Bridge state information analysis method and system

Abstract

The application discloses a bridge state information analysis method and system, the method comprises the following steps: acquiring multi-source time series monitoring data of multiple monitoring points, calculating and fusing the modal information entropy of each data sequence to obtain the comprehensive modal information entropy sequence of each point; calculating the phase locking value between the sequences, constructing a phase synchronization network and dividing multiple phase synchronization clusters by using a fuzzy clustering algorithm; selecting a core monitoring point in each cluster and taking the data of the core monitoring point as a reference time series mode; using a multi-scale time series prediction model to perform parallel prediction on the reference mode, and obtaining an accurate prediction result of the core point by wavelet transform fusion; finally, according to the phase lag and amplitude proportional relationship between other points in the cluster and the core point, the core prediction result is calibrated in time and space to generate the state prediction result of all points in the cluster. The application realizes efficient and accurate overall bridge state prediction, and significantly improves the monitoring efficiency and early warning capability.

Description

Technical Field

[0001] This invention belongs to the field of bridge structural health monitoring technology, and in particular relates to a method and system for analyzing bridge condition information. Background Technology

[0002] During long-term operation, large bridge structures are subject to the combined effects of environmental erosion, material aging, fatigue loads, and other factors, leading to gradual performance degradation and potential safety hazards. Timely and accurate understanding of the structural condition of bridges, assessment of their health status, and prediction of their evolution trends are crucial for ensuring operational safety and guiding scientific maintenance.

[0003] Traditional bridge condition monitoring methods primarily rely on the independent analysis of specific types of sensor data, such as modal parameter identification based on vibration response or assessment of local stress state based on strain data. These methods have the following limitations: First, single-type data cannot comprehensively reflect the complex mechanical behavior and damage mechanisms of a structure; second, there is a lack of effective fusion analysis methods for massive, multi-source monitoring data, failing to fully exploit its value; third, most methods focus on "diagnosing" the current state while lacking sufficient ability to "predict" future states, and the prediction models are often computationally complex, making them unsuitable for the real-time analysis needs of large-scale, distributed sensor networks.

[0004] While some existing technologies have attempted to utilize multi-sensor data for structural analysis, systematic solutions are still lacking in areas such as how to effectively quantify the intrinsic correlation between the dynamic responses of different monitoring points, how to perform efficient state clustering and representative analysis based on this correlation, and how to construct lightweight and accurate multi-scale prediction models. Summary of the Invention

[0005] This invention provides a method and system for analyzing bridge condition information, aiming to overcome the problems in the prior art such as insufficient fusion of multi-source data, superficial condition correlation analysis, and difficulty in balancing the efficiency and accuracy of prediction models.

[0006] In a first aspect, the present invention provides a method for analyzing bridge status information, comprising:

[0007] Multi-source time-series monitoring data from multiple monitoring points deployed on the bridge structure are acquired, and the multi-source time-series monitoring data from the same monitoring point are synchronized and aligned to obtain multiple synchronized time-series data sequences from the same monitoring point.

[0008] Calculate the modal information entropy of each synchronous time series data sequence, and fuse the modal information entropy of multiple synchronous time series data sequences corresponding to the same monitoring point to obtain the comprehensive modal information entropy sequence corresponding to each monitoring point.

[0009] Calculate the phase lock value between any composite modal information entropy sequences, construct a phase synchronization network with monitoring points as nodes and phase lock values ​​as edge weights, and use a fuzzy clustering algorithm to divide the phase synchronization network to obtain at least one phase synchronization cluster, wherein the same phase synchronization cluster contains at least one monitoring point.

[0010] Within a phase synchronization cluster, a core monitoring point is selected based on the centrality index of a node in the phase synchronization network, and the synchronization time series data sequence of the core monitoring point is used as the reference time series pattern of the phase synchronization cluster.

[0011] The baseline time series pattern is input into a preset multi-scale time series prediction model. The multi-scale time series prediction model generates multi-scale future state predictions for core monitoring points by running prediction sub-networks based on different time scales in parallel.

[0012] Multi-resolution fusion based on wavelet transform is performed on multi-scale future state prediction to obtain the fused state prediction results of the core monitoring points;

[0013] Based on the phase lag relationship between other monitoring points and the core monitoring point within a certain phase synchronization cluster, phase compensation and amplitude adjustment are performed on the fusion state prediction results to obtain the state prediction results of all monitoring points within a certain phase synchronization cluster.

[0014] Secondly, the present invention provides a bridge status information analysis system, comprising:

[0015] The acquisition module is configured to acquire multi-source time-series monitoring data from multiple monitoring points deployed on the bridge structure, and to perform synchronization and alignment processing on the multi-source time-series monitoring data from the same monitoring point to obtain multiple synchronized time-series data sequences from the same monitoring point.

[0016] The first fusion module is configured to calculate the modal information entropy of each synchronous time series data sequence, and to fuse the modal information entropy of multiple synchronous time series data sequences corresponding to the same monitoring point to obtain a comprehensive modal information entropy sequence corresponding to each monitoring point.

[0017] The partitioning module is configured to calculate the phase lock value between any integrated modal information entropy sequences, construct a phase synchronization network with monitoring points as nodes and phase lock values ​​as edge weights, and use a fuzzy clustering algorithm to partition the phase synchronization network to obtain at least one phase synchronization cluster, wherein the same phase synchronization cluster contains at least one monitoring point.

[0018] The module is selected and configured to select a core monitoring point within a phase synchronization cluster based on the centrality index of the node in the phase synchronization network, and use the synchronization time series data sequence of the core monitoring point as the reference time series mode of a phase synchronization cluster.

[0019] The generation module is configured to input the baseline time series pattern into a preset multi-scale time series prediction model. The multi-scale time series prediction model generates multi-scale future state predictions for core monitoring points by running prediction sub-networks based on different time scales in parallel.

[0020] The second fusion module is configured to perform multi-resolution fusion based on wavelet transform on multi-scale future state predictions to obtain the fusion state prediction results of the core monitoring points.

[0021] The adjustment module is configured to perform phase compensation and amplitude adjustment on the fusion state prediction results based on the phase lag relationship between other monitoring points and the core monitoring point in a certain phase synchronization cluster, so as to obtain the state prediction results of all monitoring points in a certain phase synchronization cluster.

[0022] Thirdly, an electronic device is provided, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the steps of the bridge status information analysis method of any embodiment of the present invention.

[0023] Fourthly, the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein when the program instructions are executed by a processor, the processor performs the steps of the bridge status information analysis method of any embodiment of the present invention.

[0024] The bridge condition information analysis method and system of this application have the following specific advantages:

[0025] Deep integration of multi-source information: By calculating and integrating the modal information entropy of multi-source time series data, a feature sequence that can comprehensively reflect the complex dynamic state of monitoring points is constructed, overcoming the one-sidedness of a single data source;

[0026] Precise quantification of state correlation: The introduction of phase lock value (PLV) quantifies the synchronicity of state evolution of different monitoring points and constructs a phase synchronization network, which can identify clusters of monitoring points with close internal correlation from complex data;

[0027] Intelligent Clustering and Core Selection: The network is adaptively partitioned using a fuzzy clustering algorithm, and core points of the cluster are intelligently selected based on the network centrality index, which optimizes the analysis units and significantly reduces the computational complexity of subsequent modeling.

[0028] Highly efficient and accurate multi-scale prediction: For the benchmark model of core points, a parallel multi-scale prediction sub-network and a wavelet transform-based fusion strategy are adopted to simultaneously capture short-term fluctuations, medium-term cycles and long-term trends, which improves computational efficiency while ensuring prediction accuracy.

[0029] Highly efficient cluster prediction generalization: Based on the phase lag and amplitude ratio relationship between points within the cluster, the prediction results of core points are calibrated in a spatiotemporal manner with clear physical meaning, realizing "one-point prediction, multiple-point application", which greatly improves the efficiency of overall state prediction of large-scale monitoring networks. Attached Figure Description

[0030] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0031] Figure 1 A flowchart of a bridge status information analysis method provided in an embodiment of the present invention;

[0032] Figure 2 This is a structural block diagram of a bridge status information analysis system provided in an embodiment of the present invention;

[0033] Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0034] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0035] Please see Figure 1 The diagram shows a flowchart of a bridge status information analysis method according to this application.

