A long-span bridge vortex-induced vibration early warning method and system based on a multi-time resolution double-channel deep neural network

Abstract

This invention proposes a method and system for early warning of vortex-induced vibration in long-span bridges based on a multi-temporal-resolution dual-channel deep neural network. The invention utilizes a bridge health monitoring system to collect real-time wind environment and vibration acceleration data from multiple measuring points on the main girder. Low-temporal-resolution long-series features, high-temporal-resolution short-series features, and time-period features are extracted and generated. These are input into a pre-trained dual-branch LSTM model, which outputs the probability of vortex-induced vibration occurring in the next 10-60 minutes using a direct multi-step method. Combined with a graded early warning threshold, corresponding early warning signals are triggered, and matching control measures are pushed out. This invention can achieve vortex-induced vibration early warning up to 60 minutes in advance, achieving 100% recall for large-scale vortex-induced vibration events. It can be directly adapted to the real-time operation of existing bridge health monitoring systems, reserving sufficient control time for vortex-induced vibration prevention and control, and ensuring the safe operation of the bridge.

Description

Technical Field

[0001] This invention relates to the field of civil engineering structural health monitoring and artificial intelligence application technology, and in particular to a method and system for early warning of vortex-induced vibration of long-span bridges based on a multi-time resolution dual-channel deep neural network. Background Technology

[0002] With the development of transportation infrastructure construction, the application of long-span bridges (such as suspension bridges and cable-stayed bridges) is becoming increasingly widespread. Due to the high flexibility and low damping of long-span bridge structures, they are extremely sensitive to wind loads. Vortex-induced vibration (VIV) is a common form of wind-induced vibration in bridges. Although its amplitude is usually limited, frequent VIV can lead to fatigue damage to components, affecting driving safety and comfort.

[0003] Existing vortex-induced vibration early warning methods mainly rely on monitoring data at a single time scale (such as using only 10-minute average wind speed) for threshold judgment or shallow machine learning regression. However, the occurrence of vortex-induced vibration is not only related to the long-term trend of average wind speed, but also closely related to short-term wind field pulsations and abrupt changes in the flow field. Data input with a single time resolution often has limitations: if the time resolution is too low (e.g., 10 minutes), it is difficult to capture instantaneous inducing factors; if the time resolution is too high (e.g., 1 minute), it is difficult to learn the long-term evolution law of the wind field with a limited window length.

[0004] Therefore, there is an urgent need for a technical solution that can integrate long-term and short-term wind field characteristics and achieve high-precision, multi-step early warning through deep learning models. Summary of the Invention

[0005] The purpose of this invention is to solve the problem that existing technologies cannot simultaneously take into account both the long-term trend and short-term pulsation characteristics of wind fields, resulting in low early warning accuracy and high false alarm and missed alarm rates. This invention proposes an early warning method and system for vortex-induced vibration of long-span bridges based on a multi-time resolution dual-channel deep neural network.

[0006] This invention is achieved through the following technical solution: This invention proposes a method for early warning of vortex-induced vibration of long-span bridges based on a multi-time-resolution dual-channel deep neural network, the method comprising: Step S1, Data Acquisition and Synchronization: Real-time acquisition of anemometer data and accelerometer data at key locations of long-span bridges, followed by time synchronization and noise reduction processing. Step S2, Multi-temporal resolution feature construction: Based on the collected data, construct low-temporal resolution feature sequences respectively. and high temporal resolution feature sequences ; Step S3, Physical and Temporal Feature Encoding: Calculate and incorporate the energy concentration coefficient feature and time period encoding feature into the feature sequence; Step S4, Dual-channel parallel inference: This involves processing the low-temporal-resolution feature sequence... Input is fed into the first LSTM network channel to extract long-term trend feature vectors. High temporal resolution feature sequences The input is fed into the second LSTM network channel to extract short-term pulsation feature vectors. ; Step S5, Feature Fusion and Probability Output: and The data is spliced ​​and fused, then input into a multilayer perceptron (MLP) for nonlinear mapping. The probability value of vortex-induced vibration occurring at the Kth time in the future is output through the sigmoid activation function. Step S6, graded early warning: Compare the output probability value with the preset threshold. If it is greater than the threshold, issue the corresponding level of vortex vibration early warning signal.

[0007] Furthermore, the low temporal resolution feature sequence Includes statistical features over the past N time steps, each time step being... minute.

