Wind power submarine cable scouring monitoring device and method based on distributed vibration sensing

Abstract

The invention discloses a wind power submarine cable scouring monitoring device and method based on distributed vibration sensing. The monitoring device comprises a distributed vibration demodulator, a submarine cable built-in dark optical fiber, a data preprocessing unit, a modal / burial depth inversion unit, a source and intensity recognition unit and an alarm and visualization unit. The monitoring method comprises the following steps: continuously collecting vibration signals along the line under the conditions of not transforming a submarine cable structure and not interrupting operation; the signal-to-noise ratio is improved through processing such as coherent superposition and band-pass filtering; quantitatively inverting the local suspension length of the submarine cable and the thickness change of the covering layer under the string vibration theory and the soil-cable coupling constraint; and a double-branch multi-task neural network is utilized to identify and quantify scouring sources (wave current leading, vortex excitation, operation disturbance and the like) and intensity, and a line source and intensity integrated graph is output. According to the method, on-line identification of the scouring state is realized, and the submarine cable burial depth change and the normalized intensity score are reflected quantitatively.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of ocean engineering and power transmission state monitoring, and particularly relates to a wind power submarine cable scour monitoring device and method based on distributed acoustic sensing (DAS). BACKGROUND

[0002] With the rapid development of offshore wind power, the collection line and outgoing line in the offshore wind farm rely heavily on submarine cables as the main power transmission channel. Compared with conventional submarine cables for cross-sea power transmission or communication purposes, the environment of wind power submarine cables is more complex, and the surrounding ocean dynamic conditions are significantly affected by various factors. The pile foundation and jacket of the wind turbine generator change the local hydrodynamic distribution, which is easy to form flow around, vortex excitation and local scour around the cable; the cable also needs to cross the nearshore shallow water, slope break and continental shelf, and the flow field and sediment under different water depth and terrain conditions further aggravate the unevenness and unpredictability of scour. Under these actions, wind power submarine cables are more likely to have problems such as erosion of the covering layer, local exposure and even suspension, which greatly increases the risk of structural fatigue and failure.

[0003] The existing monitoring methods mainly rely on sonar and ROV inspection, local sensor arrangement and numerical prediction model. Sonar and ROV can provide seabed topography and cable appearance information, but the inspection cycle is long and the cost is high, which is difficult to meet the demand of continuous monitoring of wind farms; local sensors can provide information at key positions, but the coverage is insufficient, and it is difficult to reflect the overall scour state of long-distance submarine cables, and the long-term reliability is limited; numerical simulation and empirical formula can assist in predicting the scour trend, but the real-time performance and accuracy are difficult to guarantee.

[0004] The emergence of distributed optical fiber sensing technology provides a new idea to solve this problem. Distributed vibration sensing can convert ordinary communication optical fiber into a distributed vibration array, with advantages such as long distance, continuous coverage and maintenance-free. The spare dark optical fiber commonly reserved in wind power submarine cables provides a natural sensing medium for distributed vibration sensing, making it possible to achieve online monitoring without modifying the cable structure or laying additional electronic equipment. However, there are still many challenges in directly applying distributed vibration sensing for wind power submarine cable scour monitoring: the low signal-to-noise ratio caused by complex sea conditions and instrument noise, the complex coupling between cable tension changes and burial conditions, and the difficulty of traditional data processing to stably separate the changes in the suspended section and the covering layer thickness; at the same time, existing researches mostly stay at the level of anomaly detection, lack of quantitative inversion ability based on physical models, and rely on artificial experience to distinguish the source and intensity of scour, which is difficult to realize automatic alarm and risk map generation.

[0005] Therefore, there is an urgent need for an innovative method that takes into account the complex environmental characteristics of wind power submarine cables, fully utilizes existing dark optical fiber resources, combines distributed vibration sensing and physical-intelligent fusion algorithms, and realizes real-time, quantitative, and intelligent monitoring of submarine cable scouring states to effectively improve the safety and economy of wind farm operation and maintenance. SUMMARY

[0006] The present application is a wind power submarine cable scouring monitoring device and method based on distributed vibration sensing, which uses the reserved dark optical fiber in the wind power submarine cable as a sensing medium to continuously collect vibration signals along the line without modifying the submarine cable structure or interrupting operation. Through coherent superposition and band-pass filtering processing methods, the background noise of sea conditions is suppressed and the signal-to-noise ratio is improved. Under the constraints of string vibration theory and soil-cable coupling dynamics, the quantitative inversion of the length of the local suspended section and the thickness change of the covering layer of the submarine cable is realized. Further combining neural network algorithm to output scouring sources (including wave flow dominance, vortex excitation, and human disturbance) and scouring intensity scores, the "source-intensity integrated map along the line" is finally generated. The present application realizes online identification of scouring states while quantitatively reflecting the change of submarine cable burial depth and scouring intensity scores, and is suitable for long-distance, continuous online scouring monitoring of wind power submarine cables, providing a reliable basis for the operation safety and maintenance management of wind power submarine cables.

[0007] Technical solution: To achieve the above-mentioned purpose, the wind power submarine cable scouring monitoring device based on distributed vibration sensing includes a distributed vibration demodulator, a submarine cable built-in dark optical fiber, a data preprocessing unit, a modal / burial depth inversion unit, a source and intensity identification unit, and an alarm and visualization unit.

[0008] The dark optical fiber as a sensing medium is connected with the distributed vibration demodulator to continuously acquire vibration signals along the line and transmit them to the data preprocessing unit without modifying the submarine cable structure or interrupting operation. The data preprocessing unit performs coherent superposition and band-pass filtering on the collected signals to form a stable target frequency band response. The modal / burial depth inversion unit calculates the length of the local suspended section and the thickness change of the covering layer of the submarine cable based on string vibration and soil-cable coupling constraints. The source and intensity identification unit uses a neural network model to output scouring sources and intensity scores based on the above-mentioned physical characteristics and sea condition priors. The alarm and visualization unit outputs the source category along the line and outputs the intensity risk level according to the set threshold, which is used for online identification of wind power submarine cable scouring states and operation and maintenance decision support. Among them, the scouring sources include wave flow dominance, vortex excitation, and human disturbance.

