Buoy collision and sea wave impact identification method and system based on spectral features

Abstract

This invention provides a method and system for identifying buoy collisions and wave impacts based on spectral features, relating to the field of marine monitoring equipment event recognition technology. First, it acquires a set of vibration signals containing mechanical response information generated by the buoy under actual operating conditions. For each vibration waveform data in the set, it generates a spectral feature evolution trajectory composed of multiple consecutive time-point spectral features arranged in chronological order. Each time-point spectral feature includes frequency, amplitude, and morphological characteristics. Next, based on the spectral feature evolution trajectory and the correspondence between known events, it constructs a trajectory event mapping model, generating a dynamic identification threshold for collisions and wave impacts that changes over time and adapts to the temporal characteristics of the spectral features. Finally, it generates a real-time spectral feature evolution trajectory for the vibration signals collected by the buoy in real time, compares it with the dynamic identification threshold, and obtains the identification result of the buoy collision or wave impact event, effectively improving the accuracy and real-time performance of the identification.

Description

Technical Field

[0001] This invention relates to the field of marine monitoring equipment event recognition technology, and more specifically, to a buoy collision and wave impact recognition method and system based on spectral characteristics. Background Technology

[0002] In the field of marine monitoring, buoys, as important monitoring equipment, are widely deployed in the marine environment to collect data on various aspects of marine meteorology and hydrology. However, buoys are affected by a variety of external forces while operating in the ocean, among which buoy collisions and wave impacts are relatively common and have a significant impact on the normal operation of buoys and the accuracy of data. Accurately identifying buoy collision and wave impact events is crucial for taking timely maintenance measures, ensuring the stable operation of buoys, and obtaining reliable marine monitoring data.

[0003] Existing buoy event identification methods are mostly based on simple physical parameter thresholds, such as setting fixed acceleration or displacement thresholds to distinguish between collisions and wave impacts. However, these methods have significant limitations. The marine environment is complex and variable; the intensity and frequency of waves, as well as the force and direction of collisions, are all uncertain. Fixed thresholds are difficult to adapt to various complex conditions, resulting in low identification accuracy and a high likelihood of false positives and false negatives. Furthermore, some methods only consider the signal characteristics at a single moment, failing to fully account for the dynamic changes of the signal over time. This makes it impossible to accurately capture the complete characteristics of the event, hindering effective and precise identification of buoy collisions and wave impacts. Summary of the Invention

[0004] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide a buoy collision and wave impact identification method based on spectral features, the method comprising: The set of vibration signals generated by the buoy under actual operating conditions is obtained. The set of vibration signals includes vibration waveform data continuously collected by the buoy under different external forces. The vibration waveform data carries the mechanical response information of the buoy when it is hit by a collision or wave impact. For each vibration waveform data in the vibration signal set, a corresponding spectral feature evolution trajectory is generated. The spectral feature evolution trajectory is composed of spectral features at multiple consecutive time points in chronological order. The spectral features at each time point include frequency distribution features, amplitude variation features, and spectral morphology features. Based on the correspondence between the spectral feature evolution trajectory and known buoy collision events and wave impact events, a trajectory event mapping model is constructed. The trajectory event mapping model can establish the correlation between the dynamic change pattern of the spectral feature evolution trajectory and specific event types. Based on the trajectory event mapping model, dynamic identification thresholds for collision events corresponding to buoy collision events and dynamic identification thresholds for impact events corresponding to wave impact events are generated. Both the dynamic identification thresholds for collision events and the dynamic identification thresholds for impact events change over time and are adapted to the temporal characteristics of the spectral feature evolution trajectory. The real-time vibration signal collected by the buoy is used to generate a corresponding real-time spectral feature evolution trajectory. The real-time spectral feature evolution trajectory is compared with the collision event dynamic identification threshold and the impact event dynamic identification threshold, respectively. The buoy collision event identification result or wave impact event identification result is obtained based on the comparison result.

