Bridge Settlement and Deformation Monitoring System Based on Multi-Source Sensor Data Fusion

Abstract

This invention relates to the field of bridge structure monitoring technology, specifically a bridge settlement and deformation monitoring system based on multi-source sensor data fusion. The system includes: a data acquisition module that deploys multiple types of sensors at key monitoring locations on the bridge to acquire multi-source sensor data; a data processing module that performs time-series alignment and noise suppression on the data to generate a preprocessed sequence; a feature extraction module that extracts settlement and deformation feature signals to form corresponding feature vectors; a fusion evaluation module that uses an improved fusion decision algorithm combined with prior knowledge of bridge structural mechanics to collaboratively evaluate the feature vectors and generate a fused settlement and deformation state evaluation quantity; a prediction module that calculates the deformation development trend prediction trajectory based on a state evolution prediction model; and a report generation module that visualizes and encapsulates the evaluation quantity and predicted trajectory to form a monitoring report. This system achieves multi-source data fusion evaluation and trend prediction of bridge settlement and deformation, improving the comprehensiveness and foresight of deformation monitoring.

Description

Technical Field

[0001] This invention relates to the field of bridge structure monitoring technology, and in particular to a bridge settlement and deformation monitoring system based on multi-source sensor data fusion. Background Technology

[0002] Conventional bridge settlement and deformation monitoring often relies on the deployment of single-type sensors to collect data. The acquired sensor data undergoes only basic filtering, feature extraction only separates settlement or deformation-related signals, data fusion uses conventional weighted calculation methods, and deformation state prediction can only determine the numerical value at the current moment. The monitoring results are ultimately presented in simple data reports. Furthermore, the time-series matching processing of multi-source sensor data lacks standardized procedures, noise interference cannot be effectively suppressed, and a single feature signal cannot fully characterize the deformation state of the bridge structure.

[0003] Conventional data fusion methods fail to analyze the inherent mechanical properties of the bridge structure. Settlement and deformation characteristic data are simply overlaid, resulting in assessments that do not accurately reflect the actual stress and deformation state of the bridge. Deformation monitoring can only provide real-time feedback and cannot predict subsequent changes. Furthermore, the visualization of monitoring data is limited and cannot intuitively demonstrate deformation assessment and trend evolution information.

[0004] It is necessary to rely on an appropriate fusion decision-making approach combined with prior knowledge of bridge structural mechanics to conduct collaborative analysis of settlement and deformation characteristic vectors to form a comprehensive assessment. Based on the corresponding prediction model and the fusion assessment results, the subsequent development trajectory of bridge settlement and deformation needs to be deduced. Finally, the assessment and prediction results are integrated to form a standardized monitoring report. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and propose a bridge settlement and deformation monitoring system based on multi-source sensor data fusion.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a bridge settlement and deformation monitoring system based on multi-source sensor data fusion, comprising:

[0007] The data acquisition module is used to deploy various types of sensors at key monitoring locations of the target bridge to acquire multi-source sensor data of the bridge structure under load.

[0008] The data processing module is used to perform time-series alignment and noise suppression preprocessing on the multi-source sensor data to generate a preprocessed multi-source sensor data sequence.

[0009] The feature extraction module is used to extract the settlement feature signal and deformation feature signal from the preprocessed multi-source sensing data sequence to form a settlement feature vector and a deformation feature vector.

[0010] The fusion evaluation module is used to input the settlement feature vector and deformation feature vector into the improved fusion decision algorithm. The improved fusion decision algorithm performs a collaborative evaluation of the settlement feature vector and deformation feature vector based on prior knowledge of bridge structural mechanics to generate a fusion settlement and deformation state evaluation quantity.

[0011] The prediction module is used to calculate the development trend prediction trajectory of bridge settlement deformation based on the integrated settlement deformation state assessment quantity and through the state evolution prediction model.

[0012] The report generation module is used to visualize and encapsulate the fused settlement deformation state assessment quantity and the predicted trajectory of bridge settlement deformation development trend to form a bridge settlement deformation monitoring report.

[0013] As a further aspect of the present invention, the deployment of various types of sensors at key monitoring locations of the target bridge to acquire multi-source sensing data of the bridge structure under load includes:

[0014] Based on the finite element model analysis results of the bridge structure, the locations where stress concentration, deformation sensitivity, or theoretical settlement value is greatest under the expected load are identified, and these locations are marked as key monitoring locations.

[0015] At the marked key monitoring locations, sensor units are deployed in groups. Each sensor unit includes a hydrostatic level, tilt sensor, strain gauge and temperature sensor. Global navigation satellite system reference stations and monitoring stations are also fixedly deployed at the piers at both ends of the bridge.

[0016] A unified sensor data acquisition cycle and trigger threshold are set. When the reading of any sensor exceeds its trigger threshold, all sensors are automatically and synchronously triggered to perform a high-frequency data acquisition, thereby obtaining high-time-resolution multi-source sensor data of the bridge under dynamic load response.

[0017] As a further aspect of the present invention, the multi-source sensing data is preprocessed with time alignment and noise suppression to generate a preprocessed multi-source sensing data sequence, including:

[0018] Establish an independent data buffer for each type of sensor data stream to receive the raw data stream from the corresponding sensor;

[0019] Based on a unified Coordinated Universal Time, the raw data streams in all data buffers are timestamped to ensure that data from different sensors have a unified time reference.

[0020] For each type of sensor data after timestamp calibration, a noise suppression filter matching the sensor's physical characteristics is applied, the noise suppression filter including a low-pass filter for high-frequency vibration data and a sliding mean filter for static strain data;

[0021] The filtered data output from the noise suppression filter is resampled at equal intervals according to a preset fixed time interval to generate the preprocessed multi-source sensor data sequence with strict time alignment and improved signal-to-noise ratio.

[0022] As a further aspect of the present invention, the settlement feature signal and deformation feature signal are extracted from the preprocessed multi-source sensing data sequence to form a settlement feature vector and a deformation feature vector, including:

[0023] The preprocessed multi-source sensor data sequence is analyzed to identify the tilt angle change time series data from the tilt sensor and the relative elevation difference time series data from the hydrostatic level. The tilt angle change time series data and the relative elevation difference time series data are used as the original settlement characteristic signals.

[0024] The preprocessed multi-source sensing data sequence is analyzed to identify micro-strain time-series data from strain sensors and relative displacement time-series data from displacement gauges. The micro-strain time-series data and relative displacement time-series data are used as the original deformation feature signals.

