Construction method of highway and waterway engineering catastrophe early warning model based on multi-source data fusion

Abstract

The present application relates to the technical field of disaster early warning, in particular to a highway and waterway engineering disaster early warning model construction method based on multi-source data fusion. The present application constructs a spatial coordination damage index by analyzing the deviation degree of the coordination relationship between the structure response data of monitoring points and combining the abnormal duration decay mechanism on time sequence, determines an environment response coupling deviation index based on the dynamic correlation of the environment monitoring data and the structure response data after time lag compensation, and comprehensively obtains local anomalies; generates a space-time evolution index combining the neighborhood aggregation characteristics and time sequence change trend characteristics of the monitoring site, dynamically adjusts Kalman filtering to obtain structure state estimation output disaster early warning results. The present application analyzes the space-time dynamic trend through the spatial coordination and environment coupling of multi-source data, adjusts the Kalman filtering parameters through a dynamic adaptive mechanism, reduces the filtering divergence, improves the structure state estimation accuracy, and improves the disaster early warning reliability of highway and waterway engineering in complex environment.

Description

Technical Field

[0001] This invention relates to the field of disaster early warning technology, specifically to a method for constructing a disaster early warning model for highway and waterway engineering based on multi-source data fusion. Background Technology

[0002] Health monitoring of highway and waterway engineering projects is crucial for ensuring infrastructure safety. Currently, multi-source data fusion techniques based on displacement and environmental sensors are commonly used to assess structural condition. Kalman filtering, as a classic state estimation method, is frequently used to fuse heterogeneous data to estimate key structural parameters in real time. However, when applying Kalman filtering, fixed process and measurement noise covariance matrices are typically preset, and these parameters rely on offline calibration or empirical settings.

[0003] In real-world field environments, drastic temperature changes and heavy rainfall can cause significant fluctuations in sensor measurements. Filters with fixed parameters struggle to adapt to dynamically changing environmental disturbances, easily leading to Kalman filter divergence and distorted state estimation, which in turn can result in missed or false alarms in disaster warnings. Furthermore, many adaptive filter adjustments rely solely on the judgment of a single indicator, lacking a comprehensive understanding of the inherent spatial correlations between multi-source data and the degree of environmental sensitivity, thus reducing the accuracy of state estimation. Summary of the Invention

[0004] To address the technical problem in existing technologies where Kalman filtering suffers from filter divergence and distorted state estimation due to the inability of the preset noise covariance matrix to adapt to dynamic changes in the field environment, thus affecting the accuracy of disaster early warning, this invention aims to provide a method for constructing a disaster early warning model for highway and waterway engineering based on multi-source data fusion. The specific technical solution adopted is as follows: This invention provides a method for constructing a disaster early warning model for highway and waterway engineering based on multi-source data fusion, the method comprising: Acquire multi-source data from different monitoring locations in highway and waterway engineering projects. The multi-source data includes structural response data and environmental monitoring data. Each monitoring location contains at least one set of adjacent monitoring points. Based on historical normal working condition data analysis, the degree of deviation of the structural response data coordination relationship between monitoring points in the same monitoring location is analyzed. Combined with the time-series abnormal continuous decay mechanism, the spatial coordination failure index of the monitoring location is determined. For each monitoring location, based on the dynamic correlation between environmental monitoring data and structural response data after time delay compensation, the current degree of correlation deviation is analyzed to obtain the environmental response coupling deviation index. By combining the spatial synergistic damage indicators and environmental response coupling deviation indicators of each monitoring site, local anomaly indicators are determined; based on the local anomaly indicators, and according to the neighborhood clustering characteristics and temporal trend characteristics of the monitoring sites, spatiotemporal evolution indicators of each monitoring site are obtained. Based on the spatiotemporal evolution indicators of all monitored locations, feedback parameters are determined; the measurement noise covariance matrix of the Kalman filter is dynamically adjusted using the feedback parameters to obtain an optimized structural state estimate; and disaster early warning results are output based on the optimized structural state estimate.

[0005] Furthermore, the method for obtaining the spatial synergy disruption index includes: For structural response data of each dimension, for any monitoring point within the same monitoring location, the monitoring point is paired with another monitoring point to form an analysis monitoring pair; in historical normal operating data, the incremental changes in structural response data of the analysis monitoring pair are fitted to obtain the fitted regression model of the analysis monitoring pair, and the fluctuation benchmark value is determined based on the deviation between the fitted regression model and the measured value. During the current monitoring period, for the monitored pair, based on the deviation between the increment of the measured structural response data of the monitoring point and the fitted value of the regression model for the monitoring point, the coordinated deviation value of the monitoring point is obtained by normalizing the fluctuation benchmark value; if the number of times the coordinated deviation value is continuously higher than the preset abnormal threshold exceeds the preset identification number, the period when it is continuously higher than the preset abnormal threshold is determined as the abnormal duration period. If the current period is an abnormal duration, the coordination deviation value is used as the indicator of the coordination disruption of the monitoring pair at that monitoring point. If the current period is not an abnormal duration, the coordination deviation value is adjusted exponentially with time length based on the duration between each abnormal duration and the current time to obtain the indicator of the coordination disruption of the monitoring pair at that monitoring point. The largest collaborative failure index among all structural response data within the monitored area is taken as the spatial collaborative failure index.

[0006] Furthermore, the method for obtaining the fluctuation benchmark value includes: The structural response data of the monitoring points are differentially processed to obtain the response increment sequence; the response increment sequence of the monitoring pair under historical normal operating conditions is used to fit the model to obtain the fitted regression model of the monitoring pair; based on the deviation between the fitted value of the monitoring point in the fitted regression model and the measured structural response data increment of the monitoring point in the historical data, the median is taken as the fluctuation benchmark value.

