Highway bridge comprehensive monitoring system based on big data

Abstract

The invention provides a highway bridge comprehensive monitoring system based on big data, and relates to the field of intelligent monitoring. A data processing technology based on big data is adopted to carry out data splitting and time sequence feature extraction on a time data set of key health data of a bridge so as to obtain dynamic parameter time sequence correlation implicit coding features and environmental parameter time sequence correlation implicit coding features; and obtaining an intelligent auxiliary evaluation result based on heterogeneous multi-source parameter feature interaction coding representation between the dynamic parameter time sequence correlation implicit coding feature and the environment parameter time sequence correlation implicit coding feature. Thus, by processing the bridge monitoring data in real time and fully considering the non-linear relationship between different bridge monitoring data during processing, the health condition of the bridge can be evaluated more accurately, and the requirements of accuracy, timeliness and comprehensiveness of highway bridge monitoring are met.

Description

Technical Field

[0001] This application relates to the field of intelligent monitoring, and more specifically, to an integrated highway bridge monitoring system based on big data. Background Art

[0002] As an important part of transportation infrastructure, the health status of highway bridges is directly related to traffic safety and transportation efficiency. During long-term use, bridges are affected by various factors such as temperature, load, and climate change. The accumulation of these factors may cause damage to the bridge structure, affecting its normal use and even leading to safety accidents. By monitoring highway bridges, the health status of the bridges can be grasped, potential problems can be discovered in a timely manner, so as to take intervention or maintenance measures to ensure the safe operation of the bridges and extend their service life.

[0003] However, traditional monitoring systems often face challenges of heterogeneous data types and inconsistent sampling frequencies when processing data collected by various sensors (such as stress sensors, displacement sensors, temperature and humidity sensors, etc.). Specifically, most current data analysis methods rely on simple statistical methods, and it is difficult to accurately dynamically model complex and changeable bridge conditions. This makes it impossible to fully consider the non-linear relationships between numerous monitoring data during analysis, which will in turn affect the accuracy, timeliness, and comprehensiveness of highway bridge condition monitoring. This limitation will cause the monitoring system to fail to meet the high real-time requirements of highway bridges for timely intervention or maintenance measures.

[0004] Therefore, an integrated highway bridge monitoring system based on big data is expected. Summary of the Invention

[0005] To solve the above technical problems, this application is proposed. Embodiments of this application provide an integrated highway bridge monitoring system based on big data.

[0006] According to one aspect of this application, an integrated highway bridge monitoring system based on big data is provided, which includes:

[0007] A data acquisition module for collecting a time dataset of key health data of the bridge in real time through a sensor network, where the key health data includes dynamic parameters and environmental parameters;

[0008] A data analysis module for performing data interaction analysis based on time series features on the time data set of the key health data to obtain deep interaction features of dynamic parameters - environmental parameters. Among them, the data analysis module includes: a data splitting and encoding unit for splitting the time data set of the key health data and extracting time series features to obtain dynamic parameter time series correlation encoding features and environmental parameter time series correlation encoding features; a feature interaction unit for performing heterogeneous multi-source parameter feature interaction encoding on the dynamic parameter time series correlation encoding features and the environmental parameter time series correlation encoding features to obtain the deep interaction features of dynamic parameters - environmental parameters.

[0009] An evaluation result generation module for obtaining an intelligent auxiliary evaluation result based on the deep interaction features of dynamic parameters - environmental parameters.

[0010] Compared with the prior art, the big data-based highway bridge comprehensive monitoring system provided by this application uses big data-based data processing technology to split the time data set of the key health data of the bridge and extract time series features to obtain dynamic parameter time series correlation implicit encoding features and environmental parameter time series correlation implicit encoding features, and obtains an intelligent auxiliary evaluation result based on the heterogeneous multi-source parameter feature interaction encoding representation between the dynamic parameter time series correlation implicit encoding features and the environmental parameter time series correlation implicit encoding features. In this way, by real-time processing of bridge monitoring data and fully considering the non-linear relationship between different bridge monitoring data during processing, the health status of the bridge can be evaluated more accurately, meeting the requirements of accuracy, timeliness, and comprehensiveness for highway bridge monitoring. Description of the Drawings

[0011] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings. In the drawings:

[0012] Figure 1 It is a system block diagram of a big data-based highway bridge comprehensive monitoring system according to an embodiment of the present application.

[0013] Figure 2 It is a block diagram of the data analysis module in a big data-based highway bridge comprehensive monitoring system according to an embodiment of the present application.

[0014] Figure 3 It is a block diagram of the data splitting and encoding unit in a big data-based highway bridge comprehensive monitoring system according to an embodiment of the present application.

[0015] Figure 4Block diagram of the feature interaction unit in the big data-based comprehensive highway bridge monitoring system according to an embodiment of the present application. Detailed implementation manners

[0016] Hereinafter, exemplary embodiments of the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all embodiments of the present application. It should be understood that the present application is not limited by the exemplary embodiments described herein.

[0017] Highway bridges are an important part of transportation infrastructure, and their health status is directly related to traffic safety and transportation efficiency. However, during long-term use, due to the influence of various factors such as temperature, load, and climate change, the bridge structure may be damaged, even posing potential safety hazards. Therefore, by monitoring the health status of the bridge, potential problems can be detected in a timely manner, and intervention or maintenance measures can be taken to ensure the safe operation of the bridge and extend its service life.

[0018] However, traditional monitoring systems face challenges in processing multi-source sensor data (such as stress, displacement, temperature, humidity, etc.), mainly manifested as the diversity of data types and the differences in sampling frequencies. Existing data analysis methods mostly rely on simple statistical means, and it is difficult to accurately and dynamically reflect the complex and changeable state of the bridge, and it is impossible to fully analyze the non-linear correlations between various monitoring data. This limits the accuracy, real-time performance, and comprehensiveness of the monitoring system, resulting in its inability to fully meet the requirements of timely intervention or maintenance.

