surrounding buildings within the same geographical area to analyze the linkage between the target building and surrounding buildings in terms of changes in their health status. For example, if the health index of the pile foundations of surrounding high-rise buildings remains stable over a long period, and the settlement and inclination data of the target building's pile foundation also show no abnormal fluctuations, it indicates that the overall stress is stable under structural correlation. If the pile foundations of surrounding buildings show slight stress changes, and the target building's pile foundation also shows a similar trend during the same period, then the potential impact of this structural correlation on the health status needs to be identified.

[0061] When assessing the correlation between the pile foundation of a target building and surrounding construction activities, the cloud center integrates construction activity records (such as construction location, construction type, and construction progress) within a preset period in the geographical area, comparing the spatiotemporal information of the construction activities with the monitoring data and initial assessment results of the target building's pile foundation. For example, during underground pipeline construction at a construction site in a certain area, the cloud center will focus on analyzing the settlement and stress changes of the pile foundations of buildings at different distances from the construction site. If the settlement trend of the target building's pile foundation closer to the construction site is slightly higher than that of the buildings farther away, and the settlement trend returns to stability after construction, it can be determined that the construction activities have a short-term correlation with the target building's pile foundation.

[0062] When assessing the correlation between the pile foundations of a target building and surrounding geological activities, the cloud center will access geological monitoring data for the geographic area (such as ground settlement and ground vibration frequency) and compare it with the initial assessment results of the pile foundations of multiple target buildings in the same area. If the ground data for a certain area shows no significant recent settlement or vibration, and the health index of all building pile foundations in that area is within a stable range, it indicates that geological activities have no impact on the pile foundations. If a certain area experiences slight local ground settlement, and the pile foundations of multiple buildings in that area simultaneously show similar settlement trends, then it can be confirmed that there is a correlation between geological activities and the health status of the pile foundations.

[0063] After analyzing the results of the three types of correlation trends, the cloud data processing center will transform the correlation impacts into quantified correction parameters. For example, if construction activities in a certain geographical area have a short-term, minor impact on the pile foundations of surrounding buildings, a "construction correlation correction item" will be generated. If the geological conditions in the area are stable and the overall pile foundation is stable under structural correlation, "geological correlation correction items" and "structural correlation correction items" will be generated. These correction items will clarify the adjustment weights of various correlation factors on the pile foundation health status assessment results, and will ultimately be integrated into the correction parameters corresponding to the geographical area, providing a basis for subsequently issuing updated assessment results to the monitoring platform.

[0064] Taking a residential area in a certain city as an example, the cloud data processing center receives the initial pile foundation assessment results and monitoring data of 10 residential buildings in the area and archives them into the "XX City XX Area" database according to the building number. When assessing structural correlation, it was found that the pile foundation structural parameters of the 10 buildings were similar and the health index was stable, indicating that the overall stress was balanced under structural correlation. When assessing surrounding construction activities, it was found that road renovation was underway on the east side of the area. After comparing the data, it was found that the pile foundation settlement trend of 3 buildings within 300 meters of the construction site was slightly higher than that of other buildings, but did not exceed the safe range, indicating that the construction had a short-term correlation impact. When assessing geological activities, the geological data of the area showed that the strata were stable and there was no abnormal settlement or vibration. Based on this, the cloud center generated correction parameters for the area, including "construction correlation correction items" (fine-tuning the settlement assessment results of buildings near the construction site) and "structural and geological correlation correction items" (with little impact and low weight). Subsequently, these correction parameters were sent to the area monitoring platform to update the target pile foundation assessment results of each building.

[0065] As an optional embodiment, step S103 involves comprehensively evaluating the correlation trends among target buildings within each geographical region to obtain correction parameters corresponding to each geographical region. This includes: launching a particle swarm optimization algorithm within the cloud data processing center to determine the core variables for comprehensive evaluation. These core variables include structural correlation parameters between the target building's pile foundation and surrounding buildings, correlation parameters between the target building's pile foundation and surrounding construction activities, and correlation parameters between the target building's pile foundation and surrounding geological activities within each geographical region. The particle swarm is initialized by setting its size, initial position, and initial velocity. Each particle's position corresponds to a set of core variable values, representing an evaluation scheme for correlation trends. The particle's velocity corresponds to the adjustment direction and magnitude of the core variable values. A fitness function is constructed, combining historical pile foundation health status change data and current monitoring conditions for target buildings within each geographical region. The fitness function measures the degree of matching between the core variable value combination corresponding to each particle and the actual pile foundation health change trend; a higher matching degree results in a higher fitness value for the particle. The particle swarm is iteratively updated. In each iteration, the fitness value of each particle is calculated, and the historical best position of each particle and the global best position of the entire particle swarm are determined. Based on the historical best position and the global best position, the velocity and position of each particle are adjusted to optimize the combination of core variable values. The iteration is stopped if a preset convergence condition is met. If the number of iterations reaches a preset number or the fitness value of the global best position of the particle swarm remains stable for multiple consecutive iterations, the iteration is stopped. The combination of core variable values ​​corresponding to the global best position of the particle swarm is used as the optimal evaluation result of the correlation trend between target buildings in each geographical region. Finally, the optimal evaluation result is converted into correction parameters corresponding to each geographical region. The correction parameters for each geographical region include correction terms for structural correlation, correction terms for correlation with surrounding construction activities, and correction terms for correlation with surrounding geological activities. Each correction term corresponds to the influence weight of one of the correlation trends on the pile foundation health status evaluation result.

[0066] Specifically, in step S103, the process of comprehensively evaluating correlation trends to obtain correction parameters begins by initiating the particle swarm optimization algorithm and identifying core variables at the cloud data processing center. The cloud data processing center first retrieves relevant data about target buildings stored in various geographical areas. This data includes initial pile foundation assessment results uploaded by the monitoring platform (including multi-dimensional health indices and short-term trend curves), raw monitoring data from multi-source sensors (settlement, tilt angle, stress, and deformation time-series data), and building structural archives, construction activity records, and geological monitoring reports within the area. Based on this data, the core variables are identified into three categories: First, structural correlation parameters between the target building's pile foundation and surrounding buildings, covering indicators directly related to the structure, such as building spacing, foundation stiffness matching, and load transfer coefficient. Second, correlation parameters between the target building's pile foundation and surrounding construction activities, including the distance between the construction location and the pile foundation, the magnitude of the construction load, construction duration, and construction stage. Third, correlation parameters between the target building's pile foundation and surrounding geological activities, including ground settlement, ground vibration frequency, and vibration amplitude. The determination of these core variables is closely based on the pile foundation monitoring dimensions and regional correlation factors mentioned in the document, ensuring that the algorithm evaluation direction is consistent with the actual monitoring needs.

