Highway geological disaster identification method and system based on distributed optical fiber sensing das

Abstract

This invention discloses a method and system for identifying geological hazards on highways based on distributed optical fiber sensing (DAS), relating to the field of highway monitoring. The method includes: real-time optical fiber vibration signals, followed by filtering and denoising, amplitude normalization, time-series alignment preprocessing, construction interference suppression, vehicle interference suppression, abnormal signal enhancement preprocessing, and multi-scale time-frequency feature extraction to obtain multi-scale time-frequency features; inputting these multi-scale time-frequency features into a two-stage anomaly detection framework to obtain anomaly identification results; when the anomaly identification result is abnormal and includes at least a geological hazard category, removing the vibration components corresponding to other hazard categories from the optical fiber vibration signal, and extracting features from the optical fiber vibration signal to obtain a feature set; based on the feature set and corresponding preset conditions, obtaining the geological and structural hazard anomaly classification results. This invention can accurately identify and classify the geological and structural hazard anomaly classification results.

Description

Technical Field

[0001] This invention relates to the field of highway geological and structural disaster monitoring technology, and in particular to a method and system for identifying highway geological disasters based on distributed optical fiber sensing (DAS). Background Technology

[0002] Geological and structural hazards along highways pose significant threats to road safety. Disasters such as roadbed collapses, landslides, rockfalls, and bridge damage are characterized by their high degree of concealment, suddenness, and destructive power, easily causing road interruptions, vehicle damage, and casualties. Traditional highway disaster monitoring technologies mainly rely on manual inspections, GNSS displacement monitoring, synthetic aperture radar, and strain gauges. These technologies suffer from limitations such as limited coverage, low monitoring frequency, inability to achieve continuous 24 / 7 monitoring, and high deployment and maintenance costs, making it difficult to automatically identify and issue alerts immediately after a disaster occurs.

[0003] Distributed fiber acoustic sensing (DAS) technology provides a novel technical means for the automatic identification and real-time alarm of geological and structural hazards along highways. Currently, existing DAS-based technologies for identifying geological and structural hazards along highways suffer from problems such as incomplete coverage of hazard types and inability to effectively distinguish between hazard vibrations and environmental interference. Summary of the Invention

[0004] To address the technical problems existing in the background art, this invention proposes a method and system for identifying highway geological hazards based on distributed optical fiber sensing (DAS).

[0005] In a first aspect, the present invention proposes a method for identifying highway geological hazards based on distributed optical fiber sensing (DAS), comprising: Real-time acquisition of fiber optic vibration signals collected by a DAS system pre-laid in a predetermined highway section; The fiber vibration signal is sequentially filtered for noise reduction, amplitude normalization, time alignment preprocessing, construction interference suppression, and vehicle interference suppression to obtain the first intermediate fiber vibration signal. The vibration signal of the first intermediate optical fiber is subjected to anomalous signal enhancement preprocessing and multi-scale time-frequency feature extraction to obtain multi-scale time-frequency features; a two-stage anomaly detection framework and multi-scale time-frequency features are used to perform two-stage anomaly identification to obtain anomaly identification results; among them, the anomaly identification results are divided into normal and abnormal, and the anomalies include one or more of the following: vehicle accident, geological disaster, structural damage, and other anomaly categories; When the anomaly identification result is abnormal and includes one or more of the following categories in addition to geological disasters: vehicle accidents, structural damage, and other anomalies, the vibration components corresponding to one or more of the following disaster categories are removed from the first optical fiber vibration signal to obtain the second intermediate optical fiber vibration signal. Feature extraction is performed on the vibration signal of the second intermediate fiber or the vibration signal of the first intermediate fiber that is abnormal and only includes geological disasters, to obtain a feature set; If the feature set meets the first preset condition, it is determined to be a roadbed collapse disaster; if the feature set meets the second preset condition, it is determined to be a slope landslide disaster; if the feature set meets the third preset condition, it is determined to be a rockfall disaster; if the feature set meets the fourth preset condition, it is determined to be a bridge structure damage disaster.

[0006] Preferably, the first preset condition is: at least one optical fiber channel has a channel RMS energy peak exceeding a preset low-frequency energy threshold in a preset low-frequency band and the duration of exceeding the preset low-frequency energy threshold (i.e., the impact duration) is 0.5 to 2 seconds; or, at least one optical fiber channel exhibits signal abnormality characteristics of optical fiber breakage; wherein, the preset low-frequency band is 10 to 500 Hz. The second preset condition is: there are several adjacent optical fiber channels along the slope extension direction whose channel RMS energy peak values ​​in the full frequency band successively exceed the preset full frequency band energy threshold, and the impact duration is 1 to 10 seconds, and the time delay of the adjacent channel RMS energy peak values ​​exceeding the preset full frequency band energy threshold is less than the preset trigger delay threshold; wherein, the full frequency band is 10 to 2000 Hz. The third preset condition is: there is at least one optical fiber channel whose channel RMS energy peak in the preset mid-to-high frequency band exceeds the preset rockfall impact threshold, and the duration of exceeding the rockfall impact threshold (impact duration) is less than 100ms, and there is a bouncing signal characteristic after the channel RMS energy peak; wherein, the preset mid-to-high frequency band is 50~1000Hz. The fourth preset condition is: if there is at least one optical fiber channel whose channel RMS energy peak change is greater than the preset energy change threshold in the preset high frequency band, and the change duration (impact duration) is less than 50ms, and the change rate of the natural frequency of the bridge structure reaches the preset frequency change threshold or the vibration attenuation rate reaches the preset vibration attenuation rate; wherein, the preset high frequency band is 20~2000Hz.

