A method for reconstructing highway infrastructure monitoring data

Abstract

The present application relates to the field of highway infrastructure health monitoring, and particularly relates to a highway infrastructure monitoring data reconstruction method and device and storage medium; wherein the highway infrastructure monitoring data reconstruction method comprises the following steps: obtaining monitoring data of at least two highway infrastructure monitoring points, any one of the at least two highway infrastructure monitoring points is spatially associated with other highway infrastructure monitoring points; based on the monitoring data, a monitoring data two-dimensional matrix is constructed; based on the monitoring data two-dimensional matrix, a failure structure matrix is constructed; according to the failure structure matrix, the failure monitoring data in the monitoring data two-dimensional matrix is identified, and a data reconstruction input matrix is generated; using a data reconstruction model, based on the time continuity feature and the spatial correlation feature of the valid monitoring data in the data reconstruction input matrix, the failure monitoring data identified by the failure structure matrix is directionally predicted and compensated, and a data reconstruction result matrix is generated.

Description

Technical Field

[0001] This invention relates to the field of highway infrastructure health monitoring, and in particular to a method, apparatus, and storage medium for reconstructing highway infrastructure monitoring data. Background Technology

[0002] In actual operation, existing highway infrastructure health monitoring systems are susceptible to factors such as sensor failure, energy fluctuations, abnormal data transmission, and human error, leading to anomalies such as missing data, outliers, trend drift, and interference noise. This reduces the accuracy and reliability of structural damage identification and condition assessment. Summary of the Invention

[0003] The first aspect of this invention provides a method for reconstructing highway infrastructure monitoring data, comprising the following steps: Acquire monitoring data from at least two highway infrastructure monitoring points, wherein any one of the at least two highway infrastructure monitoring points is spatially associated with the other highway infrastructure monitoring points; Based on the monitoring data, a two-dimensional matrix of monitoring data is constructed, wherein one dimension of the two-dimensional matrix of monitoring data is used to identify the monitoring point and the other dimension is used to identify the sampling time. Based on the two-dimensional matrix of monitoring data, a failure structure matrix is ​​constructed, which is used to characterize the spatial-temporal distribution characteristics of the failure monitoring data. Based on the failure structure matrix, the failure monitoring data in the two-dimensional monitoring data matrix is ​​identified, and a data reconstruction input matrix is ​​generated. The data reconstruction input matrix is ​​used to guide the data reconstruction model to predict and compensate only for the failure monitoring data. Using a data reconstruction model, based on the temporal continuity and spatial correlation characteristics of the effective monitoring data in the data reconstruction input matrix, targeted prediction compensation is performed on the failure monitoring data identified by the failure structure matrix to generate a data reconstruction result matrix.

[0004] In some embodiments, the spatial association between any two monitoring points is determined by the spatial positional relationship and / or structural topological relationship between the two monitoring points.

[0005] In some embodiments, constructing a two-dimensional matrix of monitoring data based on the monitoring data includes the following steps: Define the preset time window; Extract synchronous sampling data from each monitoring point within the preset time window; The synchronously sampled data is arranged in a structured manner according to the monitoring point dimension and the time dimension to generate a two-dimensional matrix of monitoring data.

[0006] In some embodiments, the preset time window is a fixed time window, a sliding time window, or a dynamically determined time window based on missing data.

[0007] In some embodiments, constructing a failure structure matrix based on the two-dimensional matrix of monitoring data includes the following steps: Using the two-dimensional matrix of the monitoring data as the identification object, a data validity detection mechanism is established so that each matrix element is a data unit to be identified. According to the data validity judgment rules, each data unit to be identified in the two-dimensional matrix of monitoring data is tested for validity, and the monitoring point location and sampling time node corresponding to the invalid data are identified. Based on the identification results, a failure structure matrix with the same dimensions as the two-dimensional matrix of the monitoring data is generated. Each element in the failure structure matrix is ​​used to characterize the data status of the corresponding monitoring point at the corresponding sampling time node.

[0008] In some embodiments, the step of identifying the failure monitoring data in the two-dimensional monitoring data matrix based on the failure structure matrix and generating a data reconstruction input matrix includes the following steps: Based on the state identification results of the failure structure matrix, a failure location mapping relationship is established to determine the spatial-temporal location of the corresponding failure monitoring data in the two-dimensional matrix of monitoring data in the monitoring point dimension and the sampling time dimension. Based on the failure location mapping relationship, the monitoring data two-dimensional matrix is ​​filtered: for locations identified as valid data, the corresponding original sampled values ​​are retained; for locations identified as failed data, the original sampled values ​​are replaced with preset placeholder values. Based on the screening results, the data reconstruction input matrix is ​​generated. The data reconstruction input matrix has the same structure as the two-dimensional matrix of the monitoring data. It contains a preset placeholder value at the failure location and retains the original sampled value at the valid location.

[0009] In some embodiments, the data reconstruction model includes an input unit, a temporal feature extraction unit, a spatial feature extraction network, a feature fusion unit, and an output unit; The temporal feature extraction unit consists of at least one or more one-dimensional convolutional layers, global convolutional layers, or convolutional layers with dilated convolutional structures, used to extract the temporally continuous features; the spatial feature extraction network consists of at least one or more cross-channel convolutional layers, two-dimensional convolutional layers, or feature rearrangement fusion layers, used to extract the spatially related features; and the feature fusion unit consists of at least one or more feature splicing layers, convolutional mapping layers, or fully connected mapping layers, used to fuse the temporally continuous features and the spatially related features.

[0010] In some embodiments, the data reconstruction model further includes a reconstruction optimization unit; The reconstruction optimization unit is connected to the output unit and is used to perform distribution consistency constraint optimization on the prediction compensation results generated by the output unit.

