A method for identifying bridge structural anomalies based on temperature attention-LSTM

Abstract

This invention relates to the field of bridge structure monitoring data processing and anomaly identification technology, specifically to a method for identifying bridge structure anomalies based on temperature attention-LSTM. The method includes: acquiring raw beam displacement monitoring data and raw temperature monitoring data from multiple monitoring points; simultaneously processing the raw beam displacement and temperature monitoring data at the same frequency to obtain beam displacement sampling data and temperature sampling data; and identifying bridge structure anomalies based on the beam displacement sampling data and temperature sampling data to obtain the bridge structure anomaly identification result. Compared to existing technologies that only use displacement monitoring data for bridge structure anomaly identification, this solution utilizes raw beam displacement monitoring data and raw temperature monitoring data from multiple monitoring points to obtain the bridge structure anomaly identification result, improving accuracy and achieving automatic identification of bridge structure anomalies.

Description

Technical Field

[0001] This invention relates to the field of bridge structure monitoring data processing and abnormal state identification technology, specifically to a method for identifying abnormalities in bridge structures based on temperature attention-LSTM. Background Technology

[0002] Bridge structural health monitoring systems, as an important means of understanding the service status of bridge structures, offer advantages such as real-time monitoring and long-term observation compared to scheduled inspections and manual inspections, and are therefore receiving increasing attention from the industry. Beam displacement is a crucial monitoring item in bridge structural health monitoring systems, reflecting abnormalities in the bridge structure, such as beam tilting, support slippage, beam torsion, and beam end jamming.

[0003] Existing bridge structural health monitoring systems identify anomalies solely by checking whether the displacement data from a single beam exceeds a threshold. Even after the system detects that the displacement data from a single beam exceeds the threshold, manual structural calculations and analysis of data from multiple monitoring points are typically required to comprehensively determine whether the bridge structure is truly abnormal. Therefore, this method of identifying bridge structural anomalies based solely on whether the displacement data from a single beam exceeds a threshold suffers from drawbacks such as low accuracy, reliance on manual experience, and low levels of automation. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention proposes a method for identifying anomalies in bridge structures based on temperature attention-LSTM, which improves accuracy.

[0005] The technical solution adopted in this invention is a method for identifying anomalies in bridge structures based on temperature attention-LSTM.

[0006] In the first feasible approach, a method for identifying bridge structural anomalies based on temperature attention-LSTM includes: acquiring raw monitoring data of beam displacement and raw monitoring data of temperature from multiple monitoring points; simultaneously processing the raw monitoring data of beam displacement and raw monitoring data of temperature to obtain beam displacement sampling data and temperature sampling data; and identifying bridge structural anomalies based on the beam displacement sampling data and temperature sampling data to obtain bridge structural anomaly identification results.

[0007] In conjunction with the first feasible method, the second feasible method acquires raw monitoring data of beam displacement and raw monitoring data of temperature from multiple monitoring points, including: setting at least one temperature monitoring point and at least two displacement monitoring points at the two beam ends of the bridge respectively; acquiring raw temperature monitoring data through temperature sensors deployed at each temperature monitoring point, and acquiring raw beam displacement monitoring data through displacement sensors deployed at each displacement monitoring point.

[0008] In the third feasible method, in conjunction with the first feasible method, the original beam displacement monitoring data and the original temperature monitoring data are processed simultaneously at the same frequency to obtain beam displacement sampling data and temperature sampling data. This includes: sampling the original beam displacement monitoring data and the original temperature monitoring data at the same frequency during the same acquisition period to obtain two sets of sampling data; each set of sampling data includes multiple beam displacement sampling values ​​or multiple temperature sampling values; obtaining the valid monitoring values ​​of each set of sampling data; determining the beam displacement sampling data and temperature sampling data according to each valid monitoring value, and using the acquisition time as the label for the beam displacement sampling data and temperature sampling data.

[0009] In conjunction with the third feasible method, the fourth feasible method involves obtaining the effective monitoring value of each group of sampled data, including: sorting the sampled values ​​in each group of sampled data in descending order; removing the maximum and minimum sampled values ​​in each group of sampled data, leaving several sampled values; obtaining the average value of the remaining sampled values, and using the average value as the effective monitoring value.