[0007] Furthermore, the method for obtaining the environmental response coupling deviation index includes: In historical normal operating condition data, for each monitoring point, based on the lag time between the environmental monitoring data sequence of each environmental dimension and the structural response data sequence of each environmental dimension, time lag compensation is performed on the environmental monitoring data sequence to obtain the compensated environmental data sequence. A sliding window is used to perform local linear fitting on the compensated environmental data sequence and structural response data to obtain the baseline values ​​of the influence coefficients and prediction residuals of each environmental dimension on the structural response data of each dimension. The real-time influence coefficients and real-time prediction residuals of each environmental dimension in the current monitoring period are extracted. The coefficient deviation value is determined based on the deviation between the real-time influence coefficient and the baseline value of the influence coefficient. The residual deviation value is determined based on the deviation between the real-time prediction residual and the baseline value of the prediction residual. The analysis of persistently high coefficient deviation and residual deviation values ​​was conducted separately to determine the periods of abnormal coefficient and residual deviation. Based on the temporal distribution relationship between the current time and the periods of abnormal coefficient and residual deviation, the coefficient coupling deviation index and residual coupling deviation index for each environmental dimension were determined. The maximum value among all coefficient coupling deviation indices and residual coupling deviation indices within the monitored area is taken as the environmental response coupling deviation index.

[0008] Furthermore, the method for determining the coefficient coupling deviation index and the residual coupling deviation index includes: If the current period is within a period of coefficient abnormality, the coefficient deviation value is used as the coefficient coupling deviation index; if the current period is not within a period of coefficient abnormality, the coefficient deviation value is adjusted exponentially with time length based on the duration of each period of coefficient abnormality and the current time to obtain the coefficient coupling deviation index. If the current period is within the duration of residual abnormality, the residual deviation value is used as the residual coupling deviation index; if the current period is not within the duration of residual abnormality, the residual deviation value is adjusted exponentially with time length based on the duration of each residual abnormality period and the current time to obtain the residual coupling deviation index.

[0009] Furthermore, the method for determining the local anomaly index includes: At each monitoring location, the maximum value of the spatial synergy disruption index and the environmental response coupling deviation index is used as the local anomaly index for that monitoring location.

[0010] Furthermore, the method for obtaining the spatiotemporal evolution index includes: Within a predefined spatial neighborhood of each monitoring location, the neighborhood clustering characteristics are determined based on the degree of deviation between the local anomaly index of that monitoring location and the mean of the local anomaly indexes of all monitoring locations in the defined spatial neighborhood. Within a preset time window for each monitoring location, the trend characteristics are determined based on the degree of deviation between the slope of the linearly fitted local anomaly index of that monitoring location and the baseline slope of the local anomaly index under historical normal operating conditions. By combining neighborhood clustering characteristics and trend characteristics, spatiotemporal evolution indicators for each monitoring location are obtained.

[0011] Furthermore, the feedback parameter is taken as the maximum value of the spatiotemporal evolution index of all monitored locations at the current moment.

[0012] Furthermore, the method for obtaining the optimized structural state estimate includes: The feedback parameter is multiplied by a preset scaling factor and truncated at the upper and lower limits to obtain the adaptive scaling factor; the measurement noise covariance matrix at the current time is obtained by multiplying it by the adaptive scaling factor based on the basic measurement noise covariance matrix obtained from the sensor accuracy calibration. Substituting the adjusted measurement noise covariance matrix into the Kalman filter, the optimized structural state estimate is output.

[0013] Furthermore, the disaster early warning result output based on the optimized structural state estimation includes: The optimized structural state estimate is used as the input to the pre-trained disaster early warning model, which outputs the early warning risk value. The model then performs graded state early warnings based on the risk value and trend early warnings based on the increasing trend of the continuous early warning risk value over time.

[0014] The present invention has the following beneficial effects: This invention constructs a spatial collaborative damage index by analyzing the deviation of the collaborative relationship between structural response data at monitoring points and combining it with a temporal anomaly attenuation mechanism. This filters out instantaneous noise interference and more accurately extracts the persistent anomaly damage characteristics of local structures. Secondly, based on the dynamic correlation between environmental monitoring data and structural response data after time-delay compensation, an environmental response coupling deviation index is determined. This captures deviations in the causal transmission law of the response caused by sudden environmental changes, reducing misjudgments caused by environmental fluctuations. Furthermore, by comprehensively considering the deviations between space and the environment, local anomalies are identified. Combined with the neighborhood clustering characteristics and temporal change trend characteristics of the monitored location, a spatiotemporal evolution index is generated. The measurement noise covariance matrix of the Kalman filter is dynamically adjusted, improving the accuracy of disaster precursor judgment from both spatial clustering and temporal acceleration dimensions. This allows the filter to adaptively reduce its confidence in abnormal observations when facing complex interference, effectively suppressing filter divergence and outputting high-precision state estimates. This invention analyzes spatiotemporal dynamic trends through spatial collaboration and environmental coupling of multi-source data, realizing a dynamic adaptive mechanism to adjust Kalman filter parameters, reducing filter divergence, improving the accuracy of structural state estimation, and enhancing the reliability of disaster early warning for highway and waterway engineering in complex environments. Attached Figure Description

[0015] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 The flowchart illustrates a method for constructing a disaster early warning model for highway and waterway engineering based on multi-source data fusion, as provided in one embodiment of the present invention. Detailed Implementation

[0017] Please see Figure 1 The diagram illustrates a flowchart of a method for constructing a disaster early warning model for highway and waterway engineering based on multi-source data fusion, according to an embodiment of the present invention. The method includes the following steps: S1: Acquire multi-source data from different monitoring locations of highway and waterway engineering projects. The multi-source data includes structural response data and environmental monitoring data. Each monitoring location contains at least one set of adjacent monitoring points.

[0018] In this embodiment of the invention, for structures requiring monitoring in highway and waterway engineering, such as bridges, tunnels, and high slopes, different monitoring locations are defined based on their structural characteristics and force transmission paths. Bridges can be divided into monitoring locations per span or per beam segment; tunnels can be divided into monitoring locations per cross-section; and high slopes can be divided into monitoring locations per monitoring profile. Each monitoring location contains at least one set of adjacent monitoring points, ensuring spatial correlation among points within the same monitoring location. Sensors within each group are spatially close together, enabling the capture of geometrically coordinated deformation characteristics of the local structure, such as the continuity of the bridge deflection curve and the constraint of the tunnel cross-sectional profile. As a specific example, taking a bridge as an example, the division can be based on each span or per beam segment, grouping multiple displacement sensors installed on the same beam segment into one group; for tunnels, it can be based on each cross-section, grouping convergence gauges, arch settlement gauges, etc., on the same cross-section into one group; and for high slopes, it can be based on each monitoring profile, grouping displacement gauges at different elevations on the same profile into one group.