[0067] Next, the particle swarm initialization phase begins. The cloud data processing center sets the particle swarm size based on the historical data characteristics of the geographical area. For example, for an area containing 10 target buildings, a reasonable number of particles is set to cover a sufficient number of evaluation schemes. Each particle's position is assigned an initial combination of core variable values. These values ​​are not randomly set but are determined based on the area's past correlation trend data (such as historical construction impact records on pile foundations and historical geological stability) and current preliminary monitoring results (such as no obvious geological correlation anomalies in the initial pile foundation evaluation). For example, if the area's historical geological data is stable, the initial values ​​of the geological correlation parameters will be set in a low-impact range. The particle velocity is set according to the possible adjustment range of the core variables. For example, the adjustment range of construction load correlation parameters will refer to the fluctuation range of common construction load impacts on pile foundations in the industry, ensuring that the velocity can support effective variable optimization without causing convergence difficulties due to excessive amplitude.

[0068] Next, a fitness function is constructed. The core logic of this function is to substitute the core variable values ​​of each particle into a correlation trend assessment and then compare it with the actual changes in pile foundation health. The cloud data processing center will call upon historical pile foundation health status change data of the target building within the geographical area (such as multi-dimensional health index change curves over the past year, short-term trend curves for different time periods) and current monitoring conditions (such as whether there is ongoing construction in the surrounding area, and current geological monitoring data) to calculate the degree of matching between the correlation trend of the particle assessment and the actual health changes. For example, if the core variable value combination of a particle predicts that "surrounding construction will cause a slight decrease in the pile foundation settlement health index," and the actual monitoring shows that the pile foundation settlement index does indeed show a corresponding trend during the construction period in that area, and the decrease is close to the prediction, then the fitness value of that particle is high. If another particle predicts that "geological activity will cause stress index fluctuations," but the actual geology is stable and the stress index is not abnormal, then the fitness value of that particle is low. This function construction method fully combines the full life-cycle monitoring data of the pile foundation with the regional conditions, ensuring the accuracy of the assessment.

[0069] The particle swarm optimization then enters the iterative update phase. The cloud data processing center initiates each iteration according to a preset cycle. First, it calculates the fitness value of each particle. Then, it records the position corresponding to the best fitness of each particle during its historical iterations (individual optimal position), and simultaneously determines the position corresponding to the current best fitness of the entire particle swarm (global optimal position). Subsequently, based on the individual optimal and global optimal positions, combined with preset inertia weights, cognitive coefficients, and social coefficients, the velocity and position of each particle are adjusted. For example, if a particle's individual optimal position shows "construction-related parameter values ​​are slightly low," its velocity will be adjusted to move towards increasing the construction-related parameter values, and its position will be updated accordingly to optimize the accuracy of the assessment of the construction impact. During the iteration process, the cloud data processing center continuously processes real-time uploaded data from multiple target buildings within the geographical area (such as monitoring data updated periodically by a monitoring platform) to ensure that fitness calculations are always based on the latest working conditions, avoiding assessment bias due to data lag.

[0070] When the iteration meets the preset convergence condition (such as the number of iterations reaching the preset upper limit, or the fitness value of the global optimal position remaining stable for 5 consecutive rounds without significant fluctuations), the iteration stops. At this point, the combination of core variable values ​​corresponding to the global optimal position is the optimal evaluation result of the correlation trend within that geographical area. Finally, the cloud data processing center transforms this optimal result into correction parameters. For the core variable values ​​related to structural correlation, a "structural correlation correction term" is generated. If the value indicates that the structural correlation has a small impact on the health of the pile foundation, the weight of this correction term is low. For the core variable values ​​related to surrounding construction activities, a "construction correlation correction term" is generated. If the value indicates that the current construction has a significant impact on the pile foundation, the weight of this correction term is high. For the core variable values ​​related to surrounding geological activities, a "geological correlation correction term" is generated. If the value indicates that the geology is stable, the weight of this correction term is low. These correction terms together constitute the correction parameters corresponding to this geographical area, providing a basis for subsequently updating the target pile foundation evaluation results on the monitoring platform.

[0071] Taking a mixed-use residential and commercial area in a certain city as an example, this area includes 8 residential buildings and 2 commercial buildings. There has been recent underground pipeline construction in the surrounding area, and geological monitoring shows that the strata in this area have been stable for a long time. After the cloud data processing center launched the particle swarm optimization algorithm, the core variables were determined to be "structural correlation (building spacing, foundation stiffness)," "construction correlation (construction distance, construction load)," and "geological correlation (soil settlement, vibration)." When initializing the particle swarm, based on the record of no past geological anomalies in this area, the initial values ​​of the geological correlation parameters were set low. Based on the moderate pipeline construction load in the construction records, the initial values ​​of the construction correlation parameters were set in the moderate influence range. When constructing the fitness function, the particle evaluation results were compared with the changes in pile foundation stress during historical construction in this area (such as the stress index change trend during another construction period 3 years ago) and the current settlement data. After multiple iterations, the global optimal position showed that "construction correlation parameters have a significant impact, structural correlation parameters have a moderate impact, and geological correlation parameters have a minimal impact." Based on this, among the converted correction parameters, the "construction correlation correction item" had the highest weight, and the "geological correlation correction item" had the lowest weight. The corrected parameters will then be sent to the area monitoring platform. When the platform updates the target pile foundation assessment results, it will take into account the impact of surrounding construction on the health status of the pile foundation, so that the assessment results are more in line with the actual working conditions.

