Bridge pier body settlement monitoring method and system based on multi-source data and time series analysis

Abstract

The application discloses a bridge pier body settlement monitoring method and system based on multi-source data and time series analysis, and relates to the technical field of settlement monitoring. In the monitoring process of the bridge pier body, a directed acyclic graph is constructed based on the mechanical evolution mechanism of the bridge pier body settlement, and a corresponding key settlement feature set is determined in combination with multi-source feature variables extracted from a multi-field coupling representation space of the bridge pier body settlement, thereby improving the accuracy of the sensitive feature topology network. The time series prediction result of the bridge pier body settlement is determined according to the dynamic preview of the space-time correlation embedding vector. The settlement trend item is determined based on the identification of the time series prediction result, and the actual settlement deviation content is determined in combination with the dynamic baseline of the environmental temperature and the water level. The multiple settlement influencing factors are determined according to the identification of the actual settlement deviation content, and the settlement monitoring event of the bridge pier body is determined in combination with the real-time image of the bridge pier body, thereby improving the accuracy of the settlement monitoring event of the bridge pier body.

Description

Technical Field

[0001] This invention relates to the technical field of settlement monitoring, and in particular to a method and system for monitoring bridge pier settlement based on multi-source data and time-series analysis. Background Technology

[0002] As a key foundation component supporting the superstructure and transferring loads to the foundation, the settlement and deformation of bridge piers directly affect the operational safety of the entire bridge. As bridges develop towards larger spans, heavier loads, and more complex structures, the stress environment and service boundaries of bridge piers are becoming increasingly complex, placing higher demands on their settlement monitoring and early warning.

[0003] Currently, monitoring of bridge pier settlement mainly relies on single or discrete sensors deployed on the pier and foundation surfaces to obtain multi-source monitoring data, including those for stress, temperature, seepage, and deformation. Bridge pier settlement is the result of the coupling effects of multiple fields: stress, temperature, seepage, and deformation. Existing technologies often only perform simple splicing or independent analysis of multi-field data, failing to effectively eliminate scale differences between different physical dimensions and the coupling interference of environmental fields such as temperature and seepage on the deformation field. Furthermore, existing technologies lack a rigorous registration mechanism based on spatiotemporal anchor points, leading to data misalignment in the spatiotemporal dimensions. This results in the extraction of settlement features containing a large amount of redundancy and pseudo-related information, affecting the accuracy of the sensitive feature topology network and ultimately leading to low accuracy in monitoring bridge pier settlement events. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art. This invention provides a method and system for monitoring bridge pier settlement based on multi-source data and time series analysis.

[0005] This invention provides a method for monitoring bridge pier settlement based on multi-source data and time-series analysis, including: During the monitoring of bridge piers, multiple monitoring data of the bridge piers are acquired, the multiple monitoring data are aligned, and a corresponding multi-field coupled characterization space for bridge pier settlement is constructed by combining the registration mechanism. Based on the mechanical evolution mechanism of bridge pier settlement, a directed acyclic graph is constructed, and the multi-source feature variables extracted from the multi-field coupling characterization space of bridge pier settlement are combined to determine the corresponding key settlement feature set, and then a sensitive feature topology network characterizing the spatiotemporal evolution path of load-environment-settlement is constructed. The temporal monitoring sequence of the bridge pier is obtained and combined with the sensitive feature topology network input into the preset dynamic temporal space to further determine the spatiotemporal correlation embedding vector of the bridge pier settlement; the temporal prediction result of the bridge pier settlement is determined based on the dynamic pre-simulation of the spatiotemporal correlation embedding vector. Settlement trend terms are identified based on time-series prediction results, and actual settlement deviations are determined by combining ambient temperature and water level dynamic baselines. Multiple settlement influencing factors are identified based on the identification of actual settlement deviations, and settlement monitoring events of bridge piers are determined by combining real-time images of bridge piers.

[0006] This invention provides a bridge pier settlement monitoring system based on multi-source data and time-series analysis. This system is applied to the aforementioned bridge pier settlement monitoring method based on multi-source data and time-series analysis. The multi-source data and time-series analysis bridge pier settlement monitoring system includes: The monitoring data module is used to acquire multiple monitoring data of the bridge pier during the monitoring process, align the multiple monitoring data, and construct the corresponding multi-field coupled characterization space of bridge pier settlement by combining the registration mechanism. The topology network module is used to construct a directed acyclic graph based on the mechanical evolution mechanism of bridge pier settlement, and to determine the corresponding key settlement feature set by combining the multi-source feature variables extracted from the multi-field coupling characterization space of bridge pier settlement, thereby constructing a sensitive feature topology network that characterizes the spatiotemporal evolution path of load-environment-settlement. The time-series prediction module is used to acquire the time-series monitoring sequence of the bridge pier body, and combine it with the sensitive feature topology network input into the preset dynamic time-series space to further determine the spatiotemporal correlation embedding vector of the bridge pier body settlement; the time-series prediction result of the bridge pier body settlement is determined based on the dynamic pre-simulation of the spatiotemporal correlation embedding vector. The settlement monitoring module is used to identify settlement trend items based on the time-series prediction results, and to determine the actual settlement deviations by combining the dynamic baseline of ambient temperature and water level. Based on the identification of the actual settlement deviations, multiple settlement influencing factors are identified, and the settlement monitoring events of the bridge piers are determined by combining real-time images of the bridge piers.

[0007] Compared with the prior art, the beneficial effects of the present invention are: (1) During the monitoring of bridge piers, multiple monitoring data of bridge piers are acquired, the multiple monitoring data are aligned, and the corresponding bridge pier settlement multi-field coupling characterization space is constructed in combination with the registration mechanism; a directed acyclic graph is constructed based on the mechanical evolution mechanism of bridge pier settlement, and the corresponding key settlement feature set is determined by combining the multi-source feature variables extracted from the bridge pier settlement multi-field coupling characterization space, and then a sensitive feature topology network representing the load-environment-settlement spatiotemporal evolution path is constructed. The bridge pier settlement multi-field coupling characterization space is introduced, the key settlement feature set is further controlled, and the accuracy of the sensitive feature topology network is improved.

[0008] (2) Obtain the time-series monitoring sequence of the bridge pier and input it into the preset dynamic time-series space in combination with the sensitive feature topology network to further determine the spatiotemporal correlation embedding vector of the bridge pier settlement; determine the time-series prediction result of the bridge pier settlement based on the dynamic pre-simulation of the spatiotemporal correlation embedding vector; determine the settlement trend term based on the identification of the time-series prediction result, and determine the actual settlement deviation content in combination with the dynamic baseline of ambient temperature and water level; determine multiple settlement influencing factors based on the identification of the actual settlement deviation content, and determine the settlement monitoring event of the bridge pier in combination with the real-time image of the bridge pier. Further control the time-series prediction result of the bridge pier settlement, fully consider multiple settlement influencing factors and the real-time image of the bridge pier, and improve the accuracy of the settlement monitoring event of the bridge pier. Attached Figure Description

[0009] Figure 1 This is a flowchart illustrating the bridge pier settlement monitoring method based on multi-source data and time-series analysis in an embodiment of the present invention. Figure 2 This is a flowchart illustrating step S11 in the bridge pier settlement monitoring method based on multi-source data and time series analysis in this embodiment of the invention. Figure 3 This is a flowchart illustrating step S12 in the bridge pier settlement monitoring method based on multi-source data and time series analysis in this embodiment of the invention. Figure 4 This is a flowchart illustrating step S13 in the bridge pier settlement monitoring method based on multi-source data and time series analysis in this embodiment of the invention. Figure 5 This is a flowchart illustrating step S14 in the bridge pier settlement monitoring method based on multi-source data and time series analysis in this embodiment of the invention. Figure 6 This is a schematic diagram of the structural composition of a bridge pier settlement monitoring system based on multi-source data and time-series analysis in an embodiment of the present invention. Detailed Implementation

[0010] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0011] Please see Figures 1 to 6 A method for monitoring bridge pier settlement based on multi-source data and time-series analysis is proposed and applied to settlement monitoring scenarios. The method includes: Step S11: During the monitoring of the bridge pier, multiple monitoring data of the bridge pier are acquired, the multiple monitoring data are aligned, and the corresponding bridge pier settlement multi-field coupling characterization space is constructed by combining the registration mechanism. Step S12: Construct a directed acyclic graph based on the mechanical evolution mechanism of bridge pier settlement, and determine the corresponding key settlement feature set by combining the multi-source feature variables extracted from the multi-field coupling characterization space of bridge pier settlement, and then construct a sensitive feature topology network that characterizes the spatiotemporal evolution path of load-environment-settlement. Step S13: Obtain the time-series monitoring sequence of the bridge pier body, and input it into the preset dynamic time-series space in combination with the sensitive feature topology network to further determine the spatiotemporal correlation embedding vector of the bridge pier body settlement; determine the time-series prediction result of the bridge pier body settlement based on the dynamic pre-simulation of the spatiotemporal correlation embedding vector. Step S14: Based on the identification of time-series prediction results, determine the settlement trend item, and combine the ambient temperature and water level dynamic baseline to determine the actual settlement deviation. Based on the identification of the actual settlement deviation, determine multiple settlement influencing factors, and combine the real-time images of the bridge pier to determine the settlement monitoring events of the bridge pier.

