Bridge health monitoring abnormal data point identification method and device of multi-granularity time window

Abstract

The application discloses a bridge abnormal point identification method and device based on a multi-granularity time window, and the method comprises the following steps: acquiring a plurality of sample data of a bridge by using a real-time monitoring system; constructing a sample initial feature vector representation of a sliding time window, and fusing the sample initial feature vector representation to construct a sample initial feature matrix representation; fitting a linear relationship and a nonlinear relationship of sample neighborhood features of the sliding time window; based on the linear relationship and the nonlinear relationship, fusing linear residual and nonlinear residual block features to identify an abnormal probability value of a sample, and calculating the abnormal probability value of the sample to label an abnormal sample; and identifying abnormal values in the plurality of sample data by using the labeled abnormal sample to identify a bridge abnormal point. The application predicts an abnormal probability of a sample from potential neighborhood features of the sample in a multi-granularity sliding time window, realizes identification of abnormal data points, and provides great support in bridge structure health monitoring data anomaly diagnosis.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of bridge structure health monitoring anomaly identification, and in particular to a bridge health monitoring anomaly data point identification method and device based on a multi-granularity time window. BACKGROUND

[0002] In the process of bridge structure health monitoring, the data collected by sensors often have missing, noise and abnormal data, and the identification of abnormal data is crucial for the structural safety diagnosis and evaluation of bridges. Due to the large scale of data collected by sensors, the difficulty of manually identifying abnormal data is increased. How to preprocess the time series monitoring data automatically collected by sensors, identify abnormal data, and realize the identification and trend analysis of bridge structure damage is an important part of intelligent bridge health monitoring, which can effectively analyze the correlation between bridge response and load and reduce the cost of manual inspection in bridge structure health monitoring.

[0003] The identification of abnormal samples in time series data needs to consider the information of neighboring samples. The neighborhood features of sample points can overcome the problem of insufficient current sample information to some extent, which is helpful to expand the initial feature information of samples and is of great help to abnormal identification, which can improve the accuracy of sample abnormal identification. SUMMARY

[0004] The present application aims to at least solve one of the technical problems in the related art.

[0005] To this end, one object of the present application is to provide a bridge anomaly point identification method based on a multi-granularity time window, which can improve the accuracy of bridge structure health monitoring anomaly data point identification. The present application predicts the abnormal probability of samples from the potential neighborhood features of samples in a multi-granularity sliding time window, realizes the identification of abnormal data points, and provides great support in the aspect of bridge structure health monitoring data anomaly diagnosis.

[0006] Another object of the present application is to provide a bridge anomaly point identification device based on a multi-granularity time window.

[0007] To achieve the above-mentioned purpose, the present application provides a bridge anomaly point identification method based on a multi-granularity time window, which comprises:

[0008] The method comprises the following steps: acquiring a plurality of sample data of a bridge by using a real-time monitoring system, wherein the plurality of sample data comprises bridge displacement and strain; constructing a sample initial feature vector representation of a sliding time window based on the plurality of sample data, and fusing the sample initial feature vector representation to construct a sample initial feature matrix representation; fitting a linear relationship and a nonlinear relationship of sample neighborhood features of the sliding time window based on the sample initial feature matrix representation; identifying an abnormal probability value of the sample by fusing linear residual and nonlinear residual block features based on the linear relationship and the nonlinear relationship, and calculating the abnormal probability value of the sample to label an abnormal sample; and identifying an abnormal value in the plurality of sample data by using the labeled abnormal sample to identify a bridge abnormal point.

[0009] The bridge abnormal point identification method based on the multi-granularity time window can predict the abnormal probability of a sample from the potential neighborhood features of the sample in a multi-granularity sliding time window, identify abnormal data points, and provide great support for bridge structure health monitoring data anomaly diagnosis.

[0010] In addition, the bridge abnormal point identification method based on the multi-granularity time window according to the above-mentioned embodiments of the present application can have the following additional technical features:

[0011] Further, in an embodiment of the present application, the constructing of the sample initial feature vector representation of the sliding time window based on the plurality of sample data comprises: initializing a feature of a preset sample x at a t moment x t , and taking a neighborhood initialization feature vector representation of the sample x as x t-2 , x t-1 , x t , x t+1 , x t+2 .