[0009] The wind power submarine cable scouring monitoring method based on distributed vibration sensing includes the following steps:

[0010] Step (1), connect the single-mode dark optical fiber in the wind power submarine cable using a distributed vibration demodulator, and select a fiber segment that is buried stably and not affected by scour as a reference segment as a benchmark for subsequent monitoring.

[0011] Step (2), turn on the distributed vibration demodulator to collect distributed vibration signals along the entire wind power submarine cable, and set the sampling frequency and channel spacing according to the length of the submarine cable and the monitoring scenario.

[0012] Step (3), the data preprocessing unit divides the received time series vibration data into blocks according to time windows, divides the time series of each time window into several sub-sections, and calculates the normalized cross-correlation coefficient of each sub-section with the template under discrete delay using the first sub-section as the template. When the cross-correlation coefficient is greater than a set value, it is a consistent sub-section; after time alignment of the consistent sub-section according to the maximum correlation lag, coherent superposition is performed to generate enhanced time series data, and band-pass filtering is performed in a fixed frequency band to output band-pass enhanced time series data. The band-pass enhanced time series data is subjected to short-time Fourier transform according to the same time window to output a time-frequency matrix.

[0013] Step (4), the band-pass enhanced time series data output by the data preprocessing unit is transmitted to the modal / burial inversion unit, and the first-order main peak frequency and the spectral peak amplitude of each channel are extracted in the modal / burial unit.

[0014] Step (5), the modal / burial inversion unit further processes the candidate suspended section obtained in step (4) to extract the first-order main peak frequency of the candidate suspended section, and calculates the span length using the string vibration relationship, thereby determining the suspended position and suspended length of the real suspended section.

[0015] Step (6), after determining the position and length of the real suspended section, the modal / burial inversion unit performs feature analysis on the band-pass enhanced data according to the soil-cable coupling constraint, calculates the quality factor by extracting the first-order main peak frequency and half-power bandwidth of each channel, and calculates the overburden thickness change of each channel of the optical cable with the frequency domain attenuation rate index .

[0016] Step (7), the time-frequency matrix and the band-pass enhanced time series data output by the data preprocessing unit are used as dual-input sources for the intensity recognition unit. The intensity recognition unit adopts a time-frequency branch and time-domain branch dual-branch multi-task neural network composed of a data interpretation layer, a branch fusion layer and a task output layer to generate a probability vector of the scour source category and a normalized intensity score .

[0017] Step (8) inputs the probability vector of the source category and the normalized intensity score output by the source and intensity identification unit into the alarm and visualization unit to generate an integrated map of the source and intensity along the line.

[0018] In step (1), after connecting the dark fiber, the integrity of the fiber optic link is checked to ensure signal coverage throughout the entire line. The spatial location of each channel is determined by the relationship between the channel number output by the demodulator and the fiber length. At the same time, a stable reference segment is buried as a benchmark. The reference segment is set to provide a long-term stable reference signal, whose spectral characteristics reflect the state under normal burial conditions, thus serving as a comparison standard in subsequent spectral analysis.

[0019] In step (2), the sampling frequency affects the temporal resolution of the signal, while the channel spacing determines the spatial resolution. By combining the length of the submarine cable and the on-site environment to configure the sampling frequency and channel spacing, it is ensured that the data not only covers the target signal frequency band but also has sufficient spatial resolution and stability, providing a basis for subsequent scour feature identification.

[0020] Step (3) involves the data preprocessing unit dividing the time window signal into several segments, with a sampling rate of... The time window length is And divide the time window signal into Sub-segment Use the first sub-parameter as a reference template. Calculate the normalized cross-correlation coefficient:

[0021]

[0022] in, This indicates that within the time window, data is processed according to segment length. (seconds) and sampling rate The first segment obtained Discrete-time signals of each segment ; For reference template sub-segment (take the first sub-segment of this time window, and...) (Sampling rate is consistent with length); This refers to the discrete time delay in cross-correlation calculations; In order to delay Down and The normalized cross-correlation coefficient, according to the above formula, is... Summation, range of values The closer the value is to 1, the higher the similarity between the two sequences at that delay.

[0023] Take the maximum correlation coefficient and the corresponding optimal delay:

[0024] ,

[0025] in, For the first Sub-segment With reference template The maximum normalized cross-correlation coefficient; The optimal discrete time delay is used to obtain the maximum correlation coefficient. In the formula... Indicates to Find the maximum value. Indicates to make To obtain the maximum value .

[0026] Based on empirical thresholds After filtering, a set of highly consistent segments is obtained. :

[0027]

[0028] For sets inner sub-segment Perform time alignment, and denote the aligned sequence as... The data are then weighted by correlation coefficients and coherently superimposed to obtain the enhanced time-series data for this time window:

[0029]

[0030] in, This indicates that the enhanced time series data after alignment and coherent superposition within this time window is at the [number]th [timeframe]. The values ​​of each sampling point; For the selected Each segment is determined by its optimal delay. Discrete sequences after time alignment.

[0031] Calculate the correlation between the sub-segment and the reference template to obtain the maximum correlation coefficient. This allows for the selection of a set of highly correlated segments. Subsequently, these segments are coherently superimposed.

[0032]

[0033] Obtain enhanced time series data Then in a fixed frequency band Bandpass filtering is performed inside to obtain the bandpass enhanced sequence. The bandpass enhancement sequences of each time window are concatenated in chronological order to obtain bandpass enhancement time series data. At the same time, a short-time Fourier transform is performed on the bandpass enhancement time series data according to the same time window to output the time-frequency matrix.

[0034] In step (4), the mode / depth inversion unit is based on the reference segment parameters in step (1). Constructing a suspended index function :

[0035]

[0036] in, , For experience weights. The average first-order main peak frequency of the reference segment under the buried state. This represents the average amplitude of the first-order main peak in the reference section under buried condition. Set threshold range When at least three adjacent channels within the same time window At that time, adjacent channels are marked as candidate suspended segments. It also outputs the corresponding channel status label.