[0072] Optionally, in the aforementioned step of constructing a fitness function that combines historical pile foundation health status change data of target buildings within various geographical areas with current monitoring conditions, a dynamic multimodal data fusion evaluation mechanism is employed. This involves simultaneously retrieving two core data sources from a cloud data processing center. The first data source is a sequence of historical pile foundation health status change data for target buildings within each geographical area. This sequence includes a time-series database recording pile settlement, inclination changes, stress distribution, and deformation characteristics over a long period. The second data source is real-time monitoring data collected by the monitoring platform. This real-time data includes real-time sensor readings, instantaneous environmental load values, and dynamic information streams of surrounding construction activities. Furthermore, a fitness function is established based on both the first and second data sources. The fitness function maps the current core variable values ​​represented by particle positions to a simulated trajectory of pile foundation health status. This simulated trajectory is then matched against a historical database using multi-level deep temporal pattern matching. During this matching process, a dynamic time warping algorithm aligns the temporal patterns, calculating the morphological similarity, key feature point offsets, and gradient consistency between the simulated trajectory and real data segments under similar environmental loads and construction disturbances in the historical database. Finally, a dynamic weight allocation strategy is introduced for weighted comprehensive scoring, with the weighted comprehensive score of the multi-level matching results serving as the output value of the fitness function.

[0073] Understandably, in constructing the fitness function using a dynamic multimodal data fusion evaluation mechanism, the cloud data processing center first synchronously retrieves core data sources from two dimensions. The first data source is the historical pile foundation health status change data sequence of target buildings in various geographical areas. This data is accumulated over long-term monitoring and covers time-series records of pile foundation settlement, inclination angle changes, stress distribution, and deformation characteristics. Its monitoring dimensions are consistent with the pile foundation settlement, inclination angle, stress, and deformation monitored by the data acquisition device, ensuring that historical data can reflect key state changes throughout the entire life cycle of the pile foundation. The second data source is the monitoring data collected in real time by the monitoring platform, including real-time sensor readings (from displacement, inclination angle, stress, and deformation sensors), instantaneous values ​​of environmental loads, and dynamic information streams of surrounding construction activities. The monitoring platform receives real-time data through electrical connections with the sensors and uploads the data to the cloud center via the communication module, providing real-time information support for the fitness function based on the current working conditions.

[0074] Based on these two data sources, the cloud data processing center began establishing a fitness function. This function transforms the core variable values ​​represented by particle positions (structural correlation parameters, construction correlation parameters, and geological correlation parameters) into a simulated trajectory of pile foundation health status. This trajectory simulates the trends of pile foundation settlement, inclination, stress, and deformation over time under the current correlation conditions corresponding to the core variables, aligning with the monitoring logic emphasized in the document, which requires comprehensive consideration of the impact of multiple correlated factors on the pile foundation within the region. Subsequently, a dynamic time warping algorithm is used to align the simulated trajectory with time-series data in the historical database. Since historical data and current simulated data may differ in time scale or starting point, this algorithm effectively eliminates these differences, achieving accurate matching of time-series patterns. During the matching process, similarity is calculated from three dimensions: First, the morphological similarity between the simulated trajectory and real data segments under similar environmental loads (such as similar building loads) and construction disturbance conditions (such as similar construction distances and loads) in the historical database, determining whether the overall trend is consistent. Second, the offset of key feature points, such as the deviation between the peak settlement time in the simulated trajectory and the peak time in the historical real data. Third, consistency of change gradient: compare the rate of change of the simulated trajectory and the real data within the same time period to ensure that the matching results can reflect the degree of fit between the combination of core variable values ​​and the actual pile foundation state changes.

[0075] Finally, a dynamic weighting strategy is introduced to perform a weighted comprehensive score on the above multi-level matching results. The weighting is adjusted according to the characteristics of the current monitoring conditions. For example, if there is significant surrounding construction activity in the current area, the weight of the matching dimensions corresponding to the construction-related parameters (such as the matching results of real data under historical construction disturbances) will be appropriately increased. If the current conditions focus more on the impact of geological stability, the weight of the matching dimensions related to geological correlation will be increased. Through this dynamic adjustment, the weighted comprehensive score can more accurately reflect the rationality of the combination of core variable values ​​for particles. Ultimately, this score is the output value of the fitness function. The higher the output value, the higher the degree of matching between the core variable value combination corresponding to the particle and the actual trend of pile foundation health changes, and the more it meets the actual needs of pile foundation monitoring in the region.

[0076] Optionally, based on the real-time confidence level of each data source and the statistical significance of historical data in the current monitoring conditions, the weight ratio of each matching dimension is adaptively adjusted to quantitatively evaluate the degree of agreement between the correlation trend scheme represented by the particle and the actual pile foundation behavior pattern. It is worth noting that a higher degree of agreement results in a higher fitness function output value, indicating that the core variable combination corresponding to the particle can more accurately characterize the real interaction relationship among pile foundation groups within the geographical area.

[0077] Specifically, when implementing the steps of adjusting the matching dimension weights and quantifying the degree of consistency based on the real-time confidence level of the current monitoring data source and the statistical significance of historical data, the reliability and effectiveness of the two types of data sources must first be evaluated. For real-time data sources, the cloud data processing center will rely on the real-time information transmitted by the monitoring platform to determine the confidence level. The monitoring platform uses its built-in integrated chip to perform preliminary verification of the real-time sensor readings. If the sensor data is continuous and stable without abnormal jumps, and the instantaneous values ​​of environmental loads and the status of surrounding construction activities are recorded completely (such as clear records of construction location and load size), then the real-time data is deemed to have high confidence. Conversely, if the sensor data fluctuates frequently, has missing data, or the construction status records are unclear, then the confidence rating of the real-time data is reduced. This process fully utilizes the data processing and transmission capabilities of the monitoring platform to ensure the objectivity of the real-time data assessment. For historical data sources, their statistical significance will be analyzed, focusing on the time span, data volume, and similarity to the current working conditions of historical pile foundation health status change data. If there is a large amount of long-term data in the historical database that matches the current working conditions (such as similar surrounding construction and similar geological conditions), and the data is evenly distributed and without obvious deviation, then the historical data is judged to have high statistical significance, which meets the reliance on historical data for the whole life monitoring of pile foundations.

[0078] Subsequently, the weighting of each matching dimension is adaptively adjusted based on the above evaluation results. If the confidence level of real-time data is high, the weights of matching dimensions such as feature point offset and gradient consistency between the simulated trajectory and real-time monitoring data are increased, making the fitness function more focused on the real-time matching effect under the current working conditions. If the statistical significance of historical data is high, the weight of the morphological similarity between the simulated trajectory and historical similar data segments is increased, making full use of the reference value of historical experience data. If the reliability of a certain type of data source is insufficient (such as low confidence level of real-time data), the weight of its corresponding matching dimension is reduced accordingly to avoid interference from low-quality data on the evaluation results. For example, when the real-time records of surrounding construction activities are detailed and the sensors are working properly, the weights of matching dimensions related to construction-related parameters are increased; when there are multiple pile foundation monitoring data under similar geological conditions in the past, the influence of historical data matching dimensions is amplified to ensure that the weight allocation is always dynamically adapted to the quality of the data source.