[0036] like Figure 1 As shown, the bridge condition information analysis method specifically includes the following steps:

[0037] Step S101: Obtain multi-source time-series monitoring data from multiple monitoring points deployed on the bridge structure, and perform synchronization and alignment processing on the multi-source time-series monitoring data of the same monitoring point to obtain multiple synchronized time-series data sequences of the same monitoring point.

[0038] Step S102: Calculate the modal information entropy of each synchronous time series data sequence, and fuse the modal information entropy of multiple synchronous time series data sequences corresponding to the same monitoring point to obtain the comprehensive modal information entropy sequence corresponding to each monitoring point.

[0039] In this step, a synchronous time-series data sequence at a certain monitoring point is divided into multiple equal-length analysis time windows according to time order;

[0040] For each analysis time window of a given synchronous time-series data sequence, the following steps are performed: Empirical Mode Decomposition (EMD) is performed on the data segment within the analysis time window to obtain at least one intrinsic mode function (IMF) component; the energy of each IMF component is calculated, and the total energy of all IMF components is calculated; based on the ratio of the energy of each IMF component to the total energy, the energy proportion of each IMF component is obtained; Hilbert Transform is performed on each IMF component to obtain the corresponding analytic signal, and the instantaneous frequency is extracted from the analytic signal; based on the probability distribution of the instantaneous frequency, the Shannon entropy of each IMF component is calculated; the Shannon entropy of each IMF component is multiplied by its corresponding energy proportion, and all product results are summed to obtain the modal information entropy value of the analysis time window.

[0041] Arrange the modal information entropy values ​​calculated from all analysis time windows of a certain synchronous time series data sequence in chronological order to form a certain modal information entropy subsequence of a certain synchronous time series data sequence;

[0042] For a given monitoring point, the modal information entropy subsequences corresponding to all synchronous time-series data sequences of the given monitoring point are combined to form a multi-row modal information entropy matrix;

[0043] Each row of the modal information entropy matrix is ​​standardized.

[0044] Principal component analysis is performed on the standardized modal information entropy matrix to extract the principal component vector with the largest variance contribution, and the principal component vector with the largest variance contribution is defined as the comprehensive modal information entropy sequence corresponding to the monitoring point.

[0045] In one specific embodiment, for a specified monitoring point (e.g., point P) and a specified monitoring data type (e.g., D1, representing vibration acceleration), the time-series data sequence S_{P,D1} after synchronization and alignment processing is obtained.

[0046] The sequence S_{P,D1} is divided into a series of non-overlapping or partially overlapping analysis time windows of length L according to time order: W_1, W_2, ..., W_N.

[0047] For each data segment x(t) within an analysis time window W_k (k=1,…,N):

[0048] Empirical Mode Decomposition: The EMD algorithm is applied to decompose x(t), which is adaptively decomposed into a series of intrinsic mode function components IMF_1(t), IMF_2(t), ..., IMF_M(t) arranged from high frequency to low frequency and a residual term.

[0049] Energy and energy percentage calculation: Calculate the energy of each IMF_i(t) component E_i = ∫|IMF_i(t)|²dt. Calculate the total energy of all IMF components E_total = ΣE_i. Calculate the energy percentage of each component p_i = E_i / E_total.

[0050] Instantaneous frequency extraction and Shannon entropy calculation: For each IMF_i(t), perform a Hilbert transform to obtain its analytic signal Z_i(t) = IMF_i(t) + j*H[IMF_i(t)]. H[IMF_i(t)] is the Hilbert transform result of IMF_i(t). Extract the instantaneous frequency f_i(t) from Z_i(t). Statistically calculate the probability distribution P(f) of f_i(t) within the W_k window. Based on this distribution, calculate the Shannon entropy H_i = -ΣP(f)logP(f) of the IMF component. This entropy value quantifies the complexity or uncertainty of the instantaneous frequency of the component.

[0051] Window modal information entropy synthesis: The Shannon entropy H_i of each IMF component is multiplied by its energy proportion p_i (p_i*H_i), and then the weighted entropy values ​​of all IMF components are summed to obtain the modal information entropy value IE_k=Σ(p_i*H_i) for the time window W_k. This value comprehensively reflects the energy distribution and frequency complexity of each major vibrational mode (IMF) of the signal within the window.

[0052] Arrange the IE_1, IE_2, ..., IE_N calculated for all time windows in chronological order to form the modal information entropy subsequence SubSeq_{P,D1} of data type D1 at monitoring point P.

[0053] For monitoring point P, repeat the above process to calculate the modal information entropy subsequences for all K types of monitoring data (e.g., D1: vibration acceleration, D2: dynamic strain, D3: tilt angle).

[0054] SubSeq_{P,D1},SubSeq_{P, D2},...,SubSeq_{P, DK}.

[0055] These subsequences are combined into a K-row, N-column matrix (the number of time windows), called the modal information entropy matrix M_P for that location. Each row of the matrix represents the entropy evolution of a monitoring data type.

[0056] Z-score normalization is performed on each row of the modal information entropy matrix M_P to eliminate differences in scale and entropy range between different data sources, resulting in the normalized matrix M'_P.

[0057] Principal component analysis is performed on the standardized matrix M'_P. The loading vector corresponding to the principal component with the largest variance contribution (i.e., the first principal component) is extracted. Multiplying this loading vector with the standardized original matrix M'_P (or linearly combining the subsequences along the direction of this principal component) yields a new time series of length N. This series is the comprehensive modal information entropy sequence CIES_P for monitoring point P. Essentially, this sequence fuses and compresses multiple interconnected but different perspective information entropy evolution sequences at monitoring point P into a core feature sequence that best represents the overall state complexity evolution trend.

[0058] In summary, by performing correlation analysis on the state complexity information (entropy sequence) revealed by multi-source data such as vibration and strain, and extracting the common and most important evolutionary patterns as the comprehensive modal information entropy sequence, we not only achieve effective fusion and dimensionality reduction of multi-source information, but more importantly, the generated comprehensive modal information entropy sequence integrates the commonalities of state changes reflected by various physical quantities. It is more comprehensive, more stable, and more representative of the overall "state health" evolution trend of the monitoring point than the features of any single data source.

[0059] The integrated modal information entropy sequence is a time-series feature with uniform length, clear physical meaning (state complexity), and standardized processing. This provides ideal foundational data for the next step of calculating the phase lock value (PLV) between all monitoring points to construct a phase synchronization network. Due to the extremely high quality and consistency of the input features, the calculated phase lock value (PLV) can more realistically and reliably reflect the inherent synchronicity and correlation strength of state evolution between different points, greatly improving the accuracy and reliability of subsequent clustering analysis and state prediction.

[0060] Step S103: Calculate the phase lock value between any integrated modal information entropy sequences, construct a phase synchronization network with monitoring points as nodes and phase lock values ​​as edge weights, and use a fuzzy clustering algorithm to divide the phase synchronization network to obtain at least one phase synchronization cluster, wherein the same phase synchronization cluster contains at least one monitoring point.

[0061] In this step, the comprehensive modal information entropy sequence corresponding to the first monitoring point is denoted as the first sequence, and the comprehensive modal information entropy sequence corresponding to the second monitoring point is denoted as the second sequence, wherein the first monitoring point and the second monitoring point are any two monitoring points among all monitoring points;

[0062] Perform a Hilbert transform on the first sequence to obtain a first analytic signal, and perform a Hilbert transform on the second sequence to obtain a second analytic signal;

[0063] Extract a first instantaneous phase sequence φ1(t) from the first analyzed signal, and extract a second instantaneous phase sequence φ2(t) from the second analyzed signal;

[0064] Calculate the difference between the first instantaneous phase sequence φ1(t) and the second instantaneous phase sequence φ2(t) at each corresponding time point t to obtain the phase difference sequence Δφ(t) = φ1(t) - φ2(t);

[0065] The phase difference sequence Δφ(t) is converted into a complex number representation sequence C(t) on the unit circle, where C(t) = exp(j*Δφ(t)), and j is the imaginary unit;

[0066] Calculate the average value of the complex number representation sequence C(t) over all N sampling points within a preset observation time window T, and take the modulus of the average value to obtain the phase lock value between the first monitoring point and the second monitoring point. The calculation formula is:

[0067] .