[0037] In step (5), the span length of the candidate suspended segment is obtained according to the string vibration model, and the nth natural frequency is:

[0038]

[0039] in, For the first First natural frequency, For the span, For equivalent tension, The equivalent mass per unit length. Take the first-order frequency. As the dominant frequency Thus, the span length is obtained:

[0040]

[0041] Where, constant This is the frequency span ratio coefficient, obtained through on-site calibration. The calculated length of the suspended segment; Let be the first-order dominant peak frequency of the overall channel data within the candidate suspended segment. Let the center position of the suspended segment be . Then the boundary of the actual suspended segment is: .

[0042] in, Indicates the starting position of the candidate suspended segment. Indicates the termination position of the candidate suspended segment; This indicates the starting position of the actual suspended segment. This is the end position of the actual suspended segment.

[0043] The start and end boundaries of the actual suspended segment are determined by its suspended position, thus obtaining the suspended position and length of the actual suspended segment. To ensure the stability of the engineering results, if the calculated results exceed the original candidate area... If the first-order main peak frequencies on both sides of the candidate region differ significantly, the segment with the most prominent frequency change will be retained as the final suspended position.

[0044] In step (6), the vibration characteristics of the suspended section are not only related to the span length, but also affected by the thickness of the overburden layer. As the overburden layer thickens, energy dissipation increases and the quality factor decreases; when the overburden layer thins, the system attenuation weakens, which is manifested as a slowdown in energy attenuation.

[0045] Therefore, the first-order spectral peak of the main peak is extracted first. and read half-power bandwidth Calculation channel Quality factor within the time window :

[0046] in A larger value indicates smaller damping, and a stronger "sharpness" in the system; conversely, a smaller value indicates a weaker damping. The smaller the value, the faster the energy decays. Furthermore, to improve sensitivity to changes in the cladding layer, a frequency-domain attenuation rate index is introduced. :

[0047]

[0048] in, For channel In frequency Power spectral density at that point For channel The first-order main peak frequency; To surround The frequency band spacing.

[0049] Will and Perform joint calibration to obtain the channel Changes in the thickness of the capping layer :

[0050]

[0051] in These are weighting coefficients, calibrated through actual measurements. The change represents the amount of energy dissipation; This represents the change in high-frequency energy decay characteristics.

[0052] In step (7), the source and intensity identification unit contains a dual-branch network structure, receiving the time-frequency matrix and bandpass enhancement time-series data respectively. The time-frequency branch captures the frequency band energy distribution, peak position, and bandwidth changes, while the time-domain branch represents the amplitude, impulsivity, and duration. After integration by the fusion layer and attention mechanism, the network output includes a probability distribution of a set of source categories (such as wave-current dominance, eddy current, and anthropogenic disturbance), and a normalized intensity score (between 0 and 1), providing direct basis for risk identification and graded alarms.

[0053] In step (7), the time-frequency matrix and the bandpass enhancement time series data are aligned in the same time window: for each time window and each channel, the time-frequency matrix provides the channel × frequency spectrum distribution within the time window, and the bandpass enhancement time series data provides the time domain sequence segment within the same time window.

[0054] In step (7), a dual-branch multi-task neural network is used to fuse the dual inputs and output the results: (7.1) The time-frequency branch takes the time × frequency spectrum distribution of each time window and each channel as input to learn and encode the frequency band energy distribution, main peak position and peak shape frequency domain characteristics; (7.2) The time domain branch takes the bandpass enhanced time domain sequence segment of the same time window as input to learn and encode the amplitude evolution over time, transient impact and attenuation characteristics time domain characteristics.

[0055] In step (7), after the time-frequency branch and the time-domain branch are cascaded in the branch fusion layer to form a shared representation, the output length is... probability vector And calculate the normalized intensity score. , among which, element This indicates the probability that the corresponding source is dominant within the time window.

[0056] In step (8), the alarm and visualization unit maps the intensity scores of each channel and time window to risk intensity, and uses the category with the highest probability as the source label; based on a fixed threshold, the risk intensity is divided into four levels: safe [0,0.5], attention [0.5,0.7], alert [0.7,0.85], and danger [0.85,1], to achieve real-time early warning and assistance.

[0057] Working Principle: This invention uses a single-mode dark optical fiber embedded in the wind power submarine cable as a continuous distributed vibration sensing medium. The distributed vibration sensor demodulator emits coherent pulses and receives Rayleigh scattering echoes. After coherent demodulation and phase difference analysis, minute vibration changes along the cable axis are obtained, forming a distributed vibration signal with high spatiotemporal resolution. In the data preprocessing unit, the signal undergoes coherent superposition and bandpass filtering to obtain bandpass-enhanced time-series data and a time-frequency matrix. Then, the modal / burial depth inversion unit extracts the first-order main peak frequency, amplitude, and their derived features, constructs a suspension index function, and refines the suspension position and span length. Simultaneously, the thickness of the overburden layer and its dynamic evolution are inverted by combining spectral bandwidth and frequency domain attenuation rate indices. In parallel, the time-frequency matrix and bandpass-enhanced time-series data from the data preprocessing results are sent to the source and intensity identification unit. A dual-branch multi-task neural network is used to identify the scour source type and quantify the intensity. Finally, the alarm and visualization unit generates an integrated map of the source and intensity along the cable, thereby realizing online, quantitative, and intelligent monitoring of the scour status of the wind power submarine cable.

[0058] Beneficial effects: Compared with the prior art, the present invention has the following advantages:

[0059] (1) This invention utilizes the built-in dark optical fiber of the wind power submarine cable combined with distributed vibration sensing to achieve continuous distributed online monitoring of the entire line. It has high temporal and spatial resolution, does not require modification of the submarine cable, does not require power outages or interruption of operation, and does not require underwater electronic devices. It breaks through the limitations of traditional sonar and ROV inspections, which are intermittent and have limited points, and significantly reduces operation and maintenance costs and interference risks.

[0060] (2) This invention improves the signal-to-noise ratio by methods such as coherent superposition and bandpass filtering, and quantitatively calculates the span of the suspended section under the constraints of string vibration and soil-cable coupling. With the thickness of the cover layer change.