[0019] To address the above technical problems, the technical concept of the present application is to first collect the time dataset of the key health data of the bridge in real time through a sensor network, and then use big data-based data processing technology to split the time dataset of the key health data of the bridge and extract time series features to obtain dynamic parameter time series correlation implicit coding features and environmental parameter time series correlation implicit coding features. Finally, an intelligent auxiliary evaluation result is obtained based on the heterogeneous multi-source parameter feature interaction coding representation between the dynamic parameter time series correlation implicit coding features and the environmental parameter time series correlation implicit coding features. In this way, by processing bridge monitoring data in real time and fully considering the non-linear relationships between different bridge monitoring data during processing, the health status of the bridge can be evaluated more accurately, meeting the requirements of accuracy, timeliness, and comprehensiveness for highway bridge monitoring.

[0020] Figure 1 System block diagram of the big data-based comprehensive highway bridge monitoring system according to an embodiment of the present application. As Figure 1As shown in the figure, in the integrated highway bridge monitoring system 100 based on big data, it includes: a data acquisition module 110, which is used to collect the time dataset of the key health data of the bridge in real time through a sensor network, and the key health data includes dynamic parameters and environmental parameters; a data analysis module 120, which is used to perform data interaction analysis based on time series features on the time dataset of the key health data to obtain the deep interaction features of dynamic parameters - environmental parameters; an evaluation result generation module 130, which is used to obtain an intelligent auxiliary evaluation result based on the deep interaction features of dynamic parameters - environmental parameters.

[0021] In the embodiment of the present application, the data acquisition module 110 is used to collect the time dataset of the key health data of the bridge in real time through a sensor network, and the key health data includes dynamic parameters and environmental parameters. Specifically, the dynamic parameters include acceleration values, vibration frequency values, stress values, and strain values, and the environmental parameters include temperature values, humidity values, wind speed values, and vehicle load values. It should be understood that the structural response of the bridge is not only affected by static loads, but also varies due to time-varying factors such as changes in traffic flow and seasonal climate changes. By monitoring the changes in the key health data of the bridge over time in real time, the performance of the bridge structure under different conditions can be captured, so as to more comprehensively understand its health status. Specifically, the time dataset of the key health data of the bridge reveals the patterns of how the bridge structure responds to different loads and environmental conditions. For example, if it is observed that the stress or strain values of the bridge continue to increase over time, it may indicate potential damage to the structure; similarly, changes in temperature and humidity may affect the performance of materials, resulting in structural deformation or other damages. By analyzing these time data, the system can more accurately evaluate the current state of the bridge and predict possible problems in the future.

[0022] The following is a detailed elaboration of a specific implementation process of "collecting the time dataset of the key health data of the bridge in real time through a sensor network":

[0023] The preparatory work in the early stage is the cornerstone of the entire acquisition implementation process. First, a comprehensive evaluation and planning of the bridge need to be carried out, and professional bridge engineers are organized to carry out bridge structure analysis work to deeply understand the type of the bridge, such as beam bridges, arch bridges, cable-stayed bridges, etc., and at the same time master detailed information such as the span of the bridge, the materials used, and the design parameters. Based on these analyses, the key stress-bearing parts and areas vulnerable to environmental impacts of the bridge are determined, and then a detailed monitoring plan is formulated. This plan should clarify important parameters such as the types of data to be collected, the layout positions, quantities, and sampling frequencies of the sensors.

[0024] Then, the sensor needs to be selected. Different types of parameters need to match the appropriate sensor. For dynamic parameters, the accelerometer can use a piezoelectric accelerometer with high accuracy and fast response speed, and its range can generally reach ±2g-±200g. The vibration frequency sensor can use products based on the principle of strain gauges or fiber optic sensing technology, which can accurately measure the vibration frequency of the bridge from 0 to 100Hz. Stress sensors usually choose resistance strain sensors, which have high measurement accuracy and good stability. The range is determined according to the actual stress conditions of the bridge, generally 0-2000με. The strain sensor is preferably a fiber grating strain sensor, which has strong anti-interference ability and high measurement accuracy. It is suitable for long-term monitoring of bridge strain changes, and the range matches the stress sensor. For environmental parameters, the temperature sensor can use a high-precision platinum resistance temperature sensor with a measurement range of -40℃-120℃, which can meet the measurement of bridge temperature under different climatic conditions. The humidity sensor adopts a capacitive type with fast response speed, high accuracy, and a measurement range of 0-100%RH. The wind speed sensor uses an ultrasonic anemometer with a measurement range of 0-60m / s, which can adapt to different wind environments. The vehicle load sensor uses a dynamic weighing quartz piezoelectric sensor, which can accurately measure the vehicle weight and driving speed. The range is determined according to the design load of the bridge.

[0025] Then the sensor needs to be installed. First, the installation position must be accurately determined. For dynamic parameter sensors, acceleration sensors need to be installed at the mid-span of the bridge, the top of the pier, the end of the beam, etc. to measure the acceleration response under different working conditions; vibration frequency sensors are arranged in coordination with acceleration sensors to ensure comprehensive monitoring of vibration frequency; stress sensors are installed at key stress-bearing parts of the bridge, such as the mid-span of the main beam, near the support, and the bottom of the pier; strain sensors are arranged at the upper and lower edges of the main beam, the inclined cable, etc. For environmental parameter sensors, temperature sensors are installed at different heights and positions on the bridge, such as the surface of the main beam, the inside of the box beam, and the side of the pier; humidity sensors are installed in conjunction with temperature sensors; wind speed sensors are installed at the highest point or windward side of the bridge to avoid the influence of surrounding obstacles; vehicle load sensors are installed on the bridge deck lane, and the vehicle load is accurately measured by embedding or patch type. During the installation process, select the appropriate fixing method according to the sensor type and installation location, such as pasting, welding, bolt connection or bracket installation. At the same time, the signal line of the sensor must be correctly connected to the data acquisition instrument to ensure a firm connection and good contact, and the signal line must be reasonably wired and protected to prevent external interference and damage.