[0019] In a specific implementation of this invention, multi-source data acquisition can utilize BeiDou high-precision displacement monitoring equipment, accelerometers, strain gauges, and crack gauges. Three-dimensional displacement data, vibration acceleration data, and strain data are collected at each monitoring point as structural response data, reflecting the deformation, vibration, and stress changes of the structure under external influences. The structural response data includes multiple dimensions, with each dimension corresponding to one type of data; for example, the three dimensions are displacement, acceleration, and strain. Simultaneously, data from multiple environmental dimensions also need to be collected, such as temperature and humidity data from temperature and humidity sensors, and rainfall intensity data from rain gauges. If necessary, water level gauges and anemometers can be added as environmental monitoring data, with each environmental dimension corresponding to one type of data; for example, the three environmental dimensions are temperature, humidity, and rainfall, comprehensively reflecting the influence of environmental factors.

[0020] Understandably, before data analysis, historical data from the sensors is selected, using historical data under normal operating conditions as a benchmark. In one specific embodiment of this invention, historical periods where the wind speed variance and rainfall variance are below a preset variance threshold, and the structural response does not trigger the basic alarm lower limit, are selected as historical normal operating conditions. Simultaneously, the duration of these historical normal operating conditions is ensured to exceed the analysis period, allowing for the learning of the inherent linear relationships between sensors. Within a small deformation range, the local geometric relationship of the structure can be considered linear, and the linear model is simple, robust, and easy to fit directly from the data. The preset variance threshold can be set to 0.5, and the analysis period can be set to 7 days to ensure the validity and stability of the benchmark data.

[0021] It should be noted that multi-source data may be collected at different frequencies due to different sensors. For example, displacement data is typically collected at 1 Hz, while environmental data is collected at 0.1 Hz. Therefore, time-series alignment preprocessing is required. Linear interpolation is used to resample all low-frequency data to a unified 1 Hz time step to ensure time alignment in subsequent calculations. Simultaneously, each dimension of the multi-dimensional data and the environmental dimension are normalized to eliminate the influence of dimensions. Using historical normal operating conditions as a benchmark, maximum and minimum values ​​are used for normalization. For example, in the temperature dimension, the temperature data in historical and current monitoring data are normalized based on the maximum and minimum values ​​of temperature data under historical normal operating conditions. If the current monitored temperature exceeds the maximum historical temperature value, the normalization result is set to 1; if the current monitored temperature is less than the minimum historical temperature value, the normalization result is set to 0. The method is the same for other dimensions and will not be elaborated further. Data preprocessing is a technique well-known to those skilled in the art and is not limited here.

[0022] S2: Based on historical normal operating condition data analysis, analyze the degree of deviation of the structural response data coordination relationship between monitoring points in the same monitoring location, and combine the time-series abnormal continuous decay mechanism to determine the spatial coordination failure index of the monitoring location.

[0023] Under normal service conditions, the structural response changes of adjacent monitoring points within the same monitoring location in highway and waterway engineering structures exhibit stable collaborative correlation characteristics, such as the continuity of beam deflection curves. Instantaneous, small-amplitude data fluctuations are mostly due to random environmental disturbances and do not represent structural damage. Only continuous deviations in the collaborative relationship can be identified as local structural anomalies, indicating potential latent damage in the local structure. Therefore, based on historical normal operating condition data, a referable collaborative correlation is constructed between monitoring points to quantify the degree of collaborative deviation in the current data. Considering that abnormal disturbances in real-world environments often have persistent and cumulative effects, a temporal anomaly attenuation mechanism is introduced to jointly determine the spatial collaborative damage index of the monitoring location, quantifying the damage manifestation from the perspective of local collaborative damage in the spatial structure.

[0024] Preferably, in some embodiments of the present invention, the method for obtaining the spatial coordination disruption index includes: For structural response data in each dimension, collaborative analysis ensures that the data analyzed between monitoring points are in the same dimension. For any monitoring point within the same monitoring location, this monitoring point is paired with another monitoring point to form an analysis monitoring pair. Considering the impact of environmental interference and measurement noise, the incremental changes in the structural response data of the analysis monitoring pair are fitted to historical normal operating condition data to obtain a fitted regression model for the analysis monitoring pair. Based on the deviation between the fitted regression model and the measured values, a fluctuation benchmark value is determined. This fluctuation benchmark value is used to quantify the allowable range of inherent random noise fluctuations in the analysis monitoring pair under normal conditions.

[0025] In some embodiments of the present invention, the method for obtaining the fluctuation benchmark value includes: The structural response data of the monitoring points are differentially processed to obtain a response increment sequence, where each element in the response increment sequence is the difference between the structural response data of the monitoring point at the current time and the previous time, i.e., the response increment. The response increment sequences of the monitoring pair under historical normal operating conditions are fitted to obtain a fitted regression model for the monitoring pair. In one specific embodiment of this invention, the response increment sequences of two monitoring points are fitted with a regression model using a robust regression method (such as Theil-Sen estimation), representing the linear relationship between the response increment of one monitoring point and the response increment of the other monitoring point. As an example, the expression of the fitted regression model is: In the formula, Indicates the first The monitoring points are at the [number]th The moment is affected by the first The influence of each monitoring point on the associated response increment fitted value, Represented as the first The monitoring points are at the [number]th The response increment at each moment, Represented as regression coefficients, The value is represented as the intercept. It should be noted that the regression method is a well-known technique in the art, and will not be elaborated or limited here.