[0010] In conjunction with the third feasible method, the fifth feasible method includes: inputting beam displacement sampling data and temperature sampling data into a preset displacement data prediction model to obtain beam displacement prediction data; determining whether the beam displacement prediction data is within a preset beam displacement range; if the determination result is that the beam displacement prediction data is not within the preset beam displacement range, determining that each sensor is faulty; if the determination result is that the beam displacement prediction data is within the preset beam displacement range, determining that each sensor is not faulty.

[0011] In the sixth feasible method, in conjunction with the first feasible method, bridge structural anomaly identification is performed based on beam displacement sampling data and temperature sampling data to obtain bridge structural anomaly identification results. This includes: obtaining feature values ​​of each monitoring point of the bridge based on the beam displacement sampling data and temperature sampling data; obtaining the overall bridge structural response features based on the feature values ​​of each monitoring point of the bridge; inputting the overall bridge structural response features into the fully connected layer to obtain the fully connected layer output; using a classifier to calculate the output of the fully connected layer to obtain the probability value corresponding to each bridge structural anomaly type; and determining the bridge structural anomaly type corresponding to the highest probability value as the bridge structural anomaly identification result.

[0012] In conjunction with the sixth feasible method, the seventh feasible method obtains the feature values ​​of each monitoring point of the bridge based on the displacement sampling data and temperature sampling data of each beam, including: obtaining the feature vector of each monitoring point based on the displacement sampling data and temperature sampling data of each beam; inputting each feature vector into the attention layer to obtain the attention score of each feature vector; obtaining the weight of each attention score; and obtaining the feature value of each monitoring point based on each feature vector and the corresponding attention score weight.

[0013] Combining the seventh feasible method, in the eighth feasible method, the feature vector of each monitoring point is obtained based on the displacement sampling data and temperature sampling data of each beam, including:

[0014] h t =σ(W o [h t-1 ,x t ]+b o )*tanh(σ(W f [h t-1 ,x t ]+b f )*C t-1 +σ(W i [h t-1 ,x t ]+b i )*tanh(W c [h t-1 ,x t ]+b c ))

[0015] Among them, h t Let W be the feature vector of this sequence, σ represent the sigmoid activation function, and W be the feature vector of this sequence. o h represents the output weights of the input vector. t-1 Let x be the eigenvector of the previous sequence. t b is the sampled data for this sequence. o For output bias, tanh represents the tanh activation function, W f Let b be the weight matrix of the forget gate. f For the forget gate bias, C t-1 W represents the upper-level memory unit. i For updating the weights of the input vector, b i To update the bias, W c Let b be the weight matrix of the new candidate vector. c This is the bias term for the new candidate vector.

[0016] In conjunction with the sixth feasible method, the ninth feasible method obtains the feature values ​​of each monitoring point of the bridge based on the displacement sampling data and temperature sampling data of each beam. This includes: constructing an LSTM-Attention module using an LSTM neural network and an attention mechanism; assembling the displacement sampling data and temperature sampling data of each beam into a set sequence of each monitoring point; and inputting each set sequence into the LSTM-Attention module to obtain the feature values ​​of each monitoring point.

[0017] In conjunction with the sixth feasible method, the tenth feasible method obtains the overall bridge structural response characteristics based on the feature values ​​of each monitoring point of the bridge, including: performing local fusion analysis on the feature values ​​of each monitoring point of the bridge to obtain the displacement sequence of each monitoring point of the bridge; and performing global fusion processing on the displacement sequences of each monitoring point to obtain the overall bridge structural response characteristics.

[0018] As can be seen from the above technical solution, the beneficial technical effects of the present invention are as follows:

[0019] 1. Compared to existing technologies that rely solely on displacement monitoring data for bridge structural anomaly identification, this solution obtains raw beam displacement and temperature monitoring data from multiple monitoring points. This data is then used to identify bridge structural anomalies, improving accuracy and eliminating the need for manual intervention. This enables automatic identification of bridge structural anomalies, allowing for real-time online monitoring to ensure healthy bridge operation and enhance the level of intelligence. Based on human experience and historical monitoring data, it is known that beam displacement is affected by temperature. This solution utilizes raw beam displacement and temperature monitoring data to identify bridge structural anomalies, not only increasing accuracy by adding monitoring data but also incorporating temperature, which significantly impacts beam displacement, into the anomaly identification process, further enhancing accuracy.

