Submarine water supply pipeline anchor damage risk early warning method based on multi-modal data fusion

Abstract

The application discloses a seabed water supply pipeline anchor hazard risk early warning method based on multi-modal data fusion, the method comprises the following steps: obtaining an electromagnetic field original signal, an acoustic signal and an optical signal data caused by ship anchor movement to obtain an initial data group, and performing classification processing to obtain a potential field variation interference signal group and a normal background signal group; extracting and filtering abnormal values from the potential field variation interference signal group to obtain an interference feature set; extracting acoustic and optical signal features to generate a feature vector set, and identifying an anchor dragging behavior signal group after matching the feature vector set with the interference feature set, calculating behavior trajectory parameters to obtain a behavior trajectory description set; calculating the distance between the trajectory and the pipeline, if the distance is less than a preset safety distance threshold, then the corresponding behavior trajectory description set is determined as a high threat feature group, and displacement direction data is calibrated to obtain a refined threat trajectory group; and determining a potential threat trajectory to obtain an extended early warning identifier and an alarm record. The method can realize effective multi-modal data collaborative fusion and accurate analysis.

Description

Technical Field

[0001] This invention relates to the field of marine engineering safety monitoring technology, and in particular to a method for early warning of anchor damage risks to subsea water supply pipelines based on multimodal data fusion. Background Technology

[0002] Current technologies for monitoring the risk of anchor damage to subsea pipelines have significant limitations. They often rely on single-type sensors (such as acoustic or optical sensors) to acquire signals, making them ill-suited to the complex interference of the marine environment. This leads to high false alarm rates and an inability to consistently capture the true characteristics of anchor movement. Anchor dragging is a typical spatiotemporal dynamic process, with signals changing in real time according to the anchor's displacement direction and velocity. However, existing methods mostly focus on instantaneous static signal analysis, lacking continuous tracking of the entire anchor movement process. This makes it difficult to distinguish between brief contact and continuous dragging, resulting in delayed threat identification. Furthermore, anchor movement simultaneously triggers changes in multiple types of signals, including electromagnetic, acoustic, and optical signals. However, current technologies lack an effective multi-source data collaborative analysis mechanism. Signal timestamps are prone to deviation, and some signals (such as optical signals) are easily invalidated in low-visibility deep-sea environments, failing to form a unified threat assessment basis and further reducing the accuracy of early warnings.

[0003] In summary, existing technologies cannot achieve effective multimodal data collaborative fusion and accurate analysis. Summary of the Invention

[0004] This invention provides a method for early warning of anchor damage risks in subsea water supply pipelines based on multimodal data fusion, in order to solve the problem of the inability to achieve effective multimodal data collaborative fusion and accurate analysis.

[0005] Firstly, to address the aforementioned technical problems, this invention provides a method for early warning of anchor damage risks to subsea water supply pipelines based on multimodal data fusion, comprising:

[0006] Acquire the raw electromagnetic field signal, acoustic signal data, and optical signal data caused by the movement of the ship's anchor; preprocess the raw electromagnetic field signal to obtain an initial electromagnetic field data set.

[0007] The initial electromagnetic field data set is classified to obtain a potential field-varying interference signal set and a normal background signal set;

[0008] Based on the potential field-varying interference signal group, multidimensional feature vectors are extracted and outliers are filtered to obtain the interference feature set;

[0009] Based on the acoustic signal data and the optical signal data, a feature vector set is generated through feature extraction. The interference feature set is then synchronously matched with the feature vector set to obtain the anchor dragging behavior signal group.

[0010] Based on the anchor dragging behavior signal set, calculate the signal trajectory parameters to obtain the behavior trajectory description set;

[0011] Based on the behavior trajectory description set, the distance between the behavior trajectory and the pipeline location is calculated. If the distance is less than a preset safe distance threshold, the corresponding behavior trajectory description set is identified as a high-threat feature group.

[0012] Based on the high-threat feature group, the displacement direction data is calibrated to obtain the refined threat trajectory group;

[0013] Based on the refined threat trajectory group, potential threat trajectories are determined, and extended warning indicators and alarm records are obtained.

[0014] Preferably, the acquisition of the raw electromagnetic field signal, acoustic signal data, and optical signal data caused by the movement of the ship's anchor, and the preprocessing of the raw electromagnetic field signal to obtain an initial electromagnetic field data set, includes:

[0015] The original electromagnetic field signal caused by the ship's anchor movement was obtained by collecting the real-time time series of electromagnetic field signals, and the acoustic signal data collected by the acoustic sensor and the optical signal data collected by the optical sensor were also obtained.

[0016] The original electromagnetic field signal is denoised to obtain a denoised electromagnetic field dataset.

[0017] If the intensity value in the denoised electromagnetic field dataset exceeds a preset intensity threshold, then the frequency components are extracted by fast Fourier transform to obtain a frequency feature set.

[0018] Based on the frequency feature set, the real-time time series in the denoised electromagnetic field dataset that matches the frequency feature set is extracted to obtain the initial electromagnetic field data set.

[0019] Preferably, the step of classifying the initial electromagnetic field data set to obtain a potential field-varying interference signal set and a normal background signal set includes:

[0020] The initial electromagnetic field data set is subjected to time-domain analysis and frequency-domain transformation respectively to extract time-domain statistical features. After converting the time-domain signal into a frequency-domain signal through fast Fourier transform, frequency-domain spectral features are extracted. The time-domain statistical features and the frequency-domain spectral features are integrated to form a multi-dimensional feature vector.

[0021] The multidimensional feature vector is input into a preset signal classification model to obtain the classification result;

[0022] The signal sequences marked as abnormal disturbances in the classification results are divided into the potential field-variable interference signal group, and the remaining signal sequences marked as background noise are divided into the normal background signal group.

[0023] Preferably, the step of extracting multi-dimensional feature vectors and filtering outliers from the potential field-varying interference signal set to obtain an interference feature set includes:

[0024] The original electromagnetic field signal, acoustic signal data, and optical signal data collected during the same historical period and in the same sea area under the ship anchor static state and non-anchor interference state are obtained, a feature set of normal feature distribution is established, and feature vector data of historical normal signals are obtained.

[0025] Multidimensional feature vectors are extracted from the potential field-varying interference signal group, and a benchmark feature set of normal feature distribution is established based on the feature vector data of the historical normal signals.

[0026] Calculate the deviation of each feature vector in the current multidimensional feature vector from the benchmark feature set;

[0027] The feature vector whose deviation value exceeds the preset deviation threshold is determined as an abnormal data point, and the data points determined as abnormal are deleted from the current multidimensional feature vector to obtain the interference feature set.

[0028] Preferably, the step of generating a feature vector set through feature extraction based on the acoustic signal data and the optical signal data, and synchronously matching the interference feature set with the feature vector set to obtain the anchor dragging behavior signal group, includes:

[0029] The acoustic signal data and the optical signal data are time-axis aligned to obtain a synchronization signal dataset.

[0030] Based on the synchronization signal dataset, acoustic collision features from the acoustic signal data and electromagnetic field disturbance features from the optical signal data are extracted to generate a feature vector set.

[0031] The interference feature set and the feature vector set are synchronously matched, and the comprehensive matching degree of the electromagnetic field disturbance feature, the acoustic collision feature, and the optical motion feature is calculated. When the comprehensive matching degree exceeds the preset matching degree threshold, it is determined to be an anchor dragging behavior signal group.

[0032] Preferably, the step of calculating signal trajectory parameters based on the anchor dragging behavior signal set to obtain a behavior trajectory description set includes:

[0033] The anchor dragging behavior signal group is subjected to noise reduction preprocessing to obtain a standardized signal;

[0034] Based on the initial trajectory data of the anchor calculated by the standardized signal, the trajectory parameters are extracted after filtering and smoothing, including the displacement direction, instantaneous velocity and velocity change rate corresponding to each time stamp, to obtain smooth trajectory data;

[0035] A pipeline coordinate system is established based on the design axis of the submarine water supply pipeline and combined with GPS positioning points along the pipeline.

[0036] The smooth trajectory data is mapped to the pipeline coordinate system to obtain the real-time distance sequence to key locations on the pipeline;

[0037] By integrating the smooth trajectory data and the real-time distance sequence, a behavioral trajectory description set is obtained.