[0079] Finally, the degree of fit is quantitatively assessed through a weighted comprehensive score. The scores of each matching dimension are summed according to their adjusted weights to obtain the output value of the fitness function. A high output value indicates that the trajectory of pile foundation health changes simulated by the core variable combination represented by the particles (such as structural correlation parameters and construction correlation parameters) has a high degree of matching with real-time and historical data. It can accurately characterize the real interaction between pile foundations and surrounding buildings, construction activities, and the geological environment within the geographical area. For example, it can accurately reflect the degree of influence of construction loads on the settlement of surrounding pile foundations, or the effect of structural correlation between buildings on the stress distribution of pile foundations.

[0080] Thus, this adaptive weight adjustment mechanism solves the evaluation bias problem caused by data source quality fluctuations in the fixed weight mode, making the fitness function judgment more in line with actual working conditions and effectively improving the accuracy of core variable combination selection. The optimal particle selected based on this mechanism has corresponding correction parameters that more realistically reflect the correlation trend of pile foundations within the region, thereby making the pile foundation health assessment results updated by the subsequent monitoring platform more reliable. This provides stronger data support for the health monitoring, risk warning, and demolition / reconstruction decisions throughout the entire life cycle of building pile foundations, aligning with the core objective of intelligent pile foundation monitoring instruments to provide data support for demolition / reconstruction.

[0081] In another optional embodiment, the basic data required to construct the fitness function is retrieved from the cloud data processing center. This data includes historical pile foundation health status change data such as the initial pile foundation assessment results of the target building within the geographical area, historical multi-dimensional health indices and corresponding short-term trend curves, and historical anomaly monitoring indicator processing records. Current monitoring conditions include the current initial pile foundation assessment results of the target building within the geographical area, real-time records of current surrounding construction activities, real-time data of current geological activities, and structural correlation parameters between the current target building and surrounding buildings. The core evaluation dimensions of the fitness function are defined as follows: historical trend matching dimension, current condition adaptation dimension, and anomaly tolerance dimension. The historical trend matching dimension measures the consistency between the historical pile foundation health status change trend and the health change trend predicted by the core variable value combination. The current condition adaptation dimension measures the degree of fit between the core variable value combination and the current surrounding construction, geological, and structural conditions. The anomaly tolerance dimension considers the tolerance of the core variable value combination to occasional anomaly monitoring indicators in historical and current data. The specific calculation logic for each assessment dimension is executed. The historical trend matching dimension determines the degree of consistency by comparing the slope and rate of change of historical short-term trend curves with the trend characteristics derived from the combination of core variable values. The current working condition adaptation dimension comprehensively determines the degree of adaptation by analyzing the matching of construction loads and construction locations in current surrounding construction activity records with construction-related parameters in core variables; the matching of ground settlement and vibration frequency in current geological activity data with geological-related parameters in core variables; and the matching of current structural correlation parameters with structural-related parameters in core variables. The anomaly tolerance dimension re-verifies the stability of the core variable value combination for the health status assessment results after removing occasional outliers from historical and current data, determining the fault tolerance capability. The weighting of each assessment dimension is set. Considering the core requirements of pile foundation health monitoring, the historical trend matching dimension has a higher weight than the current working condition adaptation dimension, and the current working condition adaptation dimension has a higher weight than the anomaly tolerance dimension. The sum of the weights of all dimensions is the overall assessment weight. By integrating the judgment results of various evaluation dimensions, the fitness function outputs a single fitness value by comprehensively considering the degree of matching with historical trends, the degree of adaptation to current working conditions, and the ability to tolerate anomalies. The more consistent the historical trend matching, the better the adaptation to current working conditions, and the stronger the ability to tolerate anomalies, the higher the corresponding fitness value, and vice versa. This achieves a quantitative measurement of the degree of matching between the combination of core variable values ​​and the actual trend of changes in the health of the pile foundation.

[0082] In this way, the fitness function can comprehensively consider historical experience, current realities, and the ability to respond to anomalies, effectively avoiding the one-sidedness of single-dimensional evaluation. This allows the output fitness value to accurately quantify the rationality of the combination of core variable values. Based on the optimal combination of core variables selected by this function, the subsequent transformation and correction parameters can better reflect the actual interaction relationships of pile foundations within the geographical area. This provides a reliable basis for the monitoring platform to update the pile foundation health status assessment results, ultimately helping to achieve accurate monitoring of the entire life cycle of pile foundations. It also provides strong data support for decisions such as building demolition, reconstruction, and renovation, aligning with the core application goals of intelligent pile foundation monitoring instruments.

[0083] Step S104: The correction parameters are sent to the monitoring platform, which then dynamically updates the pile foundation health status of the target building in each geographical area based on the correction parameters, thereby obtaining the target pile foundation evaluation result of the target building.

[0084] As an optional embodiment, after obtaining the target pile foundation evaluation result of the target building in step S104, if the target pile foundation evaluation result of the target building contains abnormal monitoring indicators exceeding a preset threshold, the abnormal monitoring indicators are input into the early warning model to analyze the safety risks existing in the pile foundation of the target building. Then, based on the safety risks, a corresponding alarm strategy is automatically triggered, wherein the alarm strategy includes audible and visual alarms and early warning information push notifications.

[0085] Optionally, in the above steps, abnormal monitoring indicators are input into the early warning model to analyze the safety risks of the pile foundation in the target building. Specifically, an early warning model constructed using a spatiotemporal graph neural network is used to build a correlation map between the pile foundation of the target building and its surrounding environment. The causal correlation between the abnormal monitoring indicators and the surrounding environment is extracted from this correlation map. The surrounding environment includes surrounding buildings, construction areas, and geological layers. Then, based on the causal correlation of the abnormal monitoring indicators, a multi-dimensional time-series correlation analysis is performed to obtain the time-series correlation characteristics of the abnormal monitoring indicators. Next, combined with a preset pile foundation safety risk assessment dimension, the type of safety risk present in the pile foundation is predicted based on the time-series correlation characteristics. Then, based on the impact range and severity of the safety risk type, and referring to a preset risk level classification standard, the safety risk level of the pile foundation in the target building is determined. Finally, the safety risk type, safety risk level, and causal correlation factors are used as the safety risks present in the pile foundation of the target building.