[0007] Preferably, in the construction interference suppression, the construction reporting area is obtained, and the vibration signal corresponding to the reporting area is removed from the optical fiber vibration signal; in the vehicle interference suppression process, the vehicle spatiotemporal trajectory is calculated from the optical fiber vibration signal, and the time window and spatial monitoring range corresponding to each vehicle passage are obtained based on the vehicle spatiotemporal trajectory; the spatiotemporal mask area of ​​each vehicle passage interference is determined based on the vehicle spatiotemporal trajectory; the optical fiber vibration signal generated by vehicle rolling excitation within the spatiotemporal mask area is removed from the preprocessed optical fiber vibration signal, and the retained optical fiber vibration signal is subjected to baseline smoothing splicing processing.

[0008] Preferably, the vehicle vibration signal undergoes abnormal signal enhancement preprocessing, specifically including: A background noise statistical model is established by acquiring environmental noise data collected by a distributed fiber optic sensing system during periods when no vehicles are driving. Based on the background noise statistical model and an adaptive filtering algorithm, background noise is removed from the vehicle vibration signal. Wavelet packet transform is used to decompose the vehicle vibration signal into multiple scales, resulting in a sub-band signal set and a sub-band coefficient set, which include multiple sub-band signals of different frequency bands and scales. From all the sub-band signals in the sub-band signal set, frequency bands containing abnormal event characteristics are selected and combined to form an abnormal event signal set. The coefficients in the sub-band coefficient set that correspond to the abnormal event signal set are amplified, and the sub-band signal set is reconstructed using the sub-band coefficient set through inverse transform.

[0009] Preferably, the feature set includes the frequency band characteristics, spatiotemporal distribution characteristics, signal evolution laws, and characteristic parameters of each optical fiber channel; the frequency band characteristics include: preset low-frequency band characteristics, preset mid-to-high frequency band characteristics, preset high-frequency band characteristics, and full-band characteristics.

[0010] Preferably, the preset low-frequency band characteristics include the channel RMS energy peak, impact duration, and strain abrupt change amplitude in the preset low-frequency band; The extraction process of preset low-frequency features includes: The sliding window method is used to divide the vibration signal of the second intermediate fiber or the vibration signal of the first intermediate fiber that is identified as abnormal and only includes geological disasters into multiple first window segments; the length of the first window segment is 200ms; the intra-window RMS energy characteristics of each fiber channel in the preset low frequency band are extracted in each first window segment; based on the intra-window RMS energy characteristics of each fiber channel in the preset low frequency band in each first window segment, the channel RMS energy peak value of each fiber channel in the preset low frequency band is obtained; based on the channel RMS energy peak value of each fiber channel in the preset low frequency band and the preset low frequency energy threshold, the impact start time, impact end time and impact duration of each fiber channel are obtained.

[0011] Preferably, the preset mid-to-high frequency band characteristics include the channel RMS energy peak, impact duration, impact point positioning accuracy, and bounce signal characteristics in the preset mid-to-high frequency band; The process of extracting mid-to-high frequency features includes: The sliding window method is used to divide the vibration signal of the second intermediate fiber or the vibration signal of the first intermediate fiber that is identified as abnormal and only includes geological disasters into multiple third window segments. The length of each third window segment is 500ms. The RMS energy characteristics of each fiber channel in the preset mid-to-high frequency band are extracted in each second window segment. Based on the RMS energy characteristics of each fiber channel in the preset mid-to-high frequency band in each second window segment, the channel RMS energy peak value in the preset mid-to-high frequency band is obtained. Based on the channel RMS energy peak value in the preset mid-to-high frequency band and the rockfall impact threshold, the impact start time, impact end time and impact duration of each fiber channel triggering the rockfall disaster are obtained. The impact start time is the moment when the channel RMS energy peak value first exceeds the rockfall impact threshold, and the impact end time is the moment when the channel RMS energy peak value falls back to below the noise threshold for a preset number of consecutive windows and no longer rises. From the third window segment after the RMS energy peak of the channel, multiple consecutive short-time energy pulse sequences with amplitudes lower than the channel RMS energy peak and intervals exceeding a preset time interval are selected. Based on the selected multiple consecutive short-time energy pulse sequences, bouncing signal features are extracted. Among them, bouncing signal features include the number of pulses, the attenuation law of the RMS energy amplitude of each bouncing pulse, the time interval between two adjacent bouncings, the spatial offset distance corresponding to each bouncing, and the duration of the bouncing impact.