[0025] For the original settlement feature signal, calculate its mean, variance and cumulative offset from the historical benchmark value within a fixed time window, normalize the mean, variance and cumulative offset and concatenate them to generate the settlement feature vector.

[0026] For the original deformation feature signal, calculate its maximum value, signal energy, and slope of change within a fixed time window. Then, normalize and concatenate the maximum value, signal energy, and slope of change to generate the deformation feature vector.

[0027] As a further aspect of the present invention, the improved fusion decision algorithm is an algorithm that collaboratively evaluates the settlement feature vector and the deformation feature vector based on prior knowledge of bridge structural mechanics. Its working principle includes:

[0028] Construct a knowledge-enhanced fusion evaluation network, which includes a physical constraint layer and a data-driven layer;

[0029] The physical constraint layer stores the theoretical relationship matrix of settlement and deformation of the bridge structure under typical load conditions. The theoretical relationship matrix of settlement and deformation describes the mechanical balance and geometric compatibility relationship that should be satisfied between the settlement and each deformation component.

[0030] In the data-driven layer, the settlement feature vector and deformation feature vector are received as inputs, and the actual correlation matrix between the input features is calculated.

[0031] The actual correlation matrix is ​​compared with the settlement and deformation theoretical relationship matrix retrieved from the physical constraint layer, and the inconsistency measure between the actual correlation matrix and the settlement and deformation theoretical relationship matrix is ​​calculated.

[0032] Based on the inconsistency metric, the fusion weights of each component of the settlement feature vector and the deformation feature vector in the data-driven layer are dynamically adjusted so that the fusion process tends to satisfy the results of theoretical mechanical relationships.

[0033] Based on the adjusted fusion weights, the settlement feature vector and deformation feature vector are weighted and aggregated and confidence propagated, and finally a scalar that comprehensively considers the consistency between actual observation and physical laws is output, namely the fused settlement and deformation state assessment quantity.

[0034] As a further aspect of the present invention, based on the inconsistency metric, the fusion weights of each component of the settlement feature vector and the deformation feature vector in the data-driven layer are dynamically adjusted so that the fusion process tends to satisfy the results of theoretical mechanical relationships, including:

[0035] The inconsistency metric is defined as the weighted norm of the differences between corresponding elements of the actual correlation matrix and the theoretical relationship matrix between settlement and deformation;

[0036] The inconsistency metric is input into a weight adjustment function, which is designed to be a monotonically decreasing function of the inconsistency metric, such that the output global fusion weight adjustment factor decreases as the inconsistency metric increases.

[0037] Using the global fusion weight adjustment factor, the initial fusion weight matrix between each component of the settlement feature vector and the deformation feature vector, which were originally calculated based on the data self-attention mechanism in the data-driven layer, is scaled to generate a physically guided intermediate fusion weight matrix.

[0038] The prior theoretical relation vector corresponding to the current input feature is queried from the physical constraint layer, and the dot product similarity between the prior theoretical relation vector and each component of the settlement feature vector and deformation feature vector is calculated.

[0039] The dot product similarity is normalized to generate a theoretical consistency weight vector;

[0040] The physical-guided intermediate fusion weight matrix is ​​multiplied by the theoretical consistency weight vector using Hadamard multiplication. This enhances the weights of components with high theoretical consistency and suppresses the weights of components with low theoretical consistency, thereby generating the final dynamically adjusted fusion weight matrix.

[0041] As a further aspect of the present invention, based on the integrated settlement deformation state assessment quantity, the development trend prediction trajectory of bridge settlement deformation is calculated through a state evolution prediction model, including:

[0042] Collect historical data on the fusion settlement and deformation state assessment to form a time-series historical sequence of assessment data;

[0043] The historical sequence of the evaluation quantity is analyzed to identify its long-term trend component, periodic component, and random fluctuation component.

[0044] Construct a sequence modeling-based predictor, which takes the historical sequence of the evaluation quantity of the previous time window as input and maps the input sequence to the hidden state through an encoder;

[0045] In the decoder of the predictor, the latent state is gradually decoded by combining the extrapolation results of the long-term trend component and the periodic component separated from the historical sequence of the evaluation quantity, so as to generate the predicted values ​​of the state evaluation quantity at multiple consecutive time points in the future.

[0046] By connecting the predicted values ​​of the state assessment at multiple consecutive future time points, a development trend prediction trajectory reflecting the future change path of the bridge structure state is formed.

[0047] As a further aspect of the present invention, the historical sequence of the evaluation quantity is analyzed to identify its long-term trend component, periodic component, and random fluctuation component, including:

[0048] The historical sequence of the evaluation quantity is smoothed using a local weighted regression method to extract the long-term trend component that changes slowly and monotonically over time.

[0049] From the residual sequence after removing long-term trend components, the main period lengths are identified through spectral analysis, and a sine-cosine basis function set is constructed based on the identified period lengths.

[0050] The residual sequence is projected and fitted onto the sine-cosine basis function set to obtain the amplitude and phase parameters of the periodic component, thereby reconstructing the periodic component.

[0051] The random fluctuation component is the portion that cannot be explained by trends and cycles after subtracting the long-term trend component and the periodic component from the historical sequence of the original assessment quantity.

[0052] As a further aspect of the present invention, the fused settlement deformation state assessment quantity and the predicted trajectory of bridge settlement deformation development trend are visualized and encapsulated to form a bridge settlement deformation monitoring report, including:

[0053] Create an interactive visualization template, which includes a timeline, a status indicator area, and a trend preview area;

[0054] The fused settlement deformation state assessment quantity is mapped to the pointer angle and color of the dashboard in the state indication area, wherein the pointer angle is proportional to the value of the assessment quantity, and the color changes according to the predefined threshold range in which the assessment quantity is located.

[0055] The predicted trajectory of the bridge settlement deformation development trend is plotted in the trend preview area to form a prediction curve with a confidence interval, and the fused settlement deformation state assessment at the current time point is highlighted as the starting point of the prediction curve.

[0056] The visualization elements of the timeline, status indicator area, and trend preview area are combined and embedded with text descriptions containing data timestamps and monitoring location identifiers to jointly render and generate a graphical bridge settlement and deformation monitoring report.