[0020] 2. By simultaneously processing the original beam displacement monitoring data and the original temperature monitoring data at the same frequency, other interfering data can be eliminated, and beam displacement sampling data and temperature sampling data collected at the same time can be obtained. Correlation features between beam displacement sampling data and temperature sampling data are constructed so as to identify bridge structural anomalies based on beam displacement sampling data and temperature sampling data. Attached Figure Description

[0021] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the accompanying drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to scale.

[0022] Figure 1 A schematic diagram of a method for identifying bridge structural anomalies based on temperature attention-LSTM provided by the present invention;

[0023] Figure 2 This invention provides a displacement monitoring curve of a bridge end.

[0024] Figure 3 This is a schematic diagram of the layout of bridge monitoring points provided by the present invention. Detailed Implementation

[0025] The embodiments of the technical solution of the present invention will now be described in detail with reference to the accompanying drawings. These embodiments are merely illustrative of the technical solution of the present invention and are therefore intended to limit the scope of protection of the present invention.

[0026] It should be noted that, unless otherwise stated, the technical or scientific terms used in this application should have the ordinary meaning understood by those skilled in the art to which this invention pertains. The terms "first," "second," etc., in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data used can be interchanged where appropriate for the embodiments of this disclosure described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. Unless otherwise stated, the term "a plurality of" means two or more. In this disclosure, the character " / " indicates that the preceding and following objects are in an "or" relationship. For example, A / B means: A or B. The term "and / or" describes an association relationship between objects, indicating that three relationships can exist. For example, A and / or B means: A or B, or, A and B. The term "corresponding" can refer to an association or binding relationship; A corresponding to B means that there is an association or binding relationship between A and B.

[0027] Combination Figure 1 As shown, this embodiment provides a method for identifying anomalies in bridge structures based on temperature attention-LSTM, including:

[0028] Step S01: Obtain raw monitoring data of beam displacement and raw monitoring data of temperature from multiple monitoring points;

[0029] Step S02: Simultaneously process the original monitoring data of beam displacement and the original monitoring data of temperature at the same frequency to obtain beam displacement sampling data and temperature sampling data.

[0030] Step S03: Based on the beam displacement sampling data and temperature sampling data, identify bridge structural anomalies and obtain the bridge structural anomaly identification results.

[0031] In some embodiments, Figure 2This document presents the monitoring curves for the end displacement of a bridge from February 14, 2022 to August 2, 2022. The monitoring data shows that temperature is the primary factor affecting the monitoring of bridge beam end displacement, focusing mainly on the impact of temperature monitoring data on beam end displacement monitoring. During the monitoring period, when the monitoring data for the displacement of each beam expansion joint (i.e., beam body displacement) fluctuated with temperature, on-site investigation revealed that the bridge structure was in a normal state. However, when the monitoring data for the expansion joint displacement (i.e., beam body displacement) did not change with temperature, on-site investigation revealed that the bridge had experienced beam end displacement jamming. The monitoring points and measurement points had the same meaning. Based on extensive manual experience and historical monitoring data, it is known that the displacement of each beam body is affected by temperature, and there is a certain mapping relationship between each beam end displacement monitoring point and temperature. This scheme utilizes the original beam body displacement monitoring data and the original temperature monitoring data to obtain the bridge structural anomaly identification results. This not only improves accuracy by increasing the monitoring data but also further enhances the accuracy by incorporating temperature, which has a significant impact on beam body displacement, into the bridge structural anomaly identification process.

[0032] In some embodiments, the formula for expressing the monitoring data of each beam end displacement monitoring point is as follows:

[0033] distance s1 = f1(temperature,t) + δ1(t)

[0034] distance s2 = f2(temperature,t) + δ2(t)

[0035] distance s3 = f3(temperature,t) + δ3(t)

[0036] distance s4 = f4(temperature,t) + δ4(t)

[0037] Where, distance s1 distance s2 distance s3 distance s4To correspond to the monitoring data of beam end displacement monitoring points s1, s2, s3, and s4, f1(temperature,t), f2(temperature,t), f3(temperature,t), and f4(temperature,t) represent the effects of temperature, humidity, and time on the monitoring data of monitoring points s1, s2, s3, and s4, and δ1(t), δ2(t), δ3(t), and δ4(t) represent the effects of system noise and load on the displacement monitoring data of monitoring points s1, s2, s3, and s4. Among these, temperature is the primary influence on the monitoring data.