[0068] Furthermore, each monitoring point in the phase synchronization network is defined as a data sample;

[0069] Define the similarity measure between any two data samples as the phase-locking value;

[0070] Set the number of clusters to be divided, c, where the number of clusters is an integer greater than or equal to 2 and less than the total number of monitoring points;

[0071] Initialize a membership matrix U, where the elements of the membership matrix U are... Let represent the membership degree of data sample i to cluster phase synchronization cluster k, and satisfy . and ;

[0072] The objective function J is defined as the weighted sum of squared distances from all data samples to the center point of each phase synchronization cluster, where the weights are the corresponding membership degrees, and the distances are calculated based on the dissimilarity between samples, with the dissimilarity being 1- , This is a similarity measure between data sample i and data sample j;

[0073] An iterative optimization process using the fuzzy C-means clustering algorithm alternately updates the cluster prototype and the membership matrix U to minimize the objective function J;

[0074] The iteration stops when the change in the membership matrix U is less than a preset threshold.

[0075] Based on the final membership matrix U, each data sample is assigned to... The phase synchronization cluster with the largest value is selected, thus dividing all monitoring points into c phase synchronization clusters.

[0076] In one specific embodiment, the phase-locked value is calculated as follows:

[0077] For any two different monitoring points on the bridge, their integrated modal information entropy sequences (generated by step S102) are denoted as sequence X and sequence Y, respectively. These two sequences are of equal length and are aligned in time.

[0078] First, perform Hilbert transforms on sequences X and Y respectively to obtain their analytic signals Z_X(t)=X(t)+j*H{X(t)} and Z_Y(t)=Y(t)+j*H{Y(t)}, where H{·} represents the Hilbert transform operator and j is the imaginary unit.

[0079] Then, the instantaneous phase sequence is extracted from the analytic signal. For sequence X, its instantaneous phase φ_X(t) = arctan(H{X(t)} / X(t)). Similarly, the instantaneous phase φ_Y(t) of sequence Y can be obtained.

[0080] Next, the difference between these two instantaneous phase sequences at each corresponding time point t is calculated, resulting in the phase difference sequence Δφ(t) = φ_X(t) - φ_Y(t). To represent the phase difference information uniformly, it is mapped onto the unit circle, forming a complex number sequence C(t) = exp(j*Δφ(t)). The modulus of each element C(t) in this sequence is always 1, and its argument is the phase difference Δφ(t).

[0081] Finally, within a sufficiently long preset observation time window T (e.g., including all sampling points from the most recent 24 hours or week), the average vector of the complex sequence C(t) is calculated. The phase-locked value is defined as the modulus of this average vector:

[0082] ,

[0083] Where N is the total number of sampling points within window T. The value range of PLV is [0,1]. The closer PLV is to 1, the more constant or minimally variable the phase difference between the two sequences is within the observation window, indicating that their oscillations are highly synchronized; the closer PLV is to 0, the more random or asynchronous the phase relationship between the two sequences is.

[0084] Consider all M monitoring points on the bridge as nodes. For any pair of nodes i and j (i≠j), calculate the phase lock value PLV between them. ij With this PLV ij Using the weights of the edges connecting node i and node j, an undirected weighted network called the phase synchronization network is constructed. This network comprehensively describes the strength of synchronicity in the state evolution between all monitoring points, forming the topological basis for subsequent analysis.

[0085] In summary, by calculating the phase lock value, we have surpassed the traditional correlation analysis based on signal amplitude or simple correlation coefficients, and directly captured the synergy of the state evolution of different monitoring points in terms of "time rhythm" (phase). This phase-based synchronization is a profound manifestation of the inherent coupling relationship between subsystems in complex dynamic systems (such as large bridges), and is extremely sensitive to changes in local dynamic characteristics caused by damage.

[0086] Since damage typically disrupts the mechanical continuity of a local area, affecting its dynamic coupling (synchronization) with surrounding areas, this disruption is directly reflected in a decrease in the PLV value. Therefore, by continuously monitoring the changes in edge weights or the stability of the cluster structure of the phase synchronization network, abnormal signs can be detected as early as possible, before damage causes significant amplitude changes. Furthermore, by partitioning the network into clusters, the global anomaly detection problem can be transformed into a problem of monitoring the synchronization within a few clusters, greatly narrowing the monitoring scope and improving the computational efficiency and location accuracy of anomaly identification. This provides crucial technical support for early warning and preventative maintenance of bridge structures.

[0087] Step S104: Within a certain phase synchronization cluster, based on the centrality index of the node in the phase synchronization network, a core monitoring point is selected, and the synchronization time series data sequence of the core monitoring point is used as the reference time series mode of a certain phase synchronization cluster.

[0088] In this step, for each monitoring point within a phase synchronization cluster, the eigenvector centrality EC_i of monitoring point i in the global phase synchronization network is calculated. The eigenvector centrality EC_i is obtained by solving the principal eigenvectors of the equation A*x = λ*x, where A is the adjacency matrix of the global phase synchronization network, and the elements of the adjacency matrix... The phase lock value between monitoring point i and monitoring point j is the eigenvector component x_i corresponding to the largest eigenvalue of the main eigenvector x;

[0089] Calculate the local clustering coefficient LC_i of monitoring point i in the sub-network formed by a certain phase synchronization cluster, wherein the formula for calculating the local clustering coefficient LC_i is:

[0090] LC_i = (2*E_i) / (k_i*(k_i-1)),

[0091] Where E_i represents the actual number of connection edges between neighboring nodes directly connected to monitoring point i in the sub-network, and k_i represents the total number of neighboring nodes of monitoring point i in the sub-network.

[0092] Normalize all the calculated eigenvector centralities so that the maximum value of the eigenvector centrality is 1, and obtain the normalized eigenvector centrality EC_i'.

[0093] All calculated local clustering coefficients are normalized so that the maximum value of the local clustering coefficient is 1, resulting in the normalized local clustering coefficient LC_i';

[0094] Define a comprehensive centrality index C_i, and calculate it as: C_i=w1*EC_i'+w2*LC_i', where w1 and w2 are preset weight coefficients and satisfy w1+w2=1;

[0095] The monitoring point with the highest comprehensive centrality index C_i within a certain phase synchronization cluster is selected as the core monitoring point.

[0096] In this embodiment, eigenvector centrality and local clustering coefficients are combined. The former ensures that the selected core points have a significant influence in the global synchronization network, meaning that changes in their behavioral patterns can more effectively correlate with or affect other clusters or even the entire network, thus their data contains richer "global information." The latter ensures that the core points are highly interconnected within the current cluster and are located at the center of the cluster topology, meaning that their dynamic behavior is closest to and most typical of the average behavior of other points within the cluster. The core points selected by combining these two methods have a "baseline time series pattern" that can best represent the overall state evolution characteristics of the cluster. Predictions based on this pattern have significantly improved reliability and accuracy when extrapolated to other points within the cluster.

[0097] Step S105: Input the baseline time series pattern into the preset multi-scale time series prediction model. The multi-scale time series prediction model generates multi-scale future state predictions for core monitoring points by running prediction sub-networks based on different time scales in parallel.

[0098] In this step, the multi-scale time series prediction model includes three parallel-running subnetworks: a short-term prediction subnetwork, a medium-term prediction subnetwork, and a long-term prediction subnetwork.

[0099] The baseline time series pattern is processed by a high-pass filter to obtain short-term fluctuation components, and the short-term fluctuation components are input into the short-term prediction sub-network. The short-term prediction sub-network adopts a hybrid structure of one-dimensional convolutional neural network and gated recurrent unit to output the short-term state prediction sequence Ps of the core monitoring point in the next s hours.