[0005] Furthermore, embodiments of the present invention also provide a buoy collision and wave impact identification system based on spectral characteristics, comprising: A processor; a machine-readable storage medium for storing machine-executable instructions of the processor; wherein the processor is configured to execute the above-described spectral feature-based buoy collision and wave impact identification method by executing the machine-executable instructions.

[0006] In another aspect, embodiments of the present invention also provide a computer program product, the computer program product including machine-executable instructions, the machine-executable instructions being stored in a computer-readable storage medium, a processor of a computer device reading the machine-executable instructions from the computer-readable storage medium, the processor executing the machine-executable instructions, causing the computer device to execute the above-described buoy collision and wave impact identification method based on spectral characteristics.

[0007] Based on the above, by acquiring a set of vibration signals containing mechanical response information of buoys under actual operating conditions and generating corresponding spectral feature evolution trajectories, the vibration characteristics of buoys under different external forces are characterized. A trajectory event mapping model constructed based on the correspondence between spectral feature evolution trajectories and known events can deeply establish the association between dynamic changes in spectral features and specific event types, effectively mining the implicit information in the signals. The dynamic identification threshold generated based on this model, which varies over time and is adapted to the temporal characteristics of the spectral features, overcomes the limitations of traditional fixed thresholds and can flexibly adjust the identification criteria according to the dynamic changes in the buoy's environment and stress conditions. Real-time spectral feature evolution trajectories are generated from real-time vibration signals and compared with the dynamic identification thresholds, achieving high-precision, real-time identification of buoy collisions and wave impact events, greatly improving the accuracy and reliability of identification. Attached Figure Description

[0008] Figure 1 This is a schematic diagram of the execution flow of the buoy collision and wave impact identification method based on spectral features provided in an embodiment of the present invention.

[0009] Figure 2 This is a schematic diagram of exemplary hardware and software components of the buoy collision and wave impact identification system based on spectral characteristics provided in an embodiment of the present invention. Detailed Implementation

[0010] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1 This is a flowchart illustrating a buoy collision and wave impact identification method based on spectral features provided in an embodiment of the present invention. The following is a detailed description of this buoy collision and wave impact identification method based on spectral features.

[0011] Step S110: Obtain the set of vibration signals generated by the buoy under actual operating conditions. The set of vibration signals includes vibration waveform data continuously collected by the buoy under different external forces. The vibration waveform data carries the mechanical response information of the buoy when it is hit by a collision or wave impact.

[0012] In this embodiment, taking an ocean monitoring buoy as an example, the vibration sensors (such as acceleration sensors, displacement sensors, etc.) mounted on the buoy continuously collect vibration signals at a fixed sampling frequency. When the buoy is operating in the marine environment, if a ship approaches and collides with the buoy, or if it encounters waves of different levels (such as the wave levels corresponding to Beaufort scale), the vibration sensors will record the vibration of the buoy under these external forces in the form of waveform data. The aforementioned vibration waveform data includes mechanical response information related to the buoy's displacement, acceleration, and velocity when subjected to force, such as the instantaneous acceleration change of the buoy during a collision, and the periodic displacement change of the buoy during wave impact.

[0013] Step S120: Generate a corresponding spectral feature evolution trajectory for each vibration waveform data in the vibration signal set. The spectral feature evolution trajectory is composed of spectral features at multiple consecutive time points in chronological order. The spectral features at each time point include frequency distribution features, amplitude variation features, and spectral morphology features.

[0014] Step S121: Perform time axis calibration on each vibration waveform data in the vibration signal set, and divide each calibrated vibration waveform data into multiple continuous signal segments according to a preset time interval, wherein the time interval is determined according to the change frequency of the buoy vibration signal.