[0026] Furthermore, based on the deviation between the fitted value of the monitoring point in the regression model and the incremental measured structural response data of the monitoring point in the historical data, the median is taken as the fluctuation benchmark value to eliminate the interference of outliers on the benchmark calibration. In a specific embodiment of the present invention, the difference between each measured structural response data and the previous measured structural response data is taken as the incremental measured structural response data. Based on the analysis of the incremental measured structural response data of another monitoring point in the historical data at the corresponding time, the fitted value of the monitoring point at the corresponding time can be obtained through the regression model, which is the predicted incremental structural response data of the monitoring point at the corresponding time, representing the expected change based on the synergistic relationship. By calculating the absolute value of the difference between the fitted value and the incremental measured structural response data of the actual corresponding time of the monitoring point, the fluctuation deviation value is obtained, reflecting the magnitude of the fitting error. The median of the fluctuation deviation values ​​in all historical data is taken as the fluctuation benchmark value, representing the typical amplitude of the synergistic relationship fluctuation under normal operating conditions.

[0027] During the current monitoring period, which is also the non-historical data period, the monitoring period can be determined based on the real-time acquisition frequency and the system calculation and processing cycle. In a specific embodiment of the present invention, the monitoring period can be set to update once per hour, without any limitation.

[0028] For the analysis and monitoring pair, the collaborative deviation of the monitoring point is obtained by normalizing the deviation between the increment of the measured structural response data at the monitoring point and the fitted value of the regression model for that monitoring point, using a fluctuation benchmark value. In a specific embodiment of the invention, the fitted value of the response increment at the corresponding time is obtained by using the fitted regression model based on the increment of the measured structural response data at each time step of another monitoring point in the analysis and monitoring pair. The absolute value of the difference between the increment of the measured structural response data at the monitoring point and the fitted value is used as the current increment deviation. The collaborative deviation value of the monitoring point is obtained by using the current increment deviation as the numerator and the fluctuation benchmark value as the denominator. The deviation is normalized by dividing by the benchmark value, making deviations of different orders of magnitude comparable. It should be noted that due to the complexity of the field environment, the data between monitoring points cannot be completely standardized and incrementally correlated. Therefore, the fitting cannot be perfect, and the fluctuation benchmark value cannot be zero. If the fluctuation benchmark value is zero, a historical data anomaly warning needs to be issued, and the accuracy of the historical data storage needs to be manually determined.

[0029] Furthermore, if the number of consecutive collaboration deviation values ​​exceeding the preset anomaly threshold exceeds the preset identification number, it indicates a persistent disruption of the collaboration relationship. The period of consecutive values ​​exceeding the preset anomaly threshold is then defined as the anomaly duration period. In one specific embodiment of the present invention, the preset anomaly threshold can be set to 1, and the preset identification number can be set to 3. That is, if the collaboration deviation value is continuously higher than 1 for more than 3 sampling points, this continuous period is defined as the anomaly duration period. The specific values ​​can be adjusted by the implementer and are not limited here.

[0030] In particular, a single point of instantaneous anomaly is likely caused by random noise or transient disturbance, while a true structural anomaly is often persistent. Therefore, if there is no period of anomaly persistence, the local structural spatial coordination of the current analysis is considered to be good, and the coordination disruption index is set to 0.

[0031] Because structural anomalies have persistent and residual effects, recent anomalies have a much greater impact on the current structural state than long-term anomalies. Therefore, it is necessary to attenuate historical anomaly data from periods outside the anomaly phase. If the current period is within the duration of the anomaly, it indicates that the structure may be in a state of ongoing damage. In this case, the coordination deviation value should be directly used as the coordinated damage indicator for that monitoring point in the analysis of the monitoring pair.

[0032] If the current period is not within an abnormal duration, it indicates that although the structure appears to have returned to normal, historical latent damage may still have residual effects. Based on the duration between each abnormal duration and the current time, the coordination deviation value is adjusted exponentially with time to obtain the coordination damage index of the monitoring point in the analysis monitoring pair. In one specific embodiment of the invention, for each abnormal duration before the current time, an exponential decay factor is constructed based on the duration from the end of the abnormal duration to the current time. The longer the duration, the smaller the impact and the lower the contribution, resulting in a smaller exponential decay factor. The average coordination deviation value in each abnormal duration is multiplied by the exponential decay factor and then summed to obtain the current coordination damage index. As an example, when the current period is not within an abnormal duration, the expression for the coordination damage index is: In the formula, Represented as the first The monitoring point and the first Synergistic damage indicators at each monitoring point Represented as the first Mean of coordination deviation value corresponding to each abnormal duration period This represents the total number of periods of abnormal duration. This represents the time corresponding to the current moment. Represented as the first The time corresponding to the end of each abnormal period. It is represented as the decay time constant, which can be taken as 1 hour in the embodiments of the present invention. It can be adjusted according to the structural response speed and is not limited here. Represented as natural constant An exponential function with base 0. Represented as the first The exponential decay factor corresponding to each abnormal duration period.

[0033] Finally, the largest collaborative failure index among all structural response data within the monitored area is taken as the spatial collaborative failure index. The larger the value, the more severe the local collaborative relationship failure exists within the monitored area, indicating that the structure may have suffered local damage or abnormal deformation.

[0034] S3: For each monitoring location, based on the dynamic correlation between environmental monitoring data and structural response data after time delay compensation, analyze the current degree of correlation deviation to obtain the environmental response coupling deviation index.

[0035] Considering the physical lag in the impact of environmental factors (such as temperature and rainfall) on structural response, such as the physical lag in heat conduction and seepage, and the dynamic change in the relationship between the two with structural state, further analysis of the deviation between the environmental and response influence patterns is needed. Since the transmission of environmental disturbances is time-delayed, time-delay compensation must first be applied to the monitoring data. Then, real-time deviation analysis is performed through dynamic coupling with a benchmark to obtain an environmental response coupling deviation index. A larger index indicates a more abnormal pattern in the environmental impact on the structure.

[0036] Preferably, in some embodiments of the present invention, the method for obtaining the environmental response coupling deviation index includes: In historical normal operating condition data, for each monitoring point, time lag compensation is performed on the environmental monitoring data sequence based on the lag time between the environmental monitoring data sequence and the structural response data sequence for each environmental dimension to obtain a compensated environmental data sequence, thereby achieving time alignment between environmental data and structural response data. In a specific embodiment of the present invention, the lag step corresponding to the maximum cross-correlation coefficient is extracted by calculating the cross-correlation function between environmental variables and structural response sequences, and the environmental data sequence is shifted backward by this lag step to obtain the compensated environmental data sequence.