[0100] The reference time series pattern is processed by a bandpass filter to obtain the intermediate periodic component, and the intermediate periodic component is input into the intermediate prediction sub-network. The intermediate prediction sub-network adopts a stacked temporal convolutional network structure and outputs the intermediate state prediction sequence Pm of the core monitoring point in the next m days.

[0101] The baseline time series pattern is processed by a low-pass filter to obtain the long-term trend component, and the long-term trend component is input into the long-term prediction sub-network. The long-term prediction sub-network adopts a Transformer encoder-decoder structure and outputs the long-term state trend prediction sequence P1 for the core monitoring point in the next week.

[0102] In one specific embodiment, the baseline time series pattern is the set of all types of original synchronous time series data sequences (such as vibration acceleration, dynamic strain, etc.) of the core monitoring points selected in step S104. To predict the change characteristics at different time scales, scale separation of the multi-source time series data is first required. Taking one key data point (such as the main vibration acceleration) as an example, the decomposition steps are as follows:

[0103] Short-term fluctuation component extraction: The original time-series data O(t) is processed through a high-pass digital filter. The cutoff frequency f_s of this filter is relatively high (for example, it is set to filter out fluctuations with a period greater than 4 hours), and its output S(t) is the short-term fluctuation component, which mainly captures rapid, high-frequency state changes caused by instantaneous vehicle load, wind-induced flutter, etc.

[0104] Mid-term periodic component extraction: The original time series data O(t) is processed through a bandpass digital filter. The passband frequency range of this filter [f_{m_low}, f_{m_high}] is in the middle (for example, allowing fluctuations with a period of 4 hours to 1 week to pass through). Its output M(t) is the mid-term periodic component, which mainly reflects the regularity and mid-frequency state evolution caused by daily temperature difference changes, weekly traffic flow cycles, etc.

[0105] Long-term trend component extraction: The original time series data O(t) is processed through a low-pass digital filter. The cutoff frequency f_l of this filter is extremely low (for example, filtering out fluctuations with a period of less than 1 week), and its output L(t) is the long-term trend component, which mainly characterizes the slow, low-frequency state drift trend caused by structural material aging, creep, seasonal environmental changes, etc.

[0106] The same filtering operation is performed on other types of data (such as strain) to obtain their respective multi-scale components, which constitute three reference component data packets: short-term, medium-term, and long-term.

[0107] Short-Term Prediction Subnetwork: This network is specifically designed to capture short-term, high-frequency dynamics. Its input is the short-term fluctuation component S(t) and its historical sequence. The network structure employs a hybrid architecture of a one-dimensional convolutional neural network and gated recurrent units (GRUs). The one-dimensional convolutional layers (CNNs) automatically extract local spatial features (such as pulse waveform features) within a short time series, while subsequent GRU layers learn how these features evolve over time. After training, this network can continuously predict the short-term state prediction sequence Ps of core monitoring points within the next S hours (e.g., the next 24 hours) based on historical short-term fluctuation data over the past T_s hours. This sequence has high resolution (e.g., one point per hour) and reflects the rapid fluctuation details of the state within the next day.

[0108] Mid-term Prediction Subnetwork: This network is specifically designed to predict mid-term evolution with periodicity. Its input is the mid-term periodic component M(t) and its historical sequence. The network structure employs a stacked temporal convolutional network. By expanding causal convolutions, TCNs can maintain temporal causality while possessing a very large receptive field, making them ideal for capturing periodic patterns and mid-term dependencies over several days. Based on historical mid-term periodic data from the past m days, this network predicts the mid-term state prediction sequence Pm for core monitoring points over the next M days (e.g., the next 7 days), reflecting the periodic change profile of the state within the next week.

[0109] Long-Term Prediction Subnetwork: This network is specifically designed to capture slowly changing long-term trends. Its input consists of the long-term trend component L(t) and its historical sequences. The network structure employs a Transformer encoder-decoder architecture. The Transformer's self-attention mechanism can capture global dependencies between any two time points in the sequence, making it well-suited for modeling long-range trends. The encoder encodes the historical long-term trend, while the decoder autoregressively generates the future trend. Based on historical long-term trend data from the past z weeks, this network predicts the long-term state trend prediction sequence Pl for the core monitoring points within the next l weeks (e.g., the next 4 weeks), smoothly outlining the overall state trend over the next month or so.

[0110] Step S106: Perform multi-resolution fusion based on wavelet transform on the multi-scale future state prediction to obtain the fused state prediction results of the core monitoring points.

[0111] In this step, a preset mother wavelet is used to perform discrete wavelet transform on the short-term state prediction sequence Ps, the medium-term state prediction sequence Pm, and the long-term state prediction sequence Pl respectively, to obtain the first wavelet coefficient set of the short-term state prediction sequence Ps at different scales j, the second wavelet coefficient set of the medium-term state prediction sequence Pm at different scales j, and the third wavelet coefficient set of the long-term state prediction sequence Pl at different scales j.

[0112] For each scale j, an attention-based fusion weight calculation module is designed. The input of the fusion weight calculation module is the first wavelet coefficient in the first wavelet coefficient set, the second wavelet coefficient in the second wavelet coefficient set, and the third wavelet coefficient in the third wavelet coefficient set. The output is the corresponding normalized first fusion weight, second fusion weight, and third fusion weight.

[0113] At each scale j, the first wavelet coefficient, the second wavelet coefficient, and the third wavelet coefficient are multiplied by the corresponding first fusion weight, the second fusion weight, and the third fusion weight, respectively, and then summed to obtain the fused wavelet coefficients, that is, the set of fused wavelet coefficients.

[0114] The fusion wavelet coefficient set is subjected to discrete wavelet inverse transform to reconstruct the fusion state prediction results of the core monitoring points.

[0115] In one specific embodiment, a mother wavelet function suitable for non-stationary signal analysis (e.g., Daubechies wavelet or Symlets wavelet) is selected. Discrete wavelet transform is used to perform multi-scale decomposition on the three predicted sequences respectively.

[0116] Short-term sequence decomposition: A J-level discrete wavelet transform is performed on the short-term state prediction sequence Ps (e.g., 1 point per hour for the next 24 hours, totaling 24 points). After the transform, a series of detail coefficients ds are obtained at different scales (resolutions) j (j=1 is the finest scale, j=J is the coarsest scale). j and an approximate coefficient as on the J scale J These coefficients together constitute the first wavelet coefficient set of Ps {ds1,ds2,...,ds}. J ,as J The detail coefficients capture the high-frequency fluctuations of Ps at different time resolutions, while the approximation coefficients reflect the trend of its lowest frequency.

[0117] Medium- and Long-Term Sequence Decomposition: Similarly, perform discrete wavelet transforms of the same level J on the medium-term state prediction sequence Pm (e.g., 1 PM daily for the next 7 days) and the long-term state trend prediction sequence Pl (e.g., 1 PM weekly for the next 4 weeks), respectively, to obtain the second wavelet coefficient set {dm1, dm2, ..., dm J ,am J} and the third wavelet coefficient set {dl1,dl2,...,dl J ,al J By adjusting the sampling rate of the input sequences or the scale parameter of the wavelet transform, we can ensure that the three sequences are comparable at the same scale j, i.e., they correspond to the same physical time resolution.

[0118] Attention-based cross-scale fusion weight calculation:

[0119] For each scale j after decomposition (j=1 to J, and the approximate coefficient scale J), ​​an independent fusion weight calculation module is designed. The core of this module is a lightweight neural network based on an attention mechanism.

[0120] Module input: For scale j, the input is the wavelet coefficient vector corresponding to the three sequences at that scale: ds j (or as) J ),dm j (or am) J ),dl j (or al) J ).