[0038] Preferably, the step of calculating the distance between the behavior trajectory and the pipeline location based on the behavior trajectory description set, and determining the corresponding behavior trajectory description set as a high-threat feature group if the distance is less than a preset safe distance threshold, includes:

[0039] Based on the spatial coordinate data in the behavior trajectory description set, and based on the correspondence between the spatial coordinate data and the pipeline coordinate system, the shortest distance values ​​between each sampling point on the trajectory and the pipeline centerline and pipeline protection boundary are calculated in real time to form a dynamic distance sequence.

[0040] The dynamic distance sequence is compared with a preset safe distance threshold. If the distance value of any sampling point in the dynamic distance sequence is less than the preset safe distance threshold, the behavior trajectory description set is determined to be a high-threat feature group. If the distance values ​​of all sampling points in the dynamic distance sequence are greater than the preset safe distance threshold, the behavior trajectory description set is determined to be a low-threat feature group.

[0041] Preferably, the step of calibrating the displacement direction data based on the high-threat feature group to obtain the refined threat trajectory group includes:

[0042] Extract the original displacement direction vector and corresponding timestamp and velocity parameters from the high-threat feature group to construct a direction-time associated dataset;

[0043] Based on the pipeline coordinate system, abnormal direction points in the direction-time correlation dataset are identified and corrected to eliminate direction offsets caused by measurement errors and obtain calibrated trajectory data.

[0044] Key features are extracted from the calibrated trajectory data, redundant trajectory information is removed, and a refined threat trajectory group containing precise direction parameters and core threat trajectory segments is formed.

[0045] Preferably, the step of determining potential threat trajectories based on the refined threat trajectory group to obtain extended early warning identifiers and alarm records includes:

[0046] Optical signal features are extracted from the refined threat trajectory set, and potential threat trajectories that match the anchor dragging feature are screened by frequency domain transformation.

[0047] Calculate the dynamic distance between the potential threat trajectory and the preset vulnerable points along the pipeline. If the distance between consecutive sampling points shows a decreasing trend and is less than the preset safety warning value, it is confirmed as a valid threat trajectory.

[0048] By combining the geographical coordinates of the pipeline design path and its surrounding pre-set safety buffer zone, the area of ​​threat can be located, and an extended early warning label containing the threat type, location coordinates, dynamic parameters and level can be generated.

[0049] The extended warning identifier is stored and pushed to the operation and maintenance terminal to form a structured alarm record with a timestamp.

[0050] Secondly, this invention provides a method and system for early warning of anchor damage risks to subsea water supply pipelines based on multimodal data fusion, including:

[0051] The multi-source signal acquisition module acquires the original electromagnetic field signal, acoustic signal data, and optical signal data caused by the movement of the ship's anchor; preprocesses the original electromagnetic field signal to obtain an initial electromagnetic field data set; and obtains the initial electromagnetic field data set based on the original electromagnetic field signal.

[0052] The field-varying signal classification module classifies the initial electromagnetic field data group to obtain a potential field-varying interference signal group and a normal background signal group.

[0053] The interference feature extraction module extracts multi-dimensional feature vectors and filters outliers from the potential field-varying interference signal group to obtain the interference feature set.

[0054] The multimodal feature matching module generates a feature vector set by extracting features based on the acoustic signal data and the optical signal data, and synchronously matches the interference feature set with the feature vector set to obtain the anchor dragging behavior signal group;

[0055] The trajectory parameter calculation module calculates the signal trajectory parameters based on the anchor dragging behavior signal group to obtain the behavior trajectory description set;

[0056] The threat level determination module calculates the distance between the behavior trajectory and the pipeline location based on the behavior trajectory description set. If the distance is less than a preset safe distance threshold, the corresponding behavior trajectory description set is determined to be a high-threat feature group.

[0057] The threat trajectory refinement module calibrates the displacement direction data based on the high-threat feature group to obtain a refined threat trajectory group.

[0058] The warning identifier generation module determines potential threat trajectories based on the refined threat trajectory group, and obtains extended warning identifiers and alarm records.

[0059] Thirdly, the present invention also provides an electronic device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement the method for early warning of anchor damage to submarine water supply pipelines based on multimodal data fusion as described above.

[0060] Fourthly, the present invention also provides a computer-readable storage medium comprising a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to execute the above-described method for early warning of anchorage hazards of subsea water supply pipelines based on multimodal data fusion.

[0061] Compared with the prior art, the present invention has the following beneficial effects:

[0062] (1) This invention acquires the original electromagnetic field signal caused by the movement of ship anchors in real time, and combines it with acoustic and optical sensors to obtain multi-source data to construct an initial electromagnetic field data set, thereby realizing multimodal and all-time monitoring of anchor damage behavior. By classifying and processing the initial electromagnetic field data set, potential field variation interference signals are extracted, and a benchmark feature set is established by combining it with historical normal signals to filter out outliers and form a highly reliable interference feature set. Compared with the traditional reliance on a single sensor, this invention significantly improves the richness and reliability of the data source, effectively overcoming the defects of single signals being easily interfered with and prone to failure in the marine environment.

[0063] (2) This invention achieves collaborative analysis and fusion of multi-source data by synchronously matching electromagnetic, acoustic and optical signals, calculating the comprehensive matching degree of multi-modal features, and accurately judging the anchor dragging behavior signal group. This method effectively solves the problems of large signal time deviation and matching difficulties in traditional methods, and improves the timeliness and consistency of threat behavior identification.

[0064] (3) This invention establishes a set of behavioral trajectory descriptions in the pipeline coordinate system by calculating the trajectory parameters of the anchor dragging behavior, and dynamically determines high-threat feature groups by calculating the distance between the trajectory and the pipeline position in real time. The high-threat feature groups are refined to generate a refined threat trajectory group containing precise direction and speed parameters, and then further verified by combining optical signal features to output extended warning signs. This method realizes continuous tracking and spatial relationship analysis of the entire anchor movement process, overcomes the limitation of existing technologies that only focus on instantaneous signals, significantly improves the comprehensiveness and accuracy of threat assessment, and ensures the reliability and operability of warning information through refined description and multiple verification of threat trajectories, providing a clear basis for operation and maintenance response. Attached Figure Description

[0065] Figure 1This is a schematic diagram of the process for early warning of anchor damage risk in subsea water supply pipelines based on multimodal data fusion provided in the first embodiment of the present invention;

[0066] Figure 2 This is a schematic diagram of the structure of the subsea water supply pipeline anchor damage risk early warning system based on multimodal data fusion provided in the second embodiment of the present invention. Detailed Implementation

[0067] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0068] Reference Figure 1 The first embodiment of the present invention provides a method for early warning of anchor damage risk in subsea water supply pipelines based on multimodal data fusion, including the following steps:

[0069] S11, acquire the original electromagnetic field signal, acoustic signal data and optical signal data caused by the movement of the ship's anchor, preprocess the original electromagnetic field signal to obtain the initial electromagnetic field data set;

[0070] S12, the initial electromagnetic field data group is classified to obtain a potential field-variable interference signal group and a normal background signal group;

[0071] S13, Based on the potential field-varying interference signal group, extract multi-dimensional feature vectors and filter outliers to obtain an interference feature set;

[0072] S14, Based on the acoustic signal data and the optical signal data, a feature vector set is generated by feature extraction, and the interference feature set is synchronously matched with the feature vector set to obtain the anchor dragging behavior signal group;

[0073] S15, calculate the signal trajectory parameters based on the anchor dragging behavior signal group to obtain the behavior trajectory description set;

[0074] S16, Calculate the distance between the behavior trajectory and the pipeline location based on the behavior trajectory description set. If the distance is less than a preset safe distance threshold, then the corresponding behavior trajectory description set is determined to be a high-threat feature group.

[0075] S17, Based on the high-threat feature group, calibrate the displacement direction data to obtain the refined threat trajectory group;

[0076] S18, Based on the refined threat trajectory group, determine the potential threat trajectory and obtain extended warning identifiers and alarm records.

[0077] In step S11, the original electromagnetic field signal, acoustic signal data and optical signal data caused by the movement of the ship's anchor are acquired, and the original electromagnetic field signal is preprocessed to obtain the initial electromagnetic field data set.