[0038] Optionally, the original monitoring data of beam displacement and temperature at multiple monitoring points are obtained, including: setting at least one temperature monitoring point and at least two displacement monitoring points at the two beam ends of the bridge respectively; obtaining the original temperature monitoring data through temperature sensors deployed at each temperature monitoring point, and obtaining the original beam displacement monitoring data through displacement sensors deployed at each displacement monitoring point.

[0039] In some embodiments, temperature and humidity sensors are deployed at each temperature monitoring point to monitor the temperature and obtain raw temperature monitoring data.

[0040] In some embodiments, combined with Figure 3 As shown, monitoring points are located at the dots at both ends of the bridge beams. One temperature measuring point and two beam end displacement measuring points are set on the bridge deck at each beam end. The four beam end displacement measuring points are symmetrically distributed on both sides of the bridge deck: left-side displacement monitoring point s1, left-side displacement monitoring point s2, right-side displacement monitoring point s3, and right-side displacement monitoring point s4. Two temperature measuring points are located at the two beam ends: a left temperature monitoring point and a right temperature monitoring point THM1. The temperature of the left temperature monitoring point THM1 serves as the common temperature for both left-side displacement monitoring points s1 and s2, and the temperature of the right temperature monitoring point THM2 serves as the common temperature for both right-side displacement monitoring points s3 and s4.

[0041] Optionally, the raw beam displacement monitoring data and raw temperature monitoring data are processed simultaneously at the same frequency to obtain beam displacement sampling data and temperature sampling data. This includes: sampling the raw beam displacement monitoring data and raw temperature monitoring data at the same frequency during the same acquisition period to obtain two sets of sampling data; each set of sampling data includes multiple beam displacement sampling values ​​or multiple temperature sampling values; obtaining the valid monitoring values ​​of each set of sampling data; determining the beam displacement sampling data and temperature sampling data based on each valid monitoring value, and using the acquisition time as the label for the beam displacement sampling data and temperature sampling data.

[0042] In some embodiments, since the bridge structure changes slowly, the displacement data can be considered to change little in a short period of time. According to the "Technical Specification for Monitoring Highway Bridge Structures JT / T 1037—2022," the sampling frequency table for structural response monitoring specifies a dynamic sampling frequency of 20Hz and a static sampling frequency of 1Hz for displacement. Beam end displacement is a displacement monitoring item and is typically sampled statically at a frequency of 1Hz. The temperature sampling frequency is ≤1 / 600Hz. Therefore, it is necessary to process the raw beam displacement and temperature monitoring data collected by the sensors simultaneously at the same frequency to eliminate other interfering data and obtain beam displacement and temperature sampling data collected at the same time, thus establishing the correlation characteristics between the beam displacement and temperature sampling data.

[0043] In some embodiments, within a preset acquisition time period, the original monitoring data of beam displacement is sampled at a frequency of 1 / 60Hz, and the original monitoring data of temperature is sampled at a frequency of 1 / 60Hz. Then, the uniform sampling time scale of the monitoring data is 10 minutes / time, which can be approximately considered as no change in the monitoring results.

[0044] Optionally, obtaining the valid monitoring value of each group of sampled data includes: sorting the sampled values ​​in each group of sampled data in descending order; removing the maximum and minimum sampled values ​​in each group of sampled data, leaving several sampled values; obtaining the average value of the remaining sampled values, and using the average value as the valid monitoring value.

[0045] In some embodiments, the preset acquisition time period is 10 minutes. The original monitoring data of beam displacement is sampled strictly at 1 / 60Hz, and n sample values ​​are obtained for each group. Due to hardware and transmission limitations, sampling 10 times in 10 minutes is impractical. The final sampled data is mostly obtained from 8 to 9 samples in 10 minutes, so n≤10. The n sample values ​​are arranged in descending order to obtain the sequence x1, x2, ..., x n The following formula can be used to analyze the sequence x1, x2, ..., x... n Calculations are performed to obtain valid monitoring values.

[0046] Optionally, data = (sum(x1,x2,…,x) n )-x min -x max ) / (n-2)

[0047] In the above formula, data represents the valid monitoring data for the first data collection period, and x represents the effective monitoring data for the first data collection period. min For the minimum sample value, x max For the maximum sampled value, x n This is the nth sample value.