[0026] Next, it is necessary to build a data acquisition system. According to the number, type, and sampling frequency requirements of the sensors, a data acquisition instrument with functions such as multi-channel input, high-precision sampling, data storage, and transmission should be selected. Integrate the sensors, data acquisition instrument, and communication equipment to build a complete data acquisition system. During the integration process, ensure the compatibility and stability between devices. Subsequently, debug the system, simulate different working conditions and environmental conditions, check the working status of the sensors, the sampling accuracy of the data acquisition instrument, and the transmission stability of the communication equipment to ensure the reliability and accuracy of the system.

[0027] In the data acquisition and transmission stage, start the data acquisition system according to the sampling frequency determined by the monitoring plan to ensure the continuity and stability of data acquisition and avoid data loss or errors. Transmit the acquired data to the data center using wired or wireless communication methods. Wired communication is suitable for situations with short distances and large data transmission volumes, such as Ethernet, fiber optic cables, etc.; wireless communication is suitable for situations with long distances and difficult wiring. To ensure the security and reliability of the data, encryption technology can be used to encrypt the data.

[0028] The acquired data needs to be stored. Store the data acquired by each sensor in a suitable database for subsequent data management and analysis.

[0029] The maintenance and management of the data acquisition system are also important links to ensure the continuous and stable progress of the acquisition work. Regularly inspect the sensors, data acquisition instrument, and communication equipment to check the working status, installation conditions, and line connection conditions of the equipment, and promptly discover and solve potential problems. Calibrate the sensors at regular time intervals, strictly follow relevant standards and specifications for operation to ensure the measurement accuracy and reliability of the sensors. At the same time, regularly back up the data in the database to prevent data loss or damage, and establish a data recovery mechanism to ensure that the data can be restored in a timely manner when problems occur. By implementing the above steps, a time dataset of the key health data of the bridge can be effectively collected through the sensor network.

[0030] In the embodiment of the present application, the data analysis module 120 is used to perform data interaction analysis based on time series characteristics on the time dataset of the key health data to obtain the deep interaction characteristics of dynamic parameters - environmental parameters. Specifically, Figure 2 It is a block diagram of the data analysis module in the highway bridge comprehensive monitoring system based on big data according to the embodiment of the present application. As Figure 2As shown, the data analysis module 120 includes: a data splitting and encoding unit 121, configured to perform data splitting and time series feature extraction on the time series dataset of the key health data to obtain a dynamic parameter time series correlation encoding feature and an environmental parameter time series correlation encoding feature; and a feature interaction unit 122, configured to perform heterogeneous multi-source parameter feature interaction encoding on the dynamic parameter time series correlation encoding feature and the environmental parameter time series correlation encoding feature to obtain the dynamic parameter-environment parameter deep interaction feature.

[0031] In an embodiment of the present application, the data splitting and encoding unit 121 is configured to perform data splitting and time series feature extraction on the time series dataset of the key health data to obtain a dynamic parameter time series correlation encoding feature and an environmental parameter time series correlation encoding feature. Specifically, Figure 3 FIG. is a block diagram of a data splitting and encoding unit in a highway bridge comprehensive monitoring system based on big data according to an embodiment of the present application. As Figure 3 shown, the data splitting and encoding unit 121 includes: a data splitting sub-unit 1211, configured to perform data splitting on the time series dataset of the key health data to obtain a time series dataset of dynamic parameters and a time series dataset of environmental parameters; a dynamic parameter feature encoding sub-unit 1212, configured to extract the dynamic parameter time series correlation encoding feature from the time series dataset of the dynamic parameters; and an environmental parameter feature encoding sub-unit 1213, configured to extract the environmental parameter time series correlation encoding feature from the time series dataset of the environmental parameters.

[0032] In the embodiment of the present application, the data splitting sub-unit 1211 is configured to split the time data set of the key health data to obtain a time data set of dynamic parameters and a time data set of environmental parameters. Correspondingly, considering that dynamic parameters mainly reflect the mechanical response of the bridge structure itself under various loads, while environmental parameters more reflect the influence of external environmental factors on the bridge. These two types of data have different natures, and there are significant differences in their variation laws, influencing factors, and the ways of acting on bridge risks. For example, the change of dynamic parameters may be more directly related to the instantaneous stress situation of the bridge. For instance, the vibration frequency of the bridge will change instantaneously when a vehicle passes; while the change of environmental parameters such as temperature is relatively slow, and its influence on the bridge is more long-term and gradual, such as the thermal expansion and contraction of bridge materials caused by temperature changes. That is to say, for dynamic parameters, it may be necessary to focus on analyzing their fluctuations, peaks, and characteristics related to the bridge structure in a short period of time; while for environmental parameters, more attention should be paid to their long-term statistical laws, seasonal changes, and characteristics related to bridge durability. Based on this, in the present application, the time data set of the key health data is split to obtain a time data set of dynamic parameters and a time data set of environmental parameters. This can enable more effective analysis of the characteristics of different data in subsequent processing, and avoid the interference and complexity that may occur when analyzing them mixed together.

[0033] In the embodiment of the present application, the dynamic parameter feature encoding subunit 1212 is used to extract the dynamic parameter time series correlation encoding feature from the time dataset of the dynamic parameters. Specifically, in the embodiment of the present application, the dynamic parameter feature encoding subunit is used to: use a time series encoder based on BiGRU to perform dynamic parameter time series feature extraction on the time dataset of the dynamic parameters to obtain a dynamic parameter time series correlation encoding feature vector as the dynamic parameter time series correlation encoding feature. It should be understood that dynamic parameters (such as acceleration values, vibration frequency values, stress values, and strain values) reflect the response of the bridge under different load conditions. These responses usually change over time and have strong time series characteristics. In order to capture the complex change patterns and long-term dependencies of dynamic parameters from the original time dataset, it is necessary to extract the dynamic parameter time series correlation encoding feature from the time dataset of the dynamic parameters in the present application. By extracting the time series correlation encoding feature of the dynamic parameter, it is possible to more effectively identify those patterns that may indicate structural damage or degradation. For example, if the stress value at a certain location continues to rise over time, this may be a signal of potential problems with the bridge. In particular, in a specific embodiment of the present application, a time series encoder based on BiGRU is used to perform dynamic parameter time series feature extraction on the time dataset of the dynamic parameters to obtain a dynamic parameter time series correlation encoding feature vector as the dynamic parameter time series correlation encoding feature. Those of ordinary skill in the art should know that GRU controls the flow and update of information through update gates and reset gates, so as to effectively capture long-term dependencies in sequence data. BiGRU adds a backpropagation path on the basis of GRU and enhances the understanding of the input sequence data by considering information from both the past and the future. Specifically, for the time dataset of dynamic parameters, the change of data may not only depend on the past state, but also be related to the future state. For example, the change of the vibration frequency of the bridge at a certain moment may be affected by the vehicles that have passed before, or may be a prelude to the upcoming vehicle load. BiGRU can consider both forward and backward time series information, so as to more comprehensively capture the change characteristics of dynamic parameters.