[0078] In one feasible approach, the acquisition of raw electromagnetic field signals, acoustic signal data, and optical signal data generated by the ship's anchor movement, followed by preprocessing of the raw electromagnetic field signals to obtain an initial electromagnetic field data set, includes:

[0079] The original electromagnetic field signal caused by the ship's anchor movement was obtained by collecting the real-time time series of electromagnetic field signals, and the acoustic signal data collected by the acoustic sensor and the optical signal data collected by the optical sensor were also obtained.

[0080] The original electromagnetic field signal is denoised to obtain a denoised electromagnetic field dataset.

[0081] If the intensity value in the denoised electromagnetic field dataset exceeds a preset intensity threshold, then the frequency components are extracted by fast Fourier transform to obtain a frequency feature set.

[0082] Based on the frequency feature set, the real-time time series in the denoised electromagnetic field dataset that matches the frequency feature set is extracted to obtain the initial electromagnetic field data set.

[0083] It should be noted that ship anchors are made of metal. When they are anchored or towed on the seabed, they interact with the marine environment, generating electromagnetic field disturbances and electromagnetic field signals that change with the anchor's movement. The real-time time series of electromagnetic field signals is acquired by deploying electromagnetic sensors in an array around the seabed pipeline. The sensors are required to continuously collect data at fixed time intervals of 1 second, with each acquired electromagnetic field strength value corresponding to a precise timestamp. Therefore, the real-time time series includes both the electromagnetic field strength value and the timestamp, reflecting the dynamic process of the anchor's movement. Within identical timestamps, acoustic sensors collect sound signal data generated by the anchor's friction and collision with the seabed, and optical sensors collect related optical signal data. Low-frequency optical imaging sensors collect optical imaging signal data (for target morphology and motion analysis, sampling frequency 10Hz), and high-frequency optical vibration sensors (such as fiber Bragg grating (FBG) sensors collect optical vibration signal data (for vibration characteristic analysis, sampling frequency not less than 500Hz).

[0084] Wavelet transform denoising was applied to the original electromagnetic field signal. The db4 wavelet basis was set, and the decomposition level was set to 5. The original signal was decomposed at multiple scales using the db4 wavelet basis, yielding one set of approximation coefficients (low-frequency components, corresponding to the effective signal of anchor movement and geomagnetic reference values) and four sets of detail coefficients (high-frequency components, mainly including equipment thermal noise and environmental interference). Threshold truncation was performed on the decomposed detail coefficients to suppress high-frequency noise. Normal geomagnetic fluctuations and minor environmental interference exist in the seabed background. A preset intensity threshold of 0.5 μT was used. Signals with intensity below 0.5 μT (such as residual weak ocean current disturbances and interference from fish and shrimp activity) were identified as non-anchor movement related signals, and their corresponding detail coefficients were directly set to zero. Only signals with intensity exceeding 0.5 μT were likely significant electromagnetic disturbances caused by the movement of metal anchors (anchoring, dragging), and therefore retained. The thresholded detail coefficients and approximation coefficients were reconstructed using the db4 wavelet basis to obtain the denoised electromagnetic field dataset.

[0085] For signal segments with intensity exceeding a threshold, the time-domain signal is converted to a frequency-domain signal using Fourier transform. The sensor collects data at fixed time intervals of 50 milliseconds, meaning the signal sampling frequency is 20 Hz. According to the Nyquist sampling theorem, Fourier transform can analyze frequency components in the 0-10 Hz range without aliasing, which is sufficient to reflect the frequency characteristics of anchor movement (0.1–0.3 Hz for dragging anchors, 1–5 Hz for dropping anchors). The resulting dataset containing the key frequency parameters of the signal segment, including the dominant frequency value, bandwidth, and frequency energy, is called the frequency feature set.

[0086] If it is a towed anchor (the anchor continuously rubs against the seabed), the signal fluctuates periodically. The Fourier transform will show a dominant frequency of 0.1 to 0.3 Hz (the frequency with the highest energy), and the harmonic components (0.2 Hz, 0.3 Hz) have a high energy ratio. If it is a dropped anchor (the anchor rapidly hits the seabed), it will show a dominant frequency of 1 to 5 Hz, and the signal is an instantaneous peak with a wideband component of 0.3 to 0.5 Hz. In the frequency feature set, cases matching the characteristics of anchoring and dropping are checked. To determine if a signal segment belongs to the anchoring mode, it must simultaneously meet the following two conditions: the dominant frequency of the signal segment must fall within the range of 0.1Hz to 0.3Hz; and the ratio of its second harmonic component energy to the dominant frequency energy must be greater than 0.3, to confirm the periodicity of the signal rather than random fluctuations. To determine if a signal segment belongs to the dropping mode, it must simultaneously meet the following two conditions: the dominant frequency of the signal segment must fall within the range of 1Hz to 5Hz; and the ratio of the total energy of the signal segment in the wideband range of 0.3Hz to 0.5Hz to the dominant frequency energy must be greater than 0.5, this condition is used to capture the broadband impact response characteristics induced by impact. In the frequency feature set, the frequency feature data of all signal segments are traversed, and the above rules are applied for automatic discrimination. For any signal segment determined to meet the anchoring or dropping mode, the complete real-time time series segment is located and extracted from the denoised electromagnetic field dataset based on its corresponding timestamp. All these extracted and pattern-recognized signal fragments together constitute the initial electromagnetic field data set for subsequent classification processing.

[0087] In step S12, the initial electromagnetic field data group is classified to obtain a potential field-varying interference signal group and a normal background signal group.

[0088] In one implementable manner, classifying the initial electromagnetic field data set to obtain a potential field-varying interference signal set and a normal background signal set includes:

[0089] The initial electromagnetic field data set is subjected to time-domain analysis and frequency-domain transformation respectively to extract time-domain statistical features. After converting the time-domain signal into a frequency-domain signal through fast Fourier transform, frequency-domain spectral features are extracted. The time-domain statistical features and the frequency-domain spectral features are integrated to form a multi-dimensional feature vector.

[0090] The multidimensional feature vector is input into a preset signal classification model to obtain the classification result;

[0091] The signal sequences marked as abnormal disturbances in the classification results are divided into the potential field-variable interference signal group, and the remaining signal sequences marked as background noise are divided into the normal background signal group.

[0092] It should be noted that for the real-time time series containing timestamp-intensity values ​​in the electromagnetic field data set, statistical features reflecting the temporal regularity of the signal are first extracted through time-domain analysis, including the mean intensity (reflecting the overall strength of the signal), variance (reflecting the degree of intensity fluctuation), peak value (the maximum intensity value when the anchor impacts or is towed), kurtosis (reflecting the steepness of the peak value), and signal duration. Then, the time-domain signal is converted into a frequency-domain signal through Fast Fourier Transform (FFT). Based on a sampling frequency of 20Hz, frequency-domain spectral features within 0-10Hz can be effectively extracted, including the dominant frequency (the frequency with the highest energy, mostly 0.1-0.3Hz for towed anchors and 1-5Hz for dropped anchors) and frequency bandwidth (the bandwidth of the anchor movement signal is concentrated, while the background noise bandwidth is wide). The above-mentioned time-domain statistical features (mean, variance, peak value, etc.) and frequency-domain spectral features (dominant frequency and frequency bandwidth) are integrated in sequence to form a multi-dimensional feature vector covering the temporal regularity and frequency characteristics of the signal.

[0093] The pre-defined signal classification model is a Support Vector Machine (SVM)-based model built using training data. The training data includes known characteristics of anomalous electromagnetic signals caused by anchor movement (low-frequency periodicity during anchor dragging, instantaneous peak characteristics during anchor dropping) and normal background noise characteristics (high-frequency irregularity of ocean current interference, weak fluctuations in equipment thermal noise). Through kernel function selection, RBF kernel processing of nonlinear characteristics, and hyperplane optimization, the model has the ability to distinguish between the two types of signals. After inputting a multi-dimensional feature vector integrating time-domain statistical features (such as intensity variance and peak duration) and frequency-domain spectral features (such as dominant frequency and frequency bandwidth) into the model, the model outputs the classification result corresponding to each feature vector based on the classification boundary formed during training. If the feature vector matches the anomalous signal pattern related to anchor movement (dominant frequency between 0.1 and 0.3 Hz or between 1 and 5 Hz), it is marked as an anomalous disturbance; otherwise, it is marked as background noise. Signal sequences marked as anomalous disturbances in the classification results are divided into a potential field-variable interference signal group, and signal sequences marked as background noise are divided into a normal background signal group.