[0011] Secondly, based on the highway infrastructure monitoring data reconstruction method provided in the first aspect, the present invention also provides a highway infrastructure monitoring data reconstruction device, comprising: The monitoring data acquisition module is used to acquire monitoring data from at least two highway infrastructure monitoring points, wherein any one of the at least two highway infrastructure monitoring points is spatially associated with the other highway infrastructure monitoring points. A two-dimensional matrix generation module is used to construct a two-dimensional matrix of monitoring data based on the monitoring data. One dimension of the two-dimensional matrix of monitoring data is used to identify the monitoring point, and the other dimension is used to identify the sampling time. The failure data identification module is used to construct a failure structure matrix based on the two-dimensional matrix of monitoring data. The failure structure matrix is ​​used to characterize the spatial-temporal distribution characteristics of the failure monitoring data. The reconstruction matrix input module is used to identify the failure monitoring data in the two-dimensional monitoring data matrix according to the failure structure matrix and generate a data reconstruction input matrix. The data reconstruction input matrix is ​​used to guide the data reconstruction model to perform prediction compensation only on the failure monitoring data. The reconstruction matrix output module is used to utilize the data reconstruction model to perform targeted prediction compensation on the failure monitoring data identified by the failure structure matrix based on the temporal continuity characteristics and spatial correlation characteristics of the effective monitoring data in the data reconstruction input matrix, and generate a data reconstruction result matrix.

[0012] Thirdly, based on the highway infrastructure monitoring data reconstruction method provided in the first aspect, the present invention also provides a computer storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the processor performs the steps of the highway infrastructure monitoring data reconstruction method provided in any of the above embodiments.

[0013] The highway infrastructure monitoring data reconstruction method, apparatus, and storage medium provided by this invention have gains including at least: This invention constructs a two-dimensional matrix of monitoring data and a failure structure matrix to achieve a unified structured representation and precise failure location of monitoring data in both spatial and temporal dimensions. Compared with traditional methods that make coarse-grained judgments on the entire channel or the entire time series, this invention can achieve element-by-element identification and precise marking of abnormal data, significantly improving the spatiotemporal resolution and processing accuracy of failure location. Furthermore, before data reconstruction, the present invention explicitly masks and identifies the failed data using a failure structure matrix and generates a data reconstruction input matrix to guide the data reconstruction model to perform targeted prediction and compensation only for the failure location, thereby avoiding indiscriminate modification of the original valid data, reducing the risk of erroneous repair, and improving data fidelity and physical consistency of structural response. Furthermore, this invention introduces a joint modeling mechanism of temporally continuous features and spatially correlated features in the data reconstruction stage, which can make full use of the response coordination information between spatially correlated monitoring points to improve reconstruction accuracy and prediction stability. Furthermore, by introducing a reconstruction optimization unit, the present invention optimizes the prediction compensation results through distribution consistency constraints, making the reconstructed data closer to the real monitoring data in terms of statistical distribution characteristics, spectral characteristics, and amplitude characteristics, thereby avoiding problems such as spectral distortion, loss of high-frequency information, or amplitude offset, and thus improving the engineering usability of the reconstruction results. Furthermore, the method and apparatus provided by this invention are applicable to monitoring scenarios of various types of highway infrastructure such as bridges, tunnels, and roadbeds. They can maintain the integrity and continuity of monitoring data under complex working conditions such as sensor failure, abnormal data transmission, or environmental interference, providing a more reliable data foundation for structural health assessment, damage identification, and condition prediction. Attached Figure Description

[0014] From the following description of embodiments in conjunction with the accompanying drawings, aspects, features, and advantages of the present invention will become clearer and more readily understood, in which: Figure 1 This is a schematic diagram of the highway infrastructure monitoring data reconstruction method provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the functional unit composition of the data reconstruction model provided in the embodiments of the present invention; Figure 3 This is a schematic diagram of the functional modules of the highway infrastructure monitoring data reconstruction device provided in an embodiment of the present invention. Detailed Implementation

[0015] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0016] It should be noted that the described embodiments are only a portion of the embodiments of the present invention, not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0017] In the description of this application, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary details.

[0018] In addition, the term "connection" should be interpreted broadly, where "A connects to B" includes both the case where A is directly connected to B and the case where A is indirectly connected to B via one or more middleware. "A connects between B and C" includes both the case where A is directly connected to B and C respectively and the case where A is indirectly connected between B and C via one or more middleware.

[0019] In one embodiment, see Figure 1 , Figure 1 This is a schematic diagram of the highway infrastructure monitoring data reconstruction method provided in an embodiment of the present invention.

[0020] The highway infrastructure described in this invention includes, but is not limited to, bridges, tunnels, pavements, roadbeds, piers, guardrails, drainage facilities, curtain walls, and other engineering structures requiring structural stress or operational status monitoring. The above examples are merely illustrative and do not constitute a limitation on the scope of this invention.

[0021] In the above-mentioned highway infrastructure health monitoring scenario, the highway infrastructure monitoring data of the present invention refers to the raw time-series numerical signals collected by one sensor or multiple sensors of the same type from the monitored structure at a predetermined sampling time sequence node. These data may include, but are not limited to, sensor identification information, timestamp information, and corresponding measurement values.

[0022] Furthermore, the data to be reconstructed can come from different types of monitoring sensors, including but not limited to data collected by structural health monitoring equipment such as accelerometers, vibration sensors, temperature sensors, humidity sensors, atmospheric pressure sensors, and ultrasonic detectors.

[0023] When monitoring data is lost, incomplete, or abnormal, this invention, based on existing valid data, and in accordance with... Figure 1 The illustrated process identifies and reconstructs failed data to recover missing or abnormal data, generating monitoring data that accurately reflects the actual operational status of highway infrastructure. Specifically, the highway infrastructure monitoring data reconstruction method includes: S01. Obtain monitoring data from at least two highway infrastructure monitoring points, wherein any one of the at least two highway infrastructure monitoring points is spatially associated with the other highway infrastructure monitoring points.

[0024] In this embodiment, spatial association refers to the physical coupling relationship formed between different highway infrastructure monitoring points based on spatial location or structural topology, which makes different monitoring points exhibit response transmission or structural continuity characteristics during structural stress, deformation, or vibration.

[0025] In some embodiments, the spatial association between any two monitoring points is determined by the spatial positional relationship between the two monitoring points.

[0026] Specifically, the spatial association between any two monitoring points is confirmed by the following steps: obtaining the installation coordinates of the two monitoring points in the highway infrastructure, and determining whether there is a spatial association based on the spatial distance between the two monitoring points: when the spatial distance between the two monitoring points is less than or equal to a preset distance threshold, it is determined that there is a spatial association between them.