[0121] Attention weight calculation: First, each coefficient vector is mapped to a common feature space through a shared fully connected layer. Then, an attention score vector e is calculated. j =[es j ,em j ,el j Each score reflects the confidence or importance of the prediction result of the corresponding prediction sequence at that scale (e.g., it can be implicitly learned based on the stability of the scale coefficient in historical validation or the performance of the prediction subnetwork at that scale). Finally, the Softmax function is used to evaluate e. j Normalization is performed to obtain the fusion weight vector α at this scale. j =[αs j ,αm j ,αl j ], satisfying αs j +αm j +αl j =1. This weight is dynamic and depends on the specific characteristics of the three sequence prediction features at the current scale.

[0122] Weighted fusion of wavelet coefficients and result reconstruction:

[0123] Coefficient fusion: At each scale j, the wavelet coefficients ds of the three sequences are fused together. j ,dm j ,dl j The corresponding fusion weights αs calculated respectively j ,αm j ,αl j Multiply and then sum to obtain the fused wavelet coefficients at that scale. j =αs j *ds j +αm j *dm j +αl j *dl j For the approximation coefficient as J ,am J , al J Perform the same weighted fusion operation to obtain afused J .

[0124] Result reconstruction: The detail coefficients and approximation coefficients at all scales obtained after fusion are afused. JThe process involves fusing the set of wavelet coefficients as input and performing an inverse discrete wavelet transform. Using the same mother wavelet function as during decomposition, the fused multi-resolution coefficients are faithfully reconstructed back into the time domain signal via the inverse transform algorithm. This reconstructed time-series signal is the final fused state prediction result sequence for the core monitoring points.

[0125] In summary, operations in the wavelet domain (frequency domain) effectively separate and process components at different scales. The application of the attention mechanism further endows the fusion process with adaptability, enabling it to dynamically adjust the contribution of each prediction source in different frequency bands based on historical performance and current data characteristics. This allows the final fused prediction to not only better fit the complex state curves generated by the coupling of multiple physical processes but also improves robustness to situations where a single prediction sub-network may perform poorly in certain time periods or frequency bands. The generated fused prediction results provide a more solid and refined data foundation for threshold-based early warning and maintenance decision optimization.

[0126] Step S107: Based on the phase lag relationship between other monitoring points and the core monitoring point within a certain phase synchronization cluster, perform phase compensation and amplitude adjustment on the fusion state prediction result to obtain the state prediction result of all monitoring points within a certain phase synchronization cluster.

[0127] In this step, the comprehensive modal information entropy sequence of the core monitoring point within a certain phase synchronization cluster, as well as the comprehensive modal information entropy sequences of other core monitoring points, are obtained.

[0128] The core instantaneous phase sequence is obtained by performing a Hilbert transform on the comprehensive modal information entropy sequence of the core monitoring points, and other instantaneous phase sequences are obtained by performing a Hilbert transform on the comprehensive modal information entropy sequences of other core monitoring points.

[0129] Calculate the average phase difference between the core instantaneous phase sequence and the other instantaneous phase sequences within a preset dominant oscillation frequency range;

[0130] The Fourier transform of the fusion status prediction result sequence of the core monitoring points is performed to obtain the frequency domain signal F(ω);

[0131] A linear phase shift is applied to the frequency domain signal F(ω) to obtain a phase-compensated frequency domain signal F corresponding to the other core monitoring points. adjust(ω) =F(ω)*exp(-j*ω*(Δφ / (2πf dom ))), where Δφ is the average phase difference between the core instantaneous phase sequence and other instantaneous phase sequences within the preset dominant oscillation frequency range, f dom ω is the center frequency of the preset dominant oscillation frequency range, j is the imaginary unit;

[0132] Obtain other vibration acceleration data sequences of the other core monitoring points during the historical monitoring period, as well as core vibration acceleration data sequences of the core monitoring points during the historical monitoring period;

[0133] Bandpass filtering is performed on other vibration acceleration data sequences and core vibration acceleration data sequences respectively to extract other target vibration components and core target vibration components corresponding to the center frequency of the preset dominant oscillation frequency range. The other target vibration components are the target vibration components corresponding to the other vibration acceleration data sequences, and the core target vibration components are the target vibration components corresponding to the core vibration acceleration data sequences.

[0134] Calculate the average ratio of the amplitude envelope of other target vibration components to the amplitude envelope of the core target vibration component during the historical monitoring period, and define the average ratio as the amplitude adjustment coefficient;

[0135] The amplitude of the phase compensation frequency domain signal corresponding to the other core monitoring points is multiplied by the amplitude adjustment coefficient to obtain the target phase compensation frequency domain signal;

[0136] The inverse Fourier transform of the target phase-compensated frequency domain signal is used to obtain the state prediction results of other core monitoring points.

[0137] In one specific embodiment, the comprehensive modal information entropy sequence, which characterizes the overall state evolution of each point and has been calculated in step S102, is used to obtain the sequence CIES_A for core point A and the sequence CIES_B for non-core point B.

[0138] Instantaneous phase extraction: Perform Hilbert transform on CIES_A and CIES_B respectively to obtain their analytic signals, and extract the corresponding instantaneous phase sequences φ_A(t) and φ_B(t) from them.

[0139] Calculating the average phase difference: Since the dominant modes of structural dynamic response are typically concentrated in specific frequency bands, a predefined dominant oscillation frequency range is defined (e.g., near the first or second natural frequencies of the bridge). In φ_A(t) and φ_B(t), the instantaneous phase values ​​near the center frequency f_dom of this frequency range [f_{low}, f_{high}] are extracted through filtering or direct calculation. The differences between these corresponding instantaneous phase values ​​are calculated, and their average value over the entire historical analysis period is obtained to obtain the average phase difference Δφ_{BA} between the core point and the non-core point. This value reflects the fixed phase delay caused by the time required for the signal to propagate from the core point to the non-core point.

[0140] Phase compensation is applied to the core point prediction results:

[0141] The fusion state prediction result P_fused(t) at the core point is a time-domain sequence. To adjust it to the non-core point B, the phase lag Δφ_{BA} needs to be compensated first.

[0142] Frequency domain transformation: Perform a fast Fourier transform on P_fused(t) to transform it to the frequency domain and obtain its spectrum F(ω), where ω = 2πf is the angular frequency.

[0143] Applying a linear phase shift: In the frequency domain, a fixed time delay τ corresponds to a linear phase shift -ωτ. Based on the average phase difference Δφ_{BA} (in radians) and the dominant center frequency f_{dom}, the equivalent time delay τ can be estimated as τ = Δφ_{BA} / (2πf_{dom}). Accordingly, phase compensation is applied to the spectrum F(ω) to obtain the phase-compensated frequency domain signal facing point B: F_{adj, B}(ω) = F(ω) * exp(-i * ω * τ) = F(ω) * exp(-i* ω * (Δφ_{BA} / (2π f_{dom}))), where i is the imaginary unit. This operation is equivalent to shifting the predicted waveform at the core point by τ along the time axis.

[0144] Determine the amplitude ratio and adjust the amplitude accordingly:

[0145] Since the dynamic response amplitude varies at different locations of the structure, the amplitude of the prediction signal also needs to be adjusted.

[0146] Historical vibration data acquisition and processing: Obtain the original vibration acceleration data sequences of core point A and non-core point B within the same historical monitoring period.

[0147] Extracting the dominant vibration components: Apply a bandpass filter centered at f_{dom} to the two original acceleration sequences to filter out noise and irrelevant frequency components, and extract the target vibration components a_A(t) (core point) and a_B(t) (non-core point) that reflect the dominant vibration of the structure.

[0148] Calculate the amplitude adjustment factor: Calculate the amplitude envelopes of a_A(t) and a_B(t) respectively (e.g., obtain the instantaneous amplitude using Hilbert transform). Calculate the ratio of the average values ​​of these two envelope sequences over the entire historical period: R_{BA} = mean(envelope(a_B(t))) / mean(envelope(a_A(t))). This ratio of average values, R_{BA}, is the amplitude adjustment factor, reflecting how many times larger the typical response amplitude of point B is compared to point A under the dominant vibration mode.