[0012] refer to Figure 2 In step S11, the specific steps are as follows: S111: Monitor the bridge pier body and acquire multiple monitoring data of the bridge pier body during the monitoring process. The multiple monitoring data of the bridge pier body cover the stress field, temperature field, seepage field and deformation field. Combined with the heterogeneous data registration mechanism based on spatiotemporal anchor points, the spatiotemporal alignment of multiple monitoring data is triggered, and the aligned data is output. S112: Project the aligned data into a high-dimensional latent space, extract cross-modal local manifold features, perform deep fusion of the manifold features of each part, and eliminate scale differences and coupling interference between multiple data to construct a spatiotemporally consistent multi-field coupled characterization space for bridge pier settlement.

[0013] In the embodiments of this application, the bridge pier is monitored, and multiple monitoring data of the bridge pier are acquired during the monitoring process. The multiple monitoring data of the bridge pier cover the stress field, temperature field, seepage field and deformation field. The spatiotemporal alignment of multiple monitoring data is triggered by a heterogeneous data registration mechanism based on spatiotemporal anchor points, and the aligned data is output.

[0014] At this time, during the monitoring of the bridge pier, multiple monitoring data covering stress field, temperature field, seepage field and deformation field are acquired by various types of sensing devices deployed on the pier and foundation. Due to the physical heterogeneity of the acquisition frequency, transmission delay and spatial coordinate expression of the multiple monitoring data, it is necessary to unify and sort out the above-mentioned multi-source heterogeneous data. Among them, stress field data and deformation field data are usually dynamic high-frequency sampling, while temperature field data and seepage field data are usually low-frequency slowly varying sampling. Moreover, the spatial distribution of the sensors for each field is discrete, thus forming a raw dataset that is asynchronous in the time domain and is spatially non-isomorphic.

[0015] To achieve a unified representation in the spatial domain, a heterogeneous data registration mechanism based on spatiotemporal anchor points is introduced to define and register spatial anchor points. According to the structural topology and mechanical transmission characteristics of the bridge pier, spatial anchor points are set at key sections of the pier, and the three-dimensional coordinates of these spatial anchor points are used as the spatial registration reference. The sensors of each field are transformed from local installation coordinates to a unified structural coordinate system. Using the nearest neighbor spatial projection method, stress, temperature, seepage, and deformation sensors that fall within the influence radius of the same anchor point are divided into the same spatial cluster, thereby realizing the spatial aggregation of multi-field data at physical structural nodes and triggering spatial dimension registration and alignment.

[0016] Building upon spatial registration, temporal alignment is further triggered using a spatiotemporal anchoring mechanism. The moment of significant change in environmental load or a system-defined hour is selected as the time anchor, with this anchor moment serving as the baseline time axis. For high-frequency monitoring data, a sliding window mean downsampling method is employed to extract statistical feature values ​​within the window that coincide with the time anchor. For low-frequency monitoring data, a state estimation method based on cubic spline interpolation or Kalman filtering is used to deduce and fit the transient estimates of the low-frequency data at the time anchor. This eliminates the time delay caused by acquisition frequency differences and transmission jitter in the data from different fields, achieving strict synchronization of multiple field data in the time dimension.

[0017] After spatial clustering and temporal resampling, the field data are tensor-mapped according to timestamps and spatial anchor points to eliminate boundary redundancy caused by interpolation or downsampling. The output is aligned data with a unified time step and a unified spatial node identifier, providing a spatiotemporally consistent input source for subsequent high-dimensional latent space projection.

[0018] Specifically, for the bridge pier, the acquired monitoring data includes: "stress field data: vibrating wire strain gauge, 1Hz acquisition" of the pressure zone at the root of the inclined leg of the bridge pier; "temperature field data: thermistor temperature gradient meter, 1 / 600Hz acquisition" of the surface and interior of the inclined leg; "seepage field data: pore water pressure gauge, 1 / 10Hz acquisition" of the bottom surface of the abutment of the bridge pier; and "deformation field data: hydrostatic level and tilt sensor, 1Hz acquisition" of the top of the inclined leg and the top of the pier. The above four data fields span from 1Hz to 1 / 600Hz in terms of acquisition frequency, and the sensors are located in the abutment, inclined leg and top of the pier respectively, showing significant heterogeneity.

[0019] Based on the mechanical properties of the bridge pier, the "slope root section of the bridge pier", the "center of the tie beam of the bridge pier", and the "top center of the bridge pier" are defined as three core spatial anchor points. During spatial registration, the pore water pressure gauge on the bottom surface of the abutment and the strain gauge at the slope root are spatially clustered with the "slope root section of the bridge pier" anchor point as the center. The slope temperature gradient gauge, hydrostatic level and tilt sensor are projected onto the nearest anchor point cluster, thereby anchoring the discrete four field sensors to the three key mechanical sections of the bridge pier and completing the physical alignment of the spatial dimension.

[0020] The time of the daily extreme ambient temperature (e.g., 14:00) is used as the time anchor point. For 1Hz stress and deformation field data, a sliding window of 30 seconds before and after 14:00:00 is taken as the center, and the mean value within the window is calculated as the alignment value of the time anchor point. For 1 / 600Hz temperature field and 1 / 10Hz seepage field data, based on the preceding and subsequent acquisition points, cubic spline interpolation is used to accurately calculate the interpolation estimate at 14:00:00. Thus, the stress, seepage, and temperature data at the root of the inclined leg of the pier, as well as the deformation and temperature data at the top of the inclined leg, are all strictly aligned to the time anchor point of 14:00:00.

[0021] The stress, temperature, seepage, and deformation data of the bridge pier body, after being aligned with the same time anchor point at three spatial anchor points, are recombined sequentially, and the output structure is a canonical tensor of "time stamp-anchor point number-four-field feature vector". For example, for the "slanted leg root anchor point", the output includes a spatiotemporal aligned feature vector containing [mean stress, temperature gradient, pore water pressure, tilt angle], thus completing the spatiotemporal alignment and output of the multi-source heterogeneous data of the bridge pier body.

[0022] Furthermore, the aligned data are projected into a high-dimensional latent space to extract cross-modal local manifold features. The manifold features of each part are deeply fused, and the scale differences and coupling interference between multiple data are eliminated. In this way, a spatiotemporally consistent multi-field coupling characterization space for bridge pier settlement is constructed, and a spatiotemporally consistent multi-field coupling characterization space for bridge pier settlement is introduced.

[0023] At this point, the aligned data are input into the feature mapping network of the corresponding mode, and the low-dimensional discrete data in the physical measurement space is projected to the high-dimensional latent space through nonlinear transformation. Since the multi-field response of the bridge pier often presents a linearly inseparable state in the low-dimensional physical space, the projection of the high-dimensional latent space aims to explicitly represent the distribution structure of each field data in the latent feature space by expanding the feature dimension, thereby outputting the high-dimensional latent space feature representation corresponding to each field data.

[0024] In the high-dimensional latent space, cross-modal local manifold feature extraction is performed on the feature representations of each modality. Based on the manifold assumption, the evolution of each field data in the high-dimensional space is not globally uniformly distributed, but embedded in a specific low-dimensional manifold structure. By constructing the K-nearest neighbor graph of each modality feature, the geodesic distance between sample points is calculated to replace the Euclidean distance, capturing the topological geometry of the local neighborhood within and between each modality, and then extracting cross-modal local manifold features that maintain local topological invariance to characterize the nonlinear correlation patterns between multiple field data.

[0025] After acquiring the local manifold features of each modality, scale difference elimination and deep fusion operations are performed. Due to the significant differences in the physical dimensions and numerical magnitudes of the stress field, temperature field, seepage field, and deformation field, an adaptive instance normalization mechanism is introduced to statistically align the mean and variance of the manifold features of each modality, thus eliminating the scale barrier caused by the physical dimensions. The aligned multimodal manifold features are then stitched together along the feature channel dimension, and nonlinear cross-fusion of cross-modal features is performed through a multilayer perceptron to achieve deep aggregation of multi-source information in the latent space.

[0026] To address the coupling interference generated by multi-field data during the physical evolution process, a decoupling operation is performed based on the deeply fused features. Due to the thermal expansion and contraction caused by changes in ambient temperature and the buoyancy effect caused by changes in water level in the seepage field, strong coupling occurs with the actual mechanical deformation, resulting in the deformation field features containing non-settlement interference terms. An orthogonal constraint optimization objective is constructed to force the settlement-dominant feature subspace and the temperature and pressure environmental interference feature subspace to be mutually orthogonal in the latent space. By minimizing the inner product of the features of the two subspaces, environmental coupling interference is removed. The decoupled pure features are then tensor-recombined to construct a bridge pier settlement multi-field coupling characterization space with spatiotemporal consistency, unified physical scale, and no environmental coupling interference.

[0027] Optionally, for the construction process of the multi-field coupling characterization space for bridge pier settlement, spatial anchor points are set at three key locations: the root of the inclined leg of the bridge pier, the center of the tie beam, and the top of the pier. Strain gauges, thermometers, piezometers, and levels are simultaneously deployed at each anchor point to collect stress, temperature, seepage, and deformation data, respectively. During spatial registration, the coordinates of all sensors are uniformly transformed to the structural coordinate system, and the sensors within a radius of 0.5 meters centered on each anchor point are grouped into the same spatial cluster.

[0028] Spatial anchor points were set at three key locations: the root of the inclined leg of the bridge pier, the center of the tie beam, and the top of the pier. Strain gauges, thermometers, piezometers, and levels were simultaneously deployed at each anchor point to collect stress, temperature, seepage, and deformation data. During spatial registration, the coordinates of all sensors were uniformly transformed to the structural coordinate system, and sensors within a 0.5-meter radius of each anchor point were grouped into the same spatial cluster. The aforementioned five-tuple data was input into a three-layer fully connected neural network, mapped to a 512-dimensional high-dimensional latent space, and a local linear embedding algorithm was used to extract manifold features. The neighborhood parameter K=15, and geodesic distance was used instead of Euclidean distance to preserve the nonlinear topological relationships of each physical field.