[0012] Preferably, the preset full-band characteristics include the channel RMS energy peak, impact duration, and the abrupt change rate and vibration attenuation rate of the natural frequency of the bridge structure in the preset high-frequency band. The vibration signal of the second intermediate fiber or the vibration signal of the first intermediate fiber that is identified as abnormal and only includes geological disasters is divided into multiple fourth window segments using the sliding window method. The length of each fourth window segment is 10ms. Based on the fourth window segments, the intra-window RMS energy characteristics of each fiber channel in the preset mid-to-high frequency band are calculated. Based on the intra-window RMS energy characteristics of each fiber channel in the preset high frequency band, the peak value of the channel RMS energy in the preset high frequency band, as well as the abrupt change rate and vibration attenuation rate of the natural frequency of the bridge structure, are obtained.

[0013] Preferably, the peak RMS energy of the channel, the duration of the impulse, the trigger delay of the adjacent channel, and the integral of the RMS energy are measured across the entire frequency band. The vibration signal of the second intermediate fiber or the vibration signal of the first intermediate fiber that is identified as abnormal and only includes geological disasters is divided into multiple second window segments using the sliding window method. The length of each second window segment is 500ms. Based on the second window segments, the RMS energy value of each fiber channel in the full frequency band is calculated. Based on the RMS energy value of each fiber channel in the full frequency band, the peak RMS energy, impact duration, trigger delay of adjacent channels, and RMS energy integral of each fiber channel in the full frequency band are obtained.

[0014] Secondly, this invention also proposes a DAS-based system for identifying and classifying geological and structural hazards along highways, including: The acquisition module is used to acquire, in real time, the fiber optic vibration signals collected by the DAS system pre-laid in the preset highway section; The processing module sequentially performs filtering and noise reduction, amplitude normalization, timing alignment preprocessing, construction interference suppression, and vehicle interference suppression on the fiber optic vibration signal to obtain the first intermediate fiber optic vibration signal. The vibration signal of the first intermediate optical fiber is subjected to anomalous signal enhancement preprocessing and multi-scale time-frequency feature extraction to obtain multi-scale time-frequency features; a two-stage anomaly detection framework and multi-scale time-frequency features are used to perform two-stage anomaly identification to obtain anomaly identification results; among them, the anomaly identification results are divided into normal and abnormal, and the anomalies include one or more of the following: vehicle accident, geological disaster, structural damage, and other anomaly categories; When the anomaly identification result is abnormal and includes one or more of the following categories in addition to geological disasters: vehicle accidents, structural damage, and other anomalies, the vibration components corresponding to one or more of the following disaster categories are removed from the first optical fiber vibration signal to obtain the second intermediate optical fiber vibration signal. Feature extraction is performed on the vibration signal of the second intermediate fiber or the vibration signal of the first intermediate fiber that is abnormal and only includes geological disasters, to obtain a feature set; the feature set is then compared with the first preset condition, the second preset condition, the third preset condition, and the fourth preset condition, respectively. If the feature set of the optical fiber vibration signal meets the first preset condition, it is determined to be a roadbed collapse disaster; if the feature set of the optical fiber vibration signal meets the second preset condition, it is determined to be a slope landslide disaster; if the feature set of the optical fiber vibration signal meets the third preset condition, it is determined to be a slope landslide disaster; if the feature set of the optical fiber vibration signal meets the fourth preset condition, it is determined to be a bridge structure damage disaster.

[0015] The proposed method and system for identifying highway geological hazards based on distributed optical fiber sensing (DAS) can simultaneously and accurately identify four core types of highway hazards: roadbed collapse, slope landslide, rockfall, and bridge structural damage. It covers more than 90% of the geological and structural hazard types along highways and solves the problem of insufficient ability of existing technologies to identify hidden hazards. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating a highway geological hazard identification method based on distributed optical fiber sensing (DAS) in one embodiment of the present invention. Detailed Implementation

[0017] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0018] Firstly, referring to Figure 1 The present invention proposes a method for identifying highway geological hazards based on distributed optical fiber sensing (DAS), comprising: Real-time acquisition of fiber optic vibration signals collected by a DAS system pre-laid in a predetermined highway section; The fiber vibration signal is sequentially filtered and denoised, amplitude normalized, time-series aligned preprocessing, construction interference suppression, and vehicle interference suppression to distinguish disaster signals from various interference signals, reduce false alarm rate, and obtain the first intermediate fiber vibration signal. The vibration signal of the first intermediate optical fiber is subjected to anomalous signal enhancement preprocessing and multi-scale time-frequency feature extraction to obtain multi-scale time-frequency features; a two-stage anomaly detection framework and multi-scale time-frequency features are used to perform two-stage anomaly identification to obtain anomaly identification results; among them, the anomaly identification results are divided into normal and abnormal, and the anomalies include one or more of the following: vehicle accident, geological disaster, structural damage, and other anomaly categories; When the anomaly identification result is abnormal and includes one or more of the following categories in addition to geological disasters: vehicle accidents, structural damage, and other anomalies, the vibration components corresponding to one or more of the following disaster categories are removed from the first optical fiber vibration signal to obtain the second intermediate optical fiber vibration signal, so as to eliminate the interference caused by the aliasing of multiple types of disaster features. Feature extraction is performed on the vibration signal of the second intermediate fiber or the vibration signal of the first intermediate fiber that is abnormal and only includes geological disasters, to obtain a feature set; The feature set is compared with the first preset condition, the second preset condition, the third preset condition, and the fourth preset condition respectively. If the feature set meets the first preset condition, it is determined to be a roadbed collapse disaster. If the feature set of the fiber optic vibration signal meets the second preset condition, it is determined to be a slope landslide disaster. If the feature set meets the third preset condition, it is determined to be a slope landslide disaster. If the feature set meets the fourth preset condition, it is determined to be a bridge structure damage disaster.