[0015] In this embodiment, for the vibration waveform data of the aforementioned marine monitoring buoy, time axis calibration is achieved by combining a high-precision clock (such as a crystal oscillator clock) inside the buoy with satellite timing (such as GPS timing) signals. The clock time is compared with the timing signal, the time deviation is calculated, and the time stamp of the vibration waveform data is corrected to ensure the accuracy of the time stamp for each data point. Then, the frequency variation of the buoy vibration signal is analyzed. If the frequency components of the signal change significantly within a short period (e.g., within 1 second) (e.g., rapid switching between multiple vibration components of different frequencies), the time interval is set to a smaller value (e.g., 0.01 seconds); if the frequency components change slowly (e.g., mainly periodic vibrations of a single frequency), the time interval is set to a larger value (e.g., 0.1 seconds). The calibrated vibration waveform data is divided into multiple continuous signal segments according to this time interval, each segment corresponding to a specific time interval, facilitating subsequent analysis.

[0016] Step S122: Perform a quality screening operation on each divided signal segment to obtain valid signal segments, and perform a frequency domain transformation operation on each valid signal segment to convert the valid signal segment in the time domain into frequency domain data. The frequency domain transformation operation retains the amplitude and phase information of the signal segment at different frequency points.

[0017] Step S1221: Obtain the time-domain parameters of each valid signal segment, including the sampling frequency, number of sampling points, and signal duration of the signal segment.

[0018] In this embodiment, for each segmented signal, quality screening is achieved by setting signal quality judgment criteria. For example, the signal-to-noise ratio (SNR) (the ratio of signal power to noise power) is calculated. When the SNR is greater than a preset threshold (e.g., 10dB) and the signal has no obvious distortion (e.g., by checking for data loss, jumps, etc. through waveform integrity detection), it is determined to be a valid signal segment. For valid signal segments, the sampling frequency is determined by the acquisition settings of the vibration sensor (e.g., the sensor is set to 1000Hz sampling), the number of sampling points is the number of samples contained in the signal segment (e.g., if the time interval is 0.01 seconds, the number of sampling points is 10 for 1000Hz sampling), and the signal duration is the number of sampling points divided by the sampling frequency (e.g., 10 / 1000 = 0.01 seconds).

[0019] Step S1222: Determine the frequency resolution after frequency domain conversion based on the sampling frequency and number of sampling points of the signal segment. The frequency resolution reflects the interval between adjacent frequency points in the frequency domain data.

[0020] In this embodiment, the frequency resolution is calculated by dividing the sampling frequency by the number of sampling points. For example, if the sampling frequency is 1000Hz and the number of sampling points is 10, then the frequency resolution is 1000 / 10 = 100Hz, meaning that the interval between two adjacent frequency points in the frequency domain data is 100Hz.

[0021] Step S1223: Input the time-domain data of the effective signal segment into the selected frequency-domain transformation algorithm, and perform Fourier transform operation on the time-domain data according to the operation flow of the frequency-domain transformation algorithm to obtain preliminary frequency-domain data.

[0022] In this embodiment, the Fast Fourier Transform (FFT) algorithm is selected as the frequency domain conversion algorithm. The time-domain data of the effective signal segment (such as discrete data with 10 sampling points) is input into the FFT algorithm. According to the algorithm flow, complex number operations are performed on the time-domain data to convert the discrete signal in the time domain into a complex sequence in the frequency domain, thus obtaining preliminary frequency domain data. This preliminary frequency domain data contains complex values ​​(composed of real and imaginary parts) at different frequency points.

[0023] Step S1224: Perform amplitude correction on the initial frequency domain data. Based on the amplitude scaling characteristics of the frequency domain conversion algorithm, adjust the amplitude of the initial frequency domain data to the actual physical amplitude.

[0024] In this embodiment, the FFT algorithm exhibits amplitude scaling characteristics; for an N-point FFT, the amplitude scaling factor is 1 / N. For example, if the complex amplitude of the initial frequency domain data (calculated through modulo arithmetic, i.e., √(real part² + imaginary part²)) is A, and the number of sampling points is N=10, then the actual physical amplitude is A×N=10A. The amplitude of the initial frequency domain data is corrected in this way to obtain the actual physical amplitude.