[0029] Through adaptive instance normalization, the feature distributions of the four physical fields—stress, temperature, seepage, and deformation—are uniformly adjusted to a standard distribution with a mean of 0 and a variance of 1, eliminating dimensional differences. The four normalized features are then concatenated and deeply fused using a two-layer multilayer perceptron. In the resulting 128-dimensional feature space, two orthogonal subspaces are constructed: a settlement-dominant subspace and a temperature and pressure environment interference subspace. By minimizing the sum of the squares of the inner products of the feature vectors of the two subspaces, they are forced to be orthogonal, thus removing the coupling interference of temperature and seepage on deformation. The output of the settlement-dominant subspace is then reorganized into a three-dimensional tensor according to the topological order of the three spatial anchor points. The three dimensions correspond to time, anchor point position, and settlement characteristics, respectively. This completes the construction of the multi-field coupled characterization space for bridge pier settlement.

[0030] Specifically, the output data of the bridge pier's "sloping leg root anchor point," namely the stress field, temperature field, seepage field, and deformation field, are aligned and processed by a modal encoder containing multiple one-dimensional convolutional kernels and nonlinear activation functions. For example, the 1Hz strain scalar at the sloping leg root is projected into a 256-dimensional stress latent space vector, and the interpolated temperature gradient is projected into a 256-dimensional temperature latent space vector. In this way, the complex four-field physical measurements of the bridge pier are transformed into a distributed feature representation in a high-dimensional latent space.

[0031] For the four-field high-dimensional features of the inclined leg root of the bridge pier, local manifold feature extraction is performed. Since the increase in stress and the small change in the inclination angle of the inclined leg of the bridge pier are difficult to correspond directly in low-dimensional space, but present a local smooth topological relationship in high-dimensional manifold space, the geodesic distance and manifold tangent vector in the neighborhood are extracted by calculating the K-nearest neighbor graph of the 256-dimensional stress feature vector and deformation feature vector, and the stress-deformation cross-modal local manifold features are obtained. This feature accurately preserves the nonlinear local co-evolution relationship between stress and deformation of the inclined leg of the bridge pier under stress and deflection.

[0032] Because the stress at the root of the inclined leg of the bridge pier is on the order of megapascals, while the inclination angle is on the order of radians and the pore water pressure is on the order of kilopascals, adaptive instance normalization is used to calculate the batch mean and variance of the four-modal manifold features. The distribution of stress manifold features and deformation manifold features is forcibly aligned to a unified scale space with zero mean and unit variance. The four-modal manifold features after eliminating scale differences are channel-stitched and cross-attention fusion is performed through a multilayer perceptron, so that the deformation features can fully absorb the spatial context information of stress and seepage, and achieve deep fusion of multi-source information of the bridge pier.

[0033] To address typical coupling interferences in bridge pier settlement monitoring—the tilting of the pier's inclined legs towards the shaded side due to solar radiation temperature gradients and the foundation rebound caused by the drop in water level in the abutment seepage field—a decoupling operation was performed. Settlement feature matrices and temperature and pressure environment interference feature matrices were constructed in the latent space. By minimizing the orthogonal penalty term, the thermal tilt angle feature of the inclined legs caused by the temperature gradient and the actual tilt angle feature caused by foundation settlement were made perpendicular to each other on the spatial basis. During backpropagation, the model forcibly removed the thermal deflection interference of the inclined legs, retaining the deformation features that purely reflect mechanical settlement. The pure feature tensors of the three anchor points—the root of the inclined legs, the center of the tie beam, and the top of the pier—were recombined according to the spatial topology to construct a spatiotemporally consistent multi-field coupled representation space for bridge pier settlement, providing interference-free pure input for subsequent temporal evolution analysis.

[0034] refer to Figure 3 In step S12, the specific steps are as follows: S121: Based on the traversal of the bridge pier settlement database, the mechanical evolution mechanism of bridge pier settlement is determined. Based on the analysis of the mechanical evolution mechanism of bridge pier settlement, the causal logic of load transfer, environmental erosion and ground consolidation is determined. The corresponding directed acyclic graph is constructed in combination with the three-dimensional morphology of the bridge pier. S122: Extract multi-source feature variables reflecting the multi-field coupling effect from the multi-field coupling representation space of bridge pier settlement, map the multi-source feature variables to the key nodes of the directed acyclic graph, mark the corresponding causal evolution content, and simultaneously remove redundant and pseudo-correlated features, thereby determining the key settlement feature set representing the driving core of bridge pier settlement. Using the key settlement feature set as edges and nodes, and combining the mechanical causal path in the directed acyclic graph, construct a sensitive feature topology network representing the spatiotemporal evolution path of load-environment-settlement.

[0035] In the embodiments of this application, the mechanical evolution mechanism of bridge pier settlement is determined by traversing the bridge pier settlement database. Based on the analysis of the mechanical evolution mechanism of bridge pier settlement, the causal logic of load transfer, environmental erosion and ground consolidation is determined. In addition, a corresponding directed acyclic graph is constructed in combination with the three-dimensional morphology of the bridge pier. This approach is compatible with the overall consideration of the mechanical evolution mechanism of bridge pier settlement and ensures the accuracy of the causal logic of load transfer, environmental erosion and ground consolidation.

[0036] At this point, a full traversal and in-depth mining of the pre-constructed bridge pier settlement database is performed. This database historically stores multi-source monitoring time-series data, material degradation records, and settlement instability cases. Through association rule mining and sequence pattern discovery, the evolution law of multi-field response and macroscopic instability precursor patterns within the entire settlement cycle are extracted. Based on this, the mechanical evolution mechanism of bridge pier settlement is inverted and determined. This mechanism reveals the dynamic evolution path of internal stress redistribution, material performance degradation, and external boundary condition degradation in the process from initial stress to final macroscopic settlement of the pier.

[0037] Based on the established mechanical evolution mechanism, the settlement driving force is further analyzed to determine the causal logic among load transfer, environmental erosion, and ground consolidation. Load transfer is characterized by the mechanical path of the superstructure's self-weight and dynamic load spreading to the foundation through the pier body; environmental erosion is characterized by the reduction of material modulus and effective cross-sectional area caused by temperature and humidity changes and water scouring; and ground consolidation is characterized by the compression deformation of the foundation soil caused by the dissipation of pore water. Environmental erosion is defined as the precursor that weakens the load transfer path and accelerates ground consolidation, load transfer is the direct mechanical driving force applied to ground consolidation, and ground consolidation is the physical result that ultimately leads to macroscopic settlement. This forms a causal logic chain of "environmental erosion weakens the transmission path—load transfer triggers stress concentration—ground consolidation leads to macroscopic deformation".

[0038] By combining the three-dimensional morphological parameters of the bridge pier, the aforementioned causal logic chain is mapped to a three-dimensional physical space to construct a corresponding directed acyclic graph. Using the three-dimensional geometric topology of the bridge pier as the graph structure skeleton, the key sections, foundation bearing layers, and environmental erosion surfaces in the three-dimensional morphology are defined as nodes in the graph model, and the load transfer path, the range of environmental erosion, and the area affected by soil consolidation are defined as directed edges connecting each node. The attributes of each node include the multi-field physical state parameters at that location, and the direction of the directed edges strictly follows the aforementioned causal logic, that is, from the inducing node to the result node, thereby constructing a topologically rigorous, causally clear, and physically consistent directed acyclic graph.

[0039] Optionally, the directed acyclic graph contains five nodes: the first node represents environmental erosion, the second node represents the load at the root of the inclined leg, the third node represents the stress concentration at the bottom of the pier cap, the fourth node represents the consolidation of the strata, and the fifth node represents the settlement at the top of the pier.

[0040] The rules for setting directed edges are as follows: from the environmental erosion node to the stress concentration node at the bottom of the pier cap, it indicates that erosion leads to the loss of soil resistance; from the load node at the root of the inclined leg to the stress concentration node at the bottom of the pier cap, it indicates that the load generates eccentric stress; from the stress concentration node at the bottom of the pier cap to the soil consolidation node, it indicates that the excessive stress triggers soil compression; from the soil consolidation node to the settlement node at the top of the pier, it indicates that the foundation deformation is ultimately manifested as pier settlement; the direction of all edges strictly follows the above causal order, and there are no reverse edges or cyclic paths in the graph, thus forming a directed acyclic graph with a clear structure and well-defined causal relationships.

[0041] Specifically, a full traversal of the historical settlement database of the bridge pier was conducted. This database contains records of scour during the flood season of the bridge pier over the years, monitoring data of stress concentration at the root of the inclined leg, and long-term displacement sequence of the pier top. Through temporal correlation mining, it was found that before the macroscopic settlement of the bridge pier, there must be a long-term redistribution of stress at the root of the inclined leg and a slow change in seepage pressure at the bottom of the abutment. Thus, the mechanical evolution mechanism of the bridge pier settlement was determined: that is, due to the horizontal thrust component acting on the foundation for a long time caused by the inclined leg, the settlement mechanism is manifested as a special mechanical evolution path of "accumulation of inclined leg thrust - stress concentration at the edge of the foundation - shear creep of the foundation soil - overall tilting and sinking of the pier".