[0057] As a further aspect of the present invention, the predicted trajectory of the bridge settlement deformation development trend is plotted in the trend preview area to form a prediction curve with a confidence interval, including:

[0058] While generating the predicted trajectory of the development trend, the uncertainty range of the predicted value corresponding to each future time point on the predicted trajectory is calculated through the internal uncertainty propagation mechanism of the state evolution prediction model.

[0059] Using the predicted value at each future time point as the center point, and taking half of the uncertainty range of the predicted value corresponding to the center point as the upper and lower offsets, the upper and lower boundaries of the confidence interval of the predicted value at the future time point are determined.

[0060] Connect the center points of all future time point predictions with a smooth curve to form the main prediction curve;

[0061] The area between the main prediction curve and the curves connecting the upper boundaries of all confidence intervals, as well as the curves connecting the lower boundaries of all confidence intervals, is filled with a semi-transparent color band. This semi-transparent color band is the visual representation of the confidence intervals.

[0062] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0063] The improved fusion decision algorithm uses prior knowledge of bridge structural mechanics to collaboratively evaluate settlement feature vectors and deformation feature vectors, generating a fused settlement and deformation state assessment quantity. The fusion process of settlement and deformation features conforms to the transmission logic of stress deformation in bridge structures. The analysis dimensions of feature vectors complement each other, and the assessment results can closely match the actual structural deformation state of the bridge. It avoids the state judgment deviation caused by unconstrained superposition in conventional feature fusion methods. The representation dimensions of settlement and deformation state are more comprehensive, the correlation analysis of feature data conforms to the operating law of structural mechanics, and the numerical change of the assessment quantity is synchronized with the actual deformation state of the bridge.

[0064] Based on the integrated settlement deformation state assessment, the development trend prediction trajectory of bridge settlement deformation is calculated through the state evolution prediction model. It can fully present the dynamic evolution process of bridge settlement deformation with time and load, breaking through the limitation of only monitoring the real-time deformation state. The trajectory data can continuously reflect the subsequent changes in deformation, and the temporal characteristics of deformation evolution are fully presented. The predicted trajectory can intuitively reflect the development trend of settlement deformation, and the dynamic change law of structural deformation is clearly shown. The application dimension of monitoring data extends from real-time state feedback to long-term situation prediction. Attached Figure Description

[0065] Figure 1 This is a time sequence diagram of the bridge settlement and deformation monitoring system based on multi-source sensor data fusion as described in this invention.

[0066] Figure 2 A flowchart for deploying sensors to acquire multi-source sensing data;

[0067] Figure 3 The flowchart illustrates the process of extracting feature vectors from settlement and deformation feature signals. Detailed Implementation

[0068] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0069] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0070] See Figure 1 This invention provides a bridge settlement and deformation monitoring system based on multi-source sensor data fusion, specifically comprising:

[0071] Multiple types of sensors are deployed at key monitoring locations on the target bridge. A data acquisition module acquires multi-source sensor data of the bridge structure under load. A data processing module performs time-series alignment and noise suppression preprocessing on the multi-source sensor data, generating a preprocessed multi-source sensor data sequence. A feature extraction module extracts settlement and deformation feature signals from the preprocessed multi-source sensor data sequence, constructing settlement and deformation feature vectors. A fusion evaluation module inputs the settlement and deformation feature vectors into an improved fusion decision algorithm. This algorithm, based on prior knowledge of bridge structural mechanics, collaboratively evaluates the settlement and deformation feature vectors, generating a fused settlement and deformation state assessment. A prediction module, based on the fused settlement and deformation state assessment, calculates the predicted trajectory of the bridge's settlement and deformation development trend using a state evolution prediction model. A report generation module visualizes and encapsulates the fused settlement and deformation state assessment and the predicted trajectory of the bridge's settlement and deformation development trend, forming a bridge settlement and deformation monitoring report.

[0072] In one embodiment of the present invention, see [reference] Figure 2 Based on the finite element model analysis results of the bridge structure, the locations of stress concentration, deformation sensitivity, or maximum theoretical settlement under the expected load are identified and marked as key monitoring locations. At the marked key monitoring locations, sensor units are deployed in groups, each containing a static level, tilt sensor, strain gauge, and temperature sensor. Global Navigation Satellite System reference stations and monitoring stations are fixedly deployed at the piers at both ends of the bridge. A unified sensor data acquisition cycle and trigger threshold are set. When the reading of any sensor exceeds its trigger threshold, all sensors are automatically and synchronously triggered to perform a high-frequency data acquisition, obtaining high-time-resolution multi-source sensor data of the bridge under dynamic load response. An independent data buffer is established for each type of sensor data stream to receive the raw data stream from the corresponding sensor. Based on a unified Coordinated Universal Time (UTC), the raw data streams in all data buffers are timestamped to ensure that the data from different sensors have a unified time reference. For each type of sensor data after timestamping calibration, a noise suppression filter that matches the sensor's physical characteristics is applied. This noise suppression filter includes a low-pass filter for high-frequency vibration data and a sliding mean filter for static strain data. From the filtered data output by the noise suppression filter, equal-interval resampling is performed at preset fixed time intervals to generate a preprocessed multi-source sensor data sequence with strict time alignment and improved signal-to-noise ratio.

[0073] In practical implementation, for a three-span prestressed concrete continuous beam bridge, the finite element model analysis results showed that under standard vehicle loads, the stress concentration coefficients were high at the lower edge of the mid-span section, the quarter-section of the side span, and the top of the piers, and the vertical displacement near the side supports was significant. These locations were identified as stress concentration or deformation-sensitive areas and marked as key monitoring locations. Sensing units were installed in groups at these key monitoring locations. Each sensing unit consisted of a hydrostatic level, a biaxial tilt sensor, a set of foil strain gauges, and a platinum resistance temperature sensor. Simultaneously, a global navigation satellite system (GNSS) reference station and three GNSS monitoring stations were fixedly installed on the abutments at both ends of the bridge. A unified basic data acquisition cycle of 1 second is set, and differentiated trigger thresholds are configured for various sensors: the elevation change trigger threshold for the hydrostatic level is set to 0.5 mm, the angle change trigger threshold for the tilt sensor is set to 0.02 degrees, and the micro-strain change trigger threshold for the strain gauge is set to 5 micro-strains. When the change in the real-time reading of any sensor relative to the previous sampling time exceeds its own trigger threshold, the data acquisition module will send a synchronization command to all sensors to start a high-frequency data acquisition that lasts for 10 seconds with a sampling frequency of 100 Hz, thereby capturing the dynamic response details of the bridge when sudden loads are applied or moving vehicles pass by.