[0034] The following is a detailed elaboration of a specific implementation process of "using a time series encoder based on BiGRU to perform dynamic parameter time series feature extraction on the time dataset of the dynamic parameters to obtain a dynamic parameter time series correlation encoding feature vector":

[0035] First, preprocessing operations need to be performed on the data. Specifically, the data is first cleaned by carefully checking for outliers in the data. Since the sensors may be affected by external interference or malfunction, data that deviates from the normal range may be generated, and these outliers will affect the accuracy of subsequent analysis. Statistical methods, such as the 3σ principle, can be used to identify data points that deviate too much from the mean, and then appropriate processing methods can be selected according to the actual situation, such as deleting outliers, filling them using interpolation methods, or correcting them based on the overall trend of the data. Then, the normalized operation is performed on the cleaned data because different dynamic parameters, such as acceleration values, vibration frequency values, stress values, and strain values, have large differences in dimension and numerical range. If normalization is not performed, during model training, features with large values may dominate the training process, while features with small values may be ignored. Commonly used normalization methods include min-max normalization, which maps the data to the [0,1] interval, and Z-score normalization, which makes the data have zero mean and unit variance, so as to ensure that each feature can play an equal role in model training.

[0036] After completing the data preprocessing, it enters the stage of constructing the BiGRU model. BiGRU is composed of a forward GRU and a backward GRU, and its unique structure can capture both the forward and backward information of time series data. The update gate and reset gate in GRU are the key parts. The update gate controls the degree to which the state information of the previous moment is passed to the current moment, and the reset gate determines the degree to which the state information of the previous moment is ignored. When constructing the model, the number of layers of BiGRU and the number of GRU units in each layer need to be determined. The choice of the number of layers should comprehensively consider the complexity of the data and computing resources. Generally, 1-3 layers are more appropriate. If the number of layers is too small, the model may not be able to fully mine the complex features in the data; if the number of layers is too large, it is easy to cause problems such as gradient disappearance or gradient explosion, resulting in difficult training. The number of GRU units in each layer also needs to be adjusted according to the characteristics of the actual processed data, usually between dozens and hundreds. For example, for medium-scale bridge dynamic parameter data, we can first try to set 64 or 128 GRU units in each layer. At the same time, an appropriate activation function needs to be selected. The sigmoid function is commonly used in the gating mechanism of GRU units, which can map the output value to the [0,1] interval, facilitating the control of the information transfer ratio; while when calculating the candidate hidden state, the Tanh function is more suitable for processing the positive and negative changes of the data because it can map the output value to the [-1,1] interval, so it is often selected.

[0037] After the model is constructed, it enters the crucial training stage. First, the collected dynamic parameter training data needs to be divided into a training set, a validation set, and a test set. The general division ratio is that the training set accounts for 60%-80%, the validation set accounts for 10%-20%, and the test set accounts for 10%-20%. Moreover, it is necessary to ensure that the data distributions of each data set are similar to avoid the excessive difference in data characteristics between the training set and the test set from affecting the model performance. Then, select appropriate optimizers and loss functions. The optimizer is used to adjust the model parameters to minimize the loss function. Common optimizers such as Stochastic Gradient Descent (SGD) and its variants. Among them, the Adam optimizer performs well in practical applications due to its characteristic of adaptive learning rate and is a commonly used choice. The loss function is determined according to the task type. For feature extraction tasks, the Mean Squared Error (MSE) loss function is relatively commonly used. It evaluates the performance of the model by measuring the mean of the squared errors between the model prediction values and the true values. During the training process, the training set data is input into the BiGRU model. The model calculates the output result based on the current parameters, then calculates the error through the loss function, and uses the optimizer to adjust the model parameters based on the error backpropagation algorithm. Continuously iterate the training until the performance of the model on the validation set no longer improves. For example, the value of the loss function on the validation set no longer decreases. At this time, it is considered that the model converges and the training ends. To prevent the model from overfitting, an early stopping mechanism can also be set, that is, when the performance on the validation set has not improved within a certain number of rounds (such as 10-20 rounds), stop the training. After the model training is completed, by calculating various indicators of the model on the test set, such as accuracy, recall rate, etc., accurately understand the performance of the model in practical applications. These indicators help determine whether the model meets the expected effect and whether further optimization or adjustment is needed.

[0038] Finally, input the actually collected dynamic parameter time data set into the trained BiGRU model. The model will process the data according to the parameters and structure learned during training. After being processed by the BiGRU model, its output is the dynamic parameter time series correlation coding feature vector. This vector integrates the forward and backward dependence relationships and change characteristics of the dynamic parameters in the time dimension, effectively reflecting the change laws and potential patterns of the bridge dynamic parameters over time, and can provide valuable feature support for subsequent tasks such as bridge health condition assessment.