[0042] Based on the mechanical evolution mechanism of bridge piers, the causal logic of load transfer, environmental erosion, and ground consolidation is analyzed and defined. Environmental erosion is specifically characterized by the local scouring and hollowing of the water-facing surface of the bridge pier cap by floodwater during the flood season, and the accelerated creep of the inclined leg concrete caused by temperature difference. Load transfer is specifically characterized by the transformation of the superstructure load into eccentric horizontal thrust and vertical compressive stress on the bottom surface of the cap through the inclined leg. Ground consolidation is specifically characterized by the rapid compaction and consolidation of sand caused by the increase in effective stress after the bottom of the cap is scoured and hollowed out. The causal logic is defined as follows: water scouring weakens the lateral soil resistance of the cap, causing the eccentric load transferred by the inclined leg to generate stress concentration at the foundation. After the concentrated stress exceeds the bearing capacity threshold of the foundation, it triggers local sand compaction, which is ultimately manifested as uneven settlement of the bridge pier.

[0043] Combining the three-dimensional morphology of the bridge pier, including its inclined legs and horizontal tie beams, a directed acyclic graph (DAG) is constructed. In the DAG, the "root of the inclined leg of the bridge pier," "midpoint of the horizontal tie beam," "upstream side of the abutment bottom surface," and "downstream side of the abutment bottom surface" are defined as physical nodes. The directed edges are defined as follows: a directed edge emanating from the "environmental erosion node" points to the "upstream side node of the abutment bottom surface," representing the loss of lateral constraint due to environmental erosion; a directed edge emanating from the "root of the inclined leg of the bridge pier" points to the "downstream side node of the abutment bottom surface," representing the eccentric stress generated by load transfer; and a directed edge emanating from the "downstream side node of the abutment bottom surface" points to the "virtual node of soil consolidation," representing stress concentration triggering soil consolidation. All directed edges converge to the "settlement node at the top of the bridge pier." This constructs a directed acyclic graph that strictly conforms to the spatial stress characteristics and causal transmission logic of the bridge pier, providing a mechanistic constraint framework for the subsequent extraction of sensitive feature topology networks.

[0044] Furthermore, multi-source feature variables reflecting the multi-field coupling effect are extracted from the multi-field coupling representation space of bridge pier settlement. These multi-source feature variables are mapped to key nodes of a directed acyclic graph, and the corresponding causal evolution content is marked. Redundant and pseudo-correlated features are simultaneously removed, thereby determining the key settlement feature set that represents the driving core of bridge pier settlement. Using the key settlement feature set as edges and nodes, and combining the mechanical causal path in the directed acyclic graph, a sensitive feature topology network representing the spatiotemporal evolution path of load-environment-settlement is constructed. This introduces the multi-field coupling representation space of bridge pier settlement, further controlling the key settlement feature set and improving the accuracy of the sensitive feature topology network.

[0045] At this point, multi-source feature variables reflecting the multi-field coupling effect are extracted from the multi-field coupling characterization space of bridge pier settlement along the time axis and spatial anchor point dimension. These multi-source feature variables include high-order coupling characterization parameters such as stress-deformation synergy factor and temperature-seepage offset. The extracted multi-source feature variables are precisely mapped to the corresponding key nodes of the directed acyclic graph (DAG) according to their spatial anchor point affiliation and physical properties. This allows the physical morphology nodes in the DAG to be endowed with dynamically changing latent space numerical characteristics, thereby activating the static mechanical topology into a dynamic data-driven graph.

[0046] After feature mapping is completed, the feature transmission and evolution content between key nodes is marked according to the causal pointing relationship of directed edges in the directed acyclic graph. The marked content represents the causal driving function relationship of the parent node feature to the child node feature. Redundant and pseudo-correlated feature removal operations are performed simultaneously. Based on the causal separation criterion and conditional independence test method of DAG, the information flow on non-causal paths is blocked, and pseudo-correlated features caused by common cause effects or accompanying variables, as well as redundant features with variance contribution rate below a preset threshold on the same causal path, are identified and removed. This purifies the feature space and retains driving features that strictly follow the mechanical causal law.

[0047] After mapping, labeling, and elimination operations, the feature variables that survive in the nodes of the directed acyclic graph are the core parameters driving the settlement of the bridge pier. The feature variables that have been filtered by cause and verified by causality are extracted in a set to determine the key settlement feature set that characterizes the core of the settlement driving force of the bridge pier. This set not only covers the direct causes of settlement but also includes the features of the preceding inducing terms, fully reflecting the causal characteristics of settlement from the driving source to the result.

[0048] Using each feature in the defined set of key settlement features as nodes and the causal evolution content between features as directed edges, combined with the predetermined load-environment-settlement mechanical causal path in the directed acyclic graph, the graph structure is reconstructed and weights are assigned. Based on the strength of the causal driving function relationship, the directed edges are calculated and assigned dynamic sensitive weights, thereby constructing a sensitive feature topology network that characterizes the load-environment-settlement spatiotemporal evolution path. This sensitive feature topology network realizes the leap from pure physical mechanics DAG to data-mechanism dual-driven dynamic topology, accurately quantifying the causal transmission sensitivity of multi-source features in the spatiotemporal evolution process.

[0049] To address the construction process of the sensitive feature topology network, three key settlement feature variables were extracted from the constructed bridge pier settlement multi-field coupling characterization space: the first is the eccentric stress-tilt synergy factor, which characterizes the coupling degree between the eccentric load at the root of the inclined leg and the local tilt deformation; the second is the pore pressure-scour offset factor, which characterizes the correlation between the change in seepage pressure at the bottom of the abutment and the degree of scour and erosion; and the third is the base compaction deformation factor, which characterizes the amount of compression deformation of the foundation soil under excessive stress. These three feature variables correspond to the three core nodes in the directed acyclic graph.

[0050] The three characteristic variables are mapped to their corresponding nodes in the directed acyclic graph, so that each node carries the time-series value of the corresponding characteristic variable. After mapping, the causal paths between nodes are marked with feature transfer: the directed edge from the "pore pressure-scour offset factor" node to the "base compaction deformation factor" node is marked as "scour weakening drive"; the directed edge from the "eccentric stress-tilt synergy factor" node to the "base compaction deformation factor" node is marked as "eccentric load drive".

[0051] Since changes in ambient temperature and rises in water level often occur synchronously in time, but temperature is not the mechanical cause of subsidence, based on the causal relationship of "environmental erosion → formation consolidation" in the directed acyclic graph, a conditional independence test is adopted. Given "scour and hollowing" as an intermediate variable, the information transmission path between "temperature change" and "formation consolidation" is blocked, thereby eliminating the pseudo-correlation features of thermal deformation caused by temperature and retaining only the seepage features directly related to scour.

[0052] Using the three key settlement characteristic variables retained as nodes and the causal evolution content between the characteristic variables as directed edges, a sensitive characteristic topology network is constructed. Among them, the "pore pressure-scour offset factor" node points to the "base compaction deformation factor" node through a directed edge. The sensitivity weight of this edge is dynamically calculated based on the scour intensity in historical data, with a larger weight for stronger scour. Similarly, the "eccentric stress-tilt synergy factor" node points to the "base compaction deformation factor" node through a directed edge. The sensitivity weight of this edge is dynamically adjusted based on the magnitude of the eccentric load on the inclined leg. The resulting topology network clearly represents the spatiotemporal evolution path of "environmental scour → eccentric load amplification → base compaction settlement".

[0053] Specifically, from the multi-field coupling representation space of the bridge pier, high-order coupling characteristic variables at the "anchor point at the root of the inclined leg" and the "anchor point at the bottom of the abutment" are extracted. For example, the "eccentric stress-tilt synergy factor" reflecting the coupling of the inclined leg thrust and local deformation, and the "pore pressure-scour offset factor" reflecting the coupling of scour and seepage are extracted. The "eccentric stress-tilt synergy factor" is mapped to the "anchor point at the root of the inclined leg of the bridge pier" in the DAG, and the "pore pressure-scour offset factor" is mapped to the "upstream node at the bottom of the abutment" in the DAG. In this way, the originally abstract DAG physical node is instantiated into a dynamic node carrying complex coupling state data of the bridge pier.

[0054] To address the spurious correlations that are prone to occur in bridge pier settlement monitoring, a method is used to eliminate them. For example, during the flood season, rising ambient temperature and rising water levels often occur simultaneously, showing a strong correlation at the data level. However, rising temperature is not the mechanical cause of foundation consolidation settlement. By testing the causal orientation and conditional independence of "environmental erosion → stratum consolidation" in the DAG, and given the intermediate variable of "water scouring and hollowing," the spurious correlation path from "temperature change" to "stratum consolidation" is blocked. The spurious correlation features of temperature-induced diagonal leg thermal deformation are eliminated, and only the seepage features directly related to scouring and weakening are retained. Furthermore, on the directed edge from "the root node of the diagonal leg of the bridge pier" to "the downstream side node of the abutment bottom surface," the causal evolution content is marked as "eccentric thrust amplification and transmission."

[0055] After causal separation, the “eccentric stress-tilt synergy factor,” “pore pressure-scour offset factor,” and “base edge compaction deformation factor” remaining in the bridge pier DAG are extracted in a set to form the key settlement feature set of the bridge pier settlement driving core. This set eliminates the interference of thermal expansion and contraction and accurately locks the core driving parameters of the uneven settlement of the bridge pier caused by “eccentric thrust of inclined leg + upstream scour and hollowing.”

[0056] Using features within this key settlement feature set as nodes, a sensitive feature topology network is constructed by combining the mechanical causal path of the bridge pier DAG. Among them, the "pore pressure-scour offset factor node" points to the "base edge compaction deformation factor node" through directed edges, and the directed edges are assigned a sensitive weight W1 to represent the driving sensitivity of scour weakening to consolidation. The "eccentric stress-tilt synergy factor node" points to the "base edge compaction deformation factor node", and the directed edges are assigned a sensitive weight W2 to represent the driving sensitivity of eccentric load to consolidation. Due to the existence of the inclined leg thrust of the bridge pier, W2 is significantly greater during the flood season scour period than during the non-flood season. Thus, a bridge pier sensitive feature topology network with a clear and quantifiable spatiotemporal evolution path of "water flow scour (environment) - inclined leg bias pressure (load) - foundation compaction (settlement)" is constructed, providing a causal constraint topology for the subsequent extraction of spatiotemporal correlation embedding vectors.