[0025] Step S1225: Perform frequency axis calibration on the corrected frequency domain data. Based on the sampling frequency and the number of sampling points, determine the actual frequency value corresponding to each frequency domain data point so that the scale of the frequency axis is consistent with the actual frequency.

[0026] In this embodiment, the actual frequency value corresponding to each frequency domain data point is determined based on the sampling frequency (f_s=1000Hz), the number of sampling points (N=10), and the frequency resolution (Δf=100Hz). The frequency value corresponding to the first frequency domain data point (index 0) is 0Hz, the second (index 1) is Δf×1=100Hz, the third (index 2) is Δf×2=200Hz, and so on, until the frequency domain data point with index N-1, which corresponds to a frequency value of Δf×(N-1), so that the frequency axis scale is consistent with the actual frequency.

[0027] Step S1226: Extract the effective frequency range from the calibrated frequency domain data, convert the data format of the frequency domain data within the effective frequency range, store the converted frequency domain data, and mark the corresponding effective signal segment number and acquisition time for each frequency domain data file.

[0028] In this embodiment, the extraction of the effective frequency range is achieved by analyzing the energy distribution of the signal. The energy (square of the amplitude) of each frequency point is calculated. When the energy is greater than a certain proportion (e.g., 0.1%) of the total energy, the frequency point is considered to belong to the effective frequency range. For example, the total energy is the sum of the energies of all frequency points. If the energy of a certain frequency point accounts for more than 0.1% of the total energy, it is included in the effective frequency range. The frequency domain data within the effective frequency range is converted into binary format (e.g., binary storage of a float array), and each frequency domain data file is labeled with an effective signal segment number (e.g., numbered in chronological order as S1, S2, etc.) and the acquisition time (e.g., 2024-01-01 12:00:00.000).

[0029] Step S123: Extract the spectral features of each effective signal segment from the frequency domain data. The spectral features include frequency distribution features, amplitude variation features, and spectral morphology features. The frequency distribution features reflect the distribution of signal energy in different frequency ranges, the amplitude variation features reflect the trend of amplitude value changes at different frequency points, and the spectral morphology features reflect the overall waveform profile of the frequency domain data.

[0030] In this embodiment, frequency distribution characteristics are extracted by dividing the effective frequency range into multiple sub-intervals (such as 0-100Hz, 100-200Hz, etc.) and calculating the proportion of the sum of energy in each sub-interval to the total energy. For example, if the total energy is E_total and the energy in the sub-interval 0-100Hz is E_0-100, then the energy proportion of this sub-interval is E_0-100 / E_total. Amplitude variation characteristics are extracted by analyzing the change in amplitude values ​​at different frequency points with frequency, such as plotting amplitude-frequency curves and observing the rising, falling, or fluctuating trends of the curves. Spectral morphology characteristics are extracted by observing the overall waveform of the frequency domain data, such as determining whether it is a single peak (one main frequency point with an amplitude much larger than other points), multiple peaks (multiple main frequency points), or a continuous spectrum (no obvious peaks, uniform energy distribution).

[0031] Step S124: Perform dimension completion operation on the spectral features of each extracted valid signal segment to obtain the completed spectral feature data. Then, sort the completed spectral feature data in time sequence according to the time order of the valid signal segments to form a preliminary time sequence feature sequence.

[0032] In this embodiment, dimension completion targets missing spectral feature dimensions (such as the loss of energy proportion data in a certain frequency range) and employs linear interpolation. For example, if the energy proportion in the 100-200Hz range is missing in the frequency distribution characteristics of a valid signal segment, linear interpolation is calculated based on the energy proportions of adjacent valid signal segments (the previous and the next) in this range (e.g., the previous one is 0.2, the next one is 0.4, and the time interval is T, then the current missing value is 0.2 + (0.4 - 0.2) × t / T, where t is the time difference between the current segment and the previous segment). After completion, the spectral feature data are arranged according to the chronological order of the valid signal segments to form a preliminary time-series feature sequence, such as [features of S1, features of S2, ..., features of Sn].