[0039] In the embodiment of the present application, the environmental parameter feature encoding subunit 1213 is configured to extract the environmental parameter time-series correlation encoding feature from the time dataset of the environmental parameters. Specifically, in the embodiment of the present application, the environmental parameter feature encoding subunit is configured to: use a time-series encoder based on BiLSTM to perform environmental parameter time-series feature extraction on the time dataset of the environmental parameters to obtain an environmental parameter time-series correlation encoding feature vector as the environmental parameter time-series correlation encoding feature. It should be understood that the changes in environmental parameters (temperature value, humidity value, wind speed value, and vehicle load value) over time have a long-term cumulative impact on the bridge structure, and there is also a synergistic effect among these environmental factors. For example, the changes in temperature and humidity may jointly affect the durability of bridge materials, and the combination of wind speed and vehicle load may cause the bridge to bear more complex dynamic loads. Extracting the environmental parameter time-series correlation encoding feature from the time dataset of the environmental parameters can capture the change trends and their mutual relationships of these environmental factors in the time series. In particular, in a specific embodiment of the present application, a time-series encoder based on BiLSTM is used to perform environmental parameter time-series feature extraction on the time dataset of the environmental parameters to obtain an environmental parameter time-series correlation encoding feature vector as the environmental parameter time-series correlation encoding feature. Those of ordinary skill in the art should know that LSTM can effectively learn the long-term dependencies in the sequence by introducing memory units and gating mechanisms (input gate, forget gate, and output gate). Based on LSTM, BiLSTM processes time-series data in both forward and backward directions and can consider information from both the past and the future simultaneously. Specifically, the changes in environmental parameters often have a certain periodicity and hysteresis and are affected by the interaction of multiple factors. For example, the change in temperature may be affected by multiple factors such as seasons, day and night, and weather systems. The bidirectional structure of BiLSTM can consider the changes in environmental parameters in both past and future times simultaneously, thus more comprehensively capturing the long-term trends and periodic characteristics in the environmental parameter time series. Moreover, BiLSTM can also learn the interaction relationships between different environmental parameters and encode them into the time-series correlation features. By processing the interaction information in the time-series data, BiLSTM can better reflect the comprehensive impact of environmental factors on the bridge, thereby providing richer information for accurately judging the bridge risk category.

[0040] In the embodiment of the present application, the feature interaction unit 122 is configured to perform heterogeneous multi-source parameter feature interaction encoding on the dynamic parameter time-series correlation encoding feature and the environmental parameter time-series correlation encoding feature to obtain the dynamic parameter-environment parameter deep interaction feature. Specifically, Figure 4 The block diagram of the feature interaction unit in the highway bridge comprehensive monitoring system based on big data according to the embodiment of the present application is shown in Figure 4As shown, the feature interaction unit 122 includes: a dynamic parameter - environmental parameter implicit feature extraction subunit 1221, configured to perform parameter - associated implicit feature extraction on the dynamic parameter time - series associated encoding feature vector and the environmental parameter time - series associated encoding feature vector respectively to obtain a set of dynamic parameter time - series associated implicit encoding feature vectors and environmental parameter time - series associated principal component implicit feature encoding vectors; a dynamic parameter - environmental parameter feature anchoring subunit 1222, configured to perform parameter - associated feature implicit key clue anchoring on the set of dynamic parameter time - series associated implicit encoding feature vectors and the environmental parameter time - series associated principal component implicit feature encoding vectors to obtain a {dynamic parameter time - series associated implicit encoding feature vector, anchored environmental parameter time - series associated principal component implicit feature encoding vector} feature pair; a dynamic parameter - environmental parameter deep interaction feature generation subunit 1223, configured to perform heterogeneous multi - source parameter fine - grained interaction analysis on the {dynamic parameter time - series associated implicit encoding feature vector, anchored environmental parameter time - series associated principal component implicit feature encoding vector} feature pair to obtain a dynamic parameter - environmental parameter deep interaction feature encoding vector as the dynamic parameter - environmental parameter deep interaction feature.

[0041] It should be understood that the dynamic parameter time - series associated encoding feature is obtained by encoding the time dataset of dynamic parameters, and the environmental parameter time - series associated encoding feature is obtained by encoding the time dataset of environmental parameters. Since dynamic parameters (such as acceleration, stress, etc.) and environmental parameters (such as temperature, humidity, etc.) come from different sensors, there will be an essential difference between these data types, which will further result in a lack of obvious semantic consistency between the obtained dynamic parameter time - series associated encoding feature and the environmental parameter time - series associated encoding feature. If the dynamic parameter and environmental parameter features are directly concatenated or simply combined, it may lead to ineffective interaction between features, thereby affecting the final risk assessment result of the bridge. Based on this, in this application, heterogeneous multi - source parameter feature interaction encoding is performed on the dynamic parameter time - series associated encoding feature and the environmental parameter time - series associated encoding feature to obtain the dynamic parameter - environmental parameter deep interaction feature. That is, through heterogeneous multi - source parameter feature interaction encoding processing, the input dynamic parameter and environmental parameter time - series associated encoding features are hierarchically architecturally characterized, and the semantic alignment of different parameter - associated encoding features is completed through an implicit key clue anchoring mechanism to generate a dynamic parameter - environmental parameter deep interaction feature with high semantic consistency and fine - grained interaction representation. In this way, the model can more effectively utilize the interaction information between dynamic parameters and environmental parameters to evaluate the bridge health status, thereby improving the accuracy and reliability of the assessment result.

[0042] Specifically, in the embodiments of the present application, the dynamic parameter environmental parameter implicit feature extraction subunit is used to: perform dynamic parameter association implicit feature extraction based on point convolution coding on the dynamic parameter time series correlation coding feature vector to obtain the dynamic parameter time series correlation implicit coding feature vector, and this process can be expressed by the formula:

[0043] H1 = sigmoid[Conv 1×1 (V1)]

[0044] where V1 is the dynamic parameter time series correlation coding feature vector, sigmoid is the sigmoid function, Conv 1×1 is point convolution coding, and H1 is the dynamic parameter time series correlation implicit coding feature vector;

[0045] Perform principal component analysis of the environmental parameter time series correlation coding feature vector to obtain a set of environmental parameter time series correlation principal component implicit coding vectors, and this process can be expressed by the formula:

[0046]