[0057] refer to Figure 4 In step S13, the specific steps are as follows: S131: In the settlement monitoring of bridge piers, the time-series monitoring sequence of bridge piers is obtained, and the time-series monitoring sequence and the sensitive feature topology network are input into the preset dynamic time-series space. In the dynamic time-series space, the topology structure of the sensitive feature topology network is used as the adjacency constraint. The graph attention mechanism is used to perform dynamic message passing and feature aggregation on the time-series monitoring sequence along the spatiotemporal dimension, presenting the topological state changes of multi-source feature variables in the time-series evolution, and then determining the spatiotemporal correlation embedding vector of bridge pier settlement that integrates spatiotemporal topological correlation and dynamic evolution characteristics. S132: Dynamically pre-simulate the spatiotemporal related embedding vector, simulate the multi-step forward evolution trajectory of the spatiotemporal related embedding vector, and combine it with random disturbance terms for multi-level iteration. In this process, the time-series prediction result of bridge pier settlement is determined by combining the constraint relationship of bridge pier settlement, so as to realize the probabilistic dynamic inference of the future evolution trend of settlement.

[0058] In the embodiments of this application, in the settlement monitoring of bridge piers, a time-series monitoring sequence of the bridge piers is obtained, and the time-series monitoring sequence and a sensitive feature topology network are input into a preset dynamic time-series space. In the dynamic time-series space, the topology of the sensitive feature topology network is used as an adjacency constraint, and a graph attention mechanism is used to perform dynamic message passing and feature aggregation on the time-series monitoring sequence along the spatiotemporal dimension. This presents the topological state changes of multi-source feature variables in the time-series evolution, thereby determining the spatiotemporal correlation embedding vector of bridge pier settlement that integrates spatiotemporal topological correlation and dynamic evolution characteristics. This approach is compatible with the overall consideration of the graph attention mechanism on the time-series monitoring sequence along the spatiotemporal dimension, ensuring the accuracy of the spatiotemporal correlation embedding vector of bridge pier settlement that integrates spatiotemporal topological correlation and dynamic evolution characteristics.

[0059] At this time, in the settlement monitoring of the bridge pier, the time-series monitoring sequence of the bridge pier is acquired in real time. The time-series monitoring sequence contains dynamic data of the evolution of aligned multi-field features over time. This time-series monitoring sequence and the sensitive feature topology network are synchronously input into a preset dynamic time-series space. The dynamic time-series space is a high-dimensional computing domain constructed based on graph neural networks and cyclic time-series architecture, which aims to provide a carrying environment for feature iteration and state update for subsequent spatiotemporal graph calculations.

[0060] In the dynamic temporal space, the topology of the sensitive feature topology network is used as an adjacency constraint, that is, only nodes with mechanical causal edges are allowed to interact with each other. The graph attention mechanism is used to dynamically pass messages along the spatiotemporal dimension of the time-series monitoring sequence. In the spatial dimension, based on the feature similarity and mechanical sensitivity weight of the source node and the target node at the current time step, the spatial attention coefficient is dynamically calculated. Based on this coefficient, the temporal features of the source node are aggregated to the target node along the directed edges to achieve adaptive feature fusion of the spatial neighborhood. In the temporal dimension, a temporal attention mechanism is introduced to evaluate the degree of influence of historical time step features on the current time step and dynamically capture the long-term dependence of multi-source feature variables in the temporal evolution.

[0061] Through the aforementioned dynamic message transmission in the spatiotemporal dimensions, each key node integrates the historical and current feature information of its causal upstream nodes at each time step. As the time-series monitoring sequence is continuously input, the hidden state of the nodes is updated step by step in the dynamic time-series space, thereby accurately presenting the topological state changes of multi-source feature variables in the time-series evolution. The topological state changes reflect the dynamic driving process of load transfer and environmental erosion on the settlement results on the time axis. Furthermore, the updated hidden states of each node are subjected to global feature aggregation, and an adaptive pooling or readout mechanism is used to compress the spatiotemporal evolution features of the entire graph into a fixed-dimensional graph-level representation vector.

[0062] The globally aggregated graph-level representation vector and the time-series context features of the key settlement feature set are residually concatenated and nonlinearly activated. The final output is a spatiotemporal correlation embedding vector for bridge pier settlement that integrates spatiotemporal topological correlation and dynamic evolution characteristics. This embedding vector not only contains the temporal dynamic fluctuations of multi-source monitoring data, but also encodes the spatiotemporal topological correlation between features that conforms to mechanical causal logic at a deeper level, providing a high-fidelity, mechanistic initial state representation for subsequent settlement time series simulation.

[0063] Specifically, during the real-time monitoring of the bridge pier, a time-series monitoring sequence containing "eccentric stress-tilt synergy factor", "pore pressure-scour offset factor" and "base compaction deformation factor" is acquired. This sequence is continuously input at time steps t1, t2, ..., tn. At the same time, the sensitive feature topology network of the bridge pier, which represents "water flow scour (environment) - inclined leg bias pressure (load) - foundation compaction (settlement)," is input into the preset dynamic time-series space to prepare for the initiation of spatiotemporal graph calculation.

[0064] Within the dynamic temporal space, the structure of the bridge pier's sensitive feature topology network serves as an adjacency constraint. That is, the "pore pressure-scour offset factor node" and the "eccentric stress-tilt synergy factor node" can only transmit messages to the "base compaction deformation factor node," and are isolated in the opposite direction. When the flood season arrives, in the spatial dimension, the graph attention mechanism dynamically increases the spatial attention weight of the "pore pressure-scour offset factor" transmitted to the "base compaction deformation factor" based on the sudden change characteristics of the "pore pressure-scour offset factor" at the current time step, thereby achieving strong transmission of scour weakening information. At the same time, in the temporal dimension, the temporal attention mechanism captures that the inclined leg thrust of the bridge pier has accumulated over a long period in the early stage, and assigns a higher temporal attention weight to the "eccentric stress-tilt synergy factor" in the historical time step, thereby achieving dynamic extraction of complex temporal dependencies.

[0065] Through spatiotemporal dual message transmission, the "base compaction deformation factor node" deeply integrates the dynamic information of the upstream scour weakening state and the inclined leg bias state at each time step. With the fluctuation of the flood season water level and the fluctuation of the inclined leg thrust, the hidden state of each node is updated step by step in the dynamic temporal space, clearly presenting the topological state change of the multi-source feature variable "upstream scour and erosion intensification - inclined leg eccentric thrust amplification - base compaction deformation acceleration" in the temporal evolution. Adaptive weighted pooling aggregation is performed on the hidden state of all nodes in the bridge pier topology map to integrate the discrete node-level features into global features that characterize the overall settlement evolution of the bridge pier.

[0066] The features after global pooling are residually fused with the original temporal context features to determine the spatiotemporal correlation embedding vector of bridge pier settlement. This embedding vector highly condenses the dynamic process of the evolution of "base settlement features" of bridge pier due to the spatiotemporal coupling and superposition of "tilted leg bias features" and "scouring and erosion features". It eliminates the disordered fluctuations of isolated nodes and provides an initial embedding state with a clear mechanism and close spatiotemporal correlation for the next step of accurately predicting the future settlement trend of bridge pier under complex stress and erosion coupling.

[0067] Furthermore, the spatiotemporal correlation embedding vector is dynamically pre-simulated to simulate the multi-step forward evolution trajectory of the spatiotemporal correlation embedding vector. In addition, multi-level iterations are performed in combination with random perturbation terms. Thus, in the process of iteration, the time-series prediction results of bridge pier settlement are determined in combination with the constraint relationship of bridge pier settlement, thereby realizing the probabilistic dynamic inference of the future evolution trend of settlement.

[0068] At this point, the spatiotemporal related embedding vector is dynamically pre-simulated. The spatiotemporal related embedding vector is input into a preset autoregressive dynamic pre-simulation decoder. During the pre-simulation process, the embedding vector at the current time is used as the initial hidden state. The state transition function is executed through a recurrent neural network or a temporal Transformer architecture to forward extrapolate the hidden state at time t to time t+1. This process is repeated recursively to simulate the forward evolution trajectory of the spatiotemporal related embedding vector at multiple future time steps, thereby obtaining a preliminary state sequence representation of the bridge pier settlement in the future time domain.

[0069] To address the uncontrollability of the objective environment in which bridge piers are located and the slight fluctuations in monitoring data, a random disturbance term is introduced into the trajectory simulation of multi-step forward evolution. This random disturbance term is generated based on a Gaussian or Laplace distribution and is used to characterize unknown environmental noise and system nondeterministic factors. In each forward evolution iteration, the random disturbance term is coupled and superimposed with the output of the state transition function, so that the single deterministic evolution trajectory is expanded into a diffused state space containing multiple levels of uncertainty, thereby simulating multiple possible future evolution branches and enhancing the generalization and adaptability of the pre-simulation model to extreme or sudden working conditions.