[0012] Further, in an embodiment of the present application, the fusing of the sample initial feature vector representation to construct a sample initial feature matrix representation comprises: according to a sliding time window δ, taking a δ left neighborhood initialization feature vector representation of the sample x as x - =(x t-2-δ , x t-1-δ , x t-δ , x t+1-δ , x t+2-δ , and a δ right neighborhood initialization feature vector representation of the sample x as x + =(x t-2+δ , x t-1+δ , x t+δ , x t+1+δ , x t+2+δ); fusing the left-neighborhood initialization feature and the right-neighborhood initialization feature of the sample x, constructing an initial feature matrix representation of the sample x as: x = [x0; x - ; x + ].

[0013] Further, in an embodiment of the present application, the fitting of the linear relationship and the nonlinear relationship of the sample neighborhood feature of the sliding time window based on the initial feature matrix representation of the sample includes: fitting the linear relationship of the sample x and the classification label y through a linear classifier based on the initial feature matrix representation x of the sample x; and fitting the nonlinear relationship of the sample x and the classification label y through a nonlinear classifier .

[0014] Further, in an embodiment of the present application, the fusion of the linear residual and the nonlinear residual block feature to identify the anomaly probability value of the sample based on the linear relationship and the nonlinear relationship includes: combining the linear residual and the nonlinear residual feature of the sample x, calculating the anomaly probability value of the sample x based on the anomaly probability value . The objective function of the anomaly probability value of the sample is defined as .

[0015] Further, in an embodiment of the present application, the calculation of the anomaly probability value of the sample to label the anomaly sample includes: for a sample s, according to the optimized weight parameters w1, w2 and W, calculating the anomaly probability value of the sample s. If greater than or equal to 0.5, the sample is labeled as normal, and the classification label is recorded as 1; otherwise, the sample is labeled as abnormal, and the classification label is recorded as 0.

[0016] To achieve the above purpose, another aspect of the present application provides a bridge anomaly point identification device based on a multi-granularity time window, comprising:

[0017] The data acquisition module is used for detecting and acquiring a plurality of sample data of the bridge by using a real-time monitoring system, wherein the plurality of sample data comprises bridge displacement and strain.

[0018] The bridge abnormal point identification device based on the multi-granularity time window according to the embodiment of the application predicts the abnormal probability of the sample from the potential neighborhood features of the sample in the multi-granularity sliding time window, realizes the identification of the abnormal data points, and provides great support for the bridge structure health monitoring data anomaly diagnosis.

[0019] Additional aspects and advantages of the application will be set forth in part in the description which follows, and in part will become apparent to those skilled in the art upon examination of the following and / or can be learned by practice of the application. BRIEF DESCRIPTION OF DRAWINGS

[0020] The above and / or additional aspects and advantages of the application will become apparent and be readily understood by reference to the following description, and by practicing the application, in conjunction with the accompanying drawings, in which:

[0021] Figure 1 A flowchart of the bridge abnormal point identification method based on the multi-granularity time window according to the embodiment of the application is shown in FIG. 1.

[0022] Figure 2 A labeled data sample schematic diagram according to the embodiment of the application is shown in FIG. 2.

[0023] Figure 3 A predicted abnormal sample schematic diagram according to the embodiment of the application is shown in FIG. 3.

[0024] Figure 4 A structure schematic diagram of the bridge abnormal point identification device based on the multi-granularity time window according to the embodiment of the application is shown in FIG. 4. DETAILED DESCRIPTION

[0025] It should be noted that the embodiments and features in the embodiments in the present application can be combined with each other without conflict. The present application will be described in detail below with reference to the accompanying drawings and in conjunction with the embodiments.

[0026] In the following, the technical solutions in the embodiments of the present application will be described clearly and completely with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part but not all of the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative effort should fall into the protection scope of the present application.