[0094] It is worth noting that the "preset signal classification model" is a core algorithm tool used to distinguish between abnormal disturbances caused by anchor movement and background noise in electromagnetic field signals. It is built based on support vector machine (SVM) and its design is adapted to the nonlinear distribution characteristics of signal features in complex seabed environments.

[0095] Model Structure: The model adopts a simplified structure of input layer – kernel function mapping – hyperplane classifier – output layer. The input layer receives multi-dimensional feature vectors, which need to be normalized to avoid the influence of features of different magnitudes on classification. The RBF kernel function is used to map the electromagnetic field signal features in the low-dimensional feature space to the high-dimensional space through nonlinear mapping, solving the nonlinear separation problem between anchor motion signals and background noise features. A convex quadratic programming problem is solved to find the hyperplane in the high-dimensional space that maximizes the distance to the nearest point between the two classes of samples (abnormal perturbation / background noise). The hyperplane equation is wx + b = 0, where w is the normal vector and b is the bias term. The output is a binary classification result: when a sample point satisfies wx + b ≥ 0, it is classified as an abnormal perturbation; when wx + b < 0, it is classified as background noise.

[0096] Model Training: Annotated historical data is used in this model. This data originates from long-term accumulation of data from the subsea pipeline monitoring system and is annotated through multi-source information fusion. The selection of positive samples (anchor motion signals) must simultaneously meet multiple validation conditions. First, the Automatic Identification System (AIS) data within the time period must indicate that a vessel is anchored or sailing at low speed. Second, acoustic or optical signal data within this time period must collaboratively verify the presence of anchor dragging characteristics. Negative samples (background noise) are selected from background signals in sea areas with no confirmed vessel activity or clearly identified non-anchor-related interference signals. Using the above-mentioned annotated historical data, the training and validation sets are divided in a 7:3 ratio. The hinge loss function and SMO optimizer are employed. Key parameters—the penalty parameter C (chosen as 10 to control the penalty for misclassified samples) and the kernel function γ (chosen as 0.1 to adapt to the locality of signal features)—are fine-tuned through 5-fold cross-validation to avoid overfitting or underfitting. Using the validation set F1 score (combining precision and recall) as an indicator, the model was trained until the score stabilized above 0.92, ensuring that the model could accurately identify abnormal signals (precision 0.91) and effectively capture real anchor motion signals (recall 0.93), ultimately resulting in a mature model that can be directly used for classification.

[0097] In step S13, based on the potential field-varying interference signal group, multidimensional feature vectors are extracted and outliers are filtered to obtain the interference feature set.

[0098] In one feasible approach, the step of extracting a multidimensional feature vector and filtering outliers from the potential field-varying interference signal set to obtain an interference feature set includes:

[0099] The original electromagnetic field signal, acoustic signal data, and optical signal data collected during the same historical period and in the same sea area under the ship anchor static state and non-anchor interference state are obtained, a feature set of normal feature distribution is established, and feature vector data of historical normal signals are obtained.

[0100] Multidimensional feature vectors are extracted from the potential field-varying interference signal group, and a benchmark feature set of normal feature distribution is established based on the feature vector data of the historical normal signals.

[0101] Calculate the deviation of each feature vector in the current multidimensional feature vector from the benchmark feature set;

[0102] The feature vector whose deviation value exceeds the preset deviation threshold is determined as an abnormal data point, and the data points determined as abnormal are deleted from the current multidimensional feature vector to obtain the interference feature set.

[0103] It should be noted that, due to the spatiotemporal dependence of marine environmental signals (electromagnetic field strength), only data from the same period and sea area can reflect the normal characteristics of the current scenario. Therefore, "historical period" refers to the same season and the same month. Initial electromagnetic field data, acoustic signal characteristic data, and optical signal characteristic data are acquired from the same period and sea area, under conditions of stationary ship anchors and non-anchor-related interference. Electromagnetic field characteristics include time-domain disturbance peak value, mean, variance, and frequency-domain dominant frequency and bandwidth; acoustic characteristics include sound pressure peak value and characteristic frequency bandwidth; optical characteristics include grayscale value fluctuation amplitude and target motion pixel displacement, forming feature vector data of historical normal signals. Multidimensional feature vectors are re-extracted from potential field-variant interference signal groups, and these vectors must be completely consistent with the dimensions and units of the historical normal signal feature vectors. A reasonable distribution range of risk-free signal characteristics is constructed through statistical analysis, i.e., a baseline feature set for the normal feature distribution. The method for constructing the baseline feature set using statistical analysis is to calculate statistical indicators separately for each dimension of the historical normal feature vectors to determine the normal fluctuation range for that dimension. The historical mean of the peak disturbance in the electromagnetic field's time domain is 0.3 μT, with a standard deviation of 0.1 μT. The normal range is set at the mean ± 3 times the standard deviation (i.e., 0-0.6 μT). In the frequency domain, the historical mean of the dominant frequency is 0.05 Hz, with a standard deviation of 0.02 Hz. The normal range is set at 0.01-0.09 Hz. The historical mean of the peak sound pressure level of the acoustic signal is 50 Pascals, with a standard deviation of 15 Pascals. The normal range is set at 5-95 Pascals. The historical mean of the acoustic characteristic frequency bandwidth is 20 Hz, with a standard deviation of 5 Hz. The normal range is set at 5-35 Hz. The historical mean of the grayscale fluctuation amplitude of the optical signal is 20 gray levels, with a standard deviation of 8 gray levels. The normal range is set at 0-44 gray levels. The historical mean of the optical pixel displacement is 5 pixels, with a standard deviation of 3 pixels. The normal range is set at 0-14 pixels. The normal ranges for each dimension collectively constitute a single-dimensional constraint of the baseline feature set, used to initially determine whether the current feature exceeds the normal range. The final baseline feature set is essentially the allowable range of signal features when there is no anchor hazard threat. The preset deviation threshold is 3. The formula for single-dimensional deviation is: Deviation = |Current feature value - Historical mean of this dimension| / Historical standard deviation of this dimension (if the current value is within the normal range, the deviation is less than 3; if it is not within the normal range, the deviation is greater than 3).A multi-dimensional weighted fusion judgment rule was used to screen outliers. Weights were assigned based on the sensitivity of features to anchor damage, with electromagnetic features accounting for 40% (25% in the time domain, 15% in the frequency domain), acoustic features accounting for 35% (20% peak sound pressure level, 15% frequency bandwidth), and optical features accounting for 25% (10% grayscale fluctuation, 15% pixel displacement). For each feature vector, the weights of all dimensions with a deviation greater than 3 were summed to calculate the weighted deviation sum. A comprehensive judgment threshold of 50% was set. When the weighted deviation sum was ≥50%, it was judged as an outlier (multiple dimensions jointly indicate the presence of anchor movement interference); if the weighted deviation sum was <50%, it was judged as a pseudo-outlier (individual dimension anomalies may be caused by environmental noise). After deleting pseudo-outlier data points, the remaining outlier data points constituted an interference feature set.

[0104] In step S14, based on the acoustic signal data and the optical signal data, a feature vector set is generated through feature extraction, and the interference feature set is synchronously matched with the feature vector set to obtain the anchor dragging behavior signal group.

[0105] In one implementable manner, the step of generating a feature vector set through feature extraction based on the acoustic signal data and the optical signal data, and synchronously matching the interference feature set with the feature vector set to obtain an anchor dragging behavior signal group, includes:

[0106] The acoustic signal data and the optical signal data are time-axis aligned to obtain a synchronization signal dataset.

[0107] Based on the synchronization signal dataset, acoustic collision features from the acoustic signal data and electromagnetic field disturbance features from the optical signal data are extracted to generate a feature vector set.

[0108] The interference feature set and the feature vector set are synchronously matched, and the comprehensive matching degree of the electromagnetic field disturbance feature, the acoustic collision feature, and the optical motion feature is calculated. When the comprehensive matching degree exceeds the preset matching degree threshold, it is determined to be an anchor dragging behavior signal group.