[0019] This invention can simultaneously and accurately identify four core types of highway disasters: roadbed collapse, slope landslide, rockfall, and bridge structural damage. It covers more than 90% of the geological and structural disaster types along highways and solves the problem of insufficient ability of existing technologies to identify hidden disasters.

[0020] The first preset condition is: at least one optical fiber channel has a channel RMS energy peak exceeding a preset low-frequency energy threshold in a preset low-frequency band, and the duration of its impact, i.e. the duration for which the channel RMS energy peak exceeds the preset low-frequency energy threshold, exceeds 0.5 to 2 seconds; or, at least one optical fiber channel exhibits signal abnormality characteristics of optical fiber breakage. The preset low-frequency band is 10 to 500 Hz.

[0021] The second preset condition is: there are several adjacent optical fiber channels along the slope extension direction whose channel RMS energy peak values ​​in the full frequency band successively exceed the preset full frequency band energy threshold, the impact duration is 1~10 seconds, and the time delay of the adjacent channel RMS energy peak values ​​exceeding the preset full frequency band energy threshold is less than the preset trigger delay threshold; wherein, the full frequency band is 10~2000Hz.

[0022] The third preset condition is: there is at least one optical fiber channel whose channel RMS energy peak in the preset mid-to-high frequency band exceeds the preset rockfall impact threshold, and its impact duration is less than 100ms, and there is a bouncing signal characteristic after the channel RMS energy peak; wherein, the preset mid-to-high frequency band is 50~1000Hz.

[0023] The fourth preset condition is: if the sudden change in the peak RMS energy of at least one optical fiber channel in the preset high frequency band is greater than the preset energy change threshold, and the duration of the change (impact duration) is less than 50ms, that is, an instantaneous change or a step-like increase occurs without a continuous gradual change process, and the change rate of the natural frequency of the bridge structure reaches the preset frequency change threshold or the vibration attenuation rate reaches the preset vibration attenuation rate; wherein, the preset high frequency band is 20~2000Hz.

[0024] Therefore, it can be seen that these four types of geological and structural disasters can be completely distinguished by frequency band characteristics, spatiotemporal distribution characteristics, signal evolution laws, and combination of characteristic parameters: roadbed collapse is characterized by low frequency, central diffusion, and gradual + abrupt coupling; slope landslide is characterized by full frequency band, multi-channel continuous and gradual continuous characteristics; rockfall is characterized by mid-to-high frequency, single-point transient impact, and bounce attenuation; and bridge structural damage is characterized by high-frequency instantaneous abrupt change and structural damage specific characteristics of inherent frequency shift.

[0025] In the process of suppressing construction interference, the construction reporting area is obtained, and the vibration signal corresponding to the reporting area is removed from the fiber optic vibration signal.

[0026] During vehicle interference suppression, the spatiotemporal trajectory of vehicles is calculated from the fiber optic vibration signal to obtain the time window and spatial monitoring range corresponding to each vehicle's passage. Based on the vehicle's spatiotemporal trajectory, the spatiotemporal mask region for each vehicle's passage interference is determined. The fiber optic vibration signal generated by vehicle rolling excitation within the spatiotemporal mask region is removed from the preprocessed fiber optic vibration signal to screen out the continuous background vibration signal that is not disturbed by vehicle load during periods when no vehicles pass by. The retained fiber optic vibration signal is then subjected to baseline smoothing and splicing processing.

[0027] In some embodiments, abnormal signal enhancement preprocessing is performed on the vehicle vibration signal, specifically including: A background noise statistical model is established by acquiring environmental noise data collected by a distributed fiber optic sensing system during periods when no vehicles are driving. Based on the background noise statistical model and an adaptive filtering algorithm, background noise is removed from the vehicle vibration signal. Wavelet packet transform is used to decompose the vehicle vibration signal into multiple scales, resulting in a set of sub-band signals and a set of sub-band coefficients, which include multiple sub-band signals of different frequency bands and scales. From all sub-band signals in the sub-band signal set, frequency bands containing abnormal event characteristics are selected and combined to form an abnormal event signal set; the coefficients in the sub-band coefficient set corresponding to the abnormal event signal set are amplified, and the sub-band coefficient set is used to perform inverse transform reconstruction on the sub-band signal set.