[0027] In addition, the terms "first", "second", etc. are used only for the purpose of description and should not be understood as indicating or implying relative importance or implying the number of the technical features indicated. Therefore, the features defined with "first", "second", etc. can explicitly or implicitly include one or more of the features. In the description of the present application, the meaning of "multiple" is two or more, unless otherwise explicitly specified and limited.

[0028] In the present application, unless otherwise explicitly specified and limited, "on" or "under" of the first feature to the second feature can include that the first and second features are in direct contact or that the first and second features are not in direct contact but are in contact through another feature between them. Moreover, "on", "above" and "over" of the first feature to the second feature include that the first feature is directly above and obliquely above the second feature, or only means that the first feature is higher than the second feature in horizontal height. "Under", "below" and "under" of the first feature to the second feature include that the first feature is directly above and obliquely above the second feature, or only means that the first feature is lower than the second feature in horizontal height.

[0029] The multi-granularity time window-based bridge abnormal point identification method and device proposed according to the embodiments of the present application are described below with reference to the drawings.

[0030] Figure 1 is a flowchart of the multi-granularity time window-based bridge abnormal point identification method according to an embodiment of the present application.

[0031] As shown in Figure 1 , the method includes but is not limited to the following steps:

[0032] Step S1, detecting and acquiring a plurality of sample data of the bridge by using a real-time monitoring system; wherein the plurality of sample data includes bridge displacement and strain.

[0033] It can be understood that the present application can obtain relevant data of bridge health monitoring by installing sensors and other devices on the bridge.

[0034] Specifically, in the process of acquiring actual bridge data, real-time monitoring systems such as wireless displacement sensors, acceleration sensors, and vibration sensors are mainly used to detect changes in the bridge's road surface, such as displacement and strain. In addition, images and videos can be monitored through cameras, collected through IoT gateways, and transmitted to servers and monitoring and control terminal display platforms via base stations.

[0035] Step S2: Construct the initial feature vector representation of the sample for the sliding time window based on multiple sample data, and fuse the initial feature vector representation of the sample to construct the initial feature matrix representation of the sample.

[0036] As an example, suppose the initial features of sample x at time t are... t The initial feature vector of the neighborhood of sample x is represented as x0 = (x t-2 ,x t-1 x t x t+1 x t+2 ), to construct the initial feature vector representation of the samples for the sliding time window.

[0037] As an example, based on the sliding time window δ, the feature vector of the δ-left neighbor of sample x is initialized as x. - =(x t-2-δ ,x t-1-δ x t-δ x t+1-δ x t+2-δ The initial eigenvector of the right neighbor of δ is represented as x. + =(x t-2+δ ,x t-1+δ x t+δ x t+1+δ x t+2+δ ), used to enhance the input features of sample x.

[0038] Furthermore, by fusing the δ-left neighbor initialization features and the right neighbor initialization features of sample x, an initial feature matrix representation of sample x is constructed: x = [x0; x...]. - ;x + The initial feature matrix representation of the sample is constructed by fusing the initial feature vector representation of the sample with multi-granularity sliding time windows.

[0039] Step S3: Based on the initial feature matrix representation of the samples, fit the linear and nonlinear relationships of the sample neighborhood features within the sliding time window.

[0040] Specifically, based on the initial feature matrix representation x of sample x, a linear classifier is used. The linear relationship between sample x and classification label y is fitted to perform linear relationship fitting of sample neighborhood features with multi-granularity sliding time window based on linear classifier.

[0041] Further, based on the initial feature matrix representation x of the sample x, a nonlinear classifier is fitted to fit the nonlinear relationship of the sample x and the classification label y, to fit the nonlinear relationship of the sample neighborhood features of the multi-granularity sliding time window based on the nonlinear classifier.

[0042] Step S4, based on the linear relationship and the nonlinear relationship, the linear residual and the nonlinear residual block features are fused to identify the abnormal probability value of the sample, and the abnormal probability value of the sample is calculated to label the abnormal sample.

[0043] Specifically, the linear residual of the sample x is combined with the nonlinear residual and the nonlinear residual feature, and the classification abnormal probability of the sample x is calculated

[0044] Further, based on the classification abnormal probability value of the sample , the objective function of the abnormal probability of the sample is defined as Optimize the weight parameters w1, w2 and W.