[0109] It should be noted that acoustic and optical signal data were acquired from acoustic and optical sensors respectively, with both sensors collecting data at a frequency of 10 times per second. The acoustic and optical signals were aligned using a unified timestamp to generate a synchronized signal dataset. From the acoustic signal data in the synchronized signal dataset, acoustic features related to anchor dragging behavior were extracted, mainly including peak sound pressure (reflecting collision intensity), duration (reflecting the duration of the collision event), and energy spectral density (reflecting the distribution of sound energy in the frequency domain). From the optical signal data in the synchronized signal dataset, optical features related to anchor movement were extracted, mainly including grayscale value fluctuation amplitude and target pixel displacement. Simultaneously, electromagnetic field disturbance features were extracted from the signal data acquired by the electromagnetic field sensor, forming a multi-dimensional feature vector set. Each vector represents the joint feature state of the multi-modal signal at a given time point. The disturbance feature set previously extracted from the electromagnetic field data was then time-aligned and matched with the newly constructed multi-modal feature vector set. During matching, it is crucial to ensure strict synchronization of the two feature sets on the time axis to avoid mismatches due to sensor response delays or sampling deviations. A multi-feature fusion algorithm was used to calculate the overall matching degree. Specifically, cosine similarity was used to calculate the similarity between electromagnetic field disturbance features and corresponding features in the interference feature set. Euclidean distance was used to calculate the matching degree between acoustic collision features and acoustic templates in the interference feature set. The acoustic templates are standard acoustic feature vectors, whose dimensions are completely consistent with the real-time extracted acoustic features (including three dimensions: peak sound pressure, duration, and energy spectral density). The feature values ​​of each dimension are taken from the statistical average of 500 actual anchor dragging events (average peak sound pressure 85 Pa, average duration 1.2 s, and the main peak of energy spectral density concentrated at 10 Hz). By weighted fusion of the matching results of the above features, an acoustic feature weight of 0.6 and an electromagnetic field feature weight of 0.4 were assigned to obtain the overall matching degree value. The weighting was determined based on feature importance analysis: among 100 known anchor dragging event samples, the recognition accuracy of acoustic features (92%) was significantly higher than that of electromagnetic field features (81%), and acoustic signals had stronger anti-interference capabilities in highly turbid sea areas (mean signal-to-noise ratio 18dB, higher than the 12dB of electromagnetic field signals). A preset matching degree threshold of 0.85 was set. If the overall matching degree exceeded this threshold, the signals within the current time window were determined to be anchor dragging behavior signal groups, indicating that the multimodal signals were highly consistent in time, intensity, and feature patterns, which conformed to the typical behavioral characteristics of anchor dragging.

[0110] In step S15, the signal trajectory parameters are calculated based on the anchor dragging behavior signal group to obtain the behavior trajectory description set.

[0111] In one implementable manner, calculating signal trajectory parameters based on the anchor dragging behavior signal set to obtain a behavior trajectory description set includes:

[0112] The anchor dragging behavior signal group is subjected to noise reduction preprocessing to obtain a standardized signal;

[0113] Based on the initial trajectory data of the anchor calculated by the standardized signal, the trajectory parameters are extracted after filtering and smoothing, including the displacement direction, instantaneous velocity and velocity change rate corresponding to each time stamp, to obtain smooth trajectory data;

[0114] A pipeline coordinate system is established based on the design axis of the submarine water supply pipeline and combined with GPS positioning points along the pipeline.

[0115] The smooth trajectory data is mapped to the pipeline coordinate system to obtain the real-time distance sequence to key locations on the pipeline;

[0116] By integrating the smooth trajectory data and the real-time distance sequence, a behavioral trajectory description set is obtained.

[0117] It should be noted that the anchor dragging behavior signal set is a sequence of anchor motion signals that was determined to be valid after the fusion of data from multiple sensors (electromagnetic, acoustic, and optical). These raw signals include marine environmental noise (such as water flow sound and biological activity), thermal noise from the equipment itself, and transmission interference. In the preprocessing stage, wavelet transform (using the db4 wavelet basis for decomposition) was first used to reduce noise from the signals, effectively separating noise from valid signals. Subsequently, the denoised signals were standardized by using the Z-score normalization method to convert signal values ​​from different sensors or with different dimensions to the same dimensionless scale, forming a standardized signal. Based on the standardized signal, the initial trajectory of the anchor was calculated using a multi-sensor fusion positioning method. A set of monitoring nodes was deployed every 50 meters along the pipeline axis. Each set of nodes included one acoustic sensor (sampling rate 16kHz), one electromagnetic field sensor (measurement range 0-5μT), and one optical imaging module (field of view 60°). All nodes were time-synchronized via submarine optical cable (synchronization accuracy ≤1ms). During positioning, the horizontal distance is calculated using the time difference of arrival (TDOA) of acoustic signals to adjacent nodes, and the radial distance is determined by combining this with an electromagnetic field signal intensity attenuation model (which follows an inverse square law). Simultaneously, the target size change rate from optical imaging is used to assist in correcting depth information. These three factors are then fused to calculate the anchor's real-time three-dimensional position coordinates. This method yields a series of timestamped, discrete anchor point spatial locations. , , ... This forms the initial trajectory data of the anchor. This initial trajectory data contains noise due to measurement errors and environmental disturbances, and therefore needs to be smoothed. A Kalman filter algorithm is used, and the state equation is... , among which position and speed For state variables, satisfying State transition matrix A uniform motion model was constructed; the observation equation is: Observation matrix Extracting the position component from the state variables; process noise covariance The matrix is ​​diagonal, with position noise variance set to 0.5 m² and velocity noise variance set to 0.1 (m / s)²; the observation noise covariance is... Based on dynamic adjustment of sensor accuracy, the acoustic positioning error is set to 1.2 m² when dominant and reduced to 0.3 m² when optical correction is applied. Multi-period observation data is fused through prediction and update steps to effectively suppress random errors and output a smooth and continuous sequence of displacement points. Subsequently, key trajectory parameters are extracted from the smoothed trajectory data. The displacement direction is calculated and normalized using the vector difference of consecutive position points; the instantaneous velocity is obtained as the ratio of position change to time interval; and the velocity change rate is calculated by the difference between consecutive instantaneous velocities. Thus, smooth trajectory data containing timestamps, positions, displacement directions, instantaneous velocities, and velocity change rates are obtained.

[0118] Using the design axis of the subsea water supply pipeline as the central baseline and combining the pipeline's GPS positioning points as control points, a rectangular coordinate system for the pipeline is established through coordinate transformation, with the pipeline's starting point as the origin, the pipeline axis direction as the X-axis, and the seabed plane normal as the Z-axis. The geodetic coordinates (WGS-84) in the smoothed trajectory data are mapped to this pipeline coordinate system through rotation and translation transformation. Let the coordinates of the pipeline's starting point (or the first) GPS point in the WGS-84 coordinate system be P0=(L0,λ0,h0) (longitude, latitude, elevation), and let the WGS-84 coordinates of another GPS point along the pipeline axis direction be P1=(L1,λ1,h1). The control points P0 and P1 are then converted from WGS-84 geodetic coordinates to geocentric rectangular coordinates (ECEF), denoted as X0=(X0,Y0,Z0) and X1=(X1,Y1,Z1). The conversion formula is:

[0119] ;

[0120] ;

[0121] ;

[0122] Where N is the radius of curvature of the ramusoidal circle, and e is the first eccentricity of the WGS-84 ellipsoid. Point X0 is taken as the origin of the pipeline coordinate system. The vector pointing from X0 to X1 is calculated, and this vector, after normalization, is defined as the unit vector of the X-axis (axial direction) of the pipeline coordinate system. Based on this, combining the approximately vertical direction at the pipeline starting point P0 (referencing the Earth ellipsoid normal direction) with the already obtained X-axis direction, the unit vectors of the Y-axis (lateral) and Z-axis (vertical) of the pipeline coordinate system are determined sequentially through vector projection and cross product operations. The unit vectors of the X, Y, and Z axes together constitute the rotation matrix from the geocentric geofixed coordinate system to the pipeline coordinate system. For the geodetic coordinates of any anchor point, they are first converted to geocentric geofixed coordinates using the same method, and then the origin X0 is subtracted to obtain the coordinate sequence of the anchor trajectory in the pipeline coordinate system. The Euclidean distances between each point in the trajectory and key locations of the pipeline (pipe wall, weld joints, midpoint of the cantilever section) are calculated, generating a real-time distance sequence that varies over time. Finally, the smooth trajectory data (position, speed, direction, etc.) is integrated and time-aligned with the real-time distance sequence to form a complete behavioral trajectory description set. This dataset comprehensively describes the spatiotemporal dynamic behavior of anchor dragging and its relative positional relationship with the pipeline, providing accurate input for subsequent threat level determination.