[0027] It should be noted that the spatial distance is used to objectively define the spatial relationship between monitoring points, and it can be calculated based on the installation coordinates of the monitoring points in the highway infrastructure.

[0028] In one embodiment, the three-dimensional spatial coordinates of each monitoring point are obtained, and the spatial distance between any two monitoring points is calculated using the Euclidean distance formula.

[0029] In another embodiment, for monitoring points linearly arranged along the main beam of the bridge, the axis of the tunnel, or the extension direction of the roadbed, the projected distance along the structural axis can be used as the spatial distance: The projected position of each monitoring point along the main axis of the structure is recorded as the corresponding axial coordinate value. The spatial distance between two monitoring points is defined as the absolute value of the difference between their axial coordinate values.

[0030] In a further embodiment, spatial correlation is determined by a preset distance threshold. The preset distance threshold can be set according to the structural span length of the highway infrastructure, the size of the components, or the spacing of the monitoring points. For example, it can be set to 1 to 3 times the average spacing between adjacent monitoring points.

[0031] In other embodiments, the spatial association between any two monitoring points is determined by the structural topology between the two monitoring points.

[0032] Specifically, based on the structural design drawings or structural component connection relationships of the highway infrastructure, the structural component unit to which each monitoring point belongs is determined: when multiple monitoring points are located in the same structural component, the same continuous beam segment, the same slab unit, or are connected by rigid or semi-rigid connection components, it is determined that there is a structural topological relationship between them.

[0033] For example, multiple sensors deployed within the same main beam segment in a continuous beam bridge are identified as a spatially correlated monitoring point set because they jointly participate in the stress and deformation process of that beam segment.

[0034] In other embodiments, the spatial association is determined based on both spatial location relationships and structural topological relationships. Specifically, when monitoring points satisfy both the spatial distance threshold condition and belong to the same structural stress unit, they are preferentially identified as spatially associated monitoring points.

[0035] S02. Based on the monitoring data, construct a two-dimensional matrix of monitoring data. One dimension of the two-dimensional matrix of monitoring data is used to identify the monitoring point, and the other dimension is used to identify the sampling time.

[0036] In this embodiment, step S02 organizes and structures the raw monitoring data collected from multiple spatially related but dispersed highway infrastructure monitoring points according to a unified time window to form a unified data carrier for subsequent data reconstruction.

[0037] Specifically, the two-dimensional matrix of monitoring data can be represented as a two-dimensional data structure of M×T, where M represents the number of monitoring points and T represents the number of sampling points within a preset time window.

[0038] In one implementation, the rows of the two-dimensional matrix are used to identify different highway infrastructure monitoring points, and the columns of the two-dimensional matrix are used to identify continuous sampling time points within a preset time window: that is, each row corresponds to the complete time series data of a monitoring point within the preset time window, and each column corresponds to the response data of all spatially associated monitoring points at the same sampling time.

[0039] It should be noted that the present invention does not limit the monitoring point dimension to correspond to rows and the sampling time dimension to correspond to columns; in other embodiments, the arrangement of the monitoring point dimension and the sampling time dimension can be interchanged, as long as a two-dimensional data structure that can simultaneously represent the monitoring point set and the time sampling sequence can be formed, it is within the protection scope of the present invention.

[0040] Further, step S02, which involves constructing a two-dimensional matrix of monitoring data based on the monitoring data, includes the following steps: S021. Determine the preset time window.

[0041] In this embodiment, a preset time window is used to extract data segments with a uniform time length from continuous monitoring data to ensure that monitoring data corresponding to different monitoring points can be jointly modeled within the same time range; furthermore, its length can be determined according to the sampling frequency, data reconstruction accuracy requirements, and structural response characteristics of the highway infrastructure monitoring system.

[0042] In some embodiments, the preset time window is a fixed time window. Specifically, based on the sampling frequency and data reconstruction accuracy requirements of the highway infrastructure monitoring system, a fixed time interval, such as several seconds or several minutes, is preset. Then, data segments of equal length are extracted from the continuous monitoring data to ensure that data from different monitoring points are processed within the same time range.

[0043] In other embodiments, the preset time window is a sliding time window. Specifically, the continuous monitoring data is processed by sliding segmentation according to a preset time step. Each slide generates a new time interval data segment, wherein the time step can be less than, equal to, or greater than the time window length, thereby forming multiple overlapping or non-overlapping time intervals on a continuous time axis to adapt to application scenarios where the structural response changes dynamically over time.

[0044] In some other embodiments, the preset time window is determined dynamically based on missing data. Specifically, when a monitoring point is detected to have missing or abnormal data within a continuous time period, the time interval covering the missing time period is automatically determined as the preset time window. The length of the time window is determined based on the duration of the missing data and the minimum data range required for data reconstruction, so as to achieve targeted processing of local abnormal data.

[0045] S022. Extract the synchronous sampling data of each monitoring point within the preset time window.

[0046] In some embodiments, step S022, which involves extracting synchronous sampling data from each monitoring point within the preset time window, specifically includes the following steps: S0221. Based on the preset time window, obtain the start time and end time.

[0047] S0222. Extract the monitoring data located between the start time and the end time from the continuous monitoring data of each monitoring point to obtain the corresponding sequence data segment.

[0048] In some embodiments, when there are timestamp offsets, inconsistent sampling start and end times, or inconsistent number of sampling points between different monitoring points, step S022 further includes the following steps: aligning the monitoring data of each monitoring point for time alignment processing, so that different monitoring points correspond to the same sampling time sequence within the preset time window, and making the number of sampling points of each monitoring point the same within the preset time window.

[0049] The time alignment process is implemented by matching timestamps using the sampling time sequence of a preset time window as the alignment benchmark. Furthermore, when a monitoring point is missing a small number of sampling points within the preset time window, forward hold, neighbor sampling completion, or timestamp-based interpolation can be used to ensure that the number of sampling points of the monitoring point within the window is consistent with that of other monitoring points.