[0028] This embodiment effectively improves the detection capability of weak abnormal signals through abnormal signal enhancement preprocessing, enabling the two-stage anomaly detection framework to accurately identify the type of abnormal event based on the extracted multi-scale time-frequency features, thereby reducing the false alarm rate.

[0029] The feature set in this embodiment includes the frequency band characteristics, spatiotemporal distribution characteristics, signal evolution laws, and characteristic parameters of each optical fiber channel. Among them, the frequency band characteristics include: preset low-frequency band characteristics, preset mid-to-high frequency band characteristics, preset high-frequency band characteristics, and full-band characteristics.

[0030] The preset low-frequency band characteristics in this embodiment include the channel RMS energy peak, impact duration, and strain abrupt change amplitude in the preset low-frequency band.

[0031] The extraction process of preset low-frequency features includes: The sliding window method is used to divide the vibration signal of the second intermediate fiber or the vibration signal of the first intermediate fiber that is identified as abnormal and only includes geological disasters into multiple first window segments; the length of the first window segment is 200ms; the intra-window RMS energy characteristics of each fiber channel in the preset low frequency band are extracted in each first window segment; based on the intra-window RMS energy characteristics of each fiber channel in the preset low frequency band in each first window segment, the channel RMS energy peak value of each fiber channel in the preset low frequency band is obtained; based on the channel RMS energy peak value of each fiber channel in the preset low frequency band and the preset low frequency energy threshold, the impact start time, impact end time and impact duration of each fiber channel are obtained.

[0032] The preset mid-to-high frequency band features in this embodiment include the channel RMS energy peak value, impact duration, impact point positioning accuracy, and bounce signal characteristics within the preset mid-to-high frequency band. The extraction process of the preset mid-to-high frequency band features includes: The sliding window method is used to divide the vibration signal of the second intermediate fiber or the vibration signal of the first intermediate fiber that is identified as abnormal and only includes geological disasters into multiple third window segments; the length of each third window segment is 500ms. The intra-window RMS energy characteristics of each fiber channel in the preset mid-high frequency band are extracted in each second window segment. Based on the intra-window RMS energy characteristics of each fiber channel in the preset mid-high frequency band in each second window segment, the channel RMS energy peak value of each fiber channel in the preset mid-high frequency band is obtained. Based on the channel RMS energy peak value of each fiber channel in the preset mid-high frequency band and the rockfall impact threshold, the impact start time, impact end time and impact duration of each fiber channel triggering the rockfall disaster are obtained. From the third window segment after the RMS energy peak of the channel, multiple consecutive short-time energy pulse sequences with amplitudes lower than the channel RMS energy peak and intervals exceeding a preset time interval are selected. Based on the selected multiple consecutive short-time energy pulse sequences, bouncing signal features are extracted. Among them, bouncing signal features include the number of pulses, the attenuation law of the RMS energy amplitude of each bouncing pulse, the time interval between two adjacent bouncings, the spatial offset distance corresponding to each bouncing, and the duration of the bouncing impact.

[0033] The preset high-frequency band characteristics in this embodiment include the channel RMS energy peak value, impact duration, and the abrupt change rate and vibration attenuation rate of the bridge structure's natural frequency within the preset high-frequency band. Specifically, a sliding window method is used to divide the vibration signal of the second intermediate fiber or the vibration signal of the first intermediate fiber that is identified as abnormal and only includes geological disaster-related anomalies into multiple fourth window segments; the length of each fourth window segment is 10ms. Based on the fourth window segments, the RMS energy value of each fiber channel in the preset mid-to-high frequency band is calculated. Based on the RMS energy value of each fiber channel in the preset high-frequency band, the RMS energy peak value of each fiber channel in the preset high-frequency band, as well as the abrupt change rate and vibration attenuation rate of the bridge structure's natural frequency, are obtained.

[0034] It should be understood that, in this embodiment, the impact start time is the moment when the channel RMS energy peak first exceeds the rockfall impact threshold, and the impact end time is the moment when the channel RMS energy peak falls back to below the noise threshold for a preset number of consecutive windows and no longer rises; the impact duration is the time difference between the impact end time and the impact start time.

[0035] The preset full-band features in this embodiment include the channel RMS energy peak value, impact duration, and adjacent channel trigger delay across the entire frequency band. Specifically, a sliding window method is used to divide the second intermediate fiber vibration signal or the first intermediate fiber vibration signal, which is identified as abnormal and includes only geological disaster-related anomalies, into multiple second window segments; the length of each second window segment is 500ms. Based on the second window segments, the channel RMS energy peak value of each fiber channel across the entire frequency band is calculated. Based on the RMS energy value of each fiber channel across the entire frequency band, the RMS energy peak value, impact duration, and adjacent channel trigger delay for each fiber channel across the entire frequency band are obtained. The adjacent channel trigger delay represents the time difference between the impact start time when the channel RMS energy peak value of an adjacent fiber channel exceeds the preset full-band energy threshold.

[0036] In response to the different vibration characteristics of roadbed collapse, slope landslide, rockfall, and bridge structural damage, this embodiment designs a special window length and frequency band characteristic for RMS energy, which effectively captures the weak to strong vibration signals generated when disasters occur, and solves the problem of insufficient ability of existing technologies to identify hidden disasters.