[0070] During the multi-level iterative diffusion process, the diffusion state is dynamically corrected and screened by combining the physical and mechanical constraints of bridge pier settlement. The constraints include physical extreme value constraints of settlement rate, topological coordination constraints of differential settlement between adjacent nodes, and mechanical constitutive constraints of load-settlement. It is calculated whether the candidate hidden states generated by each iteration satisfy the above hard or flexible constraint boundaries. Evolutionary branches that violate physical and mechanical laws are penalized by weight decay or directly truncated and eliminated, while evolutionary branches that meet the constraints are encouraged by weighting, thereby converging the unconstrained random diffusion into a physically reasonable mechanical evolution envelope.

[0071] After multiple levels of constraint iteration, the effective evolutionary branches that survive the final stage are extracted. The distribution characteristics of the effective branches at each future time step are statistically estimated, and their expected values ​​and confidence intervals are calculated. The expected values ​​are used as the time series prediction results of bridge pier settlement, and the confidence intervals are used as the uncertainty measure of the prediction results. This enables a probabilistic dynamic extrapolation of the future evolution trend of settlement, and outputs a time series prediction sequence that includes both deterministic trends and quantitative risk intervals.

[0072] Specifically, the "spatiotemporal correlation embedding vector of bridge pier settlement, which encodes the dynamic correlation state of inclined leg bias-scour weakening-base compaction" is input into the dynamic pre-decoder built based on a long short-term memory network. Starting from the hidden state at the current time t, an autoregressive forward calculation is performed. Based on the historical evolution inertia, the decoder infers the state transition vector of the bridge pier base compaction characteristics and inclined leg tilt characteristics at time t+1. In this recursive manner, the preliminary forward evolution trajectory of the spatiotemporal correlation embedding vector of bridge pier along the time axis in the next 7 days is simulated.

[0073] Considering the unpredictable factors that bridge piers may encounter during the flood season, such as sudden flood peak scouring or ship collisions, a random perturbation term following a Gaussian distribution is introduced in each iteration of the autoregressive deduction. For example, when simulating the hidden state on the third day, the generated random perturbation vector is superimposed on the state transition result, causing the original single settlement deduction trajectory to split into multiple parallel evolution branches. Some branches represent the adverse working conditions of intensified flood peak scouring, while others represent the favorable working conditions of stable water level receding, thus constructing a bridge pier settlement diffusion state space that covers multiple levels of uncertainty.

[0074] To prevent random disturbances from causing the simulation results to violate the physical and mechanical principles of bridge piers, a settlement constraint relationship specific to bridge piers is introduced for dynamic correction. Specifically, "differential settlement constraint" and "deformation rate constraint" are applied: Since the inclined leg of the bridge pier is extremely sensitive to differential settlement, the constraint conditions stipulate that the differential settlement of the upstream and downstream nodes of the pier cap must not exceed the extreme value of the shear deformation of the concrete of the inclined leg; at the same time, it is stipulated that the daily settlement rate must not exceed the physical limit of soft soil consolidation. In multi-level iterations, if a random evolution branch shows a sudden change in differential settlement on the 5th day and exceeds the allowable extreme value, the branch is judged to violate the mechanical compatibility constraint, its weight is reduced to an extremely low level and it is eliminated, thereby constraining the random diffusion within the envelope space allowed by the mechanical properties of the bridge pier.

[0075] Probabilistic statistics were performed on dozens of effective evolutionary branches that survived after mechanical constraint screening; the mathematical expectation and 95% confidence interval of the settlement values ​​of all effective branches at each time step in the next 7 days were calculated; the mathematical expectation sequence was determined as the time series prediction result of bridge pier settlement, and the width of the 95% confidence interval was used as the uncertainty risk indicator of future settlement evolution; thus, a probabilistic dynamic extrapolation of the future evolution trend of bridge pier settlement was realized, which not only predicted the most likely settlement trend, but also quantitatively assessed the extreme risk of settlement under complex inclined leg stress and scour disturbance.

[0076] refer to Figure 5 In step S14, the specific steps are as follows: S141: Dynamically identify the time-series prediction results, and combine the modal decomposition mechanism to determine the settlement trend of the bridge piers during the identification process. At the same time, based on the periodic content of ambient temperature and the time-series fluctuation content of water level rise and fall, dynamically construct the dynamic baseline of ambient temperature and water level that is adaptively adjusted with the season and hydrological cycle. Perform differential comparison on the settlement trend, ambient temperature and water level dynamic baseline to eliminate environmental baseline interference, thereby determining the actual settlement deviation. S142: Sensitivity decoupling is performed on the actual settlement deviation, and multiple settlement influencing factors are determined by combining causal tracing; at the same time, real-time images of the bridge pier are collected, multiple visual features are determined based on the recognition of the real-time images, and cross-modal fusion is performed by combining multiple settlement influencing factors to determine the settlement monitoring events of the bridge pier.

[0077] In the embodiments of this application, the time-series prediction results are dynamically identified, and the settlement trend of the bridge pier is determined by combining the modal decomposition mechanism during the identification process. At the same time, based on the periodicity of the ambient temperature and the temporal fluctuation of the water level, a dynamic baseline of ambient temperature and water level is dynamically constructed to adapt to the season and hydrological cycle. The settlement trend, ambient temperature and water level dynamic baseline are differentially compared to eliminate environmental baseline interference, thereby determining the actual settlement deviation. This approach takes into account the overall consideration of the settlement trend, ambient temperature and water level dynamic baseline, ensuring the accuracy of the actual settlement deviation.

[0078] At this point, the output bridge pier settlement time series prediction results are dynamically identified. In the identification process, a mode decomposition mechanism is introduced to decompose the complex time series prediction signal containing multi-band oscillations and nonlinear drift into multiple intrinsic mode functions and a residual trend component. By removing the intrinsic mode functions that reflect high-frequency vibrations and short-term fluctuations, the residual trend component that reflects the long-period, monotonic evolution law of the bridge pier is extracted. This residual trend component is determined as the settlement trend term of the bridge pier, thereby removing the influence of random fluctuations and revealing the inherent macroscopic evolution trend of settlement.

[0079] Historical and real-time monitoring data of ambient temperature and water level are acquired synchronously, and the periodicity of ambient temperature and the temporal fluctuation of water level are analyzed. Based on the daily / annual periodic Fourier expansion characteristics of ambient temperature and the hydrological cycle and rainfall-driven characteristics of water level fluctuations, a parameterized benchmark model that adaptively adjusts with seasonal changes and hydrological cycles is constructed. Through time-varying estimation of dynamic model parameters, the dynamic evolution trajectory of ambient temperature and water level under natural periodic laws is fitted, thereby generating a dynamic benchmark line of ambient temperature and water level that adaptively adjusts with time. This benchmark line accurately represents the normal threshold envelope of elastic deformation and buoyancy changes of bridge piers caused by purely environmental factors.

[0080] The extracted bridge pier settlement trend term is compared differentially with the constructed dynamic baseline of ambient temperature and water level. Based on the coefficient of thermal expansion and hydrostatic formula, the dynamic baseline is converted into temperature-induced deformation equivalent and water level-induced buoyancy deformation equivalent. The above two deformation equivalents are subtracted from the settlement trend term to achieve physical elimination of environmental baseline interference. The differential comparison process is essentially to remove the pseudo-settlement caused by structural thermal expansion and contraction due to temperature cycle changes, and the pseudo-settlement caused by elastic rebound due to changes in base buoyancy caused by water level fluctuations.

[0081] After differential comparison to eliminate environmental baseline interference, the remaining difference sequence is the true deformation expression after eliminating the influence of temperature and hydrological seasonality. This difference sequence is determined as the actual settlement deviation of the bridge pier. The actual settlement deviation purely reflects the inelastic permanent settlement offset driven by ground consolidation, structural plastic damage or mechanical instability, providing a clean data basis for subsequent causal inversion of settlement influencing factors.

[0082] Specifically, the predicted 7-day settlement probability sequence of the bridge piers is dynamically identified. A variational mode decomposition mechanism is used to decompose the predicted sequence into high-frequency modes reflecting vehicle load and wind vibration, mid-frequency modes reflecting the daily periodic temperature difference, and long-period residual components reflecting the slow deformation of the foundation. The long-period residual components are extracted as the settlement trend term of the bridge piers. This trend term excludes the instantaneous high-frequency swaying of the inclined leg structure under vehicle eccentric loading and clearly presents the macroscopic sinking trend of the bridge piers under eccentric thrust.

[0083] To address the extreme sensitivity of the inclined legs of bridge piers to temperature gradients and the significant impact of buoyancy on the abutments, this study analyzes the annual and daily Fourier coefficients of ambient temperature, as well as the non-steady fluctuations of water levels during the flood season. An adaptive benchmark model is constructed to dynamically output dynamic benchmarks for the ambient temperature and water level of the bridge piers based on the current season (e.g., the high-temperature expansion period of summer) and the flood season water level stage (e.g., the high water level during the main flood season). For example, in the midsummer afternoon, this benchmark automatically calculates the equivalent displacement of the pier top deflecting towards the shady side due to the elongation of the inclined legs on the sun-facing side, and the equivalent of the slight uplift caused by the upward buoyancy provided by the high water level to the abutments.

[0084] The settlement trend of the bridge pier is compared with the dynamic baseline using differential methods. Specifically, based on the coefficient of linear expansion of concrete, the temperature dynamic baseline is converted into the equivalent of the horizontal and vertical displacement of the pier top caused by the thermal deflection of the inclined leg. Based on Archimedes' principle, the water level dynamic baseline is converted into the equivalent of the elastic uplift caused by the reduction of the effective stress of the foundation base. The above-mentioned thermal deformation equivalent and water buoyancy equivalent are deducted from the settlement trend of the bridge pier in real time to accurately eliminate the interference of pseudo settlement caused by the elongation of the inclined leg due to high summer temperatures and pseudo uplift caused by the rebound of the foundation base due to high water levels during the flood season.