[0033] Step S125: In the preliminary time-series feature sequence, mark the specific time node corresponding to each spectral feature. The time node is the midpoint of the acquisition time of the effective signal segment to which the spectral feature belongs, forming the basic node of the trajectory.

[0034] In this embodiment, the acquisition time of the effective signal segment is from t_start to t_end, and the midpoint of the acquisition time is t_mid = (t_start + t_end) / 2. For example, the acquisition time of S1 is from 12:00:00.000 to 12:00:00.010, and the midpoint is 12:00:00.005. This time is used as the time marker of the S1 spectral feature to form the basic node (t_mid, feature data). All basic nodes are arranged in chronological order to form the basic framework of the spectral feature evolution trajectory.

[0035] Step S126: Perform temporal coherence operation on the basic nodes, calculate the feature change rate between adjacent basic nodes, and if the change rate exceeds the preset coherence standard, insert transition nodes by linear interpolation, and perform smoothness optimization operation on the trajectory containing transition nodes. Use the moving average algorithm to smooth the node feature data in the trajectory, and finally generate the spectral feature evolution trajectory.

[0036] In this embodiment, the feature change rate is calculated by dividing the difference in feature data between adjacent basic nodes by the time interval. For example, the feature data of basic node 1 is F1, and the time is t1; the feature data of basic node 2 is F2, and the time is t2, with a time interval of Δt = t2 - t1, and the feature change rate is (F2 - F1) / Δt. The preset coherence standard is that the absolute value of the feature change rate is less than V_max (e.g., determined based on a reasonable change rate range statistically derived from historical data). If the change rate exceeds V_max, a transition node is inserted between t1 and t2. The time of the transition node is t1 + Δt × k (k is a coefficient between 0 and 1, such as k = 0.2, 0.4, etc.), and the feature data is F1 + (F2 - F1) × k. Then, a moving average algorithm is used to select m consecutive nodes (e.g., m = 5), calculate their average feature data, and use it as the smoothed data for the intermediate nodes. All nodes are processed sequentially to smooth the trajectory feature changes, ultimately generating a spectral feature evolution trajectory.

[0037] Step S130: Based on the correspondence between the spectrum feature evolution trajectory and known buoy collision events and wave impact events, construct a trajectory event mapping model. The trajectory event mapping model can establish the correlation between the dynamic change pattern of the spectrum feature evolution trajectory and specific event types.

[0038] Step S131: Obtain multiple collision reference spectrum feature evolution trajectories corresponding to known buoy collision events. The collision reference spectrum feature evolution trajectories are generated from the vibration signals collected by the buoy in the controlled collision experiment. Each collision reference spectrum feature evolution trajectory is marked with corresponding collision intensity and collision direction information.

[0039] In this embodiment, the controlled collision experiment is conducted in a laboratory tank or marine test site. Objects of different masses (e.g., m1, m2, etc.) and velocities (e.g., v1, v2, etc.) are used to collide with the buoy from different directions (e.g., front, side, 45° angle, etc.). Vibration sensors collect vibration signals, and a spectral feature evolution trajectory is generated according to step S120. Each trajectory is marked with collision intensity (e.g., the momentum of the colliding object p=m×v) and collision direction information (e.g., expressed in angle), serving as a collision reference trajectory.

[0040] Step S132: Obtain multiple impact reference spectrum feature evolution trajectories corresponding to known wave impact events. The impact reference spectrum feature evolution trajectories are generated from the vibration signals collected by the buoy in different wave levels. Each impact reference spectrum feature evolution trajectory is marked with the corresponding wave level and impact duration information.