[0037] In some embodiments, multi-scale time-frequency feature extraction includes time-domain transient feature extraction, frequency-domain multi-scale feature extraction, and time-frequency domain joint feature extraction; In the process of extracting transient features in the time domain, the rise time, fall time, and duration of the fiber vibration signal are extracted, as are the peak energy, energy change rate, and number of pulses. The kurtosis, skewness, and waveform factor of the fiber vibration signal are also extracted. In the process of multi-scale feature extraction in the frequency domain, the fiber vibration signal is decomposed into sub-signals of different frequency bands by multi-scale wavelet packet decomposition; the energy proportion and energy entropy of each frequency band sub-signal are calculated, and the peak frequency, center frequency and bandwidth of each frequency band are extracted. In the joint time-frequency domain feature extraction process, Hilbert-Huang transform is used to perform time-frequency analysis on the fiber vibration signal to extract the variation law of instantaneous frequency and instantaneous amplitude; based on the energy proportion of the frequency band sub-signals of each frequency band, a time-frequency energy distribution matrix is ​​constructed to extract time-frequency feature vectors.

[0038] In some embodiments, the two-stage anomaly detection framework includes: an improved kernel density estimation anomaly detection model and a sparse adaptive convolutional neural network classifier; the improved kernel density estimation anomaly detection model is used to perform unsupervised anomaly detection on multi-scale time-frequency features to obtain preliminary anomaly detection results; when the preliminary anomaly detection result is normal, normal is output as anomaly identification result; the sparse adaptive convolutional neural network classifier is used to classify the multi-scale time-frequency features into anomaly events when the preliminary anomaly detection result is anomaly, to obtain anomaly event classification results, and the anomaly event classification results are output as anomaly identification results.

[0039] Secondly, this invention also proposes a DAS-based system for identifying and classifying geological and structural hazards along highways, including: The acquisition module is used to acquire, in real time, the fiber optic vibration signals collected by the DAS system pre-laid in the preset highway section; The processing module sequentially performs filtering and noise reduction, amplitude normalization, timing alignment preprocessing, construction interference suppression, and vehicle interference suppression on the fiber optic vibration signal to obtain the first intermediate fiber optic vibration signal. The vibration signal of the first intermediate optical fiber is subjected to anomalous signal enhancement preprocessing and multi-scale time-frequency feature extraction to obtain multi-scale time-frequency features; a two-stage anomaly detection framework and multi-scale time-frequency features are used to perform two-stage anomaly identification to obtain anomaly identification results; among them, the anomaly identification results are divided into normal and abnormal, and the anomalies include one or more of the following: vehicle accident, geological disaster, structural damage, and other anomaly categories; When the anomaly identification result is abnormal and includes one or more of the following categories in addition to geological disasters: vehicle accidents, structural damage, and other anomalies, the vibration components corresponding to one or more of the following disaster categories are removed from the first optical fiber vibration signal to obtain the second intermediate optical fiber vibration signal. Feature extraction is performed on the vibration signal of the second intermediate fiber or the vibration signal of the first intermediate fiber that is abnormal and only includes geological disasters, to obtain a feature set; the feature set is then compared with the first preset condition, the second preset condition, the third preset condition, and the fourth preset condition, respectively. If the feature set of the optical fiber vibration signal meets the first preset condition, it is determined to be a roadbed collapse disaster; if the feature set of the optical fiber vibration signal meets the second preset condition, it is determined to be a slope landslide disaster; if the feature set of the optical fiber vibration signal meets the third preset condition, it is determined to be a slope landslide disaster; if the feature set of the optical fiber vibration signal meets the fourth preset condition, it is determined to be a bridge structure damage disaster.

[0040] The first, second, third, and fourth preset conditions in this embodiment are as described in the first aspect.

[0041] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for identifying highway geological hazards based on distributed optical fiber sensing (DAS), characterized in that, include: Real-time acquisition of fiber optic vibration signals collected by a DAS system pre-laid in a predetermined highway section; The fiber vibration signal is sequentially filtered for noise reduction, amplitude normalization, time alignment preprocessing, construction interference suppression, and vehicle interference suppression to obtain the first intermediate fiber vibration signal. The vibration signal of the first intermediate fiber is subjected to anomalous signal enhancement preprocessing and multi-scale time-frequency feature extraction to obtain multi-scale time-frequency features. A two-stage anomaly detection framework and multi-scale time-frequency features are used to perform two-stage anomaly identification to obtain anomaly identification results. The anomaly identification results are divided into normal and abnormal, and the anomalies include one or more of the following categories: vehicle accidents, geological disasters, structural damage, and other anomalies. When the anomaly identification result is abnormal and includes one or more of the following categories in addition to geological disasters: vehicle accidents, structural damage, and other anomalies, the vibration components corresponding to one or more of the following disaster categories are removed from the first optical fiber vibration signal to obtain the second intermediate optical fiber vibration signal. Feature extraction is performed on the vibration signal of the second intermediate fiber or the vibration signal of the first intermediate fiber that is abnormal and only includes geological disasters to obtain a feature set; if the feature set meets the first preset condition, it is determined to be a roadbed collapse disaster; if the feature set meets the second preset condition, it is determined to be a slope landslide disaster. If the feature set meets the third preset condition, it is determined to be a rockfall disaster; if the feature set meets the fourth preset condition, it is determined to be a bridge structure damage disaster.