[0085] After differential comparison to eliminate environmental interference, the obtained residual sequence is determined as the actual settlement deviation of the bridge pier. If the actual settlement deviation continues to diverge downward after the floodwaters recede, it indicates that the bridge pier has undergone irreversible basic plastic consolidation settlement under the coupled effect of "slanted leg thrust + scour weakening". This completely eliminates the misjudgment of false settlement caused by "temperature drop and shrinkage" or "water level drop and loss of buoyancy", and provides a real and reliable deviation indicator for disaster early warning of the bridge pier.

[0086] Furthermore, sensitivity decoupling is performed on the actual settlement deviation, and multiple settlement influencing factors are identified by combining causal tracing. At the same time, real-time images of the bridge piers are collected, and multiple visual features are identified based on the recognition of these real-time images. Cross-modal fusion is then performed in combination with multiple settlement influencing factors to determine the settlement monitoring events of the bridge piers. This further controls the time-series prediction results of bridge pier settlement, fully considers multiple settlement influencing factors and real-time images of the bridge piers, and improves the accuracy of bridge pier settlement monitoring events.

[0087] At this point, sensitivity decoupling is performed on the actual settlement deviation, and the overall deviation sequence is reverse-mapped to the latent space where the sensitive feature topology network is located. Based on the directed edge weights and node feature gradients of the sensitive feature topology network, the partial derivatives of each source driving force with respect to the final deviation output are calculated as sensitivity coefficients. Through orthogonal decomposition and variance contribution rate decomposition, the actual settlement deviation is deconstructed into a linear superposition form of multiple independent driving forces, breaking the causal aliasing state caused by multi-field coupling and clarifying the independent contribution of each single factor to the macroscopic deviation.

[0088] Based on sensitivity decoupling, multiple settlement influencing factors are identified by combining causal tracing mechanism; the directed edges of the sensitive feature topology network are traced back to the source nodes of environmental erosion and load transfer, starting from the settlement deviation result node; the causal effect intensity of each source variable on the deviation result is quantified according to the cumulative product of sensitivity coefficients on the reverse tracing path; the source variables with causal effect intensity exceeding the preset significance threshold are screened out and identified as multiple settlement influencing factors that lead to the current actual settlement deviation, thus achieving accurate physical cause determination from "effect" to "cause".

[0089] Real-time images of the bridge piers are acquired synchronously and input into a preset computer vision feature extraction network. Through semantic segmentation and target detection, the structural surface and surrounding environment in the images are analyzed at the pixel level to extract and determine multiple visual features. These visual features include "structural appearance anomaly features, such as crack direction, width, and spalling area" and "environmental morphology evolution features, such as the depth and range of scour pits, waterline offset, and vegetation cover changes." This transforms the intuitive appearance of the physical world into a high-dimensional semantic feature vector.

[0090] The numerical features of multiple settlement influencing factors identified through causal tracing are fused with multiple visual features extracted from real-time images across modalities. A multimodal alignment and interactive attention mechanism is constructed to calculate the cross-modal similarity matrix between numerical factor features and visual semantic features. Attention weighting is used to achieve deep semantic complementarity and consistency verification of heterogeneous features. The fused unified representation is input into an event classifier, and settlement monitoring events of bridge piers are determined based on the physical patterns of feature co-expression. The settlement monitoring events include event type, inducing cause, and severity level, thereby completing a closed-loop decision-making process from data decoupling and tracing to physical event identification.

[0091] For settlement monitoring events, numerical factor features and visual feature vectors are concatenated to form multimodal fusion features, which are then input into a classifier. The classifier outputs the event level according to the following rules: when scour is the dominant factor, scour pits appear with a depth exceeding 0.5 meters, and settlement exceeds 3 millimeters for three consecutive days, it is judged as a "scour-induced settlement event," with a Level II orange alert; when bias pressure is the dominant factor, crack width exceeds 0.2 millimeters, and settlement exceeds 5 millimeters for three consecutive days, it is judged as a "bias pressure-induced settlement event," with a Level I red alert; when both reach the threshold simultaneously, it is judged as a "scour-bias pressure coupled settlement event," with the highest level alert.

[0092] Specifically, sensitivity decoupling is performed on the "actual settlement deviation of the bridge pier, such as the sequence of continued settlement after removing the interference of temperature and water level". This deviation sequence is then mapped inversely to the sensitive feature topology network of the bridge pier. The gradients of the "pore pressure-scour offset factor" and "eccentric stress-tilt synergy factor" with respect to the deviation are calculated based on the chain rule. Through orthogonal decomposition, it is quantified that about 70% of the variance contribution in the current deviation is caused by scour weakening and about 30% is caused by the redistribution of eccentric thrust of the inclined leg. Thus, the originally intertwined "scour-eccentric pressure" coupled deviation is decoupled into two independent driving force components.

[0093] Regarding the statement that "scour weakening contributes 70% to the variance," during the monitoring period, 70% of the fluctuations and trends in bridge pier settlement deviations can be explained by scour. Specifically, when the flood season arrives and water flow velocity increases, the scour factor rises, and the settlement deviation increases accordingly; when the flood season recedes and scour weakens, the scour factor decreases, and the settlement deviation decreases or slows down. This means that scour is currently the main dynamic driving force of settlement, and its changes directly dominate the real-time fluctuations in settlement.

[0094] Regarding the "30% contribution rate of the eccentricity redistribution variance," the remaining 30% of settlement fluctuations are explained by the redistribution of eccentric thrust from the inclined leg. This portion of settlement typically manifests as long-term, cumulative, and trending subsidence, rather than short-term fluctuations; a certain eccentricity settlement base already exists before scour occurs; after scour occurs, the eccentricity effect may be amplified, but statistically it can still be separated into independent components.

[0095] The decomposition of 70% and 30% indicates that scouring is the main factor dominating short-term fluctuations, while bias is an important background factor. The two work together, but changes in scouring are more sensitive to the real-time state of settlement.

[0096] Based on the above decoupling results, causal tracing was performed; tracing back along the directed edges of the bridge pier topology network, starting from the "base compaction deformation node", the causal source was traced to the "upstream side node of the abutment bottom surface (representing scour)" and the "root node of the pier's inclined leg (representing eccentric pressure)". The causal effect intensity of the scour path was calculated to be 0.65, and the causal effect intensity of the eccentric pressure path was 0.28, both of which exceeded the significance threshold of 0.2. Thus, the two main influencing factors leading to the current deviation of the bridge pier settlement were determined to be "local scour and hollowing of the upstream foundation" and "long-term eccentric stress accumulation at the root of the inclined leg".

[0097] The monitoring video streams of the inclined leg root and the water-facing side of the pier of the bridge were retrieved simultaneously, and real-time images were extracted at a fixed frame rate. The DeepLabV3+ semantic segmentation network and YOLOv8 defect detection network were used to identify a longitudinal pressure crack with a width of 0.8 mm on the inner side of the inclined leg root, which is an abnormal visual feature. At the same time, the obvious outline of the scour pit and the deviation of the waterline from its original position in the riverbed on the upstream side of the pier were identified, which is an environmental morphology visual feature. These visual features were encoded into corresponding semantic feature vectors.

[0098] The numerical features (scour factor and bias factor) and visual feature vectors (crack features and scour pit features) are fused across modally. Through a cross-attention mechanism, the scour factor generates a strong attention response to the visual scour pit outline, and the bias factor generates a strong response to the visual scour crack, thus achieving mutual verification of heterogeneous data. After the fused features are input into the classifier, the current pattern is identified as conforming to the physical pattern of "environmental erosion dominance + load stress catalysis". The settlement monitoring event is finally determined to be "bias instability settlement event caused by upstream foundation scour of bridge pier", and the severity level is judged as Level II, such as an orange warning. This completes the complete technical closed loop from numerical deviation perception to physical event confirmation.

[0099] Please see Figure 6The bridge pier settlement monitoring system based on multi-source data and time-series analysis is applied to the aforementioned bridge pier settlement monitoring method based on multi-source data and time-series analysis; the bridge pier settlement monitoring system based on multi-source data and time-series analysis includes: The monitoring data module 21 is used to acquire multiple monitoring data of the bridge pier during the monitoring process, align the multiple monitoring data, and construct the corresponding bridge pier settlement multi-field coupling characterization space by combining the registration mechanism. Topology network module 22 is used to construct a directed acyclic graph based on the mechanical evolution mechanism of bridge pier settlement, and combine the multi-source feature variables extracted from the multi-field coupling characterization space of bridge pier settlement to determine the corresponding key settlement feature set, and then construct a sensitive feature topology network that characterizes the spatiotemporal evolution path of load-environment-settlement. The time series prediction module 23 is used to acquire the time series monitoring sequence of the bridge pier body, and combine it with the sensitive feature topology network input to the preset dynamic time series space to further determine the spatiotemporal correlation embedding vector of the bridge pier body settlement; and determine the time series prediction result of the bridge pier body settlement based on the dynamic pre-simulation of the spatiotemporal correlation embedding vector. Settlement monitoring module 24 is used to determine settlement trend items based on the identification of time-series prediction results, and to determine the actual settlement deviation content in combination with the dynamic baseline of ambient temperature and water level. Based on the identification of the actual settlement deviation content, multiple settlement influencing factors are determined, and the settlement monitoring events of the bridge pier are determined in combination with the real-time images of the bridge pier.

[0100] It should be noted that although multiple modules are mentioned in the detailed description above, this division is not mandatory; in fact, according to the embodiments of this disclosure, the features and functions of two or more modules or described above can be embodied in one module; conversely, the features and functions of one module described above can be further divided into multiple modules to be embodied.