[0050] For example, assuming a preset time window of 10 seconds and a sampling frequency of 50Hz, the number of sampling points is T=500. Data segments within this 10-second interval are extracted from monitoring points A, B, and C respectively. If monitoring point B only obtains 498 sampling points within this time window due to transmission jitter, time alignment processing is performed to form a synchronous sequence of 500 sampling points within this time window. This ensures that monitoring points A, B, and C have a consistent number of sampling points on the same time axis and can be uniformly represented.

[0051] S023. The synchronous sampling data is arranged in a structured manner according to the monitoring point dimension and the time dimension to generate a two-dimensional matrix of monitoring data.

[0052] In some embodiments, step S023, which involves structuring the data according to the monitoring point dimension and the time dimension to form a two-dimensional matrix of monitoring data, specifically includes the following steps: S0231. Determine the length of the first dimension based on the number of monitoring points involved in the processing, and determine the length of the second dimension based on the number of sampling points within the preset time window.

[0053] In one implementation, the first dimension corresponds to the rows of the matrix, and the second dimension corresponds to the columns of the matrix; in other implementations, the arrangement of the first and second dimensions can be interchanged, i.e., the first dimension corresponds to the columns and the second dimension corresponds to the rows. This invention does not limit the specific arrangement of the monitoring point dimension and the time dimension in the matrix; as long as a two-dimensional data structure that simultaneously represents the set of monitoring points and the time sampling sequence can be formed, it falls within the scope of protection of this invention.

[0054] S0232. For the monitoring points involved in the processing, generate a monitoring point index sequence according to the monitoring point number, installation order, or preset sorting rules.

[0055] S0232. Write the sequence data segment of each monitoring point within the preset time window into the corresponding position of the two-dimensional matrix according to the monitoring point index, so that one dimension of the two-dimensional matrix is ​​used to identify the monitoring point and the other dimension is used to identify the sampling time.

[0056] For example, when the number of monitoring points M involved in the processing is 3 and the number of sampling points T within the preset time window is 5, a 3×5 two-dimensional matrix of monitoring data can be constructed, where the first row corresponds to the sampling value of monitoring point A at times t1 to t5, the second row corresponds to the sampling value of monitoring point B at times t1 to t5, and the third row corresponds to the sampling value of monitoring point C at times t1 to t5, thereby forming a structured data foundation for subsequent missing data identification and data reconstruction processing.

[0057] This invention structures the originally scattered time-series data from multiple monitoring points into a unified two-dimensional data matrix, providing a data foundation for subsequent joint data reconstruction based on spatial correlation.

[0058] S03. Based on the two-dimensional matrix of the monitoring data, construct a failure structure matrix, which is used to characterize the spatial-temporal distribution characteristics of the failure monitoring data.

[0059] In this embodiment, the failure structure matrix is ​​a state representation matrix with the same dimensional structure as the two-dimensional matrix of monitoring data, which is used to describe the failure distribution characteristics of the monitoring data in the monitoring point dimension and the time dimension.

[0060] Further, the construction of the failure structure matrix based on the two-dimensional matrix of monitoring data in step S03 includes the following steps: S031. Using the two-dimensional matrix of the monitoring data as the identification object, establish a data validity detection mechanism so that each matrix element is a data unit to be identified.

[0061] In this embodiment, the data validity detection mechanism is used to model the elements in the two-dimensional matrix of monitoring data into units, so that each element corresponds to a sampling data unit of "a certain monitoring point - a certain sampling time node".

[0062] By achieving element-by-element localization in both the monitoring point dimension and the sampling time dimension, the coarse-grained judgment of only the entire data segment or the entire channel is avoided, which helps to improve the spatiotemporal localization accuracy of failure identification and provides a unified data unit basis for the construction of the failure structure matrix.

[0063] S032. According to the data validity judgment rules, perform validity detection on each data unit to be identified in the two-dimensional matrix of monitoring data, and identify the monitoring point location and sampling time node corresponding to the invalid data.

[0064] In this embodiment, the data validity determination rule is used to determine whether the sampled data unit meets the preset validity conditions, and outputs the failure determination result accordingly.

[0065] Furthermore, the failure determination result includes at least the monitoring point index and sampling time index corresponding to the failure data, thereby realizing the location of the failure data.

[0066] In some embodiments, the data validity determination rules further include at least one of the following: missing data determination rules, out-of-bounds determination rules, mutation determination rules, and constant anomaly determination rules.

[0067] For example, when a sampled value is empty or missing, exceeds a preset physical reasonable range, the difference between adjacent samples exceeds a preset threshold, or remains unchanged for a duration exceeding a preset threshold at multiple consecutive sampling times, the corresponding data unit is determined to be invalid data.

[0068] In other embodiments, the data validity determination rule is implemented using an anomaly identification method based on a multimodal contrastive learning model.

[0069] Specifically, for any monitoring point in the two-dimensional matrix of monitoring data, a time-series data segment containing K consecutive sampling nodes is continuously extracted in the time dimension, and the time-series data segment is converted into a corresponding time-series image representation; the time-series image is input into a pre-trained multimodal contrastive learning model that aligns visual and linguistic modalities, and the matching degree between the time-series image and the preset data state type text label is calculated in the same semantic space, and the text label with the highest matching degree is output as the data state recognition result: when the output result is invalid data or an abnormal category, it is determined that the corresponding time-series segment contains invalid data.

[0070] The data status type text label includes valid data and invalid data. In some embodiments, it also includes anomaly categories such as trend drift, outlier, mutation, square wave interference, or missing data.

[0071] Furthermore, in some embodiments, a response heatmap is generated for the time series image based on the Class Activation Mapping (CAM) model, and the horizontal coordinate of the high response region in the image is mapped back to the original time series index, thereby locating the sampling time node corresponding to the failure data, and the location identification result of the failure data is output in combination with the monitoring point index.

[0072] S033. Based on the identification results, generate a failure structure matrix with the same dimensions as the two-dimensional matrix of the monitoring data. Each element in the failure structure matrix is ​​used to characterize the data status of the corresponding monitoring point at the corresponding sampling time node.