[0045] Further, for a new sample s, according to the learned weight parameters w1, w2 and W, the classification abnormal probability of the new sample is calculated

[0046] Further, based on the classification abnormal probability of the sample If greater than or equal to 0.5, the sample is normal, and the classification label is marked as 1, otherwise the sample is abnormal, and marked as 0.

[0047] Further, as shown in Figure 2 , first, manually mark part of the abnormal points based on the sigma principle, and mark the normal sample as 1 and the abnormal sample as 0. The labeled data sample is as shown in Figure 2 .

[0048] Further, through the abnormal point identification of the multi-granularity time window, the label of the abnormal point is predicted. Input 70000 samples, and the sample abnormality is marked as follows: red for abnormal samples and green for normal samples, as shown in Figure 3 .

[0049] Thus, by analyzing and processing the bridge health monitoring data, the abnormal values in the displacement monitoring data are identified, and the safety hazard of the bridge is found, providing decision support for the overall structure state evaluation, bridge damage identification and management and maintenance of the bridge. The abnormal values mainly refer to the abnormal data in data analysis.

[0050] The bridge abnormal point identification method based on the multi-granularity time window according to the embodiment of the application predicts the abnormal probability of the sample from the potential neighborhood features of the sample in the multi-granularity sliding time window, realizes the identification of the abnormal data points, and provides great support for the abnormal diagnosis of the bridge structure health monitoring data.

[0051] To realize the above-mentioned embodiments, as shown in the embodiment, the bridge abnormal point identification device 10 based on the multi-granularity time window is further provided, and the device 10 comprises a data acquisition module 100, a feature representation module 200, a relationship fitting module 300, a probability calculation module 400 and an abnormal identification module 500. Figure 4

[0052] The data acquisition module 100 is used for detecting and acquiring a plurality of sample data of the bridge by using a real-time monitoring system, wherein the plurality of sample data comprises bridge displacement and strain.

[0053] The feature representation module 200 is used for constructing a sample initial feature vector representation of a sliding time window based on the plurality of sample data, and fusing the sample initial feature vector representation to construct a sample initial feature matrix representation.

[0054] The relationship fitting module 300 is used for fitting the linear relationship and the nonlinear relationship of the sample neighborhood features of the sliding time window based on the sample initial feature matrix representation.

[0055] The probability calculation module 400 is used for fusing the linear residual and the nonlinear residual block features to identify the abnormal probability value of the sample based on the linear relationship and the nonlinear relationship, calculating the abnormal probability value of the sample to label the abnormal sample.

[0056] The abnormal identification module 500 is used for identifying the abnormal value in the plurality of sample data by the labeled abnormal sample to identify the bridge abnormal point.

[0057] The bridge abnormal point identification device based on the multi-granularity time window according to the embodiment of the application predicts the abnormal probability of the sample from the potential neighborhood features of the sample in the multi-granularity sliding time window, realizes the identification of the abnormal data points, and provides great support for the abnormal diagnosis of the bridge structure health monitoring data.

[0058] It should be noted that the aforementioned explanation and description of the bridge abnormal point identification method based on the multi-granularity time window also applies to the bridge abnormal point identification device based on the multi-granularity time window, which will not be described here.

[0059] ​In the description of the specification, the description of the terms "one embodiment", "some embodiments", "an example", "a specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In the specification, the illustrative description of the above terms does not necessarily mean the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0060] Although the embodiments of the present application have been shown and described above, it is understood that the above-described embodiments are exemplary, and should not be construed as limiting the present application, and those skilled in the art can make changes, modifications, replacements and variations to the above-described embodiments within the scope of the present application without departing from the principles and spirit of the present application.