[0008] Furthermore, the high temporal resolution feature sequence Includes statistical features for the most recent M time steps, each time step being... minutes, and .

[0009] Further, in step S2, the low temporal resolution feature sequence Time step The time resolution is 10 minutes, and the sequence length N is 12; the high temporal resolution feature sequence Time step The duration is 1 minute, and the sequence length M is 10.

[0010] Further, in step S3, the formula for calculating the energy concentration factor is:

[0011] In the formula, and These represent the magnitudes of the first and second peak values ​​in the energy spectral density image of bridge vibration data; this energy concentration coefficient feature is used to characterize the degree of concentration of vibration energy at a specific frequency.

[0012] Furthermore, in step S3, the time period encoding feature includes month features, date features, and hour features, which are defined as follows: .

[0013] Furthermore, in step S6, the warning signal includes warnings for different lead times, with the lead time range covering the next 10 to 60 minutes.

[0014] This invention also proposes an early warning system for vortex-induced vibration of long-span bridges based on a multi-time-resolution dual-channel deep neural network, the system comprising: Data sensing module: used to acquire anemometer data and accelerometer data arranged on key sections of the main beam of the bridge; Data preprocessing module: used to calculate wind speed, wind direction, wind angle of attack, turbulence intensity, root mean square value of vibration acceleration and energy concentration coefficient, and generate dual time resolution feature vectors; Intelligent inference module: Internally deployed with a pre-trained dual-channel LSTM deep neural network model, used to receive feature vectors and calculate vortex probability in real time; Early warning terminal module: Used to receive vortex-induced vibration probability. When the probability exceeds a set threshold, it sends early warning information to management personnel through a visual interface or SMS alarm.

[0015] The present invention also proposes an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the method for early warning of vortex-induced vibration of long-span bridges based on a multi-time-resolution dual-channel deep neural network.

[0016] The present invention also proposes a computer-readable storage medium for storing computer instructions, which, when executed by a processor, implement the steps of the method for early warning of vortex-induced vibration of long-span bridges based on a multi-time-resolution dual-channel deep neural network.

[0017] The beneficial effects of this invention are: 1. Multi-scale fusion: By using a dual-channel LSTM structure, the model's ability to learn about vortex-induced vibration triggering mechanisms under complex wind field conditions is significantly improved by simultaneously utilizing the trend of 10-minute statistical data and the sensitivity of 1-minute pulsation data.

[0018] 2. Enhanced physical characteristics: The introduction of an energy concentration coefficient enhances the model's ability to identify precursors of "lock-in" phenomena from a mechanistic perspective; the introduction of time coding enables the model to perceive periodic changes in the environment.

[0019] 3. High recall rate: Actual test data shows that the method described in this invention significantly improves the recall rate of sparse vortex-induced vibration events while ensuring high accuracy, effectively avoiding missed reports. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0021] Figure 1 This is a schematic diagram of the multi-time-resolution dual-channel LSTM network structure in this invention.

[0022] Figure 2 This is a schematic diagram of the confusion matrix of the single-step early warning effect of the present invention on an actual bridge test set. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] This invention proposes a method for early warning of vortex-induced vibration of long-span bridges based on a multi-time-resolution dual-channel deep neural network. The method includes: Data construction phase: Build low-time-resolution datasets (reflecting trends) in 10-minute increments, and build high-time-resolution datasets (reflecting pulsations) in 1-minute increments.

[0025] Feature engineering stage: In addition to conventional wind speed and acceleration characteristics, an "energy concentration coefficient" is creatively introduced to quantify the energy distribution characteristics in the frequency domain, and a "time period characteristic" of sine and cosine coding is introduced to quantify seasonal and diurnal variation patterns.

[0026] Model inference stage: Construct a dual LSTM (Long Short-Term Memory) network structure to process the two sets of time-series data with different resolutions, and perform non-linear mapping through feature concatenation and fully connected layers (DNN).

[0027] Early warning output stage: Outputs the probability of vortex-induced vibration occurring at different time spans in the future (e.g., 10-60 minutes).