[0074] In some embodiments, an independent circular data buffer is established for each type of sensor. The buffer depth for the hydrostatic level is set to 1000 data points, the buffer depth for the tilt sensor is set to 2000 data points, the buffer depth for the strain gauge is set to 1500 data points, and the buffer depth for the temperature sensor is set to 500 data points. Each buffer continuously receives the raw data stream with local clock timestamps uploaded by the corresponding sensor. The data preprocessing subsystem connects to a high-precision BeiDou time synchronization module, which assigns a unified time tag to the data packets in all data buffers based on Coordinated Universal Time (UTC), and corrects for millisecond-level timestamp deviations caused by internal clock drift in the sensors, ensuring that the elevation data of the hydrostatic level, the angle data of the tilt sensor, the micro-strain data of the strain gauge, and the temperature data have a consistent UTC time reference.

[0075] In practical implementation, a second-order Butterworth low-pass filter with a cutoff frequency of 10 Hz is applied to the high-frequency vibration data of the tilt sensor after timestamp calibration to filter out high-frequency noise caused by wind vibration or vehicle bumps. A Hanning window sliding mean filter with a window length of 101 points is applied to the static liquid level readings of the hydrostatic level and the static strain data of the strain gauge to suppress random measurement errors. From the filtered data sequence, equal-interval resampling is performed at fixed time intervals of 0.1 seconds. If there is no corresponding filtered output at a certain moment, linear interpolation is used to supplement it, generating a preprocessed multi-source sensor data sequence with strict time alignment and improved signal-to-noise ratio. Optionally, the trigger threshold can be adjusted according to the structural dimensions and material properties of the bridge, and its calculation relationship is expressed as:

[0076]

[0077] in: A physical unit representing the trigger threshold of a certain type of sensor, having length or deformation. Represents the typical height of the main girder section of a bridge, and has a unit of length. This represents the theoretical internal force generated at the monitoring location under the design load. This represents the elastic modulus of the bridge structural materials. The moment of inertia representing the cross section of the main beam. It is a dimensionless adjustment coefficient that is related to the sensor type and monitoring sensitivity.

[0078] In one embodiment of the present invention, see [reference] Figure 3 The preprocessed multi-source sensor data sequences were analyzed to identify tilt angle variation time-series data from the tilt sensor and relative elevation difference time-series data from the hydrostatic level. These tilt angle variation time-series data and relative elevation difference time-series data were used as the original settlement feature signals. The preprocessed multi-source sensor data sequences were also analyzed to identify micro-strain time-series data from the strain sensor and relative displacement time-series data from the displacement gauge. These micro-strain time-series data and relative displacement time-series data were used as the original deformation feature signals. For the original settlement feature signals, the mean, variance, and cumulative offset from historical benchmark values ​​within a fixed time window were calculated. These mean, variance, and cumulative offset were then normalized and concatenated to generate a settlement feature vector. For the original deformation feature signals, the maximum value, signal energy, and slope of change within a fixed time window were calculated. These maximum value, signal energy, and slope of change were then normalized and concatenated to generate a deformation feature vector.

[0079] In practical implementation, for a long-span steel box girder suspension bridge, the feature extraction module loads a pre-processed multi-source sensor data sequence. This sequence includes the relative elevation difference collected by the hydrostatic level, the tilt angle collected by the tilt sensor, the micro-strain collected by the strain sensor, and the relative displacement collected by the displacement gauge. All data have been time-aligned and the sampling interval is 0.1 seconds. The processing program traverses the entire data sequence, identifying the X-axis and Y-axis tilt angle change time-series data belonging to the tilt sensor and the relative elevation difference time-series data between the bridge tower foundation and the mid-span of the main girder belonging to the hydrostatic level, classifying these data as the original settlement characteristic signals. At the same time, it identifies the micro-strain time-series data output by the strain sensor located at the weld of the bottom plate of the steel box girder and the longitudinal and transverse relative displacement time-series data output by the displacement gauge located at the expansion joint, classifying these data as the original deformation characteristic signals.

[0080] In some embodiments, for the original settlement characteristic signal, a fixed time window of 600 seconds is selected. Within this window, the arithmetic mean of the relative elevation difference data from the hydrostatic level is calculated as the settlement mean. The mean square difference of the deviation of the data from the mean within this window is calculated as the settlement variance. The cumulative algebraic difference between the data in the current window and the initial benchmark value established during the bridge's completion and acceptance is calculated as the cumulative offset. The settlement mean, settlement variance, and cumulative offset are normalized by dividing each by its respective historical maximum absolute value. After eliminating the influence of dimensions, the three are concatenated to form a three-dimensional settlement characteristic vector. For the original deformation characteristic signal, the same operation is performed within a fixed time window of 600 seconds. The maximum absolute value of the strain sensor micro-strain time series data is found as the deformation maximum value. The sum of the squares of all micro-strain data points within the window is calculated as the signal energy. The slope of the linear change of the relative displacement data of the displacement gauge within the window over time is fitted using the least squares method as the trend slope. The deformation maximum value, signal energy, and trend slope are normalized and then concatenated to form a three-dimensional deformation characteristic vector. Alternatively, in the process of calculating signal energy, in order to more accurately characterize the signal intensity distribution, a discrete integral form of the calculation formula can be used:

[0081]

[0082] in: This represents the signal energy calculation result within a fixed time window. Represents the first in the window The original deformation feature signal values ​​at each sampling point It is the total number of sampling points contained in the entire time window. It is a constant time interval between two adjacent sampling points. It can be understood that by accumulating the square of the amplitude of each sampling point and multiplying it by the time interval, the total energy estimate of the signal within the time window can be obtained, which better reflects the cumulative intensity characteristics of the signal in the time domain than a simple sum of squares.