2. The method for identifying highway geological hazards based on distributed optical fiber sensing (DAS) according to claim 1, characterized in that, The first preset condition is: at least one optical fiber channel has a channel RMS energy peak exceeding the preset low-frequency energy threshold in the preset low-frequency band and its impact duration is 0.5 to 2 seconds, or at least one optical fiber channel exhibits signal abnormality characteristics of optical fiber breakage; wherein, the preset low-frequency band is 10 to 500 Hz. The second preset condition is: there are several adjacent optical fiber channels along the slope extension direction whose channel RMS energy peak values ​​in the full frequency band successively exceed the preset full frequency band energy threshold, the impact duration is 1~10 seconds, and the time delay of the adjacent channel RMS energy peak values ​​exceeding the preset full frequency band energy threshold is less than the preset trigger delay threshold; wherein, the full frequency band is 10~2000Hz. The third preset condition is: there is at least one optical fiber channel whose channel RMS energy peak in the preset mid-to-high frequency band exceeds the preset rockfall impact threshold and whose impact duration is less than 100ms, and there is a bouncing signal characteristic after the channel RMS energy peak; wherein, the preset mid-to-high frequency band is 50~1000Hz. The fourth preset condition is: if the change in the peak RMS energy of at least one optical fiber channel in the preset high frequency band is greater than the preset energy change threshold, and the impact duration is less than 50ms, and the change rate of the natural frequency of the bridge structure reaches the preset frequency change threshold or the vibration attenuation rate reaches the preset vibration attenuation rate; wherein, the preset high frequency band is 20~2000Hz.

3. The method for identifying highway geological hazards based on distributed optical fiber sensing (DAS) according to claim 1, characterized in that, In the construction interference suppression, the construction reporting area is obtained, and the vibration signal corresponding to the reporting area is removed from the fiber optic vibration signal. In the vehicle interference suppression process, the spatiotemporal trajectory of the vehicle is calculated from the fiber optic vibration signal. Based on the vehicle spatiotemporal trajectory, the time window and spatial monitoring range corresponding to each vehicle passage are obtained. Based on the vehicle spatiotemporal trajectory, the spatiotemporal mask area of ​​each vehicle passage interference is determined. The fiber optic vibration signal generated by vehicle rolling excitation within the spatiotemporal mask area is removed from the preprocessed fiber optic vibration signal, and the retained fiber optic vibration signal is subjected to baseline smoothing stitching.

4. The method for identifying highway geological hazards based on distributed optical fiber sensing (DAS) according to claim 1, characterized in that, The abnormal signal enhancement preprocessing of vehicle vibration signals includes: A background noise statistical model is established by acquiring environmental noise data collected by a distributed fiber optic sensing system during periods when no vehicles are driving. Based on the background noise statistical model and an adaptive filtering algorithm, background noise is removed from the vehicle vibration signal. Wavelet packet transform is used to decompose the vehicle vibration signal into multiple scales, resulting in a sub-band signal set and a sub-band coefficient set, which include multiple sub-band signals of different frequency bands and scales. From all the sub-band signals in the sub-band signal set, frequency bands containing abnormal event characteristics are selected and combined to form an abnormal event signal set. The coefficients in the sub-band coefficient set that correspond to the abnormal event signal set are amplified, and the sub-band signal set is reconstructed using the sub-band coefficient set through inverse transform.

5. The method for identifying highway geological hazards based on distributed optical fiber sensing (DAS) according to claim 1, characterized in that, The feature set includes the frequency band characteristics and spatiotemporal distribution characteristics of each fiber channel; Frequency band characteristics include: preset low-frequency band characteristics, preset mid-to-high frequency band characteristics, preset high-frequency band characteristics, and full-frequency band characteristics.

6. The method for identifying highway geological hazards based on distributed optical fiber sensing (DAS) according to claim 5, characterized in that, The preset low-frequency band characteristics include the channel RMS energy peak, impact duration, and abrupt change amplitude in the preset low-frequency band; The extraction process of preset low-frequency features includes: The sliding window method is used to divide the vibration signal of the second intermediate fiber or the vibration signal of the first intermediate fiber that is identified as abnormal and only includes geological disasters into multiple first window segments; the length of the first window segment is 200ms; the intra-window RMS energy characteristics of each fiber channel in the preset low frequency band are extracted in each first window segment; based on the intra-window RMS energy characteristics of each fiber channel in the preset low frequency band in each first window segment, the channel RMS energy peak value of each fiber channel in the preset low frequency band is obtained; based on the channel RMS energy peak value of each fiber channel in the preset low frequency band and the preset low frequency energy threshold, the impact start time, impact end time and impact duration of each fiber channel are obtained.