[0048] Optionally, the minutes of the acquisition time are forward-aligned, and the processed acquisition time is used as the label for the beam displacement sampling data and temperature sampling data. In some embodiments, forward alignment of the minutes includes processing numbers between 0 and 10 as 0, numbers between 10 and 20 as 10, numbers between 20 and 30 as 20, numbers between 30 and 40 as 30, numbers between 40 and 50 as 40, and numbers between 50 and 60 as 50. That is, if the acquisition time data is between 4:00 and 4:10, the forward-aligned acquisition time is 4:00; if the acquisition time data is between 4:10 and 4:20, the forward-aligned acquisition time is 4:10.

[0049] Optionally, after obtaining beam displacement sampling data and temperature sampling data, bridge structure anomaly identification is performed based on the beam displacement sampling data and temperature sampling data. Before obtaining the bridge structure anomaly identification result, the process includes: inputting the beam displacement sampling data and temperature sampling data into a preset displacement data prediction model to obtain beam displacement prediction data; determining whether the beam displacement prediction data is within a preset beam displacement range; if the determination result is that the beam displacement prediction data is not within the preset beam displacement range, determining that each sensor is faulty; if the determination result is that the beam displacement prediction data is within the preset beam displacement range, determining that each sensor is not faulty.

[0050] In some embodiments, beam displacement sampling data and temperature sampling data are input into a preset displacement data prediction model to obtain beam displacement prediction data. If the beam displacement prediction data is not within the preset beam displacement range, it indicates that the corresponding beam displacement exceeds the preset displacement range, i.e., the data monitored by the sensors is out of range. If only one sensor's data is out of range, while the data from the other three sensors are within range, the bridge structure is considered to be in normal condition, and the sensor that is out of range has malfunctioned, making its data unreliable. When the data from two or more sensors are out of range, the probability that both sensors have malfunctioned, resulting in all data exceeding the range, is relatively small. Therefore, it can be considered that the bridge structure state has changed, and beam displacement sampling data and temperature sampling data are used to identify bridge structural anomalies, obtaining bridge structural anomaly identification results. Compared with existing technologies that directly use sensor data for bridge structural anomaly identification, this solution eliminates the possibility of sensor malfunction before using sensor data for bridge structural anomaly identification, resulting in higher accuracy of the obtained bridge structural anomaly identification results.

[0051] Optionally, multiple historical beam displacement monitoring data and historical temperature monitoring data can be input into an LSTM neural network for training to obtain the mapping relationship between historical beam displacement monitoring data and historical temperature monitoring data at the same acquisition time and beam displacement monitoring data at the next acquisition time.

[0052] In some embodiments, beam displacement sampling data and temperature sampling data at the same acquisition time are input into a displacement data prediction model to obtain beam displacement prediction data for the next acquisition time. If the beam displacement prediction data is outside the preset beam displacement range, it is determined that each sensor is faulty, a sensor fault alarm is issued, and the beam displacement sampling data and temperature sampling data at the same acquisition time are determined to be erroneous and invalid. If the beam displacement prediction data is within the preset beam displacement range, it is determined that each sensor is not faulty, and the raw beam displacement monitoring data and raw temperature monitoring data collected by the sensors are valid data. Therefore, it is determined that the beam displacement sampling data and temperature sampling data at the same acquisition time are error-free and valid. Thus, before performing bridge structure anomaly identification based on beam displacement sampling data and temperature sampling data and obtaining the bridge structure anomaly identification result, the displacement data prediction model is used to determine whether the sensors are faulty, and then to determine whether the beam displacement sampling data and temperature sampling data are valid. This improves the accuracy of the monitoring data, greatly avoids errors in bridge structure anomaly identification caused by sensor faults, and thus improves the accuracy rate of bridge structure anomaly identification.

[0053] Optionally, bridge structural anomaly identification is performed based on beam displacement sampling data and temperature sampling data to obtain bridge structural anomaly identification results, including: obtaining feature values ​​of each monitoring point of the bridge based on the beam displacement sampling data and temperature sampling data; obtaining overall bridge structural response features based on the feature values ​​of each monitoring point of the bridge; inputting the overall bridge structural response features into the fully connected layer to obtain the fully connected layer output; using a classifier to calculate the output of the fully connected layer to obtain the probability value corresponding to each bridge structural anomaly type; and determining the bridge structural anomaly type corresponding to the highest probability value as the bridge structural anomaly identification result.