[0061] (3) The present invention introduces a source and intensity identification unit, which can intelligently distinguish the scour sources such as wave-current dominance, vortex-induced effects and human disturbance, and generate an integrated source and intensity map along the line, adapting to the complex hydrodynamic environment of the pile foundation near field and near shore-slope-continental shelf, supporting the closed-loop risk management and refined operation and maintenance decision-making of wind power submarine cables. Attached Figure Description

[0062] Figure 1 This is a schematic diagram of the wind power submarine cable scour monitoring device based on distributed vibration sensing according to the present invention.

[0063] Figure 2 This is a diagram showing the signal enhancement effect after coherent superposition and bandpass filtering of the data preprocessing unit in this embodiment of the invention.

[0064] Figure 3Time-series and time-frequency distribution diagrams of distributed vibration data for wave events, vortex-induced disturbances, and operational disturbances in the suspended section in this embodiment of the invention;

[0065] Figure 4 These are the full-field coverage layer variation curves and near-suspended section variation curves of the optical cable obtained by the mode / burial depth inversion unit in this embodiment of the invention.

[0066] Figure 5 This is an integrated map of source and intensity along the line output by the source and intensity identification unit in this embodiment of the invention. Detailed Implementation

[0067] like Figure 1 As shown, the wind power submarine cable scour monitoring and identification device based on distributed vibration sensing of the present invention includes a distributed vibration demodulator 1, a submarine cable embedded dark optical fiber 2, a data preprocessing unit 3, a mode / burial depth inversion unit 4, a source and intensity identification unit 5, and an alarm and visualization unit 6. The distributed vibration sensor demodulator 1 is connected to the submarine cable embedded single-mode dark optical fiber 2 and is used to continuously collect vibration signals along the cable during operation. The data preprocessing unit 3 receives the raw signal output by the distributed vibration sensor demodulator 1, performs coherent superposition, bandpass filtering, and time-frequency feature extraction on it, and generates... The system generates bandpass-enhanced time-series data and a time-frequency matrix. The modal / burial depth inversion unit 4 is connected to the data preprocessing unit 3 and is used to extract the first-order main peak frequency and amplitude based on the reference segment spectrum comparison, and to invert the suspended length in combination with the string vibration model, while calculating the change in the thickness of the overburden layer. The source and intensity identification unit 5 is connected to the data preprocessing unit 3 and adopts a dual-branch multi-task neural network. It takes the time-frequency matrix and the bandpass-enhanced time-series data as input, outputs the probability distribution of the scour source category and the normalized intensity score, and generates an integrated map of the source and intensity along the line by the alarm and visualization unit 6.

[0068] The wind power submarine cable scour monitoring method based on distributed vibration sensing of this invention includes the following steps:

[0069] Step (1): Use a distributed vibration demodulator 1 to connect to the single-mode dark fiber 2 in the wind power submarine cable, and select a fiber segment that is buried firmly and is not affected by scouring as a reference segment, which will serve as the benchmark for subsequent monitoring.

[0070] Step (2): Turn on the distributed vibration demodulator 1 to collect distributed vibration signals along the entire wind power submarine cable, forming time-series vibration data divided by channel. Based on the total length of the submarine cable and the overall monitoring scenario (such as typhoon arrival, strong current and wave action, or offshore construction operations), uniformly set the sampling frequency and channel spacing to ensure that the monitoring has the necessary time and spatial resolution, thereby covering and identifying typical scour events.

[0071] Step (3): The data preprocessing unit 3 receives the time-series vibration data divided by channel from the distributed vibration demodulator 1, and processes it in blocks according to a preset time window (e.g., 120 s). The time series within each time window is further divided into several sub-segments (e.g., length 8 s, step size 4 s). Using the first sub-segment within the window as a template, the normalized cross-correlation between each sub-segment and the template is calculated one by one. When the correlation coefficient is greater than 0.7, it is determined to be a consistent sub-segment. The consistent sub-segments are time-aligned according to the maximum correlation lag and then coherently superimposed to obtain the enhanced time-series data of that time window, and bandpass filtering is performed within a fixed frequency band. The bandpass enhancement sequences of each time window are spliced ​​in chronological order to output bandpass enhanced time-series data that maintains the original channel structure and time length. At the same time, the bandpass enhanced time-series data is subjected to short-time Fourier transform (STFT) according to the same time window to output a time-frequency matrix.

[0072] Step (4): The bandpass enhancement timing data output by the data preprocessing unit 3 is input into the mode / depth inversion unit 4, and the first-order main peak frequency of each channel is extracted in the mode / depth inversion unit. and amplitude Construct a suspending index function and mark candidate suspending segments.

[0073] Step (5): Modal / depth inversion unit 4 extracts the first-order main peak frequency of the candidate suspended segment obtained in step (4), and uses string vibration to calculate the span length, thereby determining the suspension position and suspension length of the real suspended segment.

[0074] Step (6): After identifying the suspended position and length, the modal / burial depth inversion unit 4 performs feature analysis on the bandpass enhancement time series data based on the soil-cable coupling constraint conditions; and calculates the quality factor by extracting the main peak frequency and half-power bandwidth of each channel. And combined with frequency domain attenuation index By using the changes in quality factor and frequency domain attenuation rate relative to the reference state, a relationship between the thickness of the cover layer and the change in the thickness of the cover layer above each channel of the optical cable is established, thereby calculating the change in the thickness of the cover layer above each channel of the optical cable.

[0075] Step (7) involves inputting the time-frequency matrix output from data preprocessing unit 3 and the bandpass-enhanced time-series data as dual inputs into source and intensity identification unit 5. The two data streams are strictly aligned within the same time window and channel: the time-frequency matrix provides the "channel × frequency" spectral distribution within the time window, and the bandpass-enhanced time-series data provides the time-domain sequence segments within the same window. Source and intensity identification unit 5 employs a dual-branch multi-task neural network consisting of a data interpretation layer, a branch fusion layer, and a task output layer: the data interpretation layer extracts frequency and time domain features respectively; the branch fusion layer weights and concatenates the two types of features to form a shared representation; and the task output layer uses this representation to achieve parallel task output, generating a probability vector for the scour source category. and normalized intensity score .