[0149] Perform amplitude adjustment: Multiply the amplitude (i.e. the magnitude of each frequency component) of the phase-compensated frequency domain signal F_{adj,B}(ω) by the amplitude adjustment coefficient R_{BA} to obtain the final target phase-compensated frequency domain signal.

[0150] In summary, the method of this application firstly constructs a unified feature sequence that comprehensively and profoundly reflects the dynamic state complexity of each monitoring point by calculating and fusing modal information entropy from multi-source time-series data. This overcomes the one-sidedness of analysis from a single data source and lays the foundation for high-sensitivity state perception. Secondly, phase-locked values ​​are introduced to quantify the synchronicity of state evolution among points, and a phase synchronization network is constructed. Fuzzy clustering algorithm is used to achieve adaptive and refined intelligent grouping of monitoring points, automatically identifying "functionally coordinated regions" with strong dynamic correlations in structure, making the analysis units more in line with physical reality. Furthermore, by integrating composite indicators of global influence and local cohesion, core points of each cluster are scientifically selected, and only the core points are subjected to parallel multi-scale deep prediction. Then, high-precision prediction results are obtained through wavelet multi-resolution fusion, realizing "deep modeling at a single point", which greatly reduces the overall prediction computation burden of massive points. Finally, a cluster generalization mechanism based on physical correlation prediction results is implemented: based on the stable phase lag and amplitude ratio relationship between each point in the cluster and the core point, the high-precision prediction of the core point is calibrated in a spatiotemporal manner with clear physical meaning, and personalized predictions for all points in the cluster are generated efficiently. This represents a leap from "one-point deep prediction" to "cluster efficient generalization," solving the computational efficiency bottleneck of large-scale sensor network state prediction while ensuring prediction accuracy. Ultimately, it provides a panoramic view of the future state evolution of bridge structures that covers the entire bridge, is spatiotemporally continuous, and includes both details and trends, providing unprecedented and powerful technical support for intelligent preventive maintenance and precise decision-making.

[0151] Please see Figure 2 The diagram shows a structural block diagram of a bridge status information analysis system according to this application.

[0152] like Figure 2 As shown, the bridge status information analysis system 200 includes an acquisition module 210, a first fusion module 220, a division module 230, a selection module 240, a generation module 250, a second fusion module 260, and an adjustment module 270.

[0153] The acquisition module 210 is configured to acquire multi-source time-series monitoring data from multiple monitoring points deployed on the bridge structure, and to synchronize and align the multi-source time-series monitoring data from the same monitoring point to obtain multiple synchronized time-series data sequences from the same monitoring point; the first fusion module 220 is configured to calculate the modal information entropy of each synchronized time-series data sequence, and to fuse the modal information entropy of multiple synchronized time-series data sequences corresponding to the same monitoring point to obtain a comprehensive modal information entropy sequence corresponding to each monitoring point; the partitioning module 230 is configured to calculate the phase lock value between any comprehensive modal information entropy sequences, construct a phase synchronization network with monitoring points as nodes and phase lock values ​​as edge weights, and partition the phase synchronization network using a fuzzy clustering algorithm to obtain at least one phase synchronization cluster, wherein the same phase synchronization cluster contains at least one monitoring point; the selection module 240 is configured to select a certain Within a phase synchronization cluster, a core monitoring point is selected based on the centrality index of a node in the phase synchronization network, and the synchronization time series data sequence of the core monitoring point is used as the reference time series pattern for a certain phase synchronization cluster. A generation module 250 is configured to input the reference time series pattern into a preset multi-scale time series prediction model. The multi-scale time series prediction model generates multi-scale future state predictions for the core monitoring point by running prediction sub-networks based on different time scales in parallel. A second fusion module 260 is configured to perform multi-resolution fusion based on wavelet transform on the multi-scale future state predictions to obtain the fused state prediction results for the core monitoring point. An adjustment module 270 is configured to perform phase compensation and amplitude adjustment on the fused state prediction results based on the phase lag relationship between other monitoring points and the core monitoring point within a certain phase synchronization cluster, to obtain the state prediction results for all monitoring points within the certain phase synchronization cluster.

[0154] It should be understood that Figure 2 The modules and references described in the document Figure 1 The steps described in the text correspond to those in the method described above. Therefore, the operations, features, and corresponding technical effects described above also apply to the method described in the text. Figure 2 The various modules in the document will not be described in detail here.

[0155] In other embodiments, the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein when the program instructions are executed by a processor, the processor performs the bridge status information analysis method in any of the above method embodiments.

[0156] In one embodiment, the computer-readable storage medium of the present invention stores computer-executable instructions, which are configured as follows:

[0157] Multi-source time-series monitoring data from multiple monitoring points deployed on the bridge structure are acquired, and the multi-source time-series monitoring data from the same monitoring point are synchronized and aligned to obtain multiple synchronized time-series data sequences from the same monitoring point.

[0158] Calculate the modal information entropy of each synchronous time series data sequence, and fuse the modal information entropy of multiple synchronous time series data sequences corresponding to the same monitoring point to obtain the comprehensive modal information entropy sequence corresponding to each monitoring point.

[0159] Calculate the phase lock value between any composite modal information entropy sequences, construct a phase synchronization network with monitoring points as nodes and phase lock values ​​as edge weights, and use a fuzzy clustering algorithm to divide the phase synchronization network to obtain at least one phase synchronization cluster, wherein the same phase synchronization cluster contains at least one monitoring point.

[0160] Within a phase synchronization cluster, a core monitoring point is selected based on the centrality index of a node in the phase synchronization network, and the synchronization time series data sequence of the core monitoring point is used as the reference time series pattern of the phase synchronization cluster.

[0161] The baseline time series pattern is input into a preset multi-scale time series prediction model. The multi-scale time series prediction model generates multi-scale future state predictions for core monitoring points by running prediction sub-networks based on different time scales in parallel.

[0162] Multi-resolution fusion based on wavelet transform is performed on multi-scale future state prediction to obtain the fused state prediction results of the core monitoring points;

[0163] Based on the phase lag relationship between other monitoring points and the core monitoring point within a certain phase synchronization cluster, phase compensation and amplitude adjustment are performed on the fusion state prediction results to obtain the state prediction results of all monitoring points within a certain phase synchronization cluster.

[0164] Computer-readable storage media may include a stored program area and a stored data area, wherein the stored program area may store an operating system and an application program required for at least one function; the stored data area may store data created based on the use of the bridge condition information analysis system, etc. Furthermore, the computer-readable storage medium may include high-speed random access memory, and may also include memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some embodiments, the computer-readable storage medium may optionally include memory remotely disposed relative to a processor, and these remote memories may be connected to the bridge condition information analysis system via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0165] Figure 3 This is a schematic diagram of the structure of the electronic device provided in the embodiment of the present invention, such as... Figure 3 As shown, the device includes a processor 310 and a memory 320. The electronic device may also include an input device 330 and an output device 340. The processor 310, memory 320, input device 330, and output device 340 can be connected via a bus or other means. Figure 3 Taking a bus connection as an example, the memory 320 is the computer-readable storage medium described above. The processor 310 executes various server functions and data processing by running non-volatile software programs, instructions, and modules stored in the memory 320, thereby implementing the bridge status information analysis method described in the above embodiment. The input device 330 can receive input digital or character information and generate key signal inputs related to user settings and function control of the bridge status information analysis system. The output device 340 may include a display screen or other display device.

[0166] The aforementioned electronic device can execute the method provided in the embodiments of the present invention, and has the corresponding functional modules and beneficial effects for executing the method. Technical details not described in detail in this embodiment can be found in the method provided in the embodiments of the present invention.

[0167] In one implementation, the above-described electronic device is applied to a bridge status information analysis system for a client, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to:

[0168] Multi-source time-series monitoring data from multiple monitoring points deployed on the bridge structure are acquired, and the multi-source time-series monitoring data from the same monitoring point are synchronized and aligned to obtain multiple synchronized time-series data sequences from the same monitoring point.