[0028] Specifically, in combination Figures 1-2 This invention proposes a method for early warning of vortex-induced vibration of long-span bridges based on a multi-time-resolution dual-channel deep neural network. The method includes: Step S1, Data Acquisition and Synchronization: Real-time acquisition of anemometer data and accelerometer data at key locations of long-span bridges, followed by time synchronization and noise reduction processing. Step S2, Multi-temporal resolution feature construction: Based on the collected data, construct low-temporal resolution feature sequences respectively. and high temporal resolution feature sequences ; Step S3, Physical and Temporal Feature Encoding: Calculate and incorporate the energy concentration coefficient feature and time period encoding feature into the feature sequence; Step S4, Dual-channel parallel inference: This involves processing the low-temporal-resolution feature sequence... Input is fed into the first LSTM network channel to extract long-term trend feature vectors. High temporal resolution feature sequences The input is fed into the second LSTM network channel to extract short-term pulsation feature vectors. ; Step S5, Feature Fusion and Probability Output: and The data is spliced ​​and fused, then input into a multilayer perceptron (MLP) for nonlinear mapping. The probability value of vortex-induced vibration occurring at the Kth time in the future is output through the sigmoid activation function. Step S6, graded early warning: Compare the output probability value with the preset threshold. If it is greater than the threshold, issue the corresponding level of vortex vibration early warning signal.

[0029] Furthermore, the low temporal resolution feature sequence Includes statistical features over the past N time steps, each time step being... minute.

[0030] Furthermore, the high temporal resolution feature sequence Includes statistical features for the most recent M time steps, each time step being... minutes, and .

[0031] Further, in step S2, the low temporal resolution feature sequence Time step The time resolution is 10 minutes, and the sequence length N is 12; the high temporal resolution feature sequence Time step The duration is 1 minute, and the sequence length M is 10.

[0032] Further, in step S3, the formula for calculating the Energy Concentration Factor is as follows:

[0033] In the formula, and These are the magnitudes of the first and second peaks in the energy spectral density (PSD) image of bridge vibration data, respectively; this energy concentration coefficient feature is used to characterize the degree of concentration of vibration energy at a specific frequency.

[0034] Furthermore, in step S3, the time period encoding feature includes month features, date features, and hour features, which are defined as follows: .

[0035] Furthermore, in step S6, the warning signal includes warnings for different lead times, with the lead time range covering the next 10 to 60 minutes.

[0036] This invention also proposes an early warning system for vortex-induced vibration of long-span bridges based on a multi-time-resolution dual-channel deep neural network, the system comprising: Data sensing module: used to acquire anemometer data and accelerometer data arranged on key sections of the main beam of the bridge; Data preprocessing module: used to calculate wind speed, wind direction, wind angle of attack, turbulence intensity, root mean square value of vibration acceleration and energy concentration coefficient, and generate dual time resolution feature vectors; Intelligent inference module: Internally deployed with a pre-trained dual-channel LSTM deep neural network model, used to receive feature vectors and calculate vortex probability in real time; Early warning terminal module: Used to receive vortex-induced vibration probability. When the probability exceeds a set threshold, it sends early warning information to management personnel through a visual interface or SMS alarm.

[0037] The deep neural network model in the intelligent reasoning module is trained on imbalanced positive and negative sample data using Focal Loss loss function based on historical monitoring data.

[0038] Example This invention proposes a method for early warning of vortex-induced vibration of long-span bridges based on a multi-time-resolution dual-channel deep neural network. The method includes: I. Data Acquisition and Feature Definition Take a long-span suspension bridge (such as the Xihoumen Bridge) as an example.

[0039] 1. Sensor deployment: Ultrasonic anemometers (sampling rate 32Hz) and accelerometers (sampling rate 100Hz) are deployed at the 1 / 4 span, 1 / 2 span and 3 / 4 span sections of the main beam.

[0040] 2. Feature Vector Construction: At runtime t, construct two sets of input vectors: Vector A (Low-frequency trend): Takes data from the past 120 minutes, calculates the average over 10 minutes, forming a sequence of length 12. Features include: average wind speed, wind direction, wind angle of attack, turbulence intensity, and RMS acceleration value.

[0041] Vector B (High-Frequency Pulsations): Takes data from the most recent 10 minutes, calculates statistical values ​​every 1 minute, forming a sequence of length 10. Features include: average wind speed, wind direction, wind angle of attack, turbulence intensity, and RMS acceleration value.

[0042] 3. Key Feature Calculation: Time-period features: To enable the model to understand the effects of diurnal temperature range and seasonal monsoons, the current time is encoded:

[0043] Where m, d, and h represent month, day, and hour, respectively.