[0123] In step S16, the distance between the behavior trajectory and the pipeline location is calculated based on the behavior trajectory description set. If the distance is less than a preset safe distance threshold, the corresponding behavior trajectory description set is determined to be a high-threat feature group.

[0124] In one feasible approach, the step of calculating the distance between the behavior trajectory and the pipeline location based on the behavior trajectory description set, and determining the corresponding behavior trajectory description set as a high-threat feature group if the distance is less than a preset safe distance threshold, includes:

[0125] Based on the spatial coordinate data in the behavior trajectory description set, and based on the correspondence between the spatial coordinate data and the pipeline coordinate system, the shortest distance values ​​between each sampling point on the trajectory and the pipeline centerline and pipeline protection boundary are calculated in real time to form a dynamic distance sequence.

[0126] The dynamic distance sequence is compared with a preset safe distance threshold. If the distance value of any sampling point in the dynamic distance sequence is less than the preset safe distance threshold, the behavior trajectory description set is determined to be a high-threat feature group. If the distance values ​​of all sampling points in the dynamic distance sequence are greater than the preset safe distance threshold, the behavior trajectory description set is determined to be a low-threat feature group.

[0127] It should be noted that the behavior trajectory description set includes smooth trajectory data anchored in the pipeline coordinate system and real-time distance sequences to key pipeline locations. Calculating the distance between the behavior trajectory and the pipeline location essentially involves finding the shortest spatial straight-line distance from each sampling point on the trajectory to the pipeline centerline and the pipeline protection boundary. The pipeline centerline is modeled as a series of continuous 3D line segments (connected by adjacent GPS positioning points). For any sampling point on the trajectory, the shortest distance to each line segment of the pipeline centerline is calculated, and the minimum value is taken as the final distance from that point to the pipeline centerline. This calculation can be transformed into a mathematical problem of finding the distance from a point to a 3D line segment, which is efficiently calculated using vector projection. The pipeline protection boundary is typically defined as a cylindrical spatial surface with a certain radius (e.g., based on the pipeline diameter, burial depth, and safety regulations, this radius can be set to 1.5 times the pipeline outer diameter plus a 1-meter buffer distance). The shortest distance from the sampling point to this cylindrical surface is calculated; if the point is outside the cylinder, the distance is positive; if it has entered the cylinder, the distance is negative. The system performs the above calculations on all sampling points in real time, generating a dynamic distance sequence arranged in chronological order.

[0128] A preset safe distance threshold is a key parameter. This threshold can be set in tiers, with a warning threshold of 5 meters and an emergency threshold of 2 meters. "Less than the preset safe distance threshold" typically refers to a distance value less than the set emergency threshold. The system compares each distance value in the dynamic distance sequence with the preset safe distance threshold in real time. The judgment logic is as follows: if any sampling point in the sequence has a distance value less than the safe distance threshold, the entire behavior trajectory description set is immediately marked as a high-threat feature group, indicating that the anchor dragging event poses an immediate and unacceptable risk to the pipeline. Conversely, if the distance values ​​of all sampling points in the sequence are consistently greater than the safe distance threshold, the behavior trajectory description set is judged as a low-threat feature group, indicating that the event does not currently pose a direct threat. The judgment result (high / low threat), along with specific timestamps, minimum distance values, and location information, will be output to trigger the corresponding warning level or enter the log for subsequent analysis.

[0129] In step S17, the displacement direction data is calibrated according to the high threat feature group to obtain the refined threat trajectory group.

[0130] In one feasible approach, the step of calibrating the displacement direction data based on the high-threat feature set to obtain a refined threat trajectory set includes:

[0131] Extract the original displacement direction vector and corresponding timestamp and velocity parameters from the high-threat feature group to construct a direction-time associated dataset;

[0132] Based on the pipeline coordinate system, abnormal direction points in the direction-time correlation dataset are identified and corrected to eliminate direction offsets caused by measurement errors and obtain calibrated trajectory data.

[0133] Key features are extracted from the calibrated trajectory data, redundant trajectory information is removed, and a refined threat trajectory group containing precise direction parameters and core threat trajectory segments is formed.

[0134] It should be noted that although the trajectory data included in the high-threat feature group is marked as high-risk, its displacement direction data may be biased due to sensor measurement errors, changes in water sound velocity profiles, or interference from complex seabed currents. Calibration aims to eliminate these non-systematic errors, improve the accuracy of trajectory direction parameters, and thus more accurately determine the anchor's movement intention and its relative position to the pipeline. First, the original displacement direction vector is extracted from the high-threat feature group, along with the corresponding timestamp and the instantaneous velocity value calculated within that time interval, thus constructing a direction-time correlated dataset containing time, direction vector, and velocity. Based on the established pipeline coordinate system, abnormal direction points in the dataset are identified and corrected. The identification process can employ a statistical method, calculating the average direction of several consecutive displacement direction vectors, and setting a dynamically permissible deviation angle threshold of 30 degrees. If the angle between a direction vector and the average direction exceeds this threshold, it is marked as an anomaly. When correcting outliers, interpolation replacement can be performed using the Kalman filter algorithm based on the effective direction vectors immediately before and after them. A two-dimensional state vector is constructed using the displacement direction angle (the angle between the displacement and the positive X-axis, ranging from 0-360°, converted to radians for calculation) and the rate of change of the direction angle in the pipe coordinate system as state variables. .in, Let be the displacement direction angle (in radians) at time k. Let be the rate of change of the orientation angle at time k (radians / second). The state transition equation adopts a uniform change model, assuming that the rate of change of the orientation angle is constant over a short period of time: State transition matrix T is the sampling time interval (consistent with the sensor sampling frequency, set to 0.1 seconds, i.e., 10Hz sampling). This is for process noise. In terms of observation, the direction angle calculated directly from the original displacement direction vector is used as the observed value. The observation noise is set according to the direction measurement accuracy of the acoustic-optical fusion positioning, typically corresponding to an angle error range of about 1°, thus effectively eliminating abrupt direction shifts caused by measurement errors and obtaining calibrated trajectory data. Key features are extracted from the calibrated trajectory data to eliminate redundant information and focus on the core trajectory segments that truly pose a threat. The angle between the tangent (direction) of each point on the trajectory and the local axis of the pipeline is calculated. Trajectory segments with a sustained angle less than 15 degrees and very close to the pipeline indicate that the anchor is dragging the pipeline almost parallel to it, representing an extremely high threat segment that should be retained. For trajectory points that are deemed high-threat but have scattered directions or extremely low instantaneous speeds (possibly just a brief pause by the anchor), the standard deviation of the direction angles of 10 consecutive trajectory points is calculated. If it exceeds 15°, it is classified as a trajectory segment with scattered directions. An instantaneous speed threshold of 0.02 m / s is set; if three consecutive points are below this value, it is classified as a trajectory segment with extremely low instantaneous speed. These two types of trajectory segments can be considered redundant information and discarded. The final refined threat trajectory group should consist of a series of core trajectory segments that are continuous in time, precisely directional, and clearly point to pose a direct physical threat to the pipeline, such as dragging, collision, or scraping. This provides the most crucial input for generating the final high-confidence warning.

[0135] In step S18, potential threat trajectories are determined based on the refined threat trajectory group, and extended warning identifiers and alarm records are obtained.

[0136] In one feasible approach, determining potential threat trajectories based on the refined threat trajectory set to obtain extended warning identifiers and alarm records includes:

[0137] Optical signal features are extracted from the refined threat trajectory set, and potential threat trajectories that match the anchor dragging feature are screened by frequency domain transformation.

[0138] Calculate the dynamic distance between the potential threat trajectory and the preset vulnerable points along the pipeline. If the distance between consecutive sampling points shows a decreasing trend and is less than the preset safety warning value, it is confirmed as a valid threat trajectory.

[0139] By combining the geographical coordinates of the pipeline design path and its surrounding pre-set safety buffer zone, the area of ​​threat can be located, and an extended early warning label containing the threat type, location coordinates, dynamic parameters and level can be generated.

[0140] The extended warning identifier is stored and pushed to the operation and maintenance terminal to form a structured alarm record with a timestamp.