[0083] In one embodiment of the present invention, a knowledge-enhanced fusion evaluation network is constructed, comprising a physical constraint layer and a data-driven layer. The physical constraint layer stores a theoretical relationship matrix of settlement and deformation of the bridge structure under typical load conditions. This matrix describes the mechanical equilibrium and geometric compatibility that should be satisfied between the settlement and each deformation component. The data-driven layer receives settlement feature vectors and deformation feature vectors as input and calculates the actual correlation matrix between the input features. The actual correlation matrix is ​​compared with the theoretical relationship matrix of settlement and deformation retrieved from the physical constraint layer, and an inconsistency measure between the two is calculated. Based on the inconsistency measure, the fusion weights of each component of the settlement and deformation feature vectors in the data-driven layer are dynamically adjusted, so that the fusion process tends to satisfy the theoretical mechanical relationship. The inconsistency metric is defined as the weighted norm of the differences between corresponding elements in the actual correlation matrix and the theoretical relationship matrix between settlement and deformation. The inconsistency metric is input into a weight adjustment function, designed to be a monotonically decreasing function of the inconsistency metric, such that the output global fusion weight adjustment factor decreases as the inconsistency metric increases. Using the global fusion weight adjustment factor, the initial fusion weight matrix between the components of the settlement feature vector and deformation feature vector, originally calculated based on the data self-attention mechanism in the data-driven layer, is scaled to generate a physically-guided intermediate fusion weight matrix. The prior theoretical relationship vector corresponding to the current input feature is queried from the physical constraint layer, and the dot product similarity between the prior theoretical relationship vector and each component of the settlement and deformation feature vectors is calculated. The dot product similarity is normalized to generate a theoretical consistency weight vector. The physically-guided intermediate fusion weight matrix and the theoretical consistency weight vector are subjected to a Hadamard product operation, which enhances the weights of components with high theoretical consistency and suppresses the weights of components with low theoretical consistency, thus generating the final dynamically adjusted fusion weight matrix.

[0084] In practical implementation, for a five-span continuous rigid frame bridge, the fusion evaluation module constructs a knowledge-enhanced fusion evaluation network. This network comprises a physical constraint layer and a data-driven layer. In the physical constraint layer, a theoretical relationship matrix between settlement and deformation of the bridge structure under three typical load conditions—dead load, lane load, and temperature gradient—is pre-stored. The rows of this matrix correspond to the normalized components of the settlement feature vector, and the columns correspond to the normalized components of the deformation feature vector. Matrix elements describe the mechanical equilibrium and geometric compatibility proportionality coefficients that should be satisfied between settlement and deformation components such as strain and displacement under specific stress states. In the data-driven layer, settlement and deformation feature vectors are received from the feature extraction module. The settlement feature vector includes three dimensions: mean settlement, variance settlement, and cumulative offset. The deformation feature vector includes three dimensions: maximum deformation, signal energy, and slope of the deformation trend. The data-driven layer calculates the Pearson correlation coefficients between each dimension of the settlement and deformation feature vectors, constructing a three-row, three-column actual correlation matrix.

[0085] In some embodiments, the actual correlation matrix is ​​compared with the settlement and deformation theoretical relationship matrix under lane load conditions retrieved from the physical constraint layer, and the inconsistency measure between the two is calculated. The inconsistency measure is defined as the Frobenius norm of the difference between corresponding elements of the two matrices, that is, the square root of the sum of the squares of the differences between the matrix elements. Based on the magnitude of the inconsistency measure, the fusion weights of each component of the settlement and deformation feature vectors in the data-driven layer are dynamically adjusted. If the inconsistency measure is small, it indicates that the measured data matches the theoretical relationship well, and the original fusion weights of the data-driven layer are maintained. If the inconsistency measure is large, the dependence on the measured abnormal components is reduced, making the fusion process tend to satisfy the direction of the theoretical mechanical relationship. The inconsistency measure is defined as the weighted norm of the difference between corresponding elements of the actual correlation matrix and the settlement and deformation theoretical relationship matrix. The weighting coefficients are determined according to the sensor measurement accuracy; higher weights are assigned to the matrix rows and columns corresponding to sensors with higher measurement accuracy. Refer to Table 1, which shows a simplified example of the settlement and deformation theoretical relationship matrix stored in the physical constraint layer under constant load conditions. The row index of the matrix corresponds to the dimension of the settlement eigenvector, the column index corresponds to the dimension of the deformation eigenvector, and the matrix element values ​​reflect the theoretical ratio of the expected unit deformation change to the unit settlement change under the ideal elasticity assumption.

[0086] Table 1: Theoretical Relationship Matrix of Settlement and Deformation under Dead Load Conditions

[0087] Mean Settlement Dimension 0.85 0.12 -0.05 Settlement variance dimension 0.03 0.78 0.15 Cumulative offset dimension 0.91 0.08 0.01

[0088] In practical implementation, the inconsistency metric is input into a weight adjustment function, which adopts a negative exponential decay form and is a monotonically decreasing function of the inconsistency metric. As the inconsistency metric increases, the output global fusion weight adjustment factor decreases. Using the global fusion weight adjustment factor, the initial fusion weight matrix between the components of the settlement feature vector and deformation feature vector, originally calculated based on the data self-attention mechanism in the data-driven layer, is scaled to generate a physically-guided intermediate fusion weight matrix. The prior theoretical relation vector corresponding to the current input feature is queried from the physical constraint layer, and the dot product similarity between the prior theoretical relation vector and each component of the settlement and deformation feature vectors is calculated. The dot product similarity is normalized to generate a theoretical consistency weight vector. The physically-guided intermediate fusion weight matrix and the theoretical consistency weight vector are multiplied using a Hadamard product operation (i.e., element-wise multiplication), thereby enhancing the weights of components with high theoretical consistency and suppressing the weights of components with low theoretical consistency in the intermediate fusion weight matrix, resulting in the final dynamically adjusted fusion weight matrix. Optionally, the weighted norm used in calculating the inconsistency metric can be specifically expressed as:

[0089]

[0090] in: Represents the final scalar value of the inconsistency metric. and These represent the number of rows and columns of the actual correlation matrix, respectively. The actual correlation matrix is ​​represented in the th... Line number The element values ​​of the column, The theoretical element values ​​at the same position represent the theoretical relationship matrix between settlement and deformation. These are the corresponding elements in a symmetric positive definite weighted coefficient matrix based on sensor confidence assignment. It can be understood that by introducing a weighted norm, the reliability differences of different sensor data can be fully considered when assessing inconsistencies, avoiding excessive interference from noisy data from low-quality sensors in inconsistency judgments.