[0037] In one specific embodiment of the invention, a sliding window is preset, with a size set to three times the lag time corresponding to each environmental dimension, to ensure complete analysis of the impact response process, which can be adjusted by the implementer. The sliding window is used to perform local linear fitting on the compensation environmental data sequence and structural response data, obtaining the baseline values ​​of the impact coefficient and predicted residual for each environmental dimension on the structural response data of each dimension. For each time point, data within the sliding window before that time point is fitted, meaning that an impact coefficient and predicted residual can be fitted for each time point. In another specific embodiment of the invention, a ridge regression model is used for fitting, extracting the local mapping relationship between environmental variables and structural responses. The impact coefficient is the slope parameter corresponding to the regression equation, and the predicted residual is the difference between the actual response value and the fitted value. The median of the impact coefficients at all times under historical normal operating conditions is selected as the baseline value of the impact coefficient, and the median of the absolute values ​​of the residuals at all times is selected as the baseline value of the predicted residual. It should be noted that due to the complexity of the real external environment, the baseline values ​​of the impact coefficient and the baseline value of the predicted residual cannot be zero. If a value of zero exists, a historical data anomaly warning needs to be issued, and the accuracy of the historical data storage needs to be manually verified.

[0038] Similarly, the real-time impact coefficient and real-time prediction residual of each environmental dimension in the current monitoring period are extracted. That is, by using the same sliding window and fitting method, the real-time impact coefficient and real-time prediction residual of each moment in the current monitoring period are obtained.

[0039] Based on the deviation between the real-time impact coefficient and the baseline value of the impact coefficient, the coefficient deviation value is determined. Specifically, the absolute value of the difference between the real-time impact coefficient and the baseline value is divided by the baseline value to obtain the coefficient deviation value, which reflects the degree of deviation of the current environmental sensitivity from normal fluctuations. Based on the deviation between the real-time prediction residual and the baseline value of the prediction residual, the residual deviation value is determined. Specifically, the absolute value of the difference between the real-time prediction residual and the baseline value is divided by the baseline value to obtain the residual deviation value, which reflects the presence of anomalous components in the current response that cannot be explained by the environment, such as interference from structural damage or sudden loads.

[0040] The abnormal coefficient deviation value and residual deviation value are analyzed separately to determine the abnormal duration of the coefficient and the abnormal duration of the residual. According to the method for obtaining the abnormal duration in S2, in a specific embodiment of the present invention, if the coefficient deviation value is higher than 1 for more than 3 consecutive sampling points, the time period formed by the consecutive sampling points is taken as the abnormal duration of the coefficient. Similarly, if the residual deviation value is higher than 1 for more than 3 consecutive sampling points, the time period formed by the consecutive sampling points is taken as the abnormal duration of the residual.

[0041] In particular, considering the persistence of structural anomalies, if there are no periods of abnormal coefficients or residual anomalies, it indicates that the current structure's response to the environment is stable and there is no deviation from the coupled causal relationship. In this case, the environmental response coupling deviation index can be directly set to 0.

[0042] Furthermore, based on the temporal distribution relationship between the current time and the periods of abnormal coefficient and residual coefficient, the coefficient coupling deviation index and residual coupling deviation index for each environmental dimension are determined. This process can be based on the acquisition of the collaborative destruction index in S2, analyzing the coefficients and residuals respectively. In this embodiment, if the current time is within a period of abnormal coefficient coefficient, the coefficient deviation value is used as the coefficient coupling deviation index. If the current time is not within a period of abnormal coefficient coefficient, the coefficient deviation value is adjusted exponentially with time length based on the duration between each period of abnormal coefficient coefficient and the current time to obtain the coefficient coupling deviation index, processed according to the same exponential decay rule as in S2. In a specific embodiment of this invention, for each period of abnormal coefficient coefficient before the current time, an exponential decay factor is constructed based on the duration from the end of the period of abnormal coefficient coefficient to the current time. The average coefficient deviation value in each period of abnormal coefficient coefficient is multiplied by the exponential decay factor and then summed to obtain the current coefficient coupling deviation index. As an example, the expression for the coefficient coupling deviation index for each monitoring point is: In the formula, Indicated as the current number Environmental monitoring data sequences of various environmental dimensions and the first Coefficient coupling deviation index of structural response data sequences in various dimensions. Represented as the first The average coefficient deviation value at the end of each abnormal period. This represents the total number of periods of abnormal coefficient duration. This represents the time corresponding to the current moment. Represented as the first The time corresponding to the end of the period of abnormality of each coefficient. Represented as the first The exponential decay factor corresponding to the period of abnormal duration of each coefficient.

[0043] If the current time is within a residual abnormality duration period, the residual deviation value is used as the residual coupling deviation index. If the current time is not within a residual abnormality duration period, the residual deviation value is adjusted exponentially with time length based on the duration between each residual abnormality duration period and the current time to obtain the residual coupling deviation index, which is then processed according to the same exponential decay rule. In a specific embodiment of the present invention, for each residual abnormality duration period before the current time, an exponential decay factor is constructed based on the duration from the end of the residual abnormality duration period to the current time. The average residual deviation value in each residual abnormality duration period is multiplied by the exponential decay factor and then summed to obtain the current residual coupling deviation index. As an example, the expression for the residual coupling deviation index for each monitoring point is: In the formula, Indicated as the current number Environmental monitoring data sequences of various environmental dimensions and the first The residual coupling deviation index of the structural response data sequence in various dimensions. Represented as the first Mean residual deviation value corresponding to the duration of each residual abnormality This represents the total number of periods of duration of disability. This represents the time corresponding to the current moment. Represented as the first The time corresponding to the end of the period of individual disability abnormality. Represented as the first The exponential decay factor corresponding to the duration of individual disability abnormalities.