[0086] Next, the monitoring platform will match the alarm intensity and transmission method according to the safety risk type (such as abnormal settlement aggravation, stress overload, tilt angle deviation, etc.) and risk level (such as general warning, important warning, emergency warning) output by the early warning model, ensuring that the alarm strategy accurately matches the actual risk situation. For audible and visual alarms, the monitoring platform will directly activate the built-in audible and visual alarm module. This module is usually installed in a location easily noticed by on-site personnel, such as the building equipment room, basement, or property duty room where the monitoring platform is located. After activation, it will emit a recognizable sound signal (such as intermittent buzzing or continuous warning tone), accompanied by high-frequency flashing warning lights (such as red or yellow warning lights). The intensity of the sound and light will be adjusted according to the risk level. In the case of an emergency risk, the sound will be louder and the flashing frequency of the lights will be higher, ensuring that on-site maintenance personnel, property management personnel, or construction supervisors can detect the abnormality immediately, quickly go to the monitoring platform to view detailed risk information, or go directly to the area where the pile foundation is located for on-site verification to prevent the risk from developing further.

[0087] For early warning information push, the monitoring platform, relying on its built-in 4G / 5G communication module, first structures and organizes the safety risk information. This organization includes the target building's specific geographical location (e.g., a building in a specific community on a specific street), the specific pile foundation number (e.g., the third pile foundation on the east side of Building 1), the specific type of abnormal monitoring indicators (e.g., excessive settlement), the type and level of safety risk, the causal factors leading to the risk (e.g., disturbance from surrounding construction or slight settlement of the geological layer), and preliminary suggested countermeasures (e.g., suspending surrounding construction, increasing monitoring frequency, or contacting a professional agency for testing). After organization, the monitoring platform automatically pushes this information to pre-set management personnel terminals, including the mobile app of engineers responsible for building maintenance in the area and the computer management system of property management personnel. If the risk level is high, it will also be simultaneously pushed to the information platform of the local building safety supervision department. At the same time, all pushed early warning information is simultaneously stored in the cloud data processing center, forming a complete alarm record, facilitating subsequent tracing of the risk handling process and its effectiveness.

[0088] This dual alarm strategy, combining audible and visual alarms with early warning information push notifications, effectively solves the problems of single alarm methods, untimely information transmission, or limited coverage in traditional pile foundation monitoring. Audible and visual alarms enable rapid on-site response, especially for emergency risks, prompting on-site personnel to intervene immediately and prevent risks from escalating into safety accidents. Early warning information push notifications overcome spatial limitations, allowing off-site managers, engineers, and even regulatory departments to obtain risk information in real time, facilitating the coordination of multiple resources (such as dispatching professional testing equipment and arranging maintenance teams) and preventing delays in risk handling due to information lag. Simultaneously, the alarm strategy is precisely matched to the level and type of safety risk, avoiding both excessive alarms for minor risks leading to resource waste and insufficient alarms for severe risks leading to inappropriate handling. For example, in general early warning situations, the audible and visual alarm intensity is moderate, and the push notification range is limited to the maintenance team. In emergency situations, high-intensity audible and visual alarms are combined with information push notifications from multiple departments. Ultimately, this strategy provides a reliable guarantee for the rapid identification, timely response, and effective handling of pile foundation safety risks, further improving the health monitoring system for the entire life cycle of pile foundations.

[0089] Optionally, before extracting the causal correlation between the abnormal monitoring indicators and the surrounding environment from the correlation map in the above steps, the abnormal monitoring indicators can be preprocessed. Noise and invalid values ​​in the abnormal monitoring indicators are removed based on preset data cleaning rules. Historical pile foundation monitoring data of the target building stored on the monitoring platform is retrieved, and a comparison relationship is established between the abnormal monitoring indicators and historical contemporaneous data and historical extreme value data in the historical pile foundation monitoring data to determine the deviation magnitude and rate of change of the abnormal monitoring indicators. Furthermore, the historical correlation trends within the geographical area where the target building is located are retrieved from the cloud data processing center. These include structural correlation parameters between the target building's pile foundation and surrounding buildings, records of surrounding construction activities within a preset period, and geological activity data. The historical correlation trends are used as supplementary input features for the early warning model. The surrounding construction activity records include construction location, construction load, and construction duration. The geological activity data includes ground settlement and vibration frequency.

[0090] Specifically, the first step is to preprocess the abnormal monitoring indicators. The monitoring platform uses preset data cleaning rules to filter the initially identified abnormal monitoring indicators. For example, when sensors experience momentary jumps in readings due to external electromagnetic interference, or when blank data appears due to temporary poor contact, the system automatically identifies and removes this type of noisy data and invalid values, ensuring that the remaining abnormal data effectively reflects the true changes in the pile foundation's condition. Subsequently, the monitoring platform retrieves historical pile foundation monitoring data for the target building from its locally stored database, focusing on extracting historical data from the same period last year (e.g., monitoring data from the same season and month last year) and historical extreme value data (e.g., the maximum reasonable fluctuation values ​​of settlement, stress, and other indicators within past monitoring cycles). By comparing the current abnormal monitoring indicators with these historical data, the numerical differences are calculated to determine the deviation range of the abnormal indicators. Simultaneously, considering the data collection time interval, the changes in the abnormal indicators per unit time are analyzed to clarify their rate of change. For example, if the comparison reveals that the current settlement data exceeds a certain range compared to historical data from the same period, and the settlement rate has significantly accelerated recently, the severity of the anomaly can be preliminarily assessed.

[0091] After preprocessing, the monitoring platform will access historical correlation trend data within the geographical area where the target building is located via a communication link with the cloud data processing center. This data specifically includes structural correlation parameters between the target building's pile foundation and surrounding buildings (such as the matching relationship between the pile foundation and the foundation stiffness of surrounding buildings recorded in past monitoring, and the influence of building spacing on the pile foundation's stress), records of surrounding construction activities in the geographical area within a preset period (covering information such as the distance between the construction location of different construction projects and the target pile foundation, the magnitude of loads during construction, and the duration of construction), and historical geological activity data for the area (such as the average rate of stratum settlement over a past period, and the frequency range and amplitude of stratum vibrations). The monitoring platform will organize this historical correlation trend data into structured feature information, serving as supplementary input features for the early warning model. This data, along with the preprocessed anomaly monitoring indicators, will participate in subsequent analysis, enabling the early warning model to make risk assessments not only based on current anomaly data but also by considering the historical correlations within the region.