[0054] Optionally, the overall bridge structural response characteristics are input into the fully connected layer to obtain the fully connected layer output, including: through calculation Obtain the output of the fully connected layer; where, bridge i This refers to the overall bridge structural response characteristics. y represents the weight of the overall bridge structural response characteristics. i This is the i-th output of the fully connected layer.

[0055] Optionally, a classifier is used to calculate the output of the fully connected layer to obtain the probability value corresponding to each bridge structural anomaly type, including: by calculating... Obtain the probability values ​​corresponding to the bridge structural anomaly types; where P c y represents the probability value of bridge structural anomaly type c. i This is the output of the fully connected layer.

[0056] Optionally, feature values ​​of each monitoring point of the bridge are obtained based on the displacement sampling data and temperature sampling data of each beam, including: obtaining feature vectors of each monitoring point based on the displacement sampling data and temperature sampling data of each beam; inputting each feature vector into the attention layer to obtain the attention score of each feature vector; obtaining the weight of each attention score; and obtaining the feature value of each monitoring point based on each feature vector and the corresponding attention score weight.

[0057] Optionally, feature vectors for each monitoring point are obtained based on displacement and temperature sampling data of each beam, including:

[0058] h t =σ(W o [h t-1 ,x t ]+b o )*tanh(σ(W f [h t-1 ,x t ]+b f )*C t-1 +σ(W i [h t-1 ,x t ]+b i )*tanh(W c [h t-1 ,x t ]+b c ))

[0059] Among them, h t Let W be the feature vector of this sequence, σ represent the sigmoid activation function, and W be the feature vector of this sequence. o h represents the output weights of the input vector. t-1 Let x be the eigenvector of the previous sequence. t b is the sampled data for this sequence. o For output bias, tanh represents the tanh activation function, W f Let b be the weight matrix of the forget gate. f For the forget gate bias, C t-1 W represents the upper-level memory unit. i For updating the weights of the input vector, b i To update the bias, W c Let b be the weight matrix of the new candidate vector. c This is the bias term for the new candidate vector.

[0060] In some embodiments, the current sequence is the current calculation round. When the current calculation round is the first calculation, the parameters of the previous sequence are all preset parameters. When the current calculation round is not the first calculation, the parameters of the previous sequence are the parameters from the previous calculation round of the current calculation round.

[0061] Optionally, the tanh activation function is:

[0062] Optionally, each feature vector is input into the attention layer to obtain the attention score for each feature vector, including: calculating m t =tanh(w t h t +b t ) Obtain the attention score for each feature vector; where m t For the feature vector h t The corresponding attention score, w t For the feature vector h t The weight, b t For bias.

[0063] Optionally, the weights of each attention score are obtained, including by calculating... Obtain the weights; where s t Attention score m t The corresponding weights are the feature vectors h. t The corresponding weights k This indicates the number of attention scores.

[0064] Optionally, the feature values ​​of each monitoring point are obtained based on the weights of each feature vector and the corresponding attention score, including: by calculating Obtain the eigenvalues; where z is the eigenvalue and s is the eigenvalue. t For the feature vector h t The corresponding weights.

[0065] Optionally, feature values ​​for each monitoring point of the bridge are obtained based on the displacement and temperature sampling data of each beam. This includes: constructing an LSTM-Attention module using an LSTM (Long Short Term Memory) neural network and an attention mechanism; assembling the displacement and temperature sampling data of each beam into a sequence of sets for each monitoring point; and inputting each sequence of sets into the LSTM-Attention module to obtain the feature values ​​for each monitoring point. The LSTM-Attention module obtains the feature values ​​for each monitoring point by: obtaining feature vectors for each monitoring point based on the displacement and temperature sampling data of each beam; inputting each feature vector into the attention layer to obtain the attention score for each feature vector; obtaining the weights of each attention score; and obtaining the feature values ​​for each monitoring point based on each feature vector and the corresponding attention score weights.

[0066] In some embodiments, temperature is the main factor affecting bridge beam end displacement monitoring. Compared with the prior art that only uses displacement data to identify bridge structural anomalies, this scheme uses displacement monitoring data and temperature monitoring data from multiple beam end displacement monitoring points, and adopts a temperature attention mechanism and LSTM neural network to identify abnormal bridge structural states, thereby improving accuracy and providing a more accurate scientific reference for technicians to manage and maintain in-service bridges and ensure operational safety.