[0073] In this embodiment, the failure structure matrix is ​​an M×T two-dimensional matrix, where M represents the number of monitoring points and T represents the number of sampling points within a preset time window. Each element in the failure structure matrix corresponds to the monitoring data status at the same location in the monitoring data two-dimensional matrix, and is used to identify whether the monitoring data at that location is valid or invalid data.

[0074] For each data unit to be identified in the two-dimensional matrix of monitoring data, when it is determined to be valid data, a first identifier value is written at the corresponding position in the failure structure matrix; when it is determined to be invalid data, a second identifier value is written at the corresponding position in the failure structure matrix.

[0075] The first and second identifier values ​​are used to distinguish between valid and invalid data. The first and second identifier values ​​can be 1 and 0, respectively, or other distinguishable numerical forms or symbolic identifiers. As long as the data status can be distinguished, they are all within the protection scope of this invention.

[0076] For example, a two-dimensional matrix of monitoring data with a dimension of 3×5 corresponds to the response data of monitoring points A, B, and C at five sampling times t1 to t5. If the validity test result indicates that the data of monitoring point B is missing at time t3, and the data of monitoring point C is abnormal at times t4 and t5, then when constructing the failure structure matrix, the position corresponding to monitoring point B at time t3 is marked as a failure state, the positions corresponding to monitoring point C at times t4 and t5 are marked as failure states, and the remaining positions are marked as valid states, thereby forming a failure structure matrix with the same dimension as the two-dimensional matrix of monitoring data.

[0077] S04. Based on the failure structure matrix, identify the failure monitoring data in the two-dimensional monitoring data matrix, and generate a data reconstruction input matrix. The data reconstruction input matrix is ​​used to guide the data reconstruction model to perform prediction compensation only on the failure monitoring data.

[0078] In this embodiment, the data reconstruction input matrix is ​​a model input data structure built based on the two-dimensional matrix of monitoring data and the failure structure matrix. It is used to distinguish and express the effective monitoring data and the failure monitoring data in the spatial-temporal dimension, so that the data reconstruction model can predict and compensate only the corresponding positions of the failure data while keeping the effective data unchanged.

[0079] Further, step S04, which involves identifying the failure monitoring data in the two-dimensional monitoring data matrix based on the failure structure matrix and generating a data reconstruction input matrix, includes: S041. Based on the state identification results of the failure structure matrix, establish a failure location mapping relationship to determine the spatial-temporal location of the corresponding failure monitoring data in the two-dimensional matrix of monitoring data in the monitoring point dimension and the sampling time dimension.

[0080] S042. Based on the failure location mapping relationship, perform filtering processing on the two-dimensional matrix of monitoring data: for locations identified as valid data, retain the corresponding original sampled values; for locations identified as failed data, replace the original sampled values ​​with preset placeholder values.

[0081] S043. Based on the screening results, generate the data reconstruction input matrix. The data reconstruction input matrix has the same structure as the two-dimensional matrix of the monitoring data. It contains a preset placeholder value at the failure location and retains the original sampled value at the valid location.

[0082] For example, a two-dimensional matrix of monitoring data with a dimension of 3×5 corresponds to the response data of monitoring point A, monitoring point B and monitoring point C at 5 sampling times t1 to t5 respectively; at the same time, the failure structure matrix is ​​represented in the form of 0 / 1 mask, where "1" indicates that the data at the corresponding position is valid data and "0" indicates that the data at the corresponding position is failed data.

[0083] If the validity test results indicate that the data at monitoring point B is missing at time t3, and the data at monitoring point C is abnormal at times t4 and t5, then the element at position B-t3 in the failure structure matrix is ​​0, the element at positions C-t4 and C-t5 is 0, and the element at all other positions is 1.

[0084] When generating the data reconstruction input matrix, the monitoring data two-dimensional matrix is ​​processed according to the 0 / 1 mask: when the mask value is 1, the original sampled value at the corresponding position in the monitoring data two-dimensional matrix is ​​retained; when the mask value is 0, the data at the corresponding position is replaced with the preset placeholder value.

[0085] Therefore, all the original sampled values ​​of monitoring point A from t1 to t5 are retained; the original sampled values ​​of monitoring point B are retained at positions t1, t2, t4, and t5, while the value at position t3 is replaced with a placeholder value; the original sampled values ​​of monitoring point C are retained at positions t1, t2, and t3, while the value at positions t4 and t5 is replaced with a placeholder value. This forms a data reconstruction input matrix with the same dimensions as the original two-dimensional matrix of monitoring data, which is used by the subsequent data reconstruction model to perform directional prediction compensation for positions with a mask of 0.

[0086] S05. Using the data reconstruction model, based on the temporal continuity characteristics and spatial correlation characteristics of the effective monitoring data in the data reconstruction input matrix, perform targeted prediction compensation on the failure monitoring data identified by the failure structure matrix to generate a data reconstruction result matrix.

[0087] In this embodiment, the data reconstruction model is a space-time joint modeling model used to predict and compensate failure monitoring data based on the temporal continuity characteristics and spatial correlation characteristics of effective monitoring data. Its input is the data reconstruction input matrix, and its output is a data reconstruction result matrix with the same dimensions as the two-dimensional matrix of the monitoring data.

[0088] Furthermore, the time continuity feature refers to the continuity, trend and dynamic evolution characteristics of adjacent sampling nodes of the same monitoring point in the time dimension, including but not limited to the smooth change characteristics, periodic vibration characteristics, amplitude evolution trend, phase change relationship and short-term fluctuation and long-term dependence relationship of the time series.

[0089] Furthermore, the spatial correlation characteristics refer to the coupling response relationship characteristics formed between different spatially correlated monitoring points during structural stress, deformation, or vibration, including but not limited to the synchronization characteristics, phase difference characteristics, amplitude ratio relationship, correlation characteristics, and response transmission characteristics caused by structural topological relationships or spatial proximity relationships between monitoring points.

[0090] Specifically, the structure of the data reconstruction result matrix is ​​M×T, where the original sampled values ​​are preserved for valid data locations, and the prediction compensation results are filled for invalid data locations.

[0091] It should be noted that the data reconstruction model can be implemented using various spatial-temporal joint modeling paradigms, including but not limited to: prediction models based on time series dependency modeling, regression models based on multivariate joint modeling, low-rank reconstruction models based on matrix completion, and deep learning models based on spatiotemporal joint feature learning.