[0044] Energy Concentration Factor: The energy concentration factor is defined to address the prominent single-frequency characteristic of vortex-induced vibrations. Power spectral density (PSD) analysis was performed on the acceleration data to extract the first and second peak values. This indicator can effectively distinguish between vortex-induced vibration (dominated by a single frequency, with a small ratio) and random chattering (multi-frequency broadband, with a large ratio).

[0045] II. Dual-channel network architecture like Figure 1 As shown, the dual-channel deep neural network used in this invention includes: The LSTM_R module has an input dimension of (Batch, 12, Feature_Dim) and is used to process low temporal resolution data. It contains two LSTM layers, each with 32 hidden units.

[0046] The LSTM_H module has an input dimension of (Batch, 10, Feature_Dim) and is used to process high temporal resolution data. It contains two LSTM layers, each with 32 hidden units.

[0047] Concatenate layer: concatenates the final output vectors of LSTM_R and LSTM_H.

[0048] Deep Neural Network (DNN): The concatenated vector is input into a 3-layer fully connected network (32-32-16 nodes). The activation function is Tanh, and Dropout (0.5) is added between layers to prevent overfitting.

[0049] Output layer: The Sigmoid function outputs a probability value P between 0 and 1, representing the probability of vortex oscillation occurring at time T in the future.

[0050] III. Online Early Warning Process This invention runs in a 1-minute cycle: Real-time reading: The system reads the latest sensor data from the database every minute.

[0051] Feature update: Update the 10-minute statistical sequence and the 1-minute statistical sequence (sliding window).

[0052] Model inference: The data is fed into the trained model. In this embodiment, the model can simultaneously output prediction results for the next 10 minutes, 20 minutes...60 minutes (i.e., steps 1-6).

[0053] Threshold discrimination: If the predicted probability is within the next 10 minutes (Threshold adjustable) The system determines that vortex-induced vibration is about to occur and issues a red alert (immediate action to be taken). If the predicted probability is within the next 60 minutes... Furthermore, the number of cases has been on the rise recently, prompting the system to issue a yellow alert (due to the long forecast period and high uncertainty, attention is advised).

[0054] IV. Verification Results Validation based on test data from the entire year of 2017 shows that the method achieves a recall rate of 96.47% and an accuracy of 99.24% in a 10-minute advance warning task; and in a 60-minute advance warning task, it achieves a high accuracy for large-amplitude vortex-induced vibrations (acceleration > 20 cm / s²). 2 The accuracy rate of early warnings remains extremely high.

[0055] This invention proposes a method and system for early warning of vortex-induced vibration in long-span bridges based on a multi-temporal-resolution dual-channel deep neural network, belonging to the field of bridge structural health monitoring and wind-induced vibration control technology. This invention collects wind environment and vibration acceleration data from multiple measuring points on the main girder in real time through a bridge health monitoring system, extracts and generates low-temporal-resolution long-series features, high-temporal-resolution short-series features, and time-period features, inputs them into a pre-trained dual-branch LSTM model, and outputs the probability of vortex-induced vibration occurrence in the next 10-60 minutes using a direct multi-step method. Combined with a graded early warning threshold, corresponding early warning signals are triggered and matching control measures are pushed. This invention can achieve vortex-induced vibration early warning up to 60 minutes in advance, achieves 100% recall for large-scale vortex-induced vibration events, and can be directly adapted to the real-time operation of existing bridge health monitoring systems, reserving sufficient control time for vortex-induced vibration prevention and control, and ensuring the safe operation of bridges.

[0056] The present invention also proposes an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the method for early warning of vortex-induced vibration of long-span bridges based on a multi-time-resolution dual-channel deep neural network.

[0057] The present invention also proposes a computer-readable storage medium for storing computer instructions, which, when executed by a processor, implement the steps of the method for early warning of vortex-induced vibration of long-span bridges based on a multi-time-resolution dual-channel deep neural network.

[0058] The memory in this application embodiment can be volatile memory or non-volatile memory, or it can include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous linked dynamic random access memory (SLDRAM), and direct rambus RAM (DR RAM). It should be noted that the memory used in the methods described in this invention is intended to include, but is not limited to, these and any other suitable types of memory.

[0059] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions. When the computer instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium accessible to a computer or a data storage device such as a server or data center that integrates one or more available media. The available media may be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., high-density digital video discs (DVDs)), or semiconductor media (e.g., solid-state disks (SSDs)).