[0067] In some embodiments, the characteristics of single beam end displacement measurement points are constructed based on the beam displacement monitoring sampling data distance and temperature monitoring sampling data THM at each monitoring point, and the corresponding mapping set sequence is obtained, namely the set sequence of each monitoring point affecting the beam end displacement at different temperatures, THM1-s1, THM1-s2, THM2-s3, THM2-s4.

[0068] In some embodiments, the set sequence of each monitoring point is {distance} s1 ,THM1},{distance s2 ,THM1},{distance s3 ,THM2},{distance s4 ,THM2}, where distance s1 The sampling data for the first displacement monitoring point, distance s2 For the sampling data of the second displacement monitoring point, distance s3 For the sampling data of the third displacement monitoring point, distance s4 THM1 is the sampling data of the fourth displacement monitoring point, THM2 is the sampling data of the first temperature monitoring point, and THM3 is the sampling data of the second temperature monitoring point.

[0069] Optionally, the overall bridge structural response characteristics are obtained based on the characteristic values ​​of each monitoring point of the bridge, including: performing local fusion analysis on the characteristic values ​​of each monitoring point of the bridge to obtain the displacement sequence of each monitoring point of the bridge; and performing global fusion processing on the displacement sequences of each monitoring point to obtain the overall bridge structural response characteristics.

[0070] Optionally, a local fusion analysis is performed on the feature values ​​of each monitoring point on the bridge to obtain the displacement sequence of each monitoring point, including: each monitoring point has its own feature value as a sequence, the difference sequence between any two monitoring points is obtained, and each difference sequence is determined as the displacement sequence of each monitoring point.

[0071] In some embodiments, the difference between the feature values ​​of monitoring points s1 and s2 is used to obtain a difference sequence between the two points, which is the displacement sequence of the beam ends of the left bridge span. The difference between the feature values ​​of monitoring points s1 and s3 is used to obtain a difference sequence between the two points, which is the displacement sequence of the beam ends of the right bridge span. The difference between the feature values ​​of monitoring points s2 and s4 is used to obtain a difference sequence between the two points, which is the displacement sequence of the beam ends of the left bridge abutment. The difference between the feature values ​​of monitoring points s3 and s4 is used to obtain a difference sequence between the two points, which is the displacement sequence of the beam ends of the right bridge abutment.

[0072] Optionally, the displacement sequences of each monitoring point are globally fused to obtain the overall bridge structural response characteristics, including: determining the displacement state of each monitoring point based on the displacement sequence of each monitoring point, and determining the overall bridge structural response characteristics by looking up a table based on the displacement state of each monitoring point.

[0073] In some embodiments, the displacement state of each monitoring point is determined based on the displacement sequence of each monitoring point. The displacement state includes rising, falling, and unchanged. For example, if the displacement sequence is positive, the displacement state is rising; if the displacement sequence is negative, the displacement state is falling; and if the displacement sequence is 0, the displacement state is unchanged.

[0074] In some embodiments, the overall bridge structural response characteristics are determined by looking up a table based on the displacement status of each monitoring point. Table 1 shows the overall bridge structural response characteristics. When monitoring points s1 and s2 rise while monitoring points s3 and s4 fall, the corresponding overall bridge structural response characteristics are: right-side tilting and support slippage. When monitoring points s1 and s3 rise while monitoring points s2 and s4 fall, the corresponding overall bridge structural response characteristics are: torsion. When the trends of monitoring points s1, s3, s2, and s4 are consistent, the corresponding overall bridge structural response characteristics are: normal.

[0075] Table 1

[0076]

[0077] Optionally, if the bridge structure anomaly identification result indicates that the bridge structure is in an abnormal state, a bridge structure anomaly warning can be issued.

[0078] By acquiring raw monitoring data of beam displacement and temperature from multiple monitoring points, and then obtaining beam displacement sampling data and temperature sampling data, it is possible to identify bridge structural anomalies based on the beam displacement sampling data and temperature sampling data, obtain bridge structural anomaly identification results, realize automatic identification of bridge structural anomalies, and at the same time, provide early warning of bridge structural anomalies when the bridge structural condition is abnormal, realize real-time online monitoring of bridge structure, and ensure the healthy operation of the bridge.

[0079] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the claims and specification of the present invention.