[0047] where V2 is the environmental parameter time series correlation coding feature vector, PCA(·) is the principal component analysis operation, C2 is the environmental parameter time series correlation feature sample covariance matrix calculated through V2, U2 is the environmental parameter time series correlation feature principal component orthogonal matrix, v 21 , v 22 , v 2i and v 2m are respectively the 1st, 2nd, ith, and mth environmental parameter time series correlation principal component implicit coding vectors in the set of environmental parameter time series correlation principal component implicit coding vectors, Λ2 is the environmental parameter time series correlation feature diagonal matrix, λ 21 , λ 2m are respectively the eigenvalues corresponding to v 21 and v 2m , and U2 T is the transpose matrix of U2;

[0048] Perform environmental parameter association implicit feature extraction based on point convolution coding on the set of environmental parameter time series correlation principal component implicit coding vectors to obtain the set of environmental parameter time series correlation principal component implicit feature coding vectors, and this process can be expressed by the formula:

[0049]

[0050] where v 2i is the ith environmental parameter time series correlation feature principal component coding vector in the set of environmental parameter time series correlation feature principal component coding vectors, Conv 1×1is point convolution encoding, sigmoid is the sigmoid function, h 21 , h 22 , h 2i and h 2m are respectively the 1st, 2nd, i-th, and m-th environmental parameter time-series correlation principal component hidden feature encoding vectors in the set of environmental parameter time-series correlation principal component hidden feature encoding vectors, and H2 is the set of the environmental parameter time-series correlation principal component hidden feature encoding vectors.

[0051] It should be understood that the time dataset of dynamic parameters (such as acceleration, stress, etc.) contains rich information. However, the previously obtained dynamic parameter time-series correlation encoding feature vectors may only be preliminary encoding results, and there are some deep features that have not been mined. There may be complex interaction relationships between different dynamic parameters, and these relationships may not be fully reflected in the original encoding features. The point convolution encoding method can integrate information between feature channels across dimensions and does not depend on the locality of the original feature space. In this way, the internal connections between different dimensions of dynamic parameters can be deeply mined, so as to extract more representative and abstract hidden features. That is, through the cross-dimensional weight mapping and non-linear activation function of point convolution, the obtained dynamic parameter time-series correlation hidden encoding feature vectors are more refined and abstract, and can better reflect the essential features and internal laws of the dynamic parameters changing with time. This can provide more valuable and distinguishable information for subsequent interaction with environmental parameter features, and contribute to improving the accuracy of bridge health state assessment.

[0052] Correspondingly, considering that the environmental parameter time-series correlation encoding feature vectors usually have a high dimension, and there may be a certain correlation between different dimensions, which will lead to redundant information in the data. At the same time, high-dimensional data will increase the complexity and computational amount of subsequent calculations and reduce the processing efficiency. Based on this, in this application, principal component analysis of the environmental parameter time-series correlation encoding feature vectors is performed to construct a new low-dimensional feature space. In this space, the high-variance information of the original environmental parameter data can be retained to the greatest extent, that is, the most important feature information in the data is retained. That is, the generated set of environmental parameter time-series correlation principal component hidden encoding vectors retains the main correlation features of the environmental parameters while reducing data redundancy and noise. Low-dimensional data can make subsequent calculations and analyses more efficient, and at the same time avoid the overfitting problem that may be caused by too high data dimensions, which is beneficial to improving the generalization ability of the model.

[0053] It should be understood that although the time-series correlated principal component implicit encoding vectors of environmental parameters obtained through principal component analysis have reduced the dimension, there may still be some implicit features that can be further explored. There may be potential interaction relationships between the principal components of different environmental parameters, and these relationships are very important for accurately evaluating the health status of the bridge. The point convolution encoding method can capture the interaction relationships between the channels in the feature space of environmental parameters, and thus can deeply explore the internal connections between the principal components of environmental parameters and extract more representative implicit features. That is, the set of time-series correlated principal component implicit feature encoding vectors of environmental parameters can more accurately reflect the law of change of environmental parameters over time and the interaction between different environmental parameters.

[0054] Then, parameter correlation feature implicit key clue anchoring is performed on the set of the time-series correlated implicit encoding feature vectors of dynamic parameters and the set of the time-series correlated principal component implicit feature encoding vectors of environmental parameters to obtain a {time-series correlated implicit encoding feature vector of dynamic parameters, anchored time-series correlated principal component implicit feature encoding vector of environmental parameters} feature pair. The above process can be expressed by the formula:

[0055]

[0056] F bestpair ={H1; h 2k}

[0057] where h 2i is the i-th time-series correlated principal component implicit feature encoding vector in the set of time-series correlated principal component implicit feature encoding vectors of environmental parameters, H1 is the time-series correlated implicit encoding feature vector of dynamic parameters, <·> represents the inner product, ‖·‖ is the Euclidean norm for calculating vectors, ε is the modulation coefficient, arg max j (·) returns the j value corresponding to the maximum value, k is the position to find the maximum approximate matching value in the set of time-series correlated principal component implicit feature encoding vectors of environmental parameters, h 2k is the anchored time-series correlated principal component implicit feature encoding vector, and F bestpair is the {time-series correlated implicit encoding feature vector of dynamic parameters, anchored time-series correlated principal component implicit feature encoding vector of environmental parameters} feature pair.

[0058] It should be understood that a parameter - associated feature implicit key clue anchoring operation is performed on the set of the dynamic - parameter time - series associated implicit coding feature vectors and the environmental - parameter time - series associated principal - component implicit feature coding vectors. By calculating the correlation between the two, the most valuable feature pairs are found to complete the semantic alignment of the dynamic - parameter and environmental - parameter features. This can endow the subsequent fine - grained feature interaction operation with a clear structural prior, eliminate the ambiguity between different feature sources, and enable the features to be associated and interact in a consistent manner. That is, the obtained {dynamic - parameter time - series associated implicit coding feature vectors, anchored environmental - parameter time - series associated principal - component implicit feature coding vectors} feature pairs are more semantically consistent and can be associated and interacted more effectively. This can avoid the chaos and ineffective calculations in the process of dynamic - parameter and environmental - parameter feature interaction, thereby improving the quality and efficiency of feature interaction.