[0076] Step (8) inputs the source category and intensity score output by the source and intensity identification unit 5 into the alarm and visualization unit 6 to generate an integrated map of source and intensity along the line.

[0077] In step (1), after connecting the dark optical fiber, an integrity check is performed on the optical fiber link to ensure that the signal can cover the entire line. After confirming the connectivity, the spatial location of each channel is located by using the channel number output by the distributed vibration demodulator 1 and the relationship between the optical fiber length, laying the foundation for the spatial presentation of subsequent scour risks. The reference segment is set to provide a long-term stable reference signal, whose spectral characteristics reflect the state under normal burial conditions, thus serving as a comparison standard in subsequent analysis.

[0078] In step (2), the sampling frequency is determined according to the principle of "covering the target frequency band and leaving sufficient margin": the main wave band is usually low frequency (<2Hz), while the eddy current / operational disturbance frequency can reach 30 Hz. It can simultaneously ensure low-frequency integrity and high-frequency event capture. The channel spacing is 2-5 m according to engineering requirements. In specific monitoring scenarios (such as typhoons or frequent construction disturbances), the sampling frequency can be increased and the channel spacing reduced to balance the overall coverage of long-distance submarine cables and the spatial resolution requirements of local scour areas.

[0079] In step (3), the time-series vibration data collected by the distributed vibration demodulator 1 and divided into channels is received and processed in blocks according to a preset time window within the data preprocessing unit 3. Let the sampling rate be... The time window length is For the timing signals within each time window, calculate based on the segment length. (seconds) and step size Segmentation is performed to obtain Sub-segment .in Number the sub-segments ; Index of discrete-time sampling points ( (Number of sampling points for each sub-segment).

[0080] Use the first segment of this time window as a reference template. Calculate the first... The segment and the template in discrete delay Normalized cross-correlation coefficient at (sampling points):

[0081]

[0082] in, This indicates that within the time window, data is processed according to segment length. (seconds) and sampling rate The first segment obtained Discrete-time signals of each segment ; For reference template segments (usually the first segment of the current time window, and...), (Sampling rate is consistent with length); This refers to the discrete time delay in cross-correlation calculations; In order to delay Down and The normalized cross-correlation coefficient, according to the above formula, is... Summation, range of values The closer the value is to 1, the higher the similarity between the two sequences at that delay.

[0083] Take the maximum correlation coefficient and the corresponding optimal delay:

[0084] ,

[0085] in, For the first Sub-segment With reference template The maximum normalized cross-correlation coefficient; The optimal discrete time delay is used to obtain the maximum correlation coefficient. In the formula... Indicates to Find the maximum value. Indicates to make To obtain the maximum value .

[0086] In this embodiment, based on an empirical threshold After filtering, a set of highly consistent segments is obtained:

[0087]

[0088] For sets inner sub-segment Perform time alignment (denote the aligned sequence as...) The data is then weighted by correlation coefficients and coherently superimposed to obtain the enhanced time series data for that time window.

[0089]

[0090] in, This indicates that the enhanced time series data after alignment and coherent superposition within this time window is at the [number]th [timeframe]. The values ​​of each sampling point; For the selected Each segment is calculated using the optimal delay. Discrete sequences after time alignment.

[0091] If a sampling point is not included in the set If any sub-segment within the window is covered, then the original signal within the window is used as a fallback signal to ensure... Full definition. Then in a fixed frequency band. Internal Apply a bandpass filter and output the bandpass-enhanced sequence for that time window. For all time windows, the results are as follows: The data is spliced ​​in chronological order to output bandpass enhanced timing data that retains the original channel structure and total duration. Then, a short-time Fourier transform is performed on the data to output a time-frequency matrix.

[0092] In step (4), the bandpass enhancement timing data output by the data preprocessing unit 3 is input into the mode / burial depth inversion unit 4, and the first-order main peak frequency of each channel is calculated. and their corresponding spectral peak amplitudes When a submarine cable forms a suspended / free span, the constraints and contact damping of the surrounding soil weaken, the equivalent span length increases, and the boundary conditions tend to be free, resulting in… The relative steady state drifts towards lower frequencies and Significantly raised; when well-buried, both are stable with minimal fluctuations. To achieve objective comparison, the long-term steady-state reference segment selected in step (1) was used to obtain... As a reference, a suspending index function is constructed accordingly. :

[0093]

[0094] in, , For experience weights. The average first-order main peak frequency of the reference section under normal burial conditions. This represents the average amplitude of the first-order main peak in the reference section under normal burial conditions.

[0095] Through the Set threshold range (in Take the reference segment The 95th percentile of the distribution Using the 99th percentile, the channel status is divided into "embedded". ,"transition" "Candidates in limbo" When at least three adjacent channels appear within the same time window At that time, adjacent channels are consecutively marked as candidate suspended segments. It also outputs the corresponding channel status labels, providing a basis for subsequent refined calculations of the suspended length and position.

[0096] That is, in step (4), for Set threshold range ,when At that time, the channel was buried; when When, the channel is in a transitional state; when At that time, the channel is a candidate suspended.

[0097] In step (5), the candidate suspended segment is a string constrained by tension, and its nth natural frequency is... for:

[0098]

[0099] in, For the first First natural frequency, For the span, For equivalent tension, The equivalent mass per unit length. In practical applications, the first-order frequency is used. As the dominant frequency Thus, the span length is obtained:

[0100]

[0101] Where, constant Frequency-span ratio factor (unit: (Obtained through on-site calibration, comprehensively reflecting the equivalent tension) equivalent mass per unit length and factors such as end boundary conditions; To accurately calculate the length of the suspended section; Let be the first-order dominant peak frequency of the overall channel data within the candidate suspended segment. Let the center position of the suspended segment be . Then the boundary of the actual suspended segment is:

[0102]

[0103] in, Indicates the starting position of the candidate suspended segment. Indicates the termination position of the candidate suspended segment; This indicates the starting position of the actual suspended segment obtained after calculation. This indicates the actual termination position of the suspended segment obtained after calculation.