[0041] In this embodiment, in a marine environment, when the wave level is level 1, 2, 3, etc., the buoy continuously collects vibration signals in this environment (e.g., for 10 minutes), and generates a spectral feature evolution trajectory according to step S120. Each trajectory is marked with the wave level (according to the Beaufort scale, such as a level 1 wave corresponding to a wave height of 0.1-0.5 meters) and the impact duration (e.g., the duration from t_start to t_end), serving as an impact reference trajectory.

[0042] Step S133: Perform data cleaning operations on the collision reference spectrum feature evolution trajectory and the impact reference spectrum feature evolution trajectory, and extract the trajectory morphology features in the cleaned collision reference spectrum feature evolution trajectory. The trajectory morphology features include the overall trend direction of the trajectory, the frequency of occurrence of feature peaks, the feature change amplitude between adjacent peaks, and the stationarity parameters of the feature data.

[0043] Step S1331: Divide the cleaned collision reference spectrum feature evolution trajectory into time nodes. Divide the trajectory into multiple continuous trajectory segments according to the preset division interval. Each trajectory segment corresponds to a fixed time length.

[0044] In this embodiment, data cleaning removes noisy data (such as outliers caused by sensor malfunctions) and duplicate data (such as trajectory points with repeated timestamps). Then, the trajectory is divided into multiple trajectory segments according to a preset interval (such as 10 seconds), with each segment corresponding to a time length of 10 seconds, which facilitates the analysis of the trajectory pattern within the segment.

[0045] Step S1332: Extract the spectral feature data of all time nodes within each trajectory segment, including frequency distribution feature data, amplitude change feature data, and spectral morphology feature data.

[0046] In this embodiment, for each trajectory segment, the frequency distribution features (such as the energy proportion of each frequency interval), amplitude change features (such as amplitude-frequency curve data), and spectral morphology features (such as waveform contour description) of all time nodes are extracted as the basic feature data of that segment.

[0047] Step S1333: Calculate the rate of change of frequency distribution characteristic data within each trajectory segment. The rate of change of frequency distribution characteristic is obtained by calculating the difference between the frequency distribution characteristics of the start time node and the end time node within the trajectory segment and then dividing it by the time length of the trajectory segment. Also, calculate the rate of change of amplitude change characteristic data and the rate of change of spectral morphology characteristic data within each trajectory segment.

[0048] In this embodiment, the rate of change of frequency distribution characteristic data is calculated as follows: Frequency distribution characteristic data (such as the energy percentage of each frequency interval) at the start time node (t_s) and end time node (t_e) of the trajectory segment are selected, and the difference (ΔF = F_e - F_s) is calculated. The trajectory segment time length is Δt = t_e - t_s, and the rate of change is ΔF / Δt. Similarly, the rate of change of amplitude change characteristic data is calculated as (ΔA = A_e - A_s) / Δt, and the rate of change of spectral morphology characteristic data is calculated as (ΔM = M_e - M_s) / Δt (ΔM is the quantization difference of the morphological characteristics, such as the quantization representation of a single peak becoming multiple peaks).

[0049] Step S1334: Extract feature data from two adjacent trajectory segments, calculate the feature data difference between the end node of the previous trajectory segment and the start node of the next trajectory segment, and use this feature data difference as the feature change amplitude between adjacent trajectory segments.

[0050] In this embodiment, the end node feature data of adjacent trajectory segment 1 are F1_e, A1_e, and M1_e, and the start node feature data of trajectory segment 2 are F2_s, A2_s, and M2_s. The differences ΔF=F2_s-F1_e, ΔA=A2_s-A1_e, and ΔM=M2_s-M1_e are calculated as the feature change amplitude of adjacent segments.

[0051] Step S1335: Perform peak detection on the spectral feature data within each trajectory segment to identify the time node when the data value in each feature dimension reaches a local maximum value. The time node is used as the feature peak position in the trajectory.