[0091] In one embodiment of the present invention, historical fusion settlement deformation state assessment quantities are collected to form a time-series assessment quantity history sequence; the assessment quantity history sequence is analyzed to identify its long-term trend component, periodic component, and random fluctuation component; a sequence modeling-based predictor is constructed, which takes the assessment quantity history sequence of the previous time window as input and maps the input sequence to the latent state through an encoder; in the predictor's decoder, the latent state is gradually decoded by combining the extrapolation results of the long-term trend component and periodic component separated from the assessment quantity history sequence, generating state assessment quantity prediction values ​​for multiple consecutive future time points; the state assessment quantity prediction values ​​for multiple consecutive future time points are connected to form a development trend prediction trajectory reflecting the future change path of the bridge structure state. A local weighted regression method is used to smooth the historical sequence of the evaluation quantity, extracting the long-term trend component that changes slowly and monotonically over time. From the residual sequence after removing the long-term trend component, the main period lengths are identified through spectral analysis, and a sine-cosine basis function set is constructed based on the identified period lengths. The residual sequence is then projected onto the sine-cosine basis function set to obtain the amplitude and phase parameters of the periodic component, thereby reconstructing the periodic component. The long-term trend component and the periodic component are subtracted from the original historical sequence of the evaluation quantity, and the remaining part that cannot be explained by trend and period is the random fluctuation component.

[0092] In practical implementation, for an operating reinforced concrete arch bridge, the prediction module collects the fused settlement deformation state assessment values ​​calculated daily over the past 90 days, totaling 90 values ​​arranged chronologically to form a time-series assessment value history sequence. This historical sequence is processed to identify long-term trend components, periodic components, and random fluctuation components. A sequence modeling predictor based on a long short-term memory network is constructed. The predictor uses the historical assessment value sequence of the past 30 days as an input window, and the encoder maps the input sequence into a 128-dimensional latent state vector. In the predictor's decoder, the latent state vector is progressively decoded by combining the extrapolation results of the long-term trend and periodic components separated from the overall historical sequence, predicting the state assessment value for each future day, iteratively generating predicted state assessment values ​​for 14 consecutive future time points. These 14 predicted state assessment values ​​are then connected chronologically to form a development trend prediction trajectory reflecting the arch bridge's structural state changes over the next two weeks.

[0093] In some embodiments, a local weighted regression method is used to smooth the 90-day historical sequence of assessment quantities, with a smoothing window width of 7 days, to extract the long-term trend component that increases slowly and monotonically over time. From the residual sequence after removing the long-term trend component, a spectral analysis is performed using Fast Fourier Transform to identify the period length corresponding to the power spectrum peak as 7 days. Based on this, a set of sine and cosine functions with a period of 7 days is constructed as basis functions. The residual sequence is fitted with least squares projection onto this set of sine-cosine basis functions to obtain the amplitude coefficient and phase shift parameter of the periodic component, thereby reconstructing the periodic component of daily fluctuations. The long-term trend component and the periodic component are subtracted one by one from the original historical sequence of assessment quantities, and the remaining residual is the random fluctuation component affected by environmental noise and accidental loads. Referring to Table 2, the decomposition results of each component for 5 consecutive days in the historical sequence of assessment quantities are shown, where the long-term trend component increases slowly with the number of days, the periodic component fluctuates in waves, and the random fluctuation component oscillates slightly around zero.

[0094] Table 2: Component Decomposition Table of Historical Series of Evaluation Quantities

[0095] 25 42.16 40.50 +1.45 +0.21 26 43.07 41.00 +1.92 +0.15 27 44.23 41.55 +2.48 +0.20 28 46.18 42.13 +3.80 +0.25 29 47.99 42.75 +4.95 +0.29

[0096] Optionally, when the decoder generates predicted values ​​for a single future time point, the influence of long-term trend increments is considered, and the recursive relationship can be expressed as follows:

[0097]

[0098] in: This represents the predicted state assessment value for the next point in time. This represents the state assessment value at the current time point or the predicted value from the previous step. It is the unit time increment obtained by extrapolation from the long-term trend components. It is the hidden state of the decoder in the current step. It is a non-linear mapping function of the fully connected layer. and This is a trainable scaling factor used to balance the contributions of trend extrapolation and neural network decoding. It can be understood that by explicitly adding a trend increment term, the predicted trajectory can maintain consistency with the historical long-term trend, reducing trend deviations that might occur if relying solely on the neural network.

[0099] In one embodiment of the present invention, an interactive visualization template is established, which includes a timeline, a status indicator area, and a trend preview area. The fused settlement deformation status assessment quantity is mapped to the gauge pointer angle and color in the status indicator area, wherein the pointer angle is proportional to the value of the assessment quantity, and the color changes according to the predefined threshold range in which the assessment quantity is located. The predicted trajectory of the development trend of bridge settlement deformation is plotted in the trend preview area, forming a prediction curve with a confidence interval, and the fused settlement deformation status assessment quantity at the current time point is highlighted as the starting point of the prediction curve. The visualization elements of the timeline, status indicator area, and trend preview area are combined and embedded with text descriptions containing data timestamps and monitoring location identifiers to jointly render and generate a graphical bridge settlement deformation monitoring report. While generating the development trend prediction trajectory, the uncertainty range of the predicted value corresponding to each future time point on the prediction trajectory is calculated through the internal uncertainty propagation mechanism of the state evolution prediction model. The predicted value of each future time point is used as the center point, and half of the uncertainty range of the predicted value corresponding to the center point is used as the upper and lower offsets to determine the upper and lower boundaries of the confidence interval of the predicted value of the future time point. The center points of all future time point prediction values ​​are connected by a smooth curve to form the main prediction curve. The area between the main prediction curve and the curves connecting the upper boundaries of all confidence intervals, as well as the curves connecting the lower boundaries of all confidence intervals, is filled with a semi-transparent color band, which is the visual representation of the confidence interval.

[0100] In practical implementation, for the task of generating a periodic monitoring report for a cable-stayed bridge, the report generation module calls a predefined interactive visualization template. This interactive visualization template, on the webpage, is divided into a timeline control at the top, a circular status indicator area on the left, and a trend preview area occupying the main area on the right. The system reads the latest calculated fusion settlement deformation status assessment value, assumed to be 58.3, and maps this value to the semi-circular dashboard in the status indicator area. The dashboard scale ranges from 0 to 100, and the pointer angle is calculated using linear interpolation; a value of 0 corresponds to a -90-degree position, and a value of 100 corresponds to a +90-degree position. Therefore, the pointer angle corresponding to the value 58.3 is approximately -4.98 degrees. Simultaneously, the system's built-in color mapping rules define the assessment value range [0,40) as a green safe zone, (40,70] as a yellow warning zone, and (70,100] as a red alarm zone. Therefore, the pointer color corresponding to the value 58.3 is rendered as yellow.