[0104] To ensure the robustness of the engineering results, if the calculation results Beyond the original candidate area Then clip back to the interval. Within the specified range, when there is a significant difference in the decrease in the first-order main peak frequency on both sides of the suspended section, the side with a larger decrease and better continuity is preferentially retained as the final suspended position. Based on the final suspended position, the start and end boundaries of the actual suspended section are determined, thereby obtaining the suspension position of the actual suspended section and its length along the cable.

[0105] In step (6), the wind power submarine cable is laid by trenching and backfilling or spraying, located below the surface sediments of the seabed. In this invention, the sedimentary medium (sand, mud, silt, etc.) covering the cable is collectively referred to as the cover layer, and its thickness refers to the vertical distance from the cable to the seabed surface along the channel position. Except for the suspended section identified in step (5), the remaining sections are all under the cover layer. Affected by wave-tidal action, topographic evolution, and operational disturbances, the thickness of the cover layer will change in time and space. Abnormal thinning or exposure serves as an early indication or warning of potential risks. As the cover layer thickens, the contact coupling between the optical cable and the sedimentary medium is enhanced, energy dissipation increases accordingly, and the quality factor decreases; when the cover layer thins or is partially suspended, the coupling weakens, signal attenuation slows down, and the quality factor increases. The relationship between the cover layer thickness and signal attenuation characteristics constitutes the soil-cable coupling constraint condition in this invention and serves as the physical basis for subsequently deriving the changes in cover layer thickness.

[0106] Extracting the first-order main peak frequency And read the half-power bandwidth of the first-order main peak frequency. Calculate the quality factor:

[0107]

[0108] in, Indicates channel The quality factor calculated within the time window reflects the damping characteristics of the channel signal. A larger value indicates smaller damping and stronger signal energy retention; conversely, a smaller value indicates weaker damping and stronger signal energy retention. The smaller the value, the faster the energy decays. Furthermore, to improve sensitivity to changes in the cladding layer, a frequency-domain attenuation rate index is introduced. :

[0109] in, For channel In frequency Power spectral density at that point For this channel The first-order main peak frequency; To surround The frequency band spacing.

[0110] This represents the quality factor of the current time window and the current channel. The change in the reciprocal of the value relative to the reference segment is used to characterize the change in damping / energy dissipation: an increase in the reciprocal indicates enhanced dissipation and a wider spectral peak; a decrease in the reciprocal indicates weakened dissipation and a sharper spectral peak. Indicates channel Frequency domain attenuation index The change relative to the reference state is used to characterize the change in high-frequency energy decay characteristics: an increase in the value means a decrease in decay, and a decrease in the value means an increase in decay.

[0111] use and The response characteristics to changes in the cover layer thickness were used to derive the channel. Variation in capping layer thickness:

[0112]

[0113] in This is a weighting coefficient used to balance the contributions of the two types of features to thickness variation. This weighting coefficient is determined using samples with externally measured overburden thickness variations.

[0114] In step (7), the time-frequency matrix and the bandpass enhanced time series data output by the data preprocessing unit 3 are sent as dual inputs to the source and intensity identification unit 5. The time-frequency matrix and the bandpass enhanced time series data are strictly aligned in the same time window: for each time window and each channel, the time-frequency matrix provides the "channel × frequency" spectrum distribution in the time window, and the bandpass enhanced time series data provides the time domain sequence segment in the same time window; the time-frequency matrix and the bandpass enhanced time series data are input synchronously according to the same time window index and channel index.

[0115] In the source and intensity identification unit 5, a dual-branch multi-task neural network consisting of a data interpretation layer, a branch fusion layer, and a task output layer is employed. The data interpretation layer includes a time-frequency branch and a time-domain branch: the time-frequency branch takes the "time × frequency" spectral distribution of each time window and each channel as input, learning and encoding frequency domain features such as band energy distribution, peak position, and peak shape; the time-domain branch takes bandpass-enhanced time-domain sequence segments of the same time window as input, automatically learning and encoding time-domain features such as amplitude evolution, transient impact, and attenuation characteristics. The features output from the two branches are weighted and concatenated in the branch fusion layer, forming a shared representation through feature splicing or attention mechanisms to achieve information fusion.

[0116] The task output layer consists of a classification branch and a regression branch: the classification branch takes a shared representation as input and has an output length of [length missing]. probability vector (This embodiment takes) (representing waves, eddy currents, and operational disturbances, respectively), where each element... This indicates the probability that the corresponding source is dominant within that time window; the regression branch jointly calculates the output of the classification branch and the fusion features obtained from the branch fusion layer. By comprehensively regressing the multidimensional information reflecting signal energy, spectral amplitude, and amplitude changes in the fusion features, and combining it with the discrimination in the classification results, a continuous normalized intensity score is calculated. The normalized intensity score reflects factors such as signal energy, main frequency amplitude, and pulse intensity; a higher value indicates a stronger scouring effect. Both inputs are strictly aligned with the same time window index and channel index, and are generated synchronously within the same window and channel. and .

[0117] In step (8), the source probability vector output by the source and intensity identification unit 5 is combined with the normalized intensity score. The alarm and visualization unit 6 receives the data and generates a combined source and intensity map along the cable route, calculated using a single time window. This map uses the channel / cable distance as the horizontal axis and normalized dimensions. Using the vertical axis as the ordinate, plot three source probability curves (probability and intensity curves for wave / tidal current, eddy current, and operational disturbance) on the same coordinate system, at each location. At each location, the largest value on the probability curve is taken as the dominant source label; three horizontal threshold lines are also provided. The vertical axis is divided into four levels: Safety [0, 0.5], Attention [0.5, 0.7], Alert [0.7, 0.85], and Danger [0.85, 1], to achieve real-time early warning and decision support.

[0118] Example

[0119] In a wind farm near the coast of Jiangsu, a 10 km long 220 kV submarine transmission cable was selected as the monitoring target. The cable is laid in a nearshore shallow sea area with an average water depth of about 10 m. The seabed is mainly composed of medium and coarse sand, and the designed burial depth is 3 m. However, there is a risk of scouring and suspended sections due to the action of waves and currents, as well as the disturbance of the local flow field of the pile foundation.