[0169] Calculate the modal information entropy of each synchronous time series data sequence, and fuse the modal information entropy of multiple synchronous time series data sequences corresponding to the same monitoring point to obtain the comprehensive modal information entropy sequence corresponding to each monitoring point.

[0170] Calculate the phase lock value between any composite modal information entropy sequences, construct a phase synchronization network with monitoring points as nodes and phase lock values ​​as edge weights, and use a fuzzy clustering algorithm to divide the phase synchronization network to obtain at least one phase synchronization cluster, wherein the same phase synchronization cluster contains at least one monitoring point.

[0171] Within a phase synchronization cluster, a core monitoring point is selected based on the centrality index of a node in the phase synchronization network, and the synchronization time series data sequence of the core monitoring point is used as the reference time series pattern of the phase synchronization cluster.

[0172] The baseline time series pattern is input into a preset multi-scale time series prediction model. The multi-scale time series prediction model generates multi-scale future state predictions for core monitoring points by running prediction sub-networks based on different time scales in parallel.

[0173] Multi-resolution fusion based on wavelet transform is performed on multi-scale future state prediction to obtain the fused state prediction results of the core monitoring points;

[0174] Based on the phase lag relationship between other monitoring points and the core monitoring point within a certain phase synchronization cluster, phase compensation and amplitude adjustment are performed on the fusion state prediction results to obtain the state prediction results of all monitoring points within a certain phase synchronization cluster.

[0175] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of various embodiments or some parts of embodiments.

[0176] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for analyzing bridge status information, characterized in that, include: Multi-source time-series monitoring data from multiple monitoring points deployed on the bridge structure are acquired, and the multi-source time-series monitoring data from the same monitoring point are synchronized and aligned to obtain multiple synchronized time-series data sequences from the same monitoring point. Calculate the modal information entropy of each synchronous time series data sequence, and fuse the modal information entropy of multiple synchronous time series data sequences corresponding to the same monitoring point to obtain the comprehensive modal information entropy sequence corresponding to each monitoring point. Calculate the phase lock value between any composite modal information entropy sequences, construct a phase synchronization network with monitoring points as nodes and phase lock values ​​as edge weights, and use a fuzzy clustering algorithm to divide the phase synchronization network to obtain at least one phase synchronization cluster, wherein the same phase synchronization cluster contains at least one monitoring point. Within a phase synchronization cluster, a core monitoring point is selected based on the centrality index of a node in the phase synchronization network, and the synchronization time series data sequence of the core monitoring point is used as the reference time series pattern of the phase synchronization cluster. The baseline time series pattern is input into a preset multi-scale time series prediction model. The multi-scale time series prediction model generates multi-scale future state predictions for core monitoring points by running prediction sub-networks based on different time scales in parallel. Multi-resolution fusion based on wavelet transform is performed on multi-scale future state prediction to obtain the fused state prediction results of the core monitoring points; Based on the phase lag relationship between other monitoring points and the core monitoring point within a certain phase synchronization cluster, phase compensation and amplitude adjustment are performed on the fusion state prediction results to obtain the state prediction results of all monitoring points within a certain phase synchronization cluster.

2. The bridge status information analysis method according to claim 1, characterized in that, The calculation of modal information entropy for each synchronized time-series data sequence, and the fusion of modal information entropy for multiple synchronized time-series data sequences corresponding to the same monitoring point to obtain a comprehensive modal information entropy sequence for each monitoring point, includes: Divide a synchronous time-series data sequence at a certain monitoring point into multiple analysis time windows of equal length according to the time sequence; For each analysis time window of the aforementioned synchronous time-series data sequence, the following steps are performed: Empirical mode decomposition is performed on the data segment within the analysis time window to obtain at least one intrinsic mode function component; Calculate the energy of each intrinsic mode function component, and calculate the total energy of all intrinsic mode function components; The energy percentage of each intrinsic mode function component is obtained based on the ratio of the energy of each intrinsic mode function component to the total energy. Perform a Hilbert transform on each intrinsic mode function component to obtain the corresponding analytic signal, and extract the instantaneous frequency from the analytic signal; Based on the probability distribution of the instantaneous frequency, calculate the Shannon entropy of each intrinsic mode function component; Multiply the Shannon entropy of each intrinsic mode function component by its corresponding energy percentage, and sum all the product results to obtain the modal information entropy value of the analysis time window; Arrange the modal information entropy values ​​calculated from all analysis time windows of a certain synchronous time series data sequence in chronological order to form a certain modal information entropy subsequence of a certain synchronous time series data sequence; For a given monitoring point, the modal information entropy subsequences corresponding to all synchronous time-series data sequences of the given monitoring point are combined to form a multi-row modal information entropy matrix; Each row of the modal information entropy matrix is ​​standardized. Principal component analysis is performed on the standardized modal information entropy matrix to extract the principal component vector with the largest variance contribution, and the principal component vector with the largest variance contribution is defined as the comprehensive modal information entropy sequence corresponding to the monitoring point.

3. The bridge status information analysis method according to claim 1, characterized in that, The calculation of the phase lock value between any synthetic modal information entropy sequences includes: The comprehensive modal information entropy sequence corresponding to the first monitoring point is denoted as the first sequence, and the comprehensive modal information entropy sequence corresponding to the second monitoring point is denoted as the second sequence, wherein the first monitoring point and the second monitoring point are any two monitoring points among all monitoring points; Perform a Hilbert transform on the first sequence to obtain a first analytic signal, and perform a Hilbert transform on the second sequence to obtain a second analytic signal; Extract a first instantaneous phase sequence φ1(t) from the first analyzed signal, and extract a second instantaneous phase sequence φ2(t) from the second analyzed signal; Calculate the difference between the first instantaneous phase sequence φ1(t) and the second instantaneous phase sequence φ2(t) at each corresponding time point t to obtain the phase difference sequence Δφ(t) = φ1(t) - φ2(t); The phase difference sequence Δφ(t) is converted into a complex number representation sequence C(t) on the unit circle, where C(t) = exp(j*Δφ(t)), and j is the imaginary unit; Calculate the average value of the complex number representation sequence C(t) over all N sampling points within a preset observation time window T, and take the modulus of the average value to obtain the phase lock value between the first monitoring point and the second monitoring point. The calculation formula is: 。 4. The bridge status information analysis method according to claim 1, characterized in that, The step of using a fuzzy clustering algorithm to divide the phase synchronization network to obtain at least one phase synchronization cluster includes: Each monitoring point in the phase synchronization network is defined as a data sample; Define the similarity measure between any two data samples as the phase-locking value; Set the number of clusters to be divided, c, where the number of clusters is an integer greater than or equal to 2 and less than the total number of monitoring points; Initialize a membership matrix U, where the elements of the membership matrix U are... Let represent the membership degree of data sample i to cluster phase synchronization cluster k, and satisfy . and ; The objective function J is defined as the weighted sum of squared distances from all data samples to the center point of each phase synchronization cluster, where the weights are the corresponding membership degrees, and the distances are calculated based on the dissimilarity between samples, with the dissimilarity being 1- , This is a similarity measure between data sample i and data sample j; An iterative optimization process using the fuzzy C-means clustering algorithm alternately updates the cluster prototype and the membership matrix U to minimize the objective function J; The iteration stops when the change in the membership matrix U is less than a preset threshold. Based on the final membership matrix U, each data sample is assigned to... The phase synchronization cluster with the largest value is selected, thus dividing all monitoring points into c phase synchronization clusters.