Claims

1. A temperature attention-LSTM based method for identifying abnormality in monitoring bridge structure, characterized in that, include: Acquire raw monitoring data of beam displacement and temperature from multiple monitoring points; Simultaneous and in-frequency processing of the raw monitoring data of beam displacement and temperature is performed to obtain beam displacement sampling data and temperature sampling data. Bridge structural anomaly identification is performed based on beam displacement sampling data and temperature sampling data to obtain bridge structural anomaly identification results; Bridge structural anomaly identification is performed based on beam displacement sampling data and temperature sampling data, and the results of the bridge structural anomaly identification are obtained, including: The characteristic values ​​of each monitoring point on the bridge are obtained based on the displacement and temperature sampling data of each beam. The overall bridge structural response characteristics are obtained based on the characteristic values ​​of each monitoring point on the bridge. The overall bridge structural response characteristics are input into the fully connected layer to obtain the fully connected layer output; A classifier is used to calculate the output of the fully connected layer to obtain the probability value corresponding to each bridge structural anomaly type; The bridge structural anomaly type corresponding to the highest probability value is determined as the bridge structural anomaly identification result. Based on the displacement and temperature sampling data of each beam, characteristic values ​​of each monitoring point on the bridge were obtained, including: The feature vectors of each monitoring point are obtained based on the displacement and temperature sampling data of each beam. Each feature vector is input into the attention layer to obtain the attention score for each feature vector; Obtain the weights of each attention score; The feature values ​​of each monitoring point are obtained based on the weights of each feature vector and the corresponding attention score; Based on the displacement and temperature sampling data of each beam, characteristic values ​​of each monitoring point on the bridge were obtained, including: An LSTM-Attention module is constructed using an LSTM neural network and an attention mechanism; displacement sampling data and temperature sampling data of each beam are combined into a set sequence of monitoring points; each set sequence is input into the LSTM-Attention module to obtain the feature value of each monitoring point; The overall bridge structural response characteristics are obtained based on the characteristic values ​​of each monitoring point on the bridge, including: Local fusion analysis was performed on the characteristic values ​​of each monitoring point on the bridge to obtain the displacement sequence of each monitoring point on the bridge. The displacement sequences of each monitoring point are globally fused to obtain the overall bridge structural response characteristics.

2. The method according to claim 1, characterized in that, Obtain raw monitoring data of beam displacement and temperature from multiple monitoring points, including: At least one temperature monitoring point and at least two displacement monitoring points are set at each of the two beam ends of the bridge. Raw temperature monitoring data is obtained through temperature sensors at each temperature monitoring point, and raw beam displacement monitoring data is obtained through displacement sensors at each displacement monitoring point.

3. The method according to claim 1, characterized in that, Simultaneous and in-frequency processing of the raw beam displacement and temperature monitoring data yields beam displacement sampling data and temperature sampling data, including: The original monitoring data of beam displacement and temperature during the same acquisition period were sampled at the same frequency to obtain two sets of sampling data; each set of sampling data includes multiple beam displacement sampling values ​​or multiple temperature sampling values; Obtain the valid monitoring values ​​for each group of sampled data; Based on each valid monitoring value, beam displacement sampling data and temperature sampling data are determined separately, and the collection time is used as the label for beam displacement sampling data and temperature sampling data.

4. The method according to claim 3, characterized in that, Obtain the valid monitoring values ​​for each group of sampled data, including: Sort the sampled values ​​in each group of sampled data in descending order; Remove the maximum and minimum sample values ​​from each group of sampled data, leaving a number of sample values; Obtain the average value of the remaining sampled values, and use the average value as the valid monitoring value.

5. The method according to claim 3, characterized in that, include: Input the beam displacement sampling data and temperature sampling data into the preset displacement data prediction model to obtain the beam displacement prediction data; Determine whether the predicted beam displacement data is within the preset beam displacement range. If the prediction data is not within the preset beam displacement range, determine that each sensor is faulty. If the prediction data is within the preset beam displacement range, determine that each sensor is not faulty.

6. The method according to claim 1, characterized in that, The feature vectors of each monitoring point are obtained based on the displacement and temperature sampling data of each beam, including: in, This is the feature vector of the current sequence. This represents the sigmoid activation function. The output weights are the input vectors. The feature vector of the previous sequence, This is the sampled data for this sequence. For output bias, This represents the tanh activation function. Here is the weight matrix for the forget gate. To offset the forget gate, Indicates the upper-level memory unit. Update the weights for the input vector. To update the bias, The weight matrix for the new candidate vectors. This is the bias term for the new candidate vector.

Images