[0120] In step (1), the distributed vibration demodulator 1 is connected to the embedded dark optical fiber 2 in the submarine cable to complete the link connectivity check and confirm that the signal covers the entire line. Then, a 200 m long optical fiber in a stable buried environment is selected as a reference segment, and its long-term spectrum is fixed as the benchmark for subsequent analysis.

[0121] In step (2), the distributed vibration demodulator 1 is turned on to collect data, and the sampling frequency is set to 100 Hz to simultaneously cover the main wave frequency (0.2–1.5 Hz) and the eddy current / operational disturbance frequency band (2–30 Hz), leaving sufficient sampling margin. The channel spacing is set to 3m, corresponding to 3300 monitoring channels along the entire line.

[0122] In step (3), such as Figure 2 As shown, the data preprocessing unit 3 receives the time-domain signal and performs segmented processing with a time window of 120 s. First, the normalized cross-correlation method is used to filter segments that are highly correlated with the reference template, and coherent superposition is performed to obtain enhanced time-series data with improved signal-to-noise ratio. Then, bandpass filtering is performed in the range of 0–60 Hz to output bandpass enhanced time-series data, and short-time Fourier transform is performed to output the time-frequency matrix.

[0123] In step (4), mode / depth inversion unit 4 performs spectral analysis on the bandpass enhancement time series data, extracts the first-order main peak frequency and amplitude of each channel, and compares them with the reference segment to construct the suspension index function. Within a continuous 60 m channel segment at a distance of 4.7 km from the shore, the suspension index exceeds the threshold. , marked as candidate suspended segments.

[0124] In step (5), the candidate suspended segment is further processed, and the span length is obtained using string vibration:

[0125]

[0126] And combined with the equivalent tension calibrated on site equivalent mass per unit length The estimated span length is approximately 35 m, and the actual suspended section is ultimately determined to be located at 4.68–4.71 km.

[0127] In step (6), such as Figure 3 As shown, after determining the suspended section (4.680–4.710 km in this example) in mode / depth inversion unit 4, the overburden thickness variation is estimated point by point only for the remaining channels that are still covered. Extract the first-order main peak frequency from each channel. half-power bandwidth Calculate the quality factor And calculate the frequency domain attenuation index. Compared with the reference section, according to

[0128]

[0129] The change in the cover layer thickness of each channel was obtained, among which... For weighting coefficients. Thickness is not calculated within the suspended section (curves are marked as missing); within the boundary influence zone of approximately 30 m on both sides of the start and end of the suspended section, due to abrupt changes in constraints, It exhibits significant thinning (negative values) and forms a steep gradient. This example yields an average gradient of 4.650–4.680 km on the left side. The average distance is -0.7m, and the distance to the right is 4.710–4.740 km. -0.45m; other sections The overall installation remains stable.

[0130] In step (7), the source and intensity identification unit 5 receives the time-frequency matrix and bandpass enhanced time-series data of the suspended section. The neural network outputs the probability distribution of the source category as follows: wave-current dominant 0.72, eddy-induced 0.18, and human disturbance 0.10. The event is determined to be wave-current dominant scour. The intensity score is 0.78, which belongs to the warning level.

[0131] In step (8), such as Figure 4 As shown, the alarm and visualization unit 6 generates a source category probability curve and intensity distribution curve along the line at a certain moment to reflect the spatial risk characteristics at that moment. In the 4.68–4.71 km section, the intensity is greater than 0.7, forming a continuous warning risk, which is judged to be wave-current dominated scour, triggering an alarm and indicating that local backfilling or reinforcement measures should be implemented in this section.

Claims

1. A wind power submarine cable scour monitoring device based on distributed vibration sensing, characterized in that: It includes a distributed vibration demodulator (1), a submarine cable embedded dark optical fiber (2), a data preprocessing unit (3), a mode / depth inversion unit (4), a source and intensity identification unit (5), and an alarm and visualization unit (6); The distributed vibration demodulator (1) is connected to the built-in dark optical fiber (2) of the wind power submarine cable and collects distributed vibration signals; the data preprocessing unit (3) is connected to the distributed vibration demodulator (1) and enhances the distributed vibration signals; the modal / burial depth inversion unit (4) is connected to the data preprocessing unit (3) and is used to invert the suspension and overburden thickness of the wind power submarine cable; the source and intensity identification unit (5) is connected to the data preprocessing unit (3) and is used to identify the source and intensity level of scour; the alarm and visualization unit (6) is connected to the source and intensity identification unit (5) and is used to generate an integrated map of the source and intensity along the line.

2. A method for monitoring scour of wind power submarine cables based on distributed vibration sensing, characterized in that: The monitoring and identification device for monitoring and identifying the scour status and intensity of wind power submarine cables based on distributed vibration sensing, as described in claim 1, includes the following steps: Step (1): Use a distributed vibration demodulator to connect the single-mode dark fiber in the wind power submarine cable, and select the fiber segment in the single-mode dark fiber that is not affected by scouring as the reference segment. Step (2): Turn on the distributed vibration demodulator to collect the vibration signal of the wind power submarine cable and form time-series vibration data divided by channel. Set the sampling frequency and channel spacing. Step (3): The data preprocessing unit divides the received time-series vibration data into blocks according to time windows, and divides the time sequence of each time window into several sub-segments. Using the first sub-segment as a template, the normalized cross-correlation coefficient between each sub-segment and the template under discrete delay is calculated. When the cross-correlation coefficient is greater than a set value, it is a consistent sub-segment. The consistent sub-segments are time-aligned according to the maximum correlation lag and then coherently superimposed to generate enhanced time-series data. Bandpass filtering is performed in a fixed frequency band to output bandpass enhanced time-series data. The bandpass enhanced time-series data is then subjected to short-time Fourier transform according to the same time window to output a time-frequency matrix. Step (4): The bandpass enhancement time series data is input into the mode / depth inversion unit. The first-order main peak frequency and spectral peak amplitude of each channel are extracted in the mode / depth unit. The suspend index function is constructed according to the reference segment, and the candidate suspend segment is marked. Step (5): The modal / depth inversion unit extracts the first-order main peak frequency of the candidate suspended segment from step (4), and uses string vibration to calculate the span length to determine the suspension position and suspension length of the actual suspended segment. Step (6): The modal / burial depth inversion unit performs feature analysis on the bandpass enhancement time series data based on the soil-cable coupling constraint. It calculates the quality factor by extracting the first-order main peak frequency and half-power bandwidth of each channel, and calculates the change in the thickness of the cover layer on each channel of the optical cable with the frequency domain attenuation rate index. Step (7) uses the time-frequency matrix output by the data preprocessing unit and the bandpass enhanced time-series data as dual-path input to the source and intensity identification unit. The source and intensity identification unit uses a dual-branch multi-task neural network consisting of a data interpretation layer, a branch fusion layer, and a task output layer, to generate a probability vector of the scour source category. With normalized intensity score ; Step (8) inputs the probability vector of the scour source category and the normalized intensity score output by the source and intensity identification unit into the alarm and visualization unit to generate an integrated map of the source and intensity along the line.