5. The bridge status information analysis method according to claim 1, characterized in that, Within a certain phase synchronization cluster, selecting a core monitoring point based on the centrality index of a node in the phase synchronization network includes: For each monitoring point within a given phase synchronization cluster, the eigenvector centrality EC_i of monitoring point i in the global phase synchronization network is calculated. This eigenvector centrality EC_i is obtained by solving the principal eigenvectors of the equation A*x = λ*x, where A is the adjacency matrix of the global phase synchronization network, and the elements of the adjacency matrix... The phase lock value between monitoring point i and monitoring point j is the eigenvector component x_i corresponding to the largest eigenvalue of the main eigenvector x; Calculate the local clustering coefficient LC_i of monitoring point i in the sub-network formed by a certain phase synchronization cluster, wherein the formula for calculating the local clustering coefficient LC_i is: LC_i = (2*E_i) / (k_i*(k_i-1)), Where E_i represents the actual number of connection edges between neighboring nodes directly connected to monitoring point i in the sub-network, and k_i represents the total number of neighboring nodes of monitoring point i in the sub-network. Normalize all the calculated eigenvector centralities so that the maximum value of the eigenvector centrality is 1, and obtain the normalized eigenvector centrality EC_i'. All calculated local clustering coefficients are normalized so that the maximum value of the local clustering coefficient is 1, resulting in the normalized local clustering coefficient LC_i'; Define a comprehensive centrality index C_i, and calculate it as: C_i=w1*EC_i'+w2*LC_i', where w1 and w2 are preset weight coefficients and satisfy w1+w2=1; The monitoring point with the highest comprehensive centrality index C_i within a certain phase synchronization cluster is selected as the core monitoring point.

6. The bridge status information analysis method according to claim 1, characterized in that, The multi-scale time series prediction model includes three parallel-running sub-networks: a short-term prediction sub-network, a medium-term prediction sub-network, and a long-term prediction sub-network. The multi-scale time-series prediction model generates multi-scale future state predictions for core monitoring points by running prediction sub-networks based on different time scales in parallel, including: The baseline time series pattern is processed by a high-pass filter to obtain short-term fluctuation components, and the short-term fluctuation components are input into the short-term prediction sub-network. The short-term prediction sub-network adopts a hybrid structure of one-dimensional convolutional neural network and gated recurrent unit to output the short-term state prediction sequence Ps of the core monitoring point in the next s hours. The reference time series pattern is processed by a bandpass filter to obtain the intermediate periodic component, and the intermediate periodic component is input into the intermediate prediction sub-network. The intermediate prediction sub-network adopts a stacked temporal convolutional network structure and outputs the intermediate state prediction sequence Pm of the core monitoring point in the next m days. The baseline time series pattern is processed by a low-pass filter to obtain the long-term trend component, and the long-term trend component is input into the long-term prediction sub-network. The long-term prediction sub-network adopts a Transformer encoder-decoder structure and outputs the long-term state trend prediction sequence P1 for the core monitoring point in the next week.

7. The bridge status information analysis method according to claim 6, characterized in that, The multi-resolution fusion based on wavelet transform for multi-scale future state prediction, resulting in the fused state prediction results for the core monitoring points, includes: Using a preset mother wavelet, discrete wavelet transform is performed on the short-term state prediction sequence Ps, the medium-term state prediction sequence Pm, and the long-term state prediction sequence Pl respectively to obtain the first wavelet coefficient set of the short-term state prediction sequence Ps at different scales j, the second wavelet coefficient set of the medium-term state prediction sequence Pm at different scales j, and the third wavelet coefficient set of the long-term state prediction sequence Pl at different scales j. For each scale j, an attention-based fusion weight calculation module is designed. The input of the fusion weight calculation module is the first wavelet coefficient in the first wavelet coefficient set, the second wavelet coefficient in the second wavelet coefficient set, and the third wavelet coefficient in the third wavelet coefficient set. The output is the corresponding normalized first fusion weight, second fusion weight, and third fusion weight. At each scale j, the first wavelet coefficient, the second wavelet coefficient, and the third wavelet coefficient are multiplied by the corresponding first fusion weight, the second fusion weight, and the third fusion weight, respectively, and then summed to obtain the fused wavelet coefficients, that is, the set of fused wavelet coefficients. The fusion wavelet coefficient set is subjected to discrete wavelet inverse transform to reconstruct the fusion state prediction results of the core monitoring points.

8. The bridge status information analysis method according to claim 1, characterized in that, The step of performing phase compensation and amplitude adjustment on the fusion state prediction results based on the phase lag relationship between other monitoring points and the core monitoring point within a certain phase synchronization cluster, to obtain the state prediction results of all monitoring points within a certain phase synchronization cluster, includes: Obtain the comprehensive modal information entropy sequence of the core monitoring point within a certain phase synchronization cluster, as well as the comprehensive modal information entropy sequences of other core monitoring points; The core instantaneous phase sequence is obtained by performing a Hilbert transform on the comprehensive modal information entropy sequence of the core monitoring points, and other instantaneous phase sequences are obtained by performing a Hilbert transform on the comprehensive modal information entropy sequences of other core monitoring points. Calculate the average phase difference between the core instantaneous phase sequence and the other instantaneous phase sequences within a preset dominant oscillation frequency range; The Fourier transform of the fusion status prediction result sequence of the core monitoring points is performed to obtain the frequency domain signal F(ω); A linear phase shift is applied to the frequency domain signal F(ω) to obtain a phase-compensated frequency domain signal F corresponding to the other core monitoring points. adjust(ω) =F(ω)*exp(-j*ω*(Δφ / (2πf dom ))), where Δφ is the average phase difference between the core instantaneous phase sequence and other instantaneous phase sequences within the preset dominant oscillation frequency range, f dom ω is the center frequency of the preset dominant oscillation frequency range, j is the imaginary unit; Obtain other vibration acceleration data sequences of the other core monitoring points during the historical monitoring period, as well as core vibration acceleration data sequences of the core monitoring points during the historical monitoring period; Bandpass filtering is performed on other vibration acceleration data sequences and core vibration acceleration data sequences respectively to extract other target vibration components and core target vibration components corresponding to the center frequency of the preset dominant oscillation frequency range. The other target vibration components are the target vibration components corresponding to the other vibration acceleration data sequences, and the core target vibration components are the target vibration components corresponding to the core vibration acceleration data sequences. Calculate the average ratio of the amplitude envelope of other target vibration components to the amplitude envelope of the core target vibration component during the historical monitoring period, and define the average ratio as the amplitude adjustment coefficient; The amplitude of the phase compensation frequency domain signal corresponding to the other core monitoring points is multiplied by the amplitude adjustment coefficient to obtain the target phase compensation frequency domain signal; The inverse Fourier transform of the target phase-compensated frequency domain signal is used to obtain the state prediction results of other core monitoring points.

9. A bridge status information analysis system, characterized in that, include: The acquisition module is configured to acquire multi-source time-series monitoring data from multiple monitoring points deployed on the bridge structure, and to perform synchronization and alignment processing on the multi-source time-series monitoring data from the same monitoring point to obtain multiple synchronized time-series data sequences from the same monitoring point. The first fusion module is configured to calculate the modal information entropy of each synchronous time series data sequence, and to fuse the modal information entropy of multiple synchronous time series data sequences corresponding to the same monitoring point to obtain a comprehensive modal information entropy sequence corresponding to each monitoring point. The partitioning module is configured to calculate the phase lock value between any integrated modal information entropy sequences, construct a phase synchronization network with monitoring points as nodes and phase lock values ​​as edge weights, and use a fuzzy clustering algorithm to partition the phase synchronization network to obtain at least one phase synchronization cluster, wherein the same phase synchronization cluster contains at least one monitoring point. The module is selected and configured to select a core monitoring point within a phase synchronization cluster based on the centrality index of the node in the phase synchronization network, and use the synchronization time series data sequence of the core monitoring point as the reference time series mode of a phase synchronization cluster. The generation module is configured to input the baseline time series pattern into a preset multi-scale time series prediction model. The multi-scale time series prediction model generates multi-scale future state predictions for core monitoring points by running prediction sub-networks based on different time scales in parallel. The second fusion module is configured to perform multi-resolution fusion based on wavelet transform on multi-scale future state predictions to obtain the fusion state prediction results of the core monitoring points. The adjustment module is configured to perform phase compensation and amplitude adjustment on the fusion state prediction results based on the phase lag relationship between other monitoring points and the core monitoring point in a certain phase synchronization cluster, so as to obtain the state prediction results of all monitoring points in a certain phase synchronization cluster.

Images