[0060] In implementation, each step of the above method can be completed by integrated logic circuits in the processor's hardware or by instructions in software. The steps of the method disclosed in the embodiments of this application can be directly implemented by a hardware processor, or by a combination of hardware and software modules in the processor. The software modules can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. This storage medium is located in memory, and the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method. To avoid repetition, detailed descriptions are omitted here.

[0061] It should be noted that the processor in the embodiments of this application can be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above method embodiments can be completed by the integrated logic circuitry in the processor's hardware or by instructions in software form. The processor can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this application can be directly embodied as execution by a hardware decoding processor, or as a combination of hardware and software modules in the decoding processor. The software modules can be located in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. This storage medium is located in memory, and the processor reads the information in the memory and, in conjunction with its hardware, completes the steps of the above methods.

[0062] The foregoing has provided a detailed description of the method and system for early warning of vortex-induced vibration of long-span bridges based on a multi-time-resolution dual-channel deep neural network proposed in this invention. Specific examples have been used to illustrate the principles and implementation methods of this invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this invention. Therefore, the content of this specification should not be construed as a limitation of this invention.

Claims

1. A long-span bridge vortex-induced vibration early warning method based on a multi-time resolution double-channel deep neural network, characterized in that, The method comprises: Step S1, data acquisition and synchronization: real-time acquisition of anemometer data and acceleration sensor data at key positions of a long-span bridge, and time synchronization and denoising processing; Step S2, multi-time resolution feature construction: based on the collected data, respectively construct a low time resolution feature sequence and a high time resolution feature sequence ; Step S3, physical and time characteristic coding: in the characteristic sequence, energy concentration coefficient characteristics and time period coding characteristics are calculated and added; Step S4, double-channel parallel inference: inputting the low-time-resolution feature sequence to the first LSTM network channel to extract a long-term trend feature vector ; inputting the high-time-resolution feature sequence to the second LSTM network channel to extract a short-term pulsation feature vector ; Step S5, feature fusion and probability output: the features are fused and output as a probability value of vortex-induced vibration at the Kth time point in the future. and are spliced and fused, input to a multi-layer perception (MLP) for non-linear mapping, and output a probability value of vortex-induced vibration at the Kth time point in the future through a Sigmoid activation function. Step S6, hierarchical early warning: the output probability value is compared with a preset threshold value, and if greater than the threshold value, a vortex vibration early warning signal of a corresponding level is issued.

2. The method of claim 1, wherein, the low temporal resolution feature sequence comprising statistical features of the past N time steps, each time step being minutes.

3. The method of claim 2, wherein, the high temporal resolution feature sequence comprising statistical features of the most recent M time steps, each time step being minutes, and .

4. The method of claim 3, wherein, In step S2, the low temporal resolution feature sequence has a time step of 10 minutes and a sequence length N of 12; the high temporal resolution feature sequence has a time step of 1 minute and a sequence length M of 10.

5. The method of claim 1, wherein, In step S3, the calculation formula of the energy concentration coefficient is: In the formula, and respectively are the size of the first peak value and the second peak value in the bridge vibration data energy spectrum density image; the energy concentration coefficient feature is used to represent the concentration degree of vibration energy at a specific frequency.

6. The method of claim 1, wherein, In step S3, the time period coding characteristics include month characteristics, date characteristics and hour characteristics, which are defined as follows: 。 7. The method of claim 1, wherein, In step S6, the early warning signal includes early warnings for different lead times, and the lead time range covers 10 minutes to 60 minutes in the future.

8. A long-span bridge vortex-induced vibration early warning system based on a multi-time resolution double-channel deep neural network, characterized in that, The system comprises: A data perception module for acquiring anemometer data and acceleration sensor data arranged at key sections of a bridge main beam; A data preprocessing module for calculating wind speed, wind direction, wind attack angle, turbulence intensity, vibration acceleration root mean square value and energy concentration coefficient, and generating a double-time resolution feature vector; An intelligent inference module internally deploying a pre-trained double-channel LSTM deep neural network model, for receiving the feature vector and calculating the vortex vibration probability in real time; An early warning terminal module for receiving the vortex vibration probability, and when the probability exceeds a set threshold value, sending early warning information to management personnel through a visual interface or a short message alarm mode. 9.An electronic device comprising a memory and a processor, the memory storing a computer program, wherein, The processor executes the computer program to realize the steps of the method of any one of claims 1-7.

10. A computer readable storage medium for storing computer instructions, characterized in that, The computer instructions are executed by the processor to realize the steps of the method of any one of claims 1-7.

Images