[0101] In some embodiments, the predicted trajectory of bridge settlement deformation is plotted in the trend preview area. This trajectory is a sequence of predicted state assessment values ​​for the next 14 days, output by the prediction module. Simultaneously, the state evolution prediction model estimates the prediction uncertainty using a Monte Carlo dropout mechanism, calculates the standard deviation of the predicted value at each future time point, and then determines the uncertainty range at a 95% confidence level. For example, for the prediction on the 7th day, the central predicted value is 62.1, with an uncertainty range of ±2.3. Therefore, the upper boundary of the confidence interval is 64.4, and the lower boundary is 59.8. When plotting, a smooth blue solid line connects the central predicted values ​​of all future time points, forming the main prediction curve. A light blue semi-transparent band fills the area between the main prediction curve and the dashed lines connecting the upper and lower boundaries of all confidence intervals, thus visually representing the reliability of the prediction results.

[0102] In practice, the current time point's integrated settlement deformation status assessment value, i.e., 58.3, is highlighted in the trend preview area as the starting point of the prediction curve. This marker is a solid red dot with a diameter larger than ordinary data points, and is accompanied by a text label displaying the specific value and timestamp "T=0:58.3". The time axis control, the yellow pointer dashboard in the status indicator area, and the main prediction curve with confidence intervals in the trend preview area are combined and arranged, and a text description block containing the monitoring date, bridge name, monitoring location number, and data update timestamp is embedded below the image. Together, they are used to generate a complete graphical bridge settlement deformation monitoring report.

[0103] Optionally, when determining the mapping relationship between pointer color and evaluation value, a piecewise linear function can be used to calculate the RGB components of the color to ensure a natural and smooth color transition. Taking the transition from green to yellow as an example, the formula for calculating the red component can be expressed as:

[0104]

[0105] in: This represents the intensity of the red channel in the output color. It is the minimum value of the red channel corresponding to the green base color. It is the maximum value of the red channel corresponding to the yellow target color. This is the input value for the integrated settlement deformation state assessment. It is the lower limit of the current color range, 40. It is the upper limit of the current color range, 70. This calculation method allows the pointer color to smoothly transition from pure green to pure yellow as the evaluation value increases, avoiding the visual abruptness of color jumps and making the status indication more refined.

[0106] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A bridge settlement and deformation monitoring system based on multi-source sensor data fusion, characterized in that, include: The data acquisition module is used to deploy various types of sensors at key monitoring locations of the target bridge to acquire multi-source sensor data of the bridge structure under load. The data processing module is used to perform time-series alignment and noise suppression preprocessing on the multi-source sensor data to generate a preprocessed multi-source sensor data sequence. The feature extraction module is used to extract the settlement feature signal and deformation feature signal from the preprocessed multi-source sensing data sequence to form a settlement feature vector and a deformation feature vector. The fusion evaluation module is used to input the settlement feature vector and deformation feature vector into an improved fusion decision algorithm. The improved fusion decision algorithm, based on prior knowledge of bridge structural mechanics, collaboratively evaluates the settlement feature vector and deformation feature vector to generate a fusion settlement and deformation state evaluation quantity. Its working principle includes: Construct a knowledge-enhanced fusion evaluation network, which includes a physical constraint layer and a data-driven layer; The physical constraint layer stores the theoretical relationship matrix of settlement and deformation of the bridge structure under typical load conditions. The theoretical relationship matrix of settlement and deformation describes the mechanical balance and geometric compatibility relationship that should be satisfied between the settlement and each deformation component. In the data-driven layer, the settlement feature vector and deformation feature vector are received as inputs, and the actual correlation matrix between the input features is calculated. The actual correlation matrix is ​​compared with the settlement and deformation theoretical relationship matrix retrieved from the physical constraint layer, and the inconsistency measure between the actual correlation matrix and the settlement and deformation theoretical relationship matrix is ​​calculated. Based on the inconsistency metric, the fusion weights of each component of the settlement feature vector and the deformation feature vector in the data-driven layer are dynamically adjusted so that the fusion process tends to satisfy the results of theoretical mechanical relationships. Based on the adjusted fusion weights, the settlement feature vector and deformation feature vector are weighted and aggregated and confidence propagated, and finally a scalar that comprehensively considers the consistency between actual observation and physical laws is output, namely the fused settlement and deformation state assessment quantity. The prediction module is used to calculate the predicted trajectory of bridge settlement deformation development trend based on the integrated settlement deformation state assessment quantity and through the state evolution prediction model, including: Collect historical data on settlement and deformation status assessment to form a historical time series of assessment data; The historical sequence of the evaluation quantity is analyzed to identify its long-term trend component, periodic component, and random fluctuation component. Construct a sequence modeling-based predictor, which takes the historical sequence of the evaluation quantity of the previous time window as input and maps the input sequence to the hidden state through an encoder; In the decoder of the predictor, the latent state is gradually decoded by combining the extrapolation results of the long-term trend component and the periodic component separated from the historical sequence of the evaluation quantity, so as to generate the predicted values ​​of the state evaluation quantity at multiple consecutive time points in the future. The predicted values ​​of the state assessment at multiple consecutive future time points are connected to form the development trend prediction trajectory that reflects the future change path of the bridge structure state. The report generation module is used to visualize and encapsulate the fused settlement deformation state assessment quantity and the predicted trajectory of bridge settlement deformation development trend to form a bridge settlement deformation monitoring report.

2. The bridge settlement and deformation monitoring system based on multi-source sensor data fusion according to claim 1, characterized in that, The deployment of various types of sensors at key monitoring locations on the target bridge to acquire multi-source sensor data on the bridge structure under load includes: Based on the finite element model analysis results of the bridge structure, the locations where stress concentration, deformation sensitivity, or theoretical settlement value is greatest under the expected load are identified, and these locations are marked as key monitoring locations. At the marked key monitoring locations, sensor units are deployed in groups. Each sensor unit includes a hydrostatic level, tilt sensor, strain gauge and temperature sensor. Global navigation satellite system reference stations and monitoring stations are also fixedly deployed at the piers at both ends of the bridge. A unified sensor data acquisition cycle and trigger threshold are set. When the reading of any sensor exceeds its trigger threshold, all sensors are automatically and synchronously triggered to perform a high-frequency data acquisition, thereby obtaining high-time-resolution multi-source sensor data of the bridge under dynamic load response.