3. The wind power submarine cable scour monitoring method based on distributed vibration sensing according to claim 2, characterized in that: In step (3), after time alignment of the consistency sub-segments according to the maximum correlation lag, coherent superposition is performed to generate enhanced time series data, and bandpass filtering is performed within a fixed frequency band. The bandpass enhancement sequences of each time window are spliced ​​together in time order to output the original channel structure and time length of the bandpass enhancement time series data. The bandpass enhancement time series data are then subjected to short-time Fourier transform according to the same time window to output the time-frequency matrix.

4. The wind power submarine cable scour monitoring method based on distributed vibration sensing according to claim 2, characterized in that: In step (3), the first sub-segment is used as a template. Calculate the first The segment and the template in discrete delay Normalized cross-correlation coefficients: ； in, Indicates the time window based on segment length With sampling rate The first segment obtained Discrete time-series signals of segments ; For template sub-segments; For discrete time delay in cross-correlation; In order to be in Down and Normalized cross-correlation coefficient; Conclusion: , ; in, For the first Sub-segment With template The maximum normalized cross-correlation coefficient; To obtain the optimal discrete time delay for maximizing the cross-correlation coefficient; Indicates to Find the maximum value. Indicates to make To obtain the maximum value ; Based on empirical thresholds The set of consistent segments is obtained by filtering: ; For sets inner sub-segment Perform time alignment, and denote the aligned sequence as... The data are then coherently superimposed to obtain the enhanced time-series data for the time window. ； in, This indicates that the enhanced time-series data after the time window is aligned and coherently superimposed is in the [missing information - likely a date or time period]. The values ​​of each sampling point; For the selected Each segment is calculated using the optimal delay. Time-aligned discrete sequences; If the sampling points are not collected If the inner sub-segment is covered, then the sampling point takes the original window signal. Definition; in a fixed frequency band Internal Perform bandpass filtering to output the bandpass enhanced sequence of the time window. ; for all time windows The bandpass enhancement time series data is concatenated in chronological order and output as the original channel structure and total duration. The bandpass enhancement time series data is then subjected to a short-time Fourier transform to output a time-frequency matrix.

5. The wind power submarine cable scour monitoring method based on distributed vibration sensing according to claim 2, characterized in that: In step (4), the reference segment obtained from step (1) is... As a reference, a suspended index function is constructed. : ; in, , For experience weights; The average first-order main peak frequency of the reference segment under the buried state. The average amplitude of the first-order main peak in the reference section under the buried state; for Set threshold range When at least three adjacent channels within the same time window At that time, adjacent channels are marked as candidate suspended segments. .

6. The wind power submarine cable scour monitoring method based on distributed vibration sensing according to claim 2, characterized in that: In step (5), the nth natural frequency of the candidate suspended segment for: ； in, For the first First natural frequency, For the span, For equivalent tension, Equivalent mass per unit length; first-order frequency is taken as the dominant frequency. The span length is obtained as follows: ; in, This is the frequency span ratio factor; The calculated length of the suspended segment; Let be the first-order dominant peak frequency of the overall channel data within the candidate suspended segment; let the center position of the suspended segment be . Then the boundary of the actual suspended segment is: ; in, Indicates the starting position of the candidate suspended segment. Indicates the termination position of the candidate suspended segment; This indicates the starting position of the actual suspended segment. This is the end position of the actual suspended segment; The start and end boundaries of the actual suspended segment are determined by the suspended position, thus obtaining the suspended position and length of the actual suspended segment.

7. The wind power submarine cable scour monitoring method based on distributed vibration sensing according to claim 2, characterized in that: In step (6), the first-order main peak frequency is extracted. and by half-power bandwidth Calculation Channel Quality factor within the time window : ; derive frequency domain attenuation index : ; in, For channel In frequency Power spectral density at that point For channel The first-order main peak frequency; To surround The frequency band spacing; Depend on and The response to changes in capping thickness yields the channel. Changes in the thickness of the capping layer : ; in, These are the weighting coefficients; The change represents the amount of energy dissipation; This represents the change in high-frequency energy decay characteristics.

8. The wind power submarine cable scour monitoring method based on distributed vibration sensing according to claim 2, characterized in that: In step (7), the time-frequency matrix and the bandpass enhancement time series data are aligned in the same time window: for each time window and each channel, the time-frequency matrix provides the channel × frequency spectrum distribution within the time window, and the bandpass enhancement time series data provides the time domain sequence segment within the same time window.

9. The wind power submarine cable scour monitoring method based on distributed vibration sensing according to claim 2, characterized in that: In step (7), a dual-branch multi-task neural network is used to fuse the dual inputs and output the results: (7.1). The time-frequency branch takes the time × frequency spectrum distribution of each time window and each channel as input to learn and encode the frequency band energy distribution, main peak position and peak shape frequency domain characteristics. (7.2) The time-domain branch takes a bandpass-enhanced time-domain sequence segment with the same time window as input, learns and encodes the time-domain features of amplitude evolution, transient impact and decay characteristics.

10. The wind power submarine cable scour monitoring method based on distributed vibration sensing according to claim 2, characterized in that: In step (7), after the time-frequency branch and the time-domain branch are cascaded in the branch fusion layer to form a shared representation, the output length is... probability vector And calculate the normalized intensity score. , among which, element This indicates the probability that the corresponding source is dominant within the time window.

Images