[0092] Therefore, the preprocessing of anomaly monitoring indicators effectively reduces the interference of noisy data and invalid values ​​on subsequent analysis, ensuring the reliability of the data entering the early warning model. By comparing with historical data, the deviation magnitude and rate of change of anomaly indicators can be quantified, avoiding misjudgment of risk due to a single data anomaly, and providing a more accurate basis for subsequent risk level determination. Furthermore, using historical correlation trends from the cloud data processing center as supplementary input features allows the early warning model to overcome the technical problem of relying solely on current data. It can combine past structural correlations, construction impacts, and geological change patterns in the region to more comprehensively analyze the background and causes of anomaly monitoring indicators. For example, if historical data shows that similar construction activities in the area have caused short-term settlement anomalies in pile foundations, the early warning model can more accurately determine whether the current anomaly is related to surrounding construction, thereby improving the accuracy of subsequent causal correlation extraction and providing more comprehensive and scientific support for the final safety risk analysis, reducing risk misjudgments or omissions caused by incomplete information.

[0093] In this embodiment, during the initial assessment phase, the monitoring platform leverages the data reception and processing capabilities of multi-source sensors, combining graph neural networks and lightweight multi-head attention time-series networks to fuse multi-dimensional data such as pile foundation settlement, inclination angle, stress, and deformation, and embeds building structural features. This avoids the limitations of single-data monitoring and enables real-time data fusion and dynamic trend analysis on resource-constrained platforms, outputting initial assessment results reflecting the current state and short-term changes of the pile foundation, laying the foundation for subsequent monitoring. Next, the monitoring platform uploads monitoring data and initial results from multiple buildings within the area to the cloud data processing center. The cloud data processing center improves single-building monitoring by comprehensively considering the correlation between the pile foundation and surrounding building structures, construction and geological connections, assessing the health correlation trend of the pile foundation within the area, and generating corrected parameters that fit the actual working conditions, avoiding the bias of traditional monitoring that ignores the surrounding environment. Finally, the cloud data processing center issues the corrected parameters, and the monitoring platform dynamically updates the target assessment results, adapting the results to environmental changes and long-term pile foundation conditions. This overcomes the limitations of static assessments at fixed times, covers the entire life cycle monitoring needs of the pile foundation, provides accurate data support for building maintenance, renovation, and demolition, and improves structural safety and reliability.

[0094] After introducing the methods of exemplary embodiments of this application, the following references are made. Figure 2This application describes an exemplary embodiment of an early warning system for pile foundation health status analysis. The system includes: a monitoring platform with a built-in integrated communication module for real-time reception and processing of various monitoring data from multi-source sensors. The various monitoring data include at least one of the following: pile foundation settlement, inclination angle, stress, and deformation data. The system uses a graph neural network and a lightweight multi-head attention temporal network combined with the building structure of the target building to perform data fusion and dynamic trend analysis on the various monitoring data to obtain an initial pile foundation assessment result. This initial pile foundation assessment result represents the pile foundation health status of the target building. The monitoring platform also uploads the initial pile foundation assessment results and various monitoring data of the target building in each geographical area to cloud data. The system comprises a cloud data processing center, which stores the initial pile foundation assessment results of target buildings within each geographical region and comprehensively evaluates the correlation trends among target buildings in each geographical region to obtain the corresponding correction parameters for each geographical region. The correlation trends include: the structural correlation between the pile foundations of the target building and surrounding buildings, the correlation between the pile foundations of the target building and surrounding construction activities, and the correlation between the pile foundations of the target building and surrounding geological activities. The cloud data processing center also distributes the correction parameters to a monitoring platform. The monitoring platform dynamically updates the pile foundation health status of target buildings in each geographical region based on the correction parameters, thereby obtaining the target pile foundation assessment results for the target buildings. The above system can implement the steps described in the above method implementation, and the specific implementation methods of each step will not be repeated here.

[0095] After introducing the methods and systems of the exemplary embodiments of this application, a terminal device of the exemplary embodiments of this application will be described next. The terminal device can implement the steps described in the above method embodiments, and the specific implementation of each step will not be repeated here.

[0096] After introducing the methods, systems, and terminal devices of exemplary embodiments of this application, the computer-readable storage medium of exemplary embodiments of this application will now be described, such as an optical disc 30 storing a computer program (i.e., a program product) thereon. When the computer program is run by a processor, it implements the steps described in the above-described method embodiments. The specific implementation methods of each step will not be repeated here.

[0097] It should be noted that the above-described embodiments are merely specific implementations of this application, used to illustrate the technical solutions of this application, and not to limit them. The protection scope of this application is not limited thereto. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some of the technical features, within the scope of the technology disclosed in this application. Therefore, the protection scope of this application should be determined by the scope of the claims.