[0059] Finally, a heterogeneous multi - source parameter fine - grained interaction analysis is performed on the {dynamic - parameter time - series associated implicit coding feature vectors, anchored environmental - parameter time - series associated principal - component implicit feature coding vectors} feature pairs to obtain a dynamic - parameter - environmental - parameter depth - interaction feature coding vector as the dynamic - parameter - environmental - parameter depth - interaction feature. The above process can be expressed by the formula:

[0060]

[0061] where H1 is the dynamic - parameter time - series associated implicit coding feature vector, h 2k T is the transpose vector of h 2k , softmax is the activation function, S is the length of h 2k , is matrix multiplication, α and β are weighted hyperparameters, and V f is the dynamic - parameter - environmental - parameter depth - interaction feature coding vector.

[0062] It should be understood that after completing the semantic alignment, it is necessary to further explore the more complex non - linear feature associations between dynamic parameters and environmental parameters. For example, temperature changes can affect the mechanical properties of bridge materials, and further affect dynamic parameters such as the stress and acceleration of the bridge; the influence of vehicle loads on the bridge is also different under different wind speed and humidity conditions. These complex relationships may not have been fully explored in the previous steps. Fine - grained interaction analysis can capture long - distance dependencies and multi - dimensional semantic interactions through the multi - head self - attention mechanism, so as to deeply explore these complex non - linear feature associations, enabling the model to more comprehensively and deeply understand the comprehensive impact of dynamic parameters and environmental parameters on the bridge health state. The finally generated dynamic - parameter - environmental - parameter depth - interaction feature coding vector contains the depth - interaction information between dynamic parameters and environmental parameters, and the model can use this interaction information to achieve a more accurate assessment of the bridge state.

[0063] Preferably, in another embodiment, the dynamic parameter - environmental parameter deep interaction feature generation subunit is configured to: perform semantic distribution modulation based on weak blurring expansion on the {dynamic parameter time - series correlation implicit coding feature vector, anchored environmental parameter time - series correlation principal component implicit feature coding vector} feature pair to obtain a modified {dynamic parameter time - series correlation implicit coding feature vector, anchored environmental parameter time - series correlation principal component implicit feature coding vector} feature pair; perform heterogeneous multi - source parameter fine - grained interaction analysis on the modified {dynamic parameter time - series correlation implicit coding feature vector, anchored environmental parameter time - series correlation principal component implicit feature coding vector} feature pair to obtain a dynamic parameter - environmental parameter deep interaction feature coding vector as the dynamic parameter - environmental parameter deep interaction feature. The above process can be expressed by the formula:

[0064]

[0065] where H 1i is the eigenvalue at the i - th position in H1, h 2ki is the eigenvalue at the i - th position in h 2k , H 1i ′ is the modified eigenvalue of H 1i , h 2ki ′ is the modified eigenvalue of h 2ki , H1′ is the modified dynamic parameter time - series correlation implicit coding feature vector, h 2k ′ is the modified anchored environmental parameter time - series correlation principal component implicit feature coding vector, and V f is the dynamic parameter - environmental parameter deep interaction feature coding vector.

[0066] Particularly, for the alignment ambiguity caused by source uncertainty of the dynamic parameter time - series correlation implicit coding feature vector and the anchored environmental parameter time - series correlation principal component implicit feature coding vector, a weakening and blurring mechanism is introduced for weak blurring power - law expansion. That is, the 3 / 8 exponent is used as the precursor prior expansion to perform power - law prior responsive fuzzification convergence on the boundary alignment condition of the dynamic parameter time - series correlation implicit coding feature parameters, and the 1 / 4 exponent is used as the main body alignment expansion to perform strict constraint on the relaxation of the power - law distribution alignment decay of the anchored environmental parameter time - series correlation principal component implicit feature. Thus, in the case where the alignment boundary condition constraints within the correlation effective range are ill - defined, the value correlation mechanism is used to avoid the prior ambiguity of the system behavior under a single mechanism, to correct the semantic distribution consistency fuzzification within the alignment interval, and to enhance the intuitive mining of the implicit fine - grained interaction correlation of the anchored features.

[0067] In the embodiment of the present application, the evaluation result generation module 130 is configured to obtain an intelligent auxiliary evaluation result based on the dynamic parameter - environmental parameter deep interaction feature. Specifically, in the embodiment of the present application, the evaluation result generation module is configured to: input the dynamic parameter - environmental parameter deep interaction feature encoding vector into the bridge comprehensive monitoring and evaluation module based on a classifier to obtain the intelligent auxiliary evaluation result, and the intelligent auxiliary evaluation result is a risk category label. That is, by using the discrimination ability of the classifier, the relatively abstract feature vector is converted into an intuitive risk category label, so as to realize the evaluation of the bridge health status. The risk category label result here can be high risk, medium risk, low risk, etc. Specifically, a classifier is a machine learning model, and its main function is to divide data into different categories according to the input data features. Its working principle is to learn the mapping relationship between features and categories based on training data. The classifier is trained with a large amount of key health data of bridges with known labels (known risk categories) so that it can learn what feature combinations correspond to high risk, medium risk, low risk and other categories. In practical applications, when a new dynamic parameter - environmental parameter deep interaction feature encoding vector is input, the classifier can make a classification judgment according to the learned knowledge. The obtained risk category label provides a key decision-making basis for bridge maintenance and management. For example, for high-risk bridges, emergency repairs or traffic flow restrictions need to be taken immediately to prevent safety accidents. For medium-risk bridges, regular inspection and maintenance plans can be arranged, and their status changes can be closely monitored. For low-risk bridges, maintenance can be carried out according to the normal maintenance cycle, and maintenance resources can be reasonably allocated.