[0092] Furthermore, in deep learning implementations, structures such as convolutional neural network models, recurrent neural network models, graph neural network models, or generative adversarial network models can be used; in non-deep learning implementations, techniques such as statistical regression models, Kalman filter models, or low-rank matrix completion models can be used.

[0093] This invention does not limit the specific structural form of the data reconstruction model. As long as it can reconstruct the effective monitoring data in the input matrix based on the data, establish a correlation between the time dimension and the monitoring point dimension, and predict and compensate for the failure location, it falls within the protection scope of this invention.

[0094] In some embodiments, see Figure 2 , Figure 2 This is a schematic diagram of the functional unit composition of the data reconstruction model provided in the embodiments of the present invention.

[0095] like Figure 2As shown, the data reconstruction model provided in this embodiment of the invention includes an input unit, a temporal feature extraction unit, a spatial feature extraction unit, a feature fusion unit, and an output unit.

[0096] Furthermore, the input unit takes the data reconstruction input matrix as input and outputs an input feature representation containing valid data features and failure location identification information. Specifically, it receives the data reconstruction input matrix and performs dimensional expansion, normalization, or mask encoding on the input data to generate a uniformly formatted input feature tensor.

[0097] Furthermore, the time feature extraction unit consists of at least one or more one-dimensional convolutional layers, global convolutional layers, or convolutional layers with dilated convolutional structures; its input is the input feature representation output by the input unit, used to extract time features from the continuous sampling sequence of each monitoring point in the time dimension; its output is a time feature representation, used to characterize the temporal continuity characteristics, trend change characteristics, and long-term dependence characteristics of the structural response.

[0098] Furthermore, the spatial feature extraction network consists of at least one or more cross-channel convolutional layers, two-dimensional convolutional layers, or feature rearrangement fusion layers; its input is the temporal feature representation or input feature representation, which is used to jointly model the response coupling relationship between different monitoring points in the monitoring point dimension; its output is the spatial feature representation, which is used to characterize the coupling features formed between monitoring points based on spatial location relationships or structural topological relationships.

[0099] Furthermore, the feature fusion unit consists of at least one or more feature splicing layers, convolutional mapping layers, or fully connected mapping layers; its input is temporal feature representation and spatial feature representation, and it further performs fusion operations through feature splicing, weighted fusion, or attention mechanism to output a fused feature representation of temporal feature representation and spatial feature representation.

[0100] Furthermore, the output unit is connected to the feature fusion unit and is used to output the predicted compensation value corresponding to the failure position according to the fused feature representation, and keep the original sampled value unchanged at the effective data position, thereby generating a data reconstruction result matrix.

[0101] In this embodiment, the time feature extraction unit performs multi-scale convolution operations on the time series of each monitoring point to obtain time-continuous features that reflect the evolution trend and periodic characteristics of structural vibration; the spatial feature extraction network performs cross-channel feature interaction operations on the monitoring point dimension to obtain spatial correlation features that reflect the response coupling relationship of different monitoring points; and the feature fusion unit performs joint mapping of time features and spatial features to obtain a fused feature representation that simultaneously contains time-continuous information and spatial coupling information.

[0102] For example, given a 3×5 two-dimensional matrix of monitoring data, and a failure structure matrix indicating that monitoring point B fails at position t3 and monitoring point C fails at positions t4 and t5, the data reconstruction model first extracts the temporal continuous features of each monitoring point in the time dimension, then models the spatial coupling relationship between A, B, and C in the monitoring point dimension, and then maps the fused features, outputting only the predicted compensation values ​​at positions B-t3, C-t4, and C-t5, and fused with the original valid data to generate a complete 3×5 data reconstruction result matrix.

[0103] Compared to existing data completion models that only model time series data based on a single monitoring point (such as one-dimensional convolutional neural networks, recurrent neural networks, or long short-term memory networks), or models that only rely on time interpolation, statistical regression, and other methods for data repair, the data reconstruction model provided in this embodiment can not only characterize the temporal continuity of structural response, but also use the coupling relationship between spatially correlated monitoring points to constrain prediction. This avoids the error accumulation problem caused by traditional deep learning models ignoring spatial correlation or reconstructing all data indiscriminately, thus having significant advantages in prediction stability, structural response consistency, and physical rationality.

[0104] In some embodiments, after completing the initial prediction compensation based on temporal continuity features and spatial correlation features, the data reconstruction model also includes a reconstruction optimization unit for further optimizing the consistency of the prediction result distribution.

[0105] Specifically, the reconstruction optimization unit is connected to the output unit and is used to optimize the distribution consistency constraint of the prediction compensation results generated by the output unit, so that the prediction data is closer to the real monitoring data in terms of statistical characteristics and spectral structure.

[0106] Furthermore, the reconstruction optimization unit includes a generator and a discriminator structure. The generator outputs the predicted compensation value corresponding to the failure location; the discriminator receives the predicted compensation result output by the generator and the actual valid monitoring data, and discriminates the differences between the two in terms of statistical distribution characteristics, spectral characteristics, or amplitude distribution characteristics.

[0107] In this embodiment, an adversarial training mechanism is constructed to enable joint optimization between the generator and the discriminator: the generator continuously adjusts its parameters so that its output can "deceive" the discriminator; the discriminator continuously improves its ability to distinguish between real data and predicted data. Through this adversarial process, the prediction compensation results output by the generator are made to approximate the distribution of real and effective monitoring data at the overall distribution level, while maintaining the consistency of temporal continuity features and spatial correlation features.

[0108] For example, monitoring point B has invalid data at position t3, and monitoring point C has invalid data at positions t4 and t5. After processing by the time feature extraction unit and the spatial feature extraction network, the output unit generates the predicted compensation value for the corresponding position. Subsequently, the data sequence containing the predicted compensation value and the actual valid monitoring data sequence are input into the discriminator, which determines whether they conform to the distribution characteristics of the actual monitoring data. If the predicted result deviates from the actual data distribution in terms of spectral distribution or amplitude statistical characteristics, the generator parameters are adjusted in reverse using an adversarial loss function to make its output closer to the statistical characteristics of the actual monitoring data.