Claims

1. An early warning method for pile foundation health status analysis, characterized in that, An early warning system for analyzing the health status of pile foundations, the system comprising multi-source sensors embedded in the pile foundation, a monitoring platform, and a cloud data processing center, the method comprising: The monitoring platform integrates a communication module to receive and process various monitoring data from multiple sources of sensors in real time. These monitoring data include at least one of the following: settlement, inclination, stress, and deformation data of the pile foundation. By combining graph neural networks and lightweight multi-head attention time-series networks with the building structure of the target building, the platform performs data fusion and dynamic trend analysis on the various monitoring data to obtain the initial pile foundation assessment results. The initial pile foundation assessment results are used to represent the health status of the pile foundation of the target building. The initial pile foundation assessment results of target buildings in each geographical area, as well as various monitoring data, are uploaded to the cloud data processing center through the monitoring platform. The cloud data processing center stores the initial pile foundation assessment results of target buildings in each geographical region and comprehensively evaluates the correlation trends between target buildings in each geographical region to obtain the corresponding correction parameters for each geographical region. The correlation trends include: the structural correlation between the pile foundation of the target building and the surrounding buildings, the correlation between the pile foundation of the target building and the surrounding construction activities, and the correlation between the pile foundation of the target building and the surrounding geological activities. The correction parameters are sent to the monitoring platform, which then dynamically updates the pile foundation health status of the target building in each geographical area based on the correction parameters, thereby obtaining the target pile foundation assessment result of the target building. The method involves combining graph neural networks and lightweight multi-head attention temporal networks with the building structure of the target building to perform data fusion and trend analysis on various monitoring data, thereby obtaining initial pile foundation evaluation results, including: Based on the classification and labeling of multiple monitoring data as dynamic monitoring features, the pre-input building structure parameters are classified and labeled as static structure features. A lightweight multi-head attention temporal network is launched, and a group query attention mechanism is used to fuse features of the multimodal input matrix. The attention calculation task is split by pre-setting the number of groups to reduce the model parameter scale and computational load. The interdependencies between different dynamic monitoring features are learned, the contribution weight of each monitoring data to the pile foundation health status assessment at different time nodes is determined, and the weighted fusion of the monitoring data from the contribution weight source is combined to obtain the fused features of the target building pile foundation. A structural relationship graph of the target building is constructed using a graph neural network. The pile foundation is the core node, and the superstructure load-bearing components and foundation cap are the related nodes. The load transfer coefficient and the connection strength parameters between the pile foundation and the related nodes in the static structural features are loaded. Through the message passing mechanism of the graph neural network, the structural relationship between the corresponding components of each node is transformed into a feature correction factor to correct the fusion features output by the lightweight multi-head attention temporal network. Dynamic trend analysis is performed on the corrected fusion characteristics to obtain the initial pile foundation evaluation results, so as to reflect the current overall state of the pile foundation and the tendency of health changes within a future preset period through the initial pile foundation evaluation results; The classification and labeling of various monitoring data into dynamic monitoring features, and the classification and labeling of pre-input building structural parameters into static structural features, include: Multiple monitoring data from multi-source sensors are preprocessed, and noise data and invalid values ​​are removed based on preset data cleaning rules. The building structure parameters of the target building are retrieved, including pile foundation depth, pile foundation diameter, concrete grade, and load transfer coefficient between the pile foundation and the superstructure. The preprocessed monitoring data are classified and labeled as dynamic monitoring features, and the building structure parameters are classified and labeled as static structural features. A lightweight multi-head attention temporal network input layer is constructed on the integrated chip of the monitoring platform. Dynamic monitoring features are divided into continuous data segments according to time series and associated with static structural features to form a multimodal input matrix. The dynamic monitoring features include at least the temporal changes of settlement, inclination angle, stress and deformation data of pile foundation. The static structural features are embedded into the input matrix as prior information.

2. The early warning method for pile foundation health status analysis according to claim 1, characterized in that, After obtaining the target pile foundation assessment results for the target building, the following is also included: If the target pile foundation assessment results of the target building contain abnormal monitoring indicators that exceed the preset threshold, the abnormal monitoring indicators will be input into the early warning model to analyze the safety risks of the pile foundation in the target building. The corresponding alarm strategy is automatically triggered based on the security risk, and the alarm strategy includes audible and visual alarms and early warning information push.

3. The early warning method for pile foundation health status analysis according to claim 2, characterized in that, The process of inputting abnormal monitoring indicators into the early warning model to analyze the safety risks of pile foundations in target buildings includes: The early warning model constructed using a spatiotemporal graph neural network builds a correlation map between the target building's pile foundation and the surrounding environment. From the correlation map, the causal correlation between abnormal monitoring indicators and the surrounding environment is extracted. The surrounding environment includes surrounding buildings, construction areas, and geological layers. Multidimensional time-series correlation analysis was conducted based on the causal correlation degree of anomaly monitoring indicators to obtain the time-series correlation characteristics of anomaly monitoring indicators. Combining the preset dimensions of pile foundation safety risk assessment, the types of safety risks existing in the pile foundation are predicted based on the temporal correlation characteristics; based on the scope of influence and degree of harm corresponding to the types of safety risks, and with reference to the preset risk level classification standards, the safety risk level of the target building's pile foundation is determined; the types of safety risks, safety risk levels, and causal correlation factors are used as the safety risks existing in the pile foundation of the target building.

4. The early warning method for pile foundation health status analysis according to claim 3, characterized in that, Before extracting the causal correlation between the abnormal monitoring indicators and the surrounding environment from the correlation map, the method further includes: The abnormal monitoring indicators are preprocessed. Noise data and invalid values ​​in the abnormal monitoring indicators are removed based on preset data cleaning rules. The historical pile foundation monitoring data of the target building stored in the monitoring platform are retrieved. The comparison relationship between the abnormal monitoring indicators and the historical data of the same period and historical extreme values ​​in the historical pile foundation monitoring data is established to determine the deviation range and change rate of the abnormal monitoring indicators. The system calls upon the historical correlation trends of the geographical area where the target building is located, stored in the cloud data processing center. These trends include the structural correlation parameters between the target building's pile foundation and surrounding buildings, records of surrounding construction activities and geological activity data within a preset period, and use these historical correlation trends as supplementary input features for the early warning model. The records of surrounding construction activities include construction location, construction load, and construction duration. The geological activity data includes ground settlement and vibration frequency.

5. The early warning method for pile foundation health status analysis according to claim 1, characterized in that, The dynamic trend analysis of the modified fusion characteristics to obtain the initial pile foundation evaluation results includes: A multi-dimensional health index is calculated through the output layer of a lightweight multi-head attention temporal network, and the values ​​of each dimension of the health index correspond to the health status level of the pile foundation. A lightweight moving average algorithm and a linear regression model are used to calculate the short-term trend curves of health indices in various dimensions. The health indices of each dimension and their corresponding trend curves together constitute the initial pile foundation assessment results, which reflect the current overall condition of the pile foundation and the tendency of its health changes within a preset period.