[0044] Finally, the maximum value among all coefficient coupling deviation indices and residual coupling deviation indices within the monitored area is taken as the environmental response coupling deviation index, which characterizes the most significant degree of environmental and response coupling deviation. The larger the value, the more severe the disruption of the causal relationship between the environment and the response.

[0045] S4: Combine the spatial synergy disruption index and environmental response coupling deviation index of each monitoring site to determine local anomaly indexes; based on the local anomaly indexes, and according to the neighborhood clustering characteristics and temporal trend characteristics of the monitoring site, obtain the spatiotemporal evolution indexes of each monitoring site.

[0046] After obtaining the spatial synergistic damage indicators and environmental response coupling deviation indicators for each monitoring location, comprehensive local anomaly indicators are determined from two dimensions: internal geometric spatial damage and external environmental causal coupling, to more comprehensively reflect the overall degree of anomaly at the monitoring location. Considering that engineering disasters typically manifest as localized spatial high-density clustering and temporal accelerated deterioration, spatiotemporal evolution indicators are obtained by extracting neighborhood clustering characteristics and temporal change trend characteristics to more accurately identify true disaster precursors.

[0047] In this embodiment of the invention, at each monitoring location, the maximum value of the spatial synergy disruption index and the environmental response coupling deviation index is taken as the local anomaly index of the monitoring location. When the spatial synergy disruption index is larger, it indicates that the disruption of the local geometric synergy relationship is dominant. When the environmental response coupling deviation index is larger, it indicates that the abnormal causal relationship between the environment and the response is dominant.

[0048] Preferably, in some embodiments of the present invention, the method for obtaining spatiotemporal evolution indicators includes: First, the spatial distribution characteristics of anomalies are analyzed. Structural disasters often begin in locally weak areas and gradually spread to the surrounding areas. Therefore, if the anomaly at a certain location is significantly higher than that at its neighboring locations, that location may be the center of the disaster. In a specific embodiment of this invention, the predefined spatial neighborhood can be determined based on the actual topological relationship of the structure. For bridges, adjacent spans on the same continuous beam can be considered as the neighborhood; for tunnels, several adjacent sections within the same mileage range can be considered as the neighborhood; for slopes, adjacent elevation measurement points above and below the same slope can be considered as the neighborhood. The definition of the neighborhood should be based on the geometric continuity of the structural force transmission path or monitoring profile, rather than simple geographical distance, to ensure that the spatial correlation has physical meaning.

[0049] Within a predefined spatial neighborhood of each monitoring location, the neighborhood clustering characteristic is determined based on the deviation between the local anomaly index of that monitoring location and the average local anomaly index of all monitoring locations within the defined spatial neighborhood. This characteristic reflects the prominence of the anomaly state of the monitoring location relative to its surrounding structure. In one specific embodiment of the invention, the local anomaly index of the monitoring location is used as the numerator, and the average local anomaly index of all monitoring locations within the defined spatial neighborhood is used as the denominator to obtain the neighborhood clustering characteristic. Specifically, if a monitoring location is an isolated node with no neighborhood, or if the average local anomaly index of all monitoring locations within its neighborhood is zero, and its own local anomaly index is greater than zero, then the denominator can be directly forced to equal the local anomaly index of the monitoring location itself, making the neighborhood clustering characteristic 1. If its own local anomaly index is zero, then the neighborhood clustering characteristic of the monitoring location is set to zero.

[0050] Within a preset time window for each monitoring location, the trend characteristics are determined by the deviation between the slope of the linearly fitted local anomaly index at that monitoring location and the baseline slope of the local anomaly index under historical normal operating conditions. The acceleration of the deterioration of the local hazard is assessed by the slope of the abnormal increase within a short time window; a larger slope indicates a faster evolution rate of the local anomaly. In one specific embodiment of the invention, the current slope can be obtained by linear fitting using the least squares method within the preset time window. The current slope is then non-negatively truncated and divided by the baseline slope to obtain the trend characteristics. The baseline slope of the local anomaly index under historical normal operating conditions can be selected as the 95th percentile of all slopes after traversing the same time window in historical normal operating conditions. If there exists a baseline slope less than or equal to zero, the current slope is directly non-negatively truncated, i.e., the maximum value between the current slope and zero is taken as the trend characteristics. The preset time window can be set to half an hour; the specific value and fitting method can be adjusted by the implementer and are not limited here.

[0051] Finally, by combining neighborhood clustering characteristics and trend characteristics, the spatiotemporal evolution index for each monitoring location is obtained. In one specific embodiment of the invention, the neighborhood clustering characteristics and trend characteristics are weighted and summed to obtain the spatiotemporal evolution index for each monitoring location. The weight of the neighborhood clustering characteristics can be set to 0.6, and the weight of the trend characteristics can be set to 0.4. The specific implementation scenario can be adjusted to accommodate the abnormal spatial clustering characteristics and temporal evolution characteristics; no restrictions are imposed here. As an example, the expression for the spatiotemporal evolution index is: In the formula, Represented as a spatiotemporal evolution index, This is represented as a neighborhood clustering feature. The weights are represented as neighborhood clustering features. This is represented by the trend of change. The weights are represented as the characteristics of the changing trend.

[0052] S5: Based on the spatiotemporal evolution indicators of all monitored locations, determine the feedback parameters; use the feedback parameters to dynamically adjust the measurement noise covariance matrix of the Kalman filter to obtain an optimized structural state estimate; output the disaster early warning result based on the optimized structural state estimate.

[0053] In order to obtain the intensity of the most dangerous abnormal evolution across the entire structure and to provide feedback on the overall catastrophic risk status of the structure, in this embodiment of the invention, the feedback parameter is taken as the maximum value of the spatiotemporal evolution index of all monitored locations at the current moment, which characterizes the intensity of the most dangerous local catastrophic precursor within the entire monitoring range.