[0141] It should be noted that although the refined threat trajectory group has eliminated most of the measurement errors and focused on the core threat segment through the aforementioned steps, it still needs to undergo final multimodal verification and precise spatial relationship analysis to confirm whether it is a real and ongoing anchor damage threat and to generate actionable early warning information. High-frequency optical vibration signal data with synchronized timestamps are extracted from the refined threat trajectory group (collected by the high-frequency optical vibration sensor described in step S11, with a sampling frequency of not less than 500Hz). The frequency domain characteristics are analyzed by performing a short-time Fourier transform (STFT) on the optical signal data. Anchor dragging behavior usually triggers vibration signals in a specific frequency range (anchor chain friction with the seabed is mainly in the range of 5–50Hz, and anchor impact is above 100Hz). The average energy of the optical signal in the 50–150Hz frequency band over the past 30 minutes is calculated in real time as the dynamic background noise baseline. When the current energy exceeds the background noise baseline energy by more than 10 times, potential threat trajectories highly correlated with the refined trajectory can be screened from the optical signal level, achieving multimodal cross-verification of acoustic, optical, and electromagnetic signals, greatly reducing the false alarm rate. The system calculates the dynamic distances between these potential threat trajectories and pre-defined vulnerable points along the pipeline (pipeline welds, flange connections, midpoints of suspended sections, weak points in the anti-corrosion layer, etc.). The vulnerable point database is built based on pipeline design drawings, construction records, and annual inspection reports, containing precise geographic coordinates (WGS-84) and corresponding safety warning values ​​for each point. The system monitors the real-time changes in the distances between trajectory points and these vulnerable points. If five consecutive sampling points show a monotonically decreasing trend in their distances from the vulnerable point, and the latest distance value is less than the pre-defined safety warning value based on pipeline structural strength and safety specifications (1.5 meters for suspended sections and 0.5 meters for buried sections), then this trajectory is confirmed as a valid threat trajectory, indicating that the anchor is continuously approaching and is highly likely to impact a weak point in the pipeline. Based on this, the system combines GIS data from the pipeline design drawings and pre-defined safety buffer zones (usually a three-dimensional geofence with the pipeline centerline as the axis) to accurately calculate the specific geographic coordinates (WGS-84 coordinates) of the threat occurrence. This automatically generates structured extended warning indicators. The identifier is encapsulated in JSON format, with key fields including: unique alert ID, timestamp of threat event and alert generation, threat type, precise location in WGS-84 and pipeline coordinate systems, instantaneous speed, direction, minimum distance from the pipeline and its changing trend, associated vulnerable points, and alert confidence level. This extended alert identifier is published in real-time to a specified message topic via the MQTT protocol, subscribed to and received by the operations and maintenance monitoring terminal, immediately triggering an audible and visual alarm corresponding to the threat level. Simultaneously, the system combines this identifier with additional management status fields (record ID, confirmation status, handling personnel, and handling result remarks) to form a complete structured alarm record, persistently stored in the central database for post-event tracing, performance evaluation, and system optimization.

[0142] Reference Figure 2 The second embodiment of the present invention provides a risk early warning system for anchor damage to subsea water supply pipelines based on multimodal data fusion, comprising:

[0143] The multi-source signal acquisition module acquires the original electromagnetic field signal, acoustic signal data, and optical signal data caused by the movement of the ship's anchor; preprocesses the original electromagnetic field signal to obtain an initial electromagnetic field data set; and obtains the initial electromagnetic field data set based on the original electromagnetic field signal.

[0144] The field-varying signal classification module classifies the initial electromagnetic field data group to obtain a potential field-varying interference signal group and a normal background signal group.

[0145] The interference feature extraction module extracts multi-dimensional feature vectors and filters outliers from the potential field-varying interference signal group to obtain the interference feature set.

[0146] The multimodal feature matching module generates a feature vector set by extracting features based on the acoustic signal data and the optical signal data, and synchronously matches the interference feature set with the feature vector set to obtain the anchor dragging behavior signal group;

[0147] The trajectory parameter calculation module calculates the signal trajectory parameters based on the anchor dragging behavior signal group to obtain the behavior trajectory description set;

[0148] The threat level determination module calculates the distance between the behavior trajectory and the pipeline location based on the behavior trajectory description set. If the distance is less than a preset safe distance threshold, the corresponding behavior trajectory description set is determined to be a high-threat feature group.

[0149] The threat trajectory refinement module calibrates the displacement direction data based on the high-threat feature group to obtain a refined threat trajectory group.

[0150] The warning identifier generation module determines potential threat trajectories based on the refined threat trajectory group, and obtains extended warning identifiers and alarm records.

[0151] It should be noted that the method for early warning of anchor damage to subsea water supply pipelines based on multimodal data fusion provided in this embodiment of the invention is used to execute all the process steps of the method for early warning of anchor damage to subsea water supply pipelines based on multimodal data fusion in the above embodiment. The working principles and beneficial effects of the two are one-to-one, so they will not be described again.

[0152] This invention also provides an electronic device. The electronic device includes a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps described in the above embodiments of the method for early warning of anchorage hazards in submarine water supply pipelines based on multimodal data fusion, for example... Figure 1 The step S11 shown. Alternatively, when the processor executes the computer program, it implements the functions of each module / unit in the above-described device embodiments, such as the data acquisition module.

[0153] For example, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the electronic device.

[0154] The electronic device may be a desktop computer, laptop, handheld computer, or smart tablet, etc. The electronic device may include, but is not limited to, a processor and memory. Those skilled in the art will understand that the above components are merely examples of electronic devices and do not constitute a limitation on the electronic device. It may include more or fewer components than described above, or combine certain components, or different components. For example, the electronic device may also include input / output devices, network access devices, buses, etc.

[0155] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the electronic device, connecting all parts of the electronic device via various interfaces and lines.

[0156] The memory can be used to store the computer programs and / or modules. The processor implements various functions of the electronic device by running or executing the computer programs and / or modules stored in the memory and by calling data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0157] Wherein, if the modules / units integrated in the electronic device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0158] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0159] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A method for early warning of anchor damage risk in subsea water supply pipelines based on multimodal data fusion, characterized in that, include: Acquire the raw electromagnetic field signal, acoustic signal data, and optical signal data caused by the movement of the ship's anchor; preprocess the raw electromagnetic field signal to obtain the initial electromagnetic field data set. The initial electromagnetic field data set is classified to obtain a potential field-varying interference signal set and a normal background signal set; Based on the potential field-varying interference signal group, multi-dimensional feature vectors are extracted and outliers are filtered to obtain the interference feature set; Based on the acoustic signal data and the optical signal data, a feature vector set is generated through feature extraction. The interference feature set is then synchronously matched with the feature vector set to obtain the anchor dragging behavior signal group. Based on the anchor dragging behavior signal set, calculate the signal trajectory parameters to obtain the behavior trajectory description set; Based on the behavior trajectory description set, the distance between the behavior trajectory and the pipeline location is calculated. If the distance is less than a preset safe distance threshold, the corresponding behavior trajectory description set is identified as a high-threat feature group. Based on the high-threat feature group, the displacement direction data is calibrated to obtain a refined threat trajectory group; Based on the refined threat trajectory group, potential threat trajectories are determined, and extended warning indicators and alarm records are obtained; The step of calculating signal trajectory parameters based on the anchor dragging behavior signal group to obtain a behavior trajectory description set includes: The anchor dragging behavior signal group is subjected to noise reduction preprocessing to obtain a standardized signal; Based on the initial trajectory data of the anchor calculated by the standardized signal, the trajectory parameters are extracted after filtering and smoothing, including the displacement direction, instantaneous velocity and velocity change rate corresponding to each time stamp, to obtain smooth trajectory data; A pipeline coordinate system is established based on the design axis of the submarine water supply pipeline and combined with GPS positioning points along the pipeline. The smooth trajectory data is mapped to the pipeline coordinate system to obtain the real-time distance sequence to key locations on the pipeline; By integrating the smooth trajectory data and the real-time distance sequence, a behavioral trajectory description set is obtained; Specifically, the Euclidean distance between each point in the trajectory and the key position of the pipeline is calculated to generate a real-time distance sequence that changes over time; the smoothed trajectory data and the real-time distance sequence are integrated and time-aligned to form a complete behavioral trajectory description set, which comprehensively describes the spatiotemporal dynamic behavior of the anchor dragging and its relative positional relationship with the pipeline. The system compares each distance value in the dynamic distance sequence with a preset safe distance threshold in real time. If the distance value of any sampling point in the sequence is found to be less than the safe distance threshold, the behavior trajectory description set is immediately marked as a high-threat feature group. If the distance values ​​of all sampling points in the sequence are continuously greater than the safe distance threshold, the behavior trajectory description set is determined to be a low-threat feature group. The system outputs the determination result along with specific timestamps, minimum distance values, and location information, triggering the corresponding warning level or entering the log for subsequent analysis.