6. The early warning method for pile foundation health status analysis according to claim 1, characterized in that, The comprehensive assessment of the correlation trends among target buildings within each geographical region yields corrected parameters for each geographical region, including: The particle swarm optimization algorithm is launched in the cloud data processing center to determine the core variables for comprehensive evaluation. The core variables include the structural correlation parameters between the target building's pile foundation and surrounding buildings in each geographical area, the correlation parameters between the target building's pile foundation and surrounding construction activities, and the correlation parameters between the target building's pile foundation and surrounding geological activities. Initialize the particle swarm, set the size of the particle swarm and the initial position and initial velocity of the particles. The position of each particle corresponds to a set of core variable values, representing an evaluation scheme for correlation trends. The velocity of the particles corresponds to the adjustment direction and adjustment magnitude of the core variable values. A fitness function is constructed, which combines the historical pile foundation health status change data of target buildings in each geographical area with the current monitoring conditions. The fitness function is used to measure the degree of matching between the core variable value combination corresponding to each particle and the actual pile foundation health change trend. The higher the matching degree, the higher the fitness value of the particle. The particle swarm is updated iteratively. In each iteration, the fitness value of each particle is calculated, the historical best position of each particle and the global best position of the entire particle swarm are determined, and the velocity and position of each particle are adjusted based on the historical best position and the global best position of the particle, thus optimizing the combination of values ​​of the core variables. Determine whether the iteration meets the preset convergence condition. If the number of iterations reaches the preset number or the fitness value of the global optimal position of the particle swarm remains stable for multiple rounds, then stop the iteration and use the combination of core variable values ​​corresponding to the global optimal position of the particle swarm as the optimal evaluation result of the correlation trend between target buildings in each geographical area. The optimal evaluation results are transformed into correction parameters corresponding to each geographical region. The correction parameters corresponding to each geographical region include correction terms for structural correlation, correction terms for correlation with surrounding construction activities, and correction terms for correlation with surrounding geological activities. Each correction term corresponds to the influence weight of one of the correlation trends on the pile foundation health status evaluation results.

7. The early warning method for pile foundation health status analysis according to claim 6, characterized in that, The fitness function is constructed by combining historical pile foundation health status change data of target buildings in each geographical region with current monitoring conditions, including: A dynamic multimodal data fusion evaluation mechanism is adopted, which synchronously retrieves two core data sources through the cloud data processing center. The first data source is the historical pile foundation health status change data sequence of the target building in each geographical area. The historical pile foundation health status change data sequence includes a time series database of pile foundation settlement, inclination angle change, stress distribution and deformation characteristics recorded over a long period. The second data source is the monitoring data collected in real time by the monitoring platform. The real-time monitoring data includes real-time sensor readings, instantaneous values ​​of environmental loads and dynamic information flow of surrounding construction activities. A fitness function is established based on the first and second data sources. The fitness function maps the current core variable values ​​represented by the particle position to a simulated change trajectory of the pile foundation health status. The simulated change trajectory is then matched with the historical database using a multi-level deep temporal pattern matching. During the deep temporal pattern matching process, a dynamic time warping algorithm is used to align the temporal pattern. The morphological similarity, key feature point offset, and change gradient consistency between the simulated change trajectory and the real data segments under similar environmental loads and construction disturbance conditions in the historical database are calculated. A dynamic weight allocation strategy is introduced to perform weighted comprehensive scoring, and the weighted comprehensive score of multi-level matching results is used as the output value of the fitness function.

8. An early warning system for pile foundation health status analysis, characterized in that, The system includes multi-source sensors embedded in the pile foundation, a monitoring platform, and a cloud data processing center, wherein... The monitoring platform has a built-in integrated communication module for receiving and processing various monitoring data from multiple sources of sensors in real time. These monitoring data include at least one of the following: settlement, inclination, stress, and deformation data of the pile foundation. By combining graph neural networks and lightweight multi-head attention temporal networks with the building structure of the target building, the platform performs data fusion and dynamic trend analysis on the various monitoring data to obtain initial pile foundation assessment results. The initial pile foundation assessment results are used to represent the health status of the pile foundation of the target building. The monitoring platform is also used to upload the initial pile foundation assessment results of target buildings in various geographical areas and various monitoring data to the cloud data processing center; The cloud data processing center is used to store the initial pile foundation evaluation results of target buildings in various geographical areas, and to comprehensively evaluate the correlation trends between target buildings in various geographical areas to obtain the correction parameters corresponding to each geographical area. The correlation trends include: the structural correlation between the pile foundations in the target building and the surrounding buildings, the correlation between the pile foundations in the target building and the surrounding construction activities, and the correlation between the pile foundations in the target building and the surrounding geological activities; The cloud data processing center is also used to send the corrected parameters to the monitoring platform; The monitoring platform is also used to dynamically update the health status of the pile foundations of target buildings in each geographical area by combining the correction parameters, so as to obtain the target pile foundation evaluation results of the target buildings; The method involves combining graph neural networks and lightweight multi-head attention temporal networks with the building structure of the target building to perform data fusion and trend analysis on various monitoring data, thereby obtaining initial pile foundation evaluation results, including: Based on the classification and labeling of multiple monitoring data as dynamic monitoring features, the pre-input building structure parameters are classified and labeled as static structure features. A lightweight multi-head attention temporal network is launched, and a group query attention mechanism is used to fuse features of the multimodal input matrix. The attention calculation task is split by pre-setting the number of groups to reduce the model parameter scale and computational load. The interdependencies between different dynamic monitoring features are learned, the contribution weight of each monitoring data to the pile foundation health status assessment at different time nodes is determined, and the weighted fusion of the monitoring data from the contribution weight source is combined to obtain the fused features of the target building pile foundation. A structural relationship graph of the target building is constructed using a graph neural network. The pile foundation is the core node, and the superstructure load-bearing components and foundation cap are the related nodes. The load transfer coefficient and the connection strength parameters between the pile foundation and the related nodes in the static structural features are loaded. Through the message passing mechanism of the graph neural network, the structural relationship between the corresponding components of each node is transformed into a feature correction factor to correct the fusion features output by the lightweight multi-head attention temporal network. Dynamic trend analysis is performed on the corrected fusion characteristics to obtain the initial pile foundation evaluation results, so as to reflect the current overall state of the pile foundation and the tendency of health changes within a future preset period through the initial pile foundation evaluation results; The classification and labeling of various monitoring data into dynamic monitoring features, and the classification and labeling of pre-input building structural parameters into static structural features, include: Multiple monitoring data from multi-source sensors are preprocessed, and noise data and invalid values ​​are removed based on preset data cleaning rules. The building structure parameters of the target building are retrieved, including pile foundation depth, pile foundation diameter, concrete grade, and load transfer coefficient between the pile foundation and the superstructure. The preprocessed monitoring data are classified and labeled as dynamic monitoring features, and the building structure parameters are classified and labeled as static structural features. A lightweight multi-head attention temporal network input layer is constructed on the integrated chip of the monitoring platform. Dynamic monitoring features are divided into continuous data segments according to time series and associated with static structural features to form a multimodal input matrix. The dynamic monitoring features include at least the temporal changes of settlement, inclination angle, stress and deformation data of pile foundation. The static structural features are embedded into the input matrix as prior information.

Images