[0054] The larger the magnitude of the feedback parameter, the more severe the interference from sudden environmental or structural damage to the current sensor's observation data, and the lower its observation reliability. Therefore, the noise variance in the observation update step of the Kalman filter is dynamically increased using the feedback parameter to actively reduce the weight ratio of abnormal measurements. Preferably, in some embodiments of the present invention, the optimized method for obtaining structural state estimation includes: First, the feedback parameter is multiplied by a preset scaling factor, and then truncated at both the upper and lower limits to obtain an adaptive scaling factor. In one specific embodiment of the invention, if the product is greater than the upper limit truncation value, the adaptive scaling factor is used as the upper limit truncation value; if the product is less than the lower limit truncation value, the adaptive scaling factor is used as the lower limit truncation value. The scaling factor is calibrated using the statistical relationship between the feedback parameter and the innovation variance in historical data, that is, it is obtained through reverse calibration and optimization by using historical simulated interference experiments with the goal of minimizing the sum of squared residuals of the filter innovation sequence. The upper limit truncation value can be set to 6 to prevent the filter from completely rejecting the measurement value due to the unbounded measurement noise covariance, and the lower limit truncation value can be set to 1 to prevent over-reliance on the measurement value and neglect of model predictions when the feedback parameter is too small. The specific values ​​can be adjusted by the implementer according to the specific implementation scenario, and no restrictions are imposed here.

[0055] Different sensors vary in their factory accuracy and deployment range, resulting in different basic measurement noise levels. Therefore, using the basic measurement noise covariance matrix obtained from sensor accuracy calibration as a benchmark, an adaptive scaling factor is multiplied to obtain the measurement noise covariance matrix at the current moment. When the adaptive scaling factor is greater than 1, the measurement noise covariance is amplified, the Kalman gain decreases, and the filter relies more on model prediction. When the adaptive scaling factor is equal to 1, the basic noise level is maintained.

[0056] Finally, the adjusted measurement noise covariance matrix is ​​substituted into the Kalman filter to output an optimized structural state estimate, such as bridge deflection, tunnel convergence value, and slope displacement.

[0057] In some embodiments of the present invention, the output of disaster early warning results based on optimized structural state estimation includes: The optimized structural state estimate is used as the input to the pre-trained disaster early warning model. The disaster early warning model outputs a warning risk value. The physical magnitude estimate is transformed into an intuitive 0 to 1 probability risk assessment through nonlinear mapping. In a specific embodiment of the present invention, the training set uses the structural state estimate and corresponding disaster labels under historical typical working conditions. The model can use a support vector machine or a deep random forest classifier. The input is an optimized multi-source state sequence matrix, and the output is a risk value between 0 and 1. The specific selection of training loss function and hyperparameter tuning are not limited or elaborated here.

[0058] Ultimately, warnings are tiered based on the risk value. For example, a risk value below 0.3 is considered green (normal), exceeding 0.3 triggers a yellow warning, and exceeding 0.85 triggers a red (critical) warning. Simultaneously, trend warnings are issued based on a continuous upward trend in the risk value over time; for example, a trend warning is issued if the risk value monotonically increases by more than 0.2 for one consecutive hour. In one specific embodiment of this invention, the warning level is linked to response measures. For instance, when a yellow warning is reached, monitoring is intensified; when a trend warning or a red warning is issued, an emergency response is initiated to intervene.

[0059] In summary, this invention constructs a spatial collaborative damage index by analyzing the deviation of the collaborative relationship between structural response data at monitoring points and combining it with a temporal anomaly attenuation mechanism. This filters out instantaneous noise interference and more accurately extracts the persistent anomaly damage characteristics of local structures. Secondly, based on the dynamic correlation between time-delay compensated environmental monitoring data and structural response data, an environmental response coupling deviation index is determined. This captures deviations in the causal transmission law of the response caused by sudden environmental changes, reducing misjudgments caused by environmental fluctuations. Furthermore, by comprehensively considering the deviations between space and the environment, local anomalies are identified. A spatiotemporal evolution index is generated by combining the neighborhood clustering characteristics and temporal change trend characteristics of the monitored location. The measurement noise covariance matrix of the Kalman filter is dynamically adjusted, improving the accuracy of disaster precursor judgment from both spatial clustering and temporal acceleration dimensions. This allows the filter to adaptively reduce its confidence in abnormal observations when facing complex interference, effectively suppressing filter divergence and outputting high-precision state estimates. This invention analyzes spatiotemporal dynamic trends by spatial collaboration and environmental coupling of multi-source data, realizes a dynamic adaptive mechanism to adjust Kalman filter parameters, reduces filter divergence, improves the accuracy of structural state estimation, and enhances the reliability of disaster early warning for highway and waterway engineering in complex environments.

Claims

1. A method for constructing a disaster early warning model for highway and waterway engineering based on multi-source data fusion, characterized in that, The method includes: Acquire multi-source data from different monitoring locations in highway and waterway engineering projects. The multi-source data includes structural response data and environmental monitoring data. Each monitoring location contains at least one set of adjacent monitoring points. Based on historical normal working condition data analysis, the degree of deviation of the structural response data coordination relationship between monitoring points in the same monitoring location is analyzed. Combined with the time-series abnormal continuous decay mechanism, the spatial coordination failure index of the monitoring location is determined. For each monitoring location, based on the dynamic correlation between environmental monitoring data and structural response data after time delay compensation, the current degree of correlation deviation is analyzed to obtain the environmental response coupling deviation index. By combining the spatial synergistic damage indicators and environmental response coupling deviation indicators of each monitoring site, local anomaly indicators are determined; based on the local anomaly indicators, and according to the neighborhood clustering characteristics and temporal trend characteristics of the monitoring sites, spatiotemporal evolution indicators of each monitoring site are obtained. Based on the spatiotemporal evolution indicators of all monitored locations, feedback parameters are determined; the measurement noise covariance matrix of the Kalman filter is dynamically adjusted using the feedback parameters to obtain an optimized structural state estimate; and disaster early warning results are output based on the optimized structural state estimate.