[0101] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the disclosure herein; this application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein; the specification and embodiments are to be considered exemplary only.

[0102] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for monitoring bridge pier settlement based on multi-source data and time-series analysis, characterized in that, include: During the monitoring of bridge piers, multiple monitoring data of the bridge piers are acquired, the multiple monitoring data are aligned, and a corresponding multi-field coupled characterization space for bridge pier settlement is constructed by combining the registration mechanism. Based on the mechanical evolution mechanism of bridge pier settlement, a directed acyclic graph is constructed, and the multi-source feature variables extracted from the multi-field coupling characterization space of bridge pier settlement are combined to determine the corresponding key settlement feature set, and then a sensitive feature topology network characterizing the spatiotemporal evolution path of load-environment-settlement is constructed. The temporal monitoring sequence of the bridge pier is obtained and combined with the sensitive feature topology network input into the preset dynamic temporal space to further determine the spatiotemporal correlation embedding vector of the bridge pier settlement; the temporal prediction result of the bridge pier settlement is determined based on the dynamic pre-simulation of the spatiotemporal correlation embedding vector. Settlement trend terms are identified based on time-series prediction results, and actual settlement deviations are determined by combining ambient temperature and water level dynamic baselines. Multiple settlement influencing factors are identified based on the identification of actual settlement deviations, and settlement monitoring events of bridge piers are determined by combining real-time images of bridge piers.

2. The bridge pier settlement monitoring method based on multi-source data and time series analysis according to claim 1, characterized in that, During the monitoring of bridge piers, multiple monitoring data points of the bridge piers are acquired. These multiple monitoring data points are aligned, and a corresponding multi-field coupled characterization space for bridge pier settlement is constructed using a registration mechanism. This includes: The bridge piers are monitored, and multiple monitoring data are acquired during the monitoring process. These multiple monitoring data cover stress field, temperature field, seepage field and deformation field. The heterogeneous data registration mechanism based on spatiotemporal anchor points is used to trigger the spatiotemporal alignment of multiple monitoring data and output the aligned data.

3. The bridge pier settlement monitoring method based on multi-source data and time series analysis according to claim 2, characterized in that, The process of monitoring bridge piers involves acquiring multiple monitoring data points, aligning these data points, and constructing a corresponding multi-field coupled characterization space for bridge pier settlement using a registration mechanism. This also includes: The aligned data are projected into a high-dimensional latent space to extract cross-modal local manifold features. The manifold features of each part are deeply fused, and the scale differences and coupling interference between multiple data are eliminated. In this way, a multi-field coupled characterization space for bridge pier settlement with spatiotemporal consistency is constructed.

4. The bridge pier settlement monitoring method based on multi-source data and time series analysis according to claim 1, characterized in that, The mechanical evolution mechanism based on bridge pier settlement is used to construct a directed acyclic graph, and multi-source feature variables extracted from the multi-field coupling characterization space of bridge pier settlement are combined to determine the corresponding key settlement feature set. This leads to the construction of a sensitive feature topology network characterizing the spatiotemporal evolution path of load-environment-settlement, including: Based on the traversal of the bridge pier settlement database, the mechanical evolution mechanism of bridge pier settlement is determined. Based on the analysis of the mechanical evolution mechanism of bridge pier settlement, the causal logic of load transfer, environmental erosion and ground consolidation is determined. The corresponding directed acyclic graph is constructed in combination with the three-dimensional morphology of the bridge pier.

5. The bridge pier settlement monitoring method based on multi-source data and time series analysis according to claim 4, characterized in that, The method constructs a directed acyclic graph based on the mechanical evolution mechanism of bridge pier settlement, and determines the corresponding key settlement feature set by combining multi-source feature variables extracted from the multi-field coupling characterization space of bridge pier settlement. This leads to the construction of a sensitive feature topology network characterizing the spatiotemporal evolution path of load-environment-settlement, and further includes: Multi-source feature variables reflecting the multi-field coupling effect are extracted from the multi-field coupling representation space of bridge pier settlement. These multi-source feature variables are mapped to key nodes of a directed acyclic graph, and the corresponding causal evolution content is marked. Redundant and pseudo-correlated features are removed simultaneously to determine the key settlement feature set that represents the driving core of bridge pier settlement. Using the key settlement feature set as edges and nodes, and combining the mechanical causal path in the directed acyclic graph, a sensitive feature topology network representing the spatiotemporal evolution path of load-environment-settlement is constructed.

6. The bridge pier settlement monitoring method based on multi-source data and time series analysis according to claim 1, characterized in that, The process involves acquiring the temporal monitoring sequence of the bridge pier body and inputting it into a preset dynamic temporal space using a sensitive feature topology network to further determine the spatiotemporal correlation embedding vector of the bridge pier body settlement. The time-series prediction results of bridge pier settlement are determined based on the dynamic pre-simulation of the spatiotemporal correlation embedding vector, including: In the settlement monitoring of bridge piers, the time-series monitoring sequence of bridge piers is acquired, and the time-series monitoring sequence and the sensitive feature topology network are input into a preset dynamic time-series space. In the dynamic time-series space, the topology structure of the sensitive feature topology network is used as the adjacency constraint. Through the graph attention mechanism, the time-series monitoring sequence is dynamically message-passed and feature-aggregated along the spatiotemporal dimension, presenting the topological state changes of multi-source feature variables in the time-series evolution. Then, the spatiotemporal correlation embedding vector of bridge pier settlement, which integrates spatiotemporal topological correlation and dynamic evolution characteristics, is determined.

7. The bridge pier settlement monitoring method based on multi-source data and time series analysis according to claim 6, characterized in that, The process of acquiring the time-series monitoring sequence of the bridge pier and inputting it into a preset dynamic time-series space in conjunction with a sensitive feature topology network to further determine the spatiotemporal correlation embedding vector of the bridge pier settlement; and determining the time-series prediction result of the bridge pier settlement based on the dynamic pre-simulation of this spatiotemporal correlation embedding vector, further includes: Dynamic pre-simulation of spatiotemporal related embedding vectors is performed to simulate the multi-step forward evolution trajectory of spatiotemporal related embedding vectors. Multi-level iteration is carried out in combination with random perturbation terms. In this process, the temporal prediction results of bridge pier settlement are determined by combining the constraint relationship of bridge pier settlement, so as to realize the probabilistic dynamic inference of the future evolution trend of settlement.

8. The method for monitoring bridge pier settlement based on multi-source data and time series analysis according to claim 1, characterized in that, The settlement trend is identified based on the time-series prediction results, and the actual settlement deviation is determined by combining the ambient temperature and water level dynamic baseline. Multiple settlement influencing factors are identified based on the identification of the actual settlement deviation, and settlement monitoring events of the bridge piers are determined by combining real-time images of the bridge piers, including: The time-series prediction results are dynamically identified, and the settlement trend of the bridge pier is determined by combining the modal decomposition mechanism during the identification process. At the same time, based on the periodic content of ambient temperature and the time-series fluctuation content of water level rise and fall, dynamic baselines of ambient temperature and water level are dynamically constructed to adapt to the season and hydrological cycle. The settlement trend, ambient temperature and water level dynamic baselines are differentially compared to eliminate environmental baseline interference, thereby determining the actual settlement deviation.

9. The bridge pier settlement monitoring method based on multi-source data and time series analysis according to claim 8, characterized in that, The process of identifying settlement trend items based on time-series prediction results, determining actual settlement deviations by combining environmental temperature and water level dynamic baselines, identifying multiple settlement influencing factors based on the identification of actual settlement deviations, and determining settlement monitoring events of bridge piers by combining real-time images of the bridge piers, also includes: Sensitivity decoupling is performed on the actual settlement deviation, and multiple settlement influencing factors are identified by combining causal tracing. At the same time, real-time images of the bridge piers are collected, and multiple visual features are identified based on the recognition of these real-time images. These features are then combined with multiple settlement influencing factors for cross-modal fusion to determine the settlement monitoring events of the bridge piers.

10. A bridge pier settlement monitoring system based on multi-source data and time-series analysis, characterized in that, The bridge pier settlement monitoring system based on multi-source data and time series analysis is applied to the bridge pier settlement monitoring method based on multi-source data and time series analysis as described in any one of claims 1-9. The bridge pier settlement monitoring system based on multi-source data and time-series analysis includes: The monitoring data module is used to acquire multiple monitoring data of the bridge pier during the monitoring process, align the multiple monitoring data, and construct the corresponding multi-field coupled characterization space of bridge pier settlement by combining the registration mechanism. The topology network module is used to construct a directed acyclic graph based on the mechanical evolution mechanism of bridge pier settlement, and to determine the corresponding key settlement feature set by combining the multi-source feature variables extracted from the multi-field coupling characterization space of bridge pier settlement, thereby constructing a sensitive feature topology network that characterizes the spatiotemporal evolution path of load-environment-settlement. The time-series prediction module is used to acquire the time-series monitoring sequence of the bridge pier body, and combine it with the sensitive feature topology network input into the preset dynamic time-series space to further determine the spatiotemporal correlation embedding vector of the bridge pier body settlement; the time-series prediction result of the bridge pier body settlement is determined based on the dynamic pre-simulation of the spatiotemporal correlation embedding vector. The settlement monitoring module is used to identify settlement trend items based on the time-series prediction results, and to determine the actual settlement deviations by combining the dynamic baseline of ambient temperature and water level. Based on the identification of the actual settlement deviations, multiple settlement influencing factors are identified, and the settlement monitoring events of the bridge piers are determined by combining real-time images of the bridge piers.

Images