[0109] In these embodiments, by introducing the reconstruction optimization unit, not only can the accuracy of the predicted compensation value be improved at the numerical level, but the reconstruction result can also be guaranteed to be consistent with the real monitoring data in terms of spectral structure, amplitude distribution and physical rationality of structural response, thereby improving the engineering applicability and reliability of the data reconstruction result.

[0110] In one embodiment, a highway infrastructure monitoring data reconstruction device is provided. This device can be deployed in the data processing server, edge computing node, or centralized monitoring platform of a highway infrastructure health monitoring system. It is used to automatically identify and reconstruct the original monitoring data when monitoring data is missing or abnormal.

[0111] Please see Figure 3 , Figure 3 This is a schematic diagram illustrating the functional modules of the highway infrastructure monitoring data reconstruction device provided in an embodiment of the present invention. Figure 3 As shown, the highway infrastructure monitoring data reconstruction device provided in this embodiment includes a monitoring data acquisition module, a two-dimensional matrix generation module, a failure data identification module, a reconstruction matrix input module, and a reconstruction matrix output module.

[0112] Furthermore, the monitoring data acquisition module is used to acquire monitoring data from at least two highway infrastructure monitoring points, which are spatially correlated. This spatial correlation can be determined based on the spatial location or structural topology between the monitoring points; for example, multiple sensors located on the same bridge main beam segment or the same structural component unit may have a physical coupling relationship during structural response. The monitoring data may include time-series data such as sensor identifiers, timestamps, and corresponding measurement values.

[0113] Furthermore, the two-dimensional matrix generation module is used to construct a two-dimensional matrix of monitoring data based on the monitoring data. Specifically, this module performs unified time window truncation and alignment processing on time-series data from multiple spatially correlated monitoring points, and arranges them in a structured manner according to the "monitoring point dimension - time dimension" to generate an M×T two-dimensional matrix of monitoring data, where M represents the number of monitoring points and T represents the number of sampling points within a preset time window. This two-dimensional matrix provides a unified data carrier for subsequent failure identification and data reconstruction.

[0114] Furthermore, the failure data identification module is used to construct a failure structure matrix based on the two-dimensional matrix of monitoring data. This module performs data validity checks on each data unit in the two-dimensional matrix, identifying missing data, out-of-bounds data, or abnormally abrupt data changes, and generates a failure structure matrix with the same dimensions as the two-dimensional matrix of monitoring data. The elements in the failure structure matrix are used to characterize the data status of the corresponding monitoring point at the corresponding sampling time node, for example, using a 0 / 1 mask format, where "1" represents valid data and "0" represents failed data.

[0115] Furthermore, the reconstruction matrix input module is used to identify the failure monitoring data in the two-dimensional matrix of monitoring data based on the failure structure matrix and generate a data reconstruction input matrix. Specifically, this module retains the original sampled values ​​at valid data locations and replaces them with preset placeholder values ​​at failure locations according to the failure location mapping relationship. This ensures that the failure locations are explicitly retained in the data structure but do not participate in direct calculation, thereby forming a data reconstruction input matrix consistent with the original two-dimensional matrix structure. This matrix guides the data reconstruction model to predict and compensate only for failure locations.

[0116] Furthermore, the reconstruction matrix output module is used to process the data reconstruction input matrix using a data reconstruction model. Based on the temporal continuity characteristics of the effective monitoring data and the spatial correlation characteristics between monitoring points, this module performs targeted prediction compensation on the failure monitoring data identified by the failure structure matrix and outputs a data reconstruction result matrix. The data reconstruction result matrix has the same dimensions as the original two-dimensional monitoring data matrix, maintaining the original sampled values ​​at the effective data locations and filling the failure data locations with prediction compensation results.

[0117] For example, the monitoring points involved in the processing include monitoring point A, monitoring point B, and monitoring point C. The number of sampling points within the preset time window is 5, so the two-dimensional matrix of monitoring data has a 3×5 structure. If the failure data identification module identifies that monitoring point B has missing data at time t3 and monitoring point C has abnormal data at times t4 and t5, the reconstruction matrix input module writes placeholder values ​​at the corresponding positions, and retains the original data at the other positions. Subsequently, the reconstruction matrix output module outputs predicted compensation values ​​only for positions B-t3, C-t4, and C-t5 based on the continuous response characteristics in the time dimension and the spatial coupling relationship between the three monitoring points, and merges them with the original valid data to generate a complete 3×5 data reconstruction result matrix.

[0118] Through the coordinated operation of the above modules, the device provided in this embodiment can achieve targeted reconstruction of failure monitoring data while keeping the effective data undisturbed, thereby improving data integrity and physical consistency of structural response. It is suitable for health monitoring scenarios of highway infrastructure such as bridges, tunnels, and roadbeds.

[0119] In one embodiment, a computer-readable storage medium is also provided, having a computer program stored thereon, which, when executed by a processor, causes the processor to perform the highway infrastructure monitoring data reconstruction method as described in any of the foregoing embodiments.

[0120] Specifically, when the computer program is executed, it performs the following steps: acquiring monitoring data from at least two highway infrastructure monitoring points with spatial correlation; constructing a two-dimensional matrix of monitoring data; constructing a failure structure matrix based on the two-dimensional matrix of monitoring data; generating a data reconstruction input matrix based on the failure structure matrix; and performing predictive compensation on the failure monitoring data based on the data reconstruction input matrix to generate a data reconstruction result matrix.

[0121] It should be noted that the computer program may exist in the form of source code, object code, executable file or any other form, and may be loaded and executed by computing devices including general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs) or other programmable logic devices.

[0122] Furthermore, the computer-readable storage medium may be, but is not limited to, a read-only memory (ROM), a random access memory (RAM), a solid-state drive, flash memory, a disk, an optical disk, or other storage media capable of storing program code and readable by a computer device. It is used to store program instructions and is not limited to a specific hardware structure. As long as it can store and execute the program code of the highway infrastructure monitoring data reconstruction method, it falls within the protection scope of this invention.