2. The method for constructing a disaster early warning model for highway and waterway engineering based on multi-source data fusion as described in claim 1, characterized in that, The method for obtaining the spatial synergy disruption index includes: For structural response data of each dimension, for any monitoring point within the same monitoring location, the monitoring point is paired with another monitoring point to form an analysis monitoring pair; in historical normal operating data, the incremental changes in structural response data of the analysis monitoring pair are fitted to obtain the fitted regression model of the analysis monitoring pair, and the fluctuation benchmark value is determined based on the deviation between the fitted regression model and the measured value. During the current monitoring period, for the monitored pair, based on the deviation between the increment of the measured structural response data of the monitoring point and the fitted value of the regression model for the monitoring point, the coordinated deviation value of the monitoring point is obtained by normalizing the fluctuation benchmark value; if the number of times the coordinated deviation value is continuously higher than the preset abnormal threshold exceeds the preset identification number, the period when it is continuously higher than the preset abnormal threshold is determined as the abnormal duration period. If the current period is an abnormal duration, the coordination deviation value is used as the indicator of the coordination disruption of the monitoring pair at that monitoring point. If the current period is not an abnormal duration, the coordination deviation value is adjusted exponentially with time length based on the duration between each abnormal duration and the current time to obtain the indicator of the coordination disruption of the monitoring pair at that monitoring point. The largest collaborative failure index among all structural response data within the monitored area is taken as the spatial collaborative failure index.

3. The method for constructing a disaster early warning model for highway and waterway engineering based on multi-source data fusion according to claim 2, characterized in that, The method for obtaining the fluctuation benchmark value includes: The structural response data of the monitoring points are differentially processed to obtain the response increment sequence; the response increment sequence of the monitoring pair under historical normal operating conditions is used to fit the model to obtain the fitted regression model of the monitoring pair; based on the deviation between the fitted value of the monitoring point in the fitted regression model and the measured structural response data increment of the monitoring point in the historical data, the median is taken as the fluctuation benchmark value.

4. The method for constructing a disaster early warning model for highway and waterway engineering based on multi-source data fusion as described in claim 1, characterized in that, The method for obtaining the environmental response coupling deviation index includes: In historical normal operating condition data, for each monitoring point, based on the lag time between the environmental monitoring data sequence of each environmental dimension and the structural response data sequence of each environmental dimension, time lag compensation is performed on the environmental monitoring data sequence to obtain the compensated environmental data sequence. A sliding window is used to perform local linear fitting on the compensated environmental data sequence and structural response data to obtain the baseline values ​​of the influence coefficients and prediction residuals of each environmental dimension on the structural response data of each dimension. The real-time influence coefficients and real-time prediction residuals of each environmental dimension in the current monitoring period are extracted. The coefficient deviation value is determined based on the deviation between the real-time influence coefficient and the baseline value of the influence coefficient. The residual deviation value is determined based on the deviation between the real-time prediction residual and the baseline value of the prediction residual. The analysis of persistently high coefficient deviation and residual deviation values ​​was conducted separately to determine the periods of abnormal coefficient and residual deviation. Based on the temporal distribution relationship between the current time and the periods of abnormal coefficient and residual deviation, the coefficient coupling deviation index and residual coupling deviation index for each environmental dimension were determined. The maximum value among all coefficient coupling deviation indices and residual coupling deviation indices within the monitored area is taken as the environmental response coupling deviation index.

5. The method for constructing a disaster early warning model for highway and waterway engineering based on multi-source data fusion according to claim 4, characterized in that, The methods for determining the coefficient coupling deviation index and the residual coupling deviation index include: If the current period is within a period of coefficient abnormality, the coefficient deviation value is used as the coefficient coupling deviation index; if the current period is not within a period of coefficient abnormality, the coefficient deviation value is adjusted exponentially with time length based on the duration of each period of coefficient abnormality and the current time to obtain the coefficient coupling deviation index. If the current period is within the duration of residual abnormality, the residual deviation value is used as the residual coupling deviation index; if the current period is not within the duration of residual abnormality, the residual deviation value is adjusted exponentially with time length based on the duration of each residual abnormality period and the current time to obtain the residual coupling deviation index.

6. The method for constructing a disaster early warning model for highway and waterway engineering based on multi-source data fusion according to claim 1, characterized in that, The method for determining the local anomaly index includes: At each monitoring location, the maximum value of the spatial synergy disruption index and the environmental response coupling deviation index is used as the local anomaly index for that monitoring location.

7. The method for constructing a disaster early warning model for highway and waterway engineering based on multi-source data fusion as described in claim 1, characterized in that, The methods for obtaining the spatiotemporal evolution indicators include: Within a predefined spatial neighborhood of each monitoring location, the neighborhood clustering characteristics are determined based on the degree of deviation between the local anomaly index of that monitoring location and the mean of the local anomaly indexes of all monitoring locations in the defined spatial neighborhood. Within a preset time window for each monitoring location, the trend characteristics are determined based on the degree of deviation between the slope of the linearly fitted local anomaly index of that monitoring location and the baseline slope of the local anomaly index under historical normal operating conditions. By combining neighborhood clustering characteristics and trend characteristics, spatiotemporal evolution indicators for each monitoring location are obtained.

8. The method for constructing a disaster early warning model for highway and waterway engineering based on multi-source data fusion according to claim 1, characterized in that, The feedback parameter is taken as the maximum value of the spatiotemporal evolution index of all monitored locations at the current moment.

9. The method for constructing a disaster early warning model for highway and waterway engineering based on multi-source data fusion according to claim 1, characterized in that, The method for obtaining the optimized structural state estimate includes: The feedback parameter is multiplied by a preset scaling factor and truncated at the upper and lower limits to obtain the adaptive scaling factor; the measurement noise covariance matrix at the current time is obtained by multiplying it by the adaptive scaling factor based on the basic measurement noise covariance matrix obtained from the sensor accuracy calibration. Substituting the adjusted measurement noise covariance matrix into the Kalman filter, the optimized structural state estimate is output.

10. The method for constructing a disaster early warning model for highway and waterway engineering based on multi-source data fusion according to claim 1, characterized in that, The disaster early warning results output by the optimized structural state estimation include: The optimized structural state estimate is used as the input to the pre-trained disaster early warning model, which outputs the early warning risk value. The model then performs graded state early warnings based on the risk value and trend early warnings based on the increasing trend of the continuous early warning risk value over time.

Images