Claims

1. A method for identifying bridge anomalies based on multi-granularity time windows, characterized in that, Includes the following steps: Multiple sample data points of the bridge are acquired using a real-time monitoring system; wherein, the multiple sample data points include bridge displacement and strain; Based on the multiple sample data, a sample initial feature vector representation of a sliding time window is constructed, and the sample initial feature vector representation is fused to construct a sample initial feature matrix representation; Based on the initial feature matrix representation of the samples, the linear and nonlinear relationships of the sample neighborhood features of the sliding time window are fitted. Based on the linear and nonlinear relationships, the abnormal probability values ​​of the samples are identified by fusing linear residual and nonlinear residual block features, and the abnormal probability values ​​of the samples are calculated to label abnormal samples. By identifying the labeled abnormal samples, outliers in the multiple sample data are identified to identify abnormal points on the bridge. The initial feature vector representation of the samples used to construct the sliding time window based on the multiple sample data includes: Preset Sample exist Initialization features of time Take the sample The neighborhood initialization feature vector is represented as: ; The process of fusing the initial feature vector representations of the samples to construct the initial feature matrix representation of the samples includes: Based on the sliding time window The sample of The initial feature vector of the left neighborhood is represented as: , The initial feature vector of the right neighborhood is represented as: ; Fuse the samples of The sample is constructed using left-neighbor initialization features and right-neighbor initialization features. The initial feature matrix is ​​represented as: .

2. The method according to claim 1, characterized in that, The process of fitting the linear and nonlinear relationships of the sample neighborhood features based on the initial feature matrix representation of the samples includes: Based on the sample The initial characteristic matrix representation Through a linear classifier Fit the sample With category tags Linear relationship; and, Through nonlinear classifiers Fit the sample With category tags The nonlinear relationship.

3. The method according to claim 2, characterized in that, The step of identifying the anomaly probability value of the sample by fusing linear residual and nonlinear residual block features based on the linear and nonlinear relationships includes: Combine the samples linear residuals Nonlinear residuals Features, calculate the sample abnormal probability value ; Based on the anomaly probability value Define the anomaly probability value of the sample. The objective function is Optimize weight parameters , and .

4. The method according to claim 3, characterized in that, The calculation of the anomaly probability value of the sample to label the anomaly sample includes: For the sample Based on the optimized weight parameters , and Calculate the sample abnormal probability value ; Compare the samples abnormal probability value If the value is greater than or equal to 0.5, the labeled sample is normal and the classification label is recorded as 1; otherwise, the labeled sample is abnormal and the classification label is recorded as 0.

5. A bridge anomaly identification device based on multi-granularity time windows, characterized in that, include: The data acquisition module is used to detect and acquire multiple sample data of the bridge using a real-time monitoring system; wherein, the multiple sample data includes bridge displacement and strain; The feature representation module is used to construct an initial feature vector representation of the sample for a sliding time window based on the multiple sample data, and to fuse the initial feature vector representation of the sample to construct an initial feature matrix representation of the sample; The relationship fitting module is used to fit the linear and nonlinear relationships of the sample neighborhood features of the sliding time window based on the initial feature matrix representation of the samples. The probability calculation module is used to identify the abnormal probability value of the sample by fusing the features of linear residuals and nonlinear residual blocks based on the linear relationship and the nonlinear relationship, and to calculate the abnormal probability value of the sample to label abnormal samples. An anomaly identification module is used to identify abnormal values ​​in the multiple sample data through the labeled abnormal samples in order to identify abnormal points on the bridge. The feature representation module is further configured to: Preset Sample exist Initialization features of time The sample The neighborhood initialization feature vector is represented as: ; The feature representation module includes: Feature representation submodule, used to represent based on sliding time window The sample of The initial feature vector of the left neighborhood is represented as: , The initial feature vector of the right neighborhood is represented as: ; A fusion building module is used to fuse the samples. of The sample is constructed using left-neighbor initialization features and right-neighbor initialization features. The initial feature matrix is ​​represented as: .

6. The apparatus according to claim 5, characterized in that, The relationship fitting module includes: Linear fitting module, used for fitting the samples The initial characteristic matrix representation Through a linear classifier Fit the sample With category tags Linear relationship; and, Nonlinear fitting module, used to fit through a nonlinear classifier Fit the sample With category tags The nonlinear relationship.

Images