[0123] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0124] It should be noted that the above embodiments can be freely combined as needed. The above are merely preferred embodiments of the present invention; it should be observed that those skilled in the art can make several improvements and modifications without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for reconstructing monitoring data of highway infrastructure, characterized in that, Includes the following steps: Acquire monitoring data from at least two highway infrastructure monitoring points, wherein any one of the at least two highway infrastructure monitoring points is spatially associated with the other highway infrastructure monitoring points; Based on the monitoring data, a two-dimensional matrix of monitoring data is constructed, wherein one dimension of the two-dimensional matrix of monitoring data is used to identify the monitoring point and the other dimension is used to identify the sampling time. Based on the two-dimensional matrix of monitoring data, a failure structure matrix is ​​constructed, which is used to characterize the spatial-temporal distribution characteristics of the failure monitoring data. Based on the failure structure matrix, the failure monitoring data in the two-dimensional monitoring data matrix is ​​identified, and a data reconstruction input matrix is ​​generated. The data reconstruction input matrix is ​​used to guide the data reconstruction model to predict and compensate only for the failure monitoring data. Using a data reconstruction model, based on the temporal continuity and spatial correlation characteristics of the effective monitoring data in the data reconstruction input matrix, targeted prediction compensation is performed on the failure monitoring data identified by the failure structure matrix to generate a data reconstruction result matrix.

2. The method for reconstructing highway infrastructure monitoring data according to claim 1, characterized in that, The spatial relationship between any two monitoring points is determined by the spatial positional relationship and / or structural topological relationship between the two monitoring points.

3. The method for reconstructing highway infrastructure monitoring data according to claim 1, characterized in that, The step of constructing a two-dimensional matrix of monitoring data based on the monitoring data includes the following steps: Define the preset time window; Extract synchronous sampling data from each monitoring point within the preset time window; The synchronously sampled data is arranged in a structured manner according to the monitoring point dimension and the time dimension to generate a two-dimensional matrix of monitoring data.

4. The method for reconstructing highway infrastructure monitoring data according to claim 3, characterized in that, The preset time window can be a fixed time window, a sliding time window, or a dynamically determined time window based on missing data.

5. The method for reconstructing highway infrastructure monitoring data according to claim 1, characterized in that, The construction of the failure structure matrix based on the two-dimensional matrix of monitoring data includes the following steps: Using the two-dimensional matrix of the monitoring data as the identification object, a data validity detection mechanism is established so that each matrix element is a data unit to be identified. According to the data validity judgment rules, each data unit to be identified in the two-dimensional matrix of monitoring data is tested for validity, and the monitoring point location and sampling time node corresponding to the invalid data are identified. Based on the identification results, a failure structure matrix with the same dimensions as the two-dimensional matrix of the monitoring data is generated. Each element in the failure structure matrix is ​​used to characterize the data status of the corresponding monitoring point at the corresponding sampling time node.

6. The method for reconstructing highway infrastructure monitoring data according to claim 1, characterized in that, The step of identifying the failure monitoring data in the two-dimensional monitoring data matrix based on the failure structure matrix and generating a data reconstruction input matrix includes the following steps: Based on the state identification results of the failure structure matrix, a failure location mapping relationship is established to determine the spatial-temporal location of the corresponding failure monitoring data in the two-dimensional matrix of monitoring data in the monitoring point dimension and the sampling time dimension. Based on the failure location mapping relationship, the monitoring data two-dimensional matrix is ​​filtered: for locations identified as valid data, the corresponding original sampled values ​​are retained; for locations identified as failed data, the original sampled values ​​are replaced with preset placeholder values. Based on the screening results, the data reconstruction input matrix is ​​generated. The data reconstruction input matrix has the same structure as the two-dimensional matrix of the monitoring data. It contains a preset placeholder value at the failure location and retains the original sampled value at the valid location.

7. The method for reconstructing highway infrastructure monitoring data according to claim 1, characterized in that, The data reconstruction model includes an input unit, a temporal feature extraction unit, a spatial feature extraction network, a feature fusion unit, and an output unit. The temporal feature extraction unit consists of at least one or more one-dimensional convolutional layers, global convolutional layers, or convolutional layers with dilated convolutional structures, used to extract the temporally continuous features; the spatial feature extraction network consists of at least one or more cross-channel convolutional layers, two-dimensional convolutional layers, or feature rearrangement fusion layers, used to extract the spatially related features; and the feature fusion unit consists of at least one or more feature splicing layers, convolutional mapping layers, or fully connected mapping layers, used to fuse the temporally continuous features and the spatially related features.

8. The method for reconstructing highway infrastructure monitoring data according to claim 7, characterized in that, The data reconstruction model also includes a reconstruction optimization unit; The reconstruction optimization unit is connected to the output unit and is used to perform distribution consistency constraint optimization on the prediction compensation results generated by the output unit.

9. A device for reconstructing monitoring data of highway infrastructure, characterized in that, include: The monitoring data acquisition module is used to acquire monitoring data from at least two highway infrastructure monitoring points, wherein any one of the at least two highway infrastructure monitoring points is spatially associated with the other highway infrastructure monitoring points. A two-dimensional matrix generation module is used to construct a two-dimensional matrix of monitoring data based on the monitoring data. One dimension of the two-dimensional matrix of monitoring data is used to identify the monitoring point, and the other dimension is used to identify the sampling time. The failure data identification module is used to construct a failure structure matrix based on the two-dimensional matrix of monitoring data. The failure structure matrix is ​​used to characterize the spatial-temporal distribution characteristics of the failure monitoring data. The reconstruction matrix input module is used to identify the failure monitoring data in the two-dimensional monitoring data matrix according to the failure structure matrix and generate a data reconstruction input matrix. The data reconstruction input matrix is ​​used to guide the data reconstruction model to perform prediction compensation only on the failure monitoring data. The reconstruction matrix output module is used to utilize the data reconstruction model to perform targeted prediction compensation on the failure monitoring data identified by the failure structure matrix based on the temporal continuity characteristics and spatial correlation characteristics of the effective monitoring data in the data reconstruction input matrix, and generate a data reconstruction result matrix.

10. A computer storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, causes the processor to perform the steps of the highway infrastructure monitoring data reconstruction method as described in any one of claims 1 to 8.

Images