7. The method for identifying highway geological hazards based on distributed optical fiber sensing (DAS) according to claim 5, characterized in that, The preset mid-to-high frequency band characteristics include the channel RMS energy peak, impact duration, impact point positioning accuracy, and bounce signal characteristics in the preset mid-to-high frequency band. The process of extracting mid-to-high frequency features includes: The sliding window method is used to divide the vibration signal of the second intermediate fiber or the vibration signal of the first intermediate fiber that is identified as abnormal and only includes geological disasters into multiple third window segments. The length of each third window segment is 500ms. The intra-window RMS energy characteristics of each fiber channel in the preset mid-high frequency band are extracted in each third window segment. Based on the intra-window RMS energy characteristics of each fiber channel in the preset mid-high frequency band, the channel RMS energy peak value of each fiber channel in the preset mid-high frequency band is obtained. Based on the channel RMS energy peak value of each fiber channel in the preset mid-high frequency band and the rockfall impact threshold, the impact start time, impact end time and impact duration of each fiber channel triggering the rockfall disaster are obtained. The impact start time is the moment when the channel RMS energy peak value first exceeds the rockfall impact threshold, and the impact end time is the moment when the channel RMS energy peak value falls back to below the noise threshold for a preset number of consecutive windows and no longer rises. From the third window segment after the RMS energy peak of the channel, multiple consecutive short-time energy pulse sequences with amplitudes lower than the channel RMS energy peak and intervals exceeding a preset time interval are selected. Based on the selected multiple consecutive short-time energy pulse sequences, bouncing signal features are extracted. Among them, bouncing signal features include the number of pulses, the attenuation law of the RMS energy amplitude of each bouncing pulse, the time interval between two adjacent bouncings, the spatial offset distance corresponding to each bouncing, and the duration of the bouncing impact.

8. The method for identifying highway geological hazards based on distributed optical fiber sensing (DAS) according to claim 7, characterized in that, The preset high-frequency band characteristics include the channel RMS energy peak, impact duration, and the abrupt change rate and vibration attenuation rate of the bridge structure's natural frequency in the preset high-frequency band. The sliding window method is used to divide the vibration signal of the second intermediate fiber or the vibration signal of the first intermediate fiber that is identified as abnormal and only includes geological disasters into multiple fourth window segments; the length of the fourth window segment is 10ms; based on the fourth window segments, the energy characteristics of each fiber channel in the window RMS frame in the preset mid-high frequency band are calculated; based on the energy characteristics of each fiber channel in the window RMS frame in the preset high frequency band, the peak value of channel RMS energy, impact duration, the abrupt change rate of the natural frequency of the bridge structure, and the vibration attenuation rate of each fiber channel in the preset high frequency band are obtained.

9. The method for identifying highway geological hazards based on distributed optical fiber sensing (DAS) according to claim 7, characterized in that, The preset full-band characteristics include the channel RMS energy peak, impact duration, and adjacent channel trigger delay in the preset full-band. The vibration signal of the second intermediate fiber or the vibration signal of the first intermediate fiber that is identified as abnormal and only includes geological disasters is divided into multiple second window segments using the sliding window method. The length of each second window segment is 500ms. Based on the second window segments, the RMS energy value of each fiber channel in the full frequency band is calculated. Based on the RMS energy value of each fiber channel in the full frequency band, the peak RMS energy, impact duration, and trigger delay of adjacent channels for each fiber channel in the full frequency band are obtained.

10. A DAS-based system for identifying and classifying geological and structural hazards along highways, characterized in that, include: The acquisition module is used to acquire, in real time, the fiber optic vibration signals collected by the DAS system pre-laid in the preset highway section; The processing module sequentially performs filtering and noise reduction, amplitude normalization, timing alignment preprocessing, construction interference suppression, and vehicle interference suppression on the fiber optic vibration signal to obtain the first intermediate fiber optic vibration signal. The vibration signal of the first intermediate fiber is subjected to anomalous signal enhancement preprocessing and multi-scale time-frequency feature extraction to obtain multi-scale time-frequency features. A two-stage anomaly detection framework and multi-scale time-frequency features are used to perform two-stage anomaly identification to obtain anomaly identification results. The anomaly identification results are divided into normal and abnormal, and the anomalies include one or more of the following categories: vehicle accidents, geological disasters, structural damage, and other anomalies. When the anomaly identification result is abnormal and includes one or more of the following categories in addition to geological disasters: vehicle accidents, structural damage, and other anomalies, the vibration components corresponding to one or more of the following disaster categories are removed from the first optical fiber vibration signal to obtain the second intermediate optical fiber vibration signal. Feature extraction is performed on the vibration signal of the second intermediate fiber or the vibration signal of the first intermediate fiber that is abnormal and only includes geological disasters to obtain a feature set; if the feature set meets the first preset condition, it is determined to be a roadbed collapse disaster; if the feature set meets the second preset condition, it is determined to be a slope landslide disaster. If the feature set meets the third preset condition, it is determined to be a slope landslide disaster; if the feature set meets the fourth preset condition, it is determined to be a bridge structure damage disaster.

Images