[0052] In this embodiment, for frequency distribution characteristics, the time nodes corresponding to the local maximum values ​​of energy proportion in each frequency range are detected; for amplitude variation characteristics, the time nodes corresponding to the local maximum values ​​of amplitude values ​​are detected; for spectral morphology characteristics, the time nodes corresponding to the local maximum values ​​of waveform contour changes (such as the number and position of peaks) are detected. For example, a peak detection algorithm (such as local maximum detection based on a sliding window) is used to identify the peak positions of each feature dimension.

[0053] Step S1336: Record the specific timestamp and feature data value corresponding to each feature peak position to form a peak information list.

[0054] In this embodiment, for each characteristic peak position, its timestamp (e.g., t_peak) and corresponding characteristic data values ​​(e.g., energy proportion of frequency distribution characteristics, amplitude value, quantization value of morphological characteristics) are recorded and organized into a peak information list, such as [(t_peak1, F1, A1, M1), (t_peak2, F2, A2, M2), ...].

[0055] Step S1337: Sort the peak timestamps in the peak information list, calculate the time difference between two adjacent peak timestamps to obtain the peak interval time, and arrange all peak interval times in chronological order to form a peak interval sequence.

[0056] In this embodiment, the peak timestamps are arranged in ascending order, and the difference between adjacent timestamps (such as t_peak2-t_peak1, t_peak3-t_peak2, etc.) is calculated to obtain the peak interval time, forming a sequence [Δt1, Δt2, ...].

[0057] Step S1338: Perform stationarity analysis on the peak interval sequence. By calculating the mean and variance of the peak interval sequence, determine whether the peak interval sequence is stationary. If the peak interval sequence is not stationary, use differencing to make the peak interval sequence stationary.

[0058] In this embodiment, the mean (μ=(Δt1+Δt2+…+Δtn) / n) and variance (σ²=Σ(Δti-μ)² / n) of the peak-interval sequence are calculated. If the mean and variance change with time (e.g., the mean is μ1 in the first half and μ2 in the second half, μ1≠μ2), the sequence is not stationary. In this case, the sequence is differentially processed, and the difference between adjacent time intervals is calculated (e.g., Δt2-Δt1, Δt3-Δt2, etc.), generating a new sequence that makes the mean and variance of the new sequence tend to be stable.

[0059] Step S1339: Statistically analyze the frequency of occurrence of different intervals in the peak interval sequence after stabilization, form an interval frequency distribution table, and determine the main interval range and typical interval value of peak occurrence based on the interval frequency distribution table, which together constitute the interval pattern of peak occurrence in the trajectory.

[0060] In this embodiment, the frequency of occurrence of each interval in the peak interval sequence after stabilization is counted, and the frequency of occurrence (number of occurrences / total number of occurrences) is calculated to form an interval frequency distribution table. For example, the frequency of occurrence of an interval of 1 second is 0.3, and that of an interval of 2 seconds is 0.5, etc. Based on the distribution table, the main interval range (such as the interval range with an occurrence frequency greater than 0.1, such as 1-3 seconds) and the typical interval value (the interval with the highest occurrence frequency, such as 2 seconds) are determined, thus forming the peak interval pattern.

[0061] Step S13310: Integrate the characteristic change rate of each trajectory segment, the characteristic change amplitude of adjacent trajectory segments, the characteristic peak position information and the peak interval pattern to form the trajectory morphology features of the collision reference spectrum feature evolution trajectory. Perform dimensional normalization operation on the integrated trajectory morphology features, and assign a unique identifier to the normalized trajectory morphology features, assigning a unique feature number to each feature dimension.

[0062] In this embodiment, the characteristic change rate of each trajectory segment (such as the change rate of frequency, amplitude, and shape), the characteristic change amplitude of adjacent segments, the characteristic peak position information (timestamp and data value), and the peak interval pattern (main range and typical value) are integrated to form trajectory shape features. Dimension normalization adopts min-max normalization, which transforms the value of each feature dimension to the interval [0, 1] (e.g., feature value x, minimum value x_min, maximum value x_max, after normalization x'=(x-x_min) / (x_max-x_min)). Then, a unique number is assigned to each normalized feature dimension (e.g., F1, F2, A1, A2, M1, M2, P1, P2, etc.) to facilitate subsequent processing.