2. The method for early warning of anchor damage risk in subsea water supply pipelines based on multimodal data fusion according to claim 1, characterized in that, The process involves acquiring the raw electromagnetic field signal, acoustic signal data, and optical signal data generated by the ship's anchor movement, and preprocessing the raw electromagnetic field signal to obtain an initial electromagnetic field data set, including: The original electromagnetic field signal caused by the ship's anchor movement was obtained by collecting the real-time time series of electromagnetic field signals, and the acoustic signal data collected by the acoustic sensor and the optical signal data collected by the optical sensor were also obtained. The original electromagnetic field signal is denoised to obtain a denoised electromagnetic field dataset. If the intensity value in the denoised electromagnetic field dataset exceeds a preset intensity threshold, then the frequency components are extracted by fast Fourier transform to obtain a frequency feature set. Based on the frequency feature set, extract the real-time time series from the denoised electromagnetic field dataset that matches the frequency feature set to obtain the initial electromagnetic field data set.

3. The method for early warning of anchor damage risk in subsea water supply pipelines based on multimodal data fusion according to claim 1, characterized in that, The classification process of the initial electromagnetic field data group to obtain a potential field-varying interference signal group and a normal background signal group includes: The initial electromagnetic field data set is subjected to time-domain analysis and frequency-domain transformation respectively to extract time-domain statistical features. After converting the time-domain signal into a frequency-domain signal through fast Fourier transform, frequency-domain spectral features are extracted. The time-domain statistical features and the frequency-domain spectral features are integrated to form a multi-dimensional feature vector. The multidimensional feature vector is input into a preset signal classification model to obtain the classification result; The signal sequences marked as abnormal disturbances in the classification results are divided into the potential field-variable interference signal group, and the remaining signal sequences marked as background noise are divided into the normal background signal group.

4. The method for early warning of anchor damage risk in subsea water supply pipelines based on multimodal data fusion according to claim 1, characterized in that, The step of extracting multidimensional feature vectors and filtering outliers based on the potential field-varying interference signal set to obtain an interference feature set includes: The original electromagnetic field signal, acoustic signal data, and optical signal data collected during the same historical period and in the same sea area under the ship anchor static state and non-anchor interference state are obtained, a feature set of normal feature distribution is established, and feature vector data of historical normal signals are obtained. Multidimensional feature vectors are extracted from the potential field-varying interference signal group, and a benchmark feature set of normal feature distribution is established based on the feature vector data of the historical normal signals. Calculate the deviation of each feature vector in the current multidimensional feature vector from the benchmark feature set; The feature vector whose deviation value exceeds the preset deviation threshold is determined as an abnormal data point, and the data points determined as abnormal are deleted from the current multidimensional feature vector to obtain the interference feature set.

5. The method for early warning of anchor damage risk in subsea water supply pipelines based on multimodal data fusion according to claim 1, characterized in that, The step of generating a feature vector set through feature extraction based on the acoustic signal data and the optical signal data, and synchronously matching the interference feature set with the feature vector set to obtain the anchor dragging behavior signal group includes: The acoustic signal data and the optical signal data are time-axis aligned to obtain a synchronization signal dataset. Based on the synchronization signal dataset, acoustic collision features from the acoustic signal data and electromagnetic field disturbance features from the optical signal data are extracted to generate a feature vector set. The interference feature set and the feature vector set are synchronously matched, and the comprehensive matching degree of the electromagnetic field disturbance feature, the acoustic collision feature, and the optical motion feature is calculated. When the comprehensive matching degree exceeds the preset matching degree threshold, it is determined to be an anchor dragging behavior signal group.

6. The method for early warning of anchor damage risk in subsea water supply pipelines based on multimodal data fusion according to claim 1, characterized in that, The step involves calculating the distance between the behavior trajectory and the pipeline location based on the behavior trajectory description set. If the distance is less than a preset safe distance threshold, the corresponding behavior trajectory description set is identified as a high-threat feature group, including: Based on the spatial coordinate data in the behavior trajectory description set, and based on the correspondence between the spatial coordinate data and the pipeline coordinate system, the shortest distance values ​​between each sampling point on the trajectory and the pipeline centerline and pipeline protection boundary are calculated in real time to form a dynamic distance sequence. The dynamic distance sequence is compared with a preset safe distance threshold. If the distance value of any sampling point in the dynamic distance sequence is less than the preset safe distance threshold, the behavior trajectory description set is determined to be a high-threat feature group. If the distance values ​​of all sampling points in the dynamic distance sequence are greater than the preset safe distance threshold, the behavior trajectory description set is determined to be a low-threat feature group.

7. The method for early warning of anchor damage risk in subsea water supply pipelines based on multimodal data fusion according to claim 6, characterized in that, The step of calibrating displacement direction data based on the high-threat feature group to obtain a refined threat trajectory group includes: Extract the original displacement direction vector and corresponding timestamp and velocity parameters from the high-threat feature group to construct a direction-time associated dataset; Based on the pipeline coordinate system, abnormal direction points in the direction-time correlation dataset are identified and corrected to eliminate direction offsets caused by measurement errors and obtain calibrated trajectory data. Key features are extracted from the calibrated trajectory data, redundant trajectory information is removed, and a refined threat trajectory group containing precise direction parameters and core threat trajectory segments is formed.

8. The method for early warning of anchor damage risk in subsea water supply pipelines based on multimodal data fusion according to claim 1, characterized in that, The step of determining potential threat trajectories based on the refined threat trajectory group, and obtaining extended early warning identifiers and alarm records, includes: Optical signal features are extracted from the refined threat trajectory set, and potential threat trajectories that match the anchor dragging feature are screened by frequency domain transformation. Calculate the dynamic distance between the potential threat trajectory and the preset vulnerable points along the pipeline. If the distance between consecutive sampling points shows a decreasing trend and is less than the preset safety warning value, it is confirmed as a valid threat trajectory. By combining the geographical coordinates of the pipeline design path and its surrounding pre-set safety buffer zone, the area of ​​threat can be located, and an extended early warning label containing the threat type, location coordinates, dynamic parameters and level can be generated. The extended warning identifier is stored and pushed to the operation and maintenance terminal to form a structured alarm record with a timestamp.

9. A risk early warning system for anchor damage to subsea water supply pipelines based on multimodal data fusion, characterized in that, For implementing the method as described in any one of claims 1-8, comprising: The multi-source signal acquisition module acquires the original electromagnetic field signal, acoustic signal data, and optical signal data caused by the movement of the ship's anchor; preprocesses the original electromagnetic field signal to obtain an initial electromagnetic field data set; and obtains the initial electromagnetic field data set based on the original electromagnetic field signal. The field-varying signal classification module classifies the initial electromagnetic field data group to obtain a potential field-varying interference signal group and a normal background signal group. The interference feature extraction module extracts multi-dimensional feature vectors and filters outliers from the potential field-varying interference signal group to obtain the interference feature set. The multimodal feature matching module generates a feature vector set by extracting features based on the acoustic signal data and the optical signal data, and synchronously matches the interference feature set with the feature vector set to obtain the anchor dragging behavior signal group; The trajectory parameter calculation module calculates the signal trajectory parameters based on the anchor dragging behavior signal group to obtain the behavior trajectory description set; The threat level determination module calculates the distance between the behavior trajectory and the pipeline location based on the behavior trajectory description set. If the distance is less than a preset safe distance threshold, the corresponding behavior trajectory description set is determined to be a high-threat feature group. The threat trajectory refinement module calibrates the displacement direction data based on the high-threat feature group to obtain a refined threat trajectory group. The warning identifier generation module determines potential threat trajectories based on the refined threat trajectory group, and obtains extended warning identifiers and alarm records.

Images