3. The bridge settlement and deformation monitoring system based on multi-source sensor data fusion according to claim 1, characterized in that, The multi-source sensor data undergoes time-series alignment and noise suppression preprocessing to generate a preprocessed multi-source sensor data sequence, including: Establish an independent data buffer for each type of sensor data stream to receive the raw data stream from the corresponding sensor; Based on a unified Coordinated Universal Time, the raw data streams in all data buffers are timestamped to ensure that data from different sensors have a unified time reference. For each type of sensor data after timestamp calibration, a noise suppression filter matching the sensor's physical characteristics is applied, the noise suppression filter including a low-pass filter for high-frequency vibration data and a sliding mean filter for static strain data; The filtered data output from the noise suppression filter is resampled at equal intervals according to a preset fixed time interval to generate the preprocessed multi-source sensor data sequence with strict time alignment and improved signal-to-noise ratio.

4. The bridge settlement and deformation monitoring system based on multi-source sensor data fusion according to claim 1, characterized in that, Extracting settlement feature signals and deformation feature signals from the preprocessed multi-source sensor data sequence to construct settlement feature vectors and deformation feature vectors includes: The preprocessed multi-source sensor data sequence is analyzed to identify the tilt angle change time series data from the tilt sensor and the relative elevation difference time series data from the hydrostatic level. The tilt angle change time series data and the relative elevation difference time series data are used as the original settlement characteristic signals. The preprocessed multi-source sensing data sequence is analyzed to identify micro-strain time-series data from strain sensors and relative displacement time-series data from displacement gauges. The micro-strain time-series data and relative displacement time-series data are used as the original deformation feature signals. For the original settlement feature signal, calculate its mean, variance and cumulative offset from the historical benchmark value within a fixed time window, normalize the mean, variance and cumulative offset and concatenate them to generate the settlement feature vector. For the original deformation feature signal, calculate its maximum value, signal energy, and slope of change within a fixed time window. Then, normalize and concatenate the maximum value, signal energy, and slope of change to generate the deformation feature vector.

5. The bridge settlement and deformation monitoring system based on multi-source sensor data fusion according to claim 4, characterized in that, Based on the aforementioned inconsistency metric, the fusion weights of each component of the settlement feature vector and deformation feature vector in the data-driven layer are dynamically adjusted to ensure that the fusion process tends to satisfy the theoretical mechanical relationship, including: The inconsistency metric is defined as the weighted norm of the differences between corresponding elements of the actual correlation matrix and the theoretical relationship matrix between settlement and deformation; The inconsistency metric is input into a weight adjustment function, which is designed to be a monotonically decreasing function of the inconsistency metric, such that the output global fusion weight adjustment factor decreases as the inconsistency metric increases. Using the global fusion weight adjustment factor, the initial fusion weight matrix between each component of the settlement feature vector and the deformation feature vector, which were originally calculated based on the data self-attention mechanism in the data-driven layer, is scaled to generate a physically guided intermediate fusion weight matrix. The prior theoretical relation vector corresponding to the current input feature is queried from the physical constraint layer, and the dot product similarity between the prior theoretical relation vector and each component of the settlement feature vector and deformation feature vector is calculated. The dot product similarity is normalized to generate a theoretical consistency weight vector; The physical-guided intermediate fusion weight matrix is ​​multiplied by the theoretical consistency weight vector using Hadamard multiplication. This enhances the weights of components with high theoretical consistency and suppresses the weights of components with low theoretical consistency, thereby generating the final dynamically adjusted fusion weight matrix.

6. The bridge settlement and deformation monitoring system based on multi-source sensor data fusion according to claim 5, characterized in that, Analyzing the historical sequence of the assessment quantity identifies its long-term trend component, periodic component, and random fluctuation component, including: The historical sequence of the evaluation quantity is smoothed using a local weighted regression method to extract the long-term trend component that changes slowly and monotonically over time. From the residual sequence after removing long-term trend components, the main period lengths are identified through spectral analysis, and a sine-cosine basis function set is constructed based on the identified period lengths. The residual sequence is projected and fitted onto the sine-cosine basis function set to obtain the amplitude and phase parameters of the periodic component, thereby reconstructing the periodic component. The random fluctuation component is the portion that cannot be explained by trends and cycles after subtracting the long-term trend component and the periodic component from the historical sequence of the original assessment quantity.

7. The bridge settlement and deformation monitoring system based on multi-source sensor data fusion according to claim 1, characterized in that, The integrated settlement deformation state assessment and the predicted trajectory of bridge settlement deformation development trend are visualized and encapsulated to form a bridge settlement deformation monitoring report, including: Create an interactive visualization template, which includes a timeline, a status indicator area, and a trend preview area; The fused settlement deformation state assessment quantity is mapped to the pointer angle and color of the dashboard in the state indication area, wherein the pointer angle is proportional to the value of the assessment quantity, and the color changes according to the predefined threshold range in which the assessment quantity is located. The predicted trajectory of the bridge settlement deformation development trend is plotted in the trend preview area to form a prediction curve with a confidence interval, and the fused settlement deformation state assessment at the current time point is highlighted as the starting point of the prediction curve. The visualization elements of the timeline, status indicator area, and trend preview area are combined and embedded with text descriptions containing data timestamps and monitoring location identifiers to jointly render and generate a graphical bridge settlement and deformation monitoring report.

8. The bridge settlement and deformation monitoring system based on multi-source sensor data fusion according to claim 7, characterized in that, The predicted trajectory of the bridge's settlement deformation development trend is plotted in the trend preview area, forming a prediction curve with a confidence interval, including: While generating the predicted trajectory of the development trend, the uncertainty range of the predicted value corresponding to each future time point on the predicted trajectory is calculated through the internal uncertainty propagation mechanism of the state evolution prediction model. Using the predicted value at each future time point as the center point, and taking half of the uncertainty range of the predicted value corresponding to the center point as the upper and lower offsets, the upper and lower boundaries of the confidence interval of the predicted value at the future time point are determined. Connect the center points of all future time point predictions with a smooth curve to form the main prediction curve; The area between the main prediction curve and the curves connecting the upper boundaries of all confidence intervals, as well as the curves connecting the lower boundaries of all confidence intervals, is filled with a semi-transparent color band. This semi-transparent color band is the visual representation of the confidence intervals.

Images