[0068] In summary, the big data-based highway bridge comprehensive monitoring system 100 according to the embodiment of the present application is clarified. It uses big data-based data processing technology to perform data splitting and time series feature extraction on the time series dataset of the key health data of the bridge to obtain the dynamic parameter time series correlation implicit coding feature and the environmental parameter time series correlation implicit coding feature, and obtains an intelligent auxiliary evaluation result based on the heterogeneous multi-source parameter feature interaction coding representation between the dynamic parameter time series correlation implicit coding feature and the environmental parameter time series correlation implicit coding feature. In this way, by processing bridge monitoring data in real time and fully considering the non-linear relationship between different bridge monitoring data during processing, the health status of the bridge can be evaluated more accurately, meeting the accuracy, timeliness and comprehensiveness requirements of highway bridge monitoring.

Claims

1. A comprehensive monitoring system for highway bridges based on big data, characterized by: include: A data acquisition module, used for collecting a time data set of key health data of the bridge in real time through a sensor network, wherein the key health data includes dynamic parameters and environmental parameters; A data analysis module, used for performing data interaction analysis based on time series features on the time data set of the key health data to obtain dynamic parameter-environmental parameter deep interaction features, wherein the data analysis module includes: a data splitting and encoding unit, used for performing data splitting and time series feature extraction on the time data set of the key health data to obtain dynamic parameter time series correlation coding features and environmental parameter time series correlation coding features; a feature interaction unit, used for performing heterogeneous multi-source parameter feature interaction encoding on the dynamic parameter time series correlation coding features and the environmental parameter time series correlation coding features to obtain the dynamic parameter-environmental parameter deep interaction features; The evaluation result generation module is used to obtain intelligent auxiliary evaluation results based on the deep interaction characteristics of the dynamic parameters and environmental parameters.

2. The highway bridge comprehensive monitoring system based on big data according to claim 1 is characterized in that: The dynamic parameters include acceleration value, vibration frequency value, stress value and strain value, and the environmental parameters include temperature value, humidity value, wind speed value and vehicle load value.

3. The highway bridge comprehensive monitoring system based on big data according to claim 2 is characterized in that: The data splitting and encoding unit comprises: A data splitting subunit, used for splitting the time data set of the key health data to obtain a time data set of dynamic parameters and a time data set of environmental parameters; A dynamic parameter feature encoding subunit, used to extract the dynamic parameter temporal correlation encoding feature from the time data set of the dynamic parameter; The environmental parameter feature encoding subunit is used to extract the environmental parameter time-series associated coding features from the environmental parameter time data set.

4. The highway bridge comprehensive monitoring system based on big data according to claim 3 is characterized in that: The dynamic parameter feature encoding subunit is used to: use a BiGRU-based temporal encoder to extract dynamic parameter temporal features from the time data set of the dynamic parameter to obtain a dynamic parameter temporal association encoding feature vector as the dynamic parameter temporal association encoding feature.

5. The highway bridge comprehensive monitoring system based on big data according to claim 4 is characterized in that: The environmental parameter feature encoding subunit is used to: use a BiLSTM-based temporal encoder to extract environmental parameter temporal features from the time data set of the environmental parameter to obtain an environmental parameter temporal association encoding feature vector as the environmental parameter temporal association encoding feature.

6. The highway bridge comprehensive monitoring system based on big data according to claim 1 is characterized in that: The feature interaction unit includes: A dynamic parameter and environmental parameter implicit feature extraction subunit is used to extract parameter association implicit features from the dynamic parameter time series association coding feature vector and the environmental parameter time series association coding feature vector respectively to obtain a set of dynamic parameter time series association implicit coding feature vectors and environmental parameter time series association principal component implicit feature coding vectors; A dynamic parameter environment parameter feature anchoring subunit is used to anchor the parameter association feature implicit key clues on the set of the dynamic parameter time series association implicit coding feature vector and the environment parameter time series association principal component implicit feature coding vector to obtain a feature pair of {dynamic parameter time series association implicit coding feature vector, anchored environment parameter time series association principal component implicit feature coding vector}; The dynamic parameter-environmental parameter deep interaction feature generation subunit is used to perform heterogeneous multi-source parameter fine-grained interaction analysis on the feature pair of {dynamic parameter time series association implicit coding feature vector, anchored environmental parameter time series association principal component implicit feature coding vector} to obtain the dynamic parameter-environmental parameter deep interaction feature coding vector as the dynamic parameter-environmental parameter deep interaction feature.

7. The highway bridge comprehensive monitoring system based on big data according to claim 6 is characterized in that: The dynamic parameter environment parameter implicit feature extraction subunit is used to: Performing point convolution coding-based dynamic parameter association implicit feature extraction on the dynamic parameter time series association coding feature vector to obtain the dynamic parameter time series association implicit coding feature vector; Performing a principal component analysis on the environmental parameter time series associated coding feature vector to obtain a set of environmental parameter time series associated principal component implicit coding vectors; The set of the environmental parameter time series associated principal component implicit coding vectors is subjected to environmental parameter associated implicit feature extraction based on point convolution coding to obtain the set of the environmental parameter time series associated principal component implicit feature coding vectors.

8. The highway bridge comprehensive monitoring system based on big data according to claim 7 is characterized in that: The dynamic parameter-environmental parameter deep interaction feature generation subunit is used to: Performing semantic distribution modulation based on weak virtualization extension on the feature pair of {dynamic parameter temporal association implicit coding feature vector, anchored environmental parameter temporal association principal component implicit feature coding vector} to obtain a modified feature pair of {dynamic parameter temporal association implicit coding feature vector, anchored environmental parameter temporal association principal component implicit feature coding vector}; A heterogeneous multi-source parameter fine-grained interaction analysis is performed on the modified {dynamic parameter time series association implicit coding feature vector, anchored environmental parameter time series association principal component implicit feature coding vector} feature pair to obtain a dynamic parameter-environmental parameter deep interaction feature coding vector as the dynamic parameter-environmental parameter deep interaction feature.

9. The highway bridge comprehensive monitoring system based on big data according to claim 8 is characterized in that: The assessment result generation module is used to: input the dynamic parameter-environmental parameter deep interaction feature coding vector into a classifier-based bridge comprehensive monitoring and assessment module to obtain the intelligent auxiliary assessment result, and the intelligent auxiliary assessment result is a risk category label.

Images