[0063] Step S134: Extract the trajectory morphology features from the cleaned impact reference spectrum feature evolution trajectory, perform cross-event difference quantization on the trajectory morphology features of the collision reference spectrum feature evolution trajectory and the impact reference spectrum feature evolution trajectory, calculate the difference coefficient of the two trajectories in each morphology feature dimension, and determine the key morphology features with the highest distinguishability for event types.

[0064] In this embodiment, the trajectory morphology features of the impact reference trajectory are extracted according to the method in step S133. Cross-event differential quantization is performed by calculating the difference coefficients of the collision and impact trajectories in each morphological feature dimension, such as Euclidean distance (for numerical features, the distance between the feature vectors of the two trajectories is calculated) or cosine similarity (for vector features, the cosine of the included angle is calculated). For example, the feature vector of the collision trajectory is [F1_c, F2_c, ...], and that of the impact trajectory is [F1_i, F2_i, ...], with a difference coefficient of √Σ(Fj_c-Fj_i)². Based on the magnitude of the difference coefficient, the key morphological features with the highest discriminative power (such as feature dimensions with a difference coefficient greater than a preset threshold) are selected.

[0065] Step S135: Quantify the key morphological features of the collision reference spectrum feature evolution trajectory with the corresponding buoy collision events, and assign correlation weights to different key morphological features according to the collision intensity and collision direction information to form a collision event feature correlation matrix.

[0066] In this embodiment, the performance of key morphological features under different collision intensities (e.g., momentum p1<p2<p3) and collision directions (e.g., front, side) is analyzed. For example, the greater the collision intensity, the larger the value of a certain key morphological feature (e.g., the change rate of frequency distribution features), so a weight positively correlated with the collision intensity is assigned to the feature (e.g., weight w=k×p, where k is the proportionality coefficient); different collision directions lead to different values of a certain key morphological feature (e.g., the change amplitude of amplitude variation features), and a weight correlated with the collision direction is assigned to the feature (e.g., the weight for front collision is w1, and that for side collision is w2). The key morphological features and their corresponding weights are organized into a matrix, where rows represent features and columns represent event parameters (collision intensity, direction), thus forming a collision event-feature association matrix.

[0067] Step S136: perform associative quantization on the key morphological features of the impact reference spectral feature evolution trajectory and the corresponding ocean wave impact events, assign association weights to different key morphological features according to ocean wave level and impact duration information, and form an impact event-feature association matrix.

[0068] In this embodiment, the performance of key morphological features under different ocean wave levels (e.g., level 1 < level 2 < level 3) and impact durations (e.g., t1<t2<t3) is analyzed. For example, the higher the ocean wave level, the larger the value of a certain key morphological feature (e.g., the change rate of spectral morphological features), so a weight positively correlated with the ocean wave level is assigned to the feature (e.g., w=k×level); the longer the impact duration, the larger the value of a certain key morphological feature (e.g., the stationarity parameter of amplitude variation features), so a weight positively correlated with the duration is assigned to the feature (e.g., w=k×t). The key morphological features and their corresponding weights are organized into a matrix, thus forming an impact event-feature association matrix.

[0069] Step S137: select a preset model framework as the base model, wherein the base model comprises a feature input layer, an association calculation layer and a result output layer, the feature input layer is configured to receive key morphological feature data, the association calculation layer is configured to implement association operation between features and events, and the result output layer is configured to output the event type probability corresponding to the trajectory.

[0070] In this embodiment, a multi-layer perceptron (MLP) is selected as the base model. The feature input layer comprises a plurality of input neurons, the number of which is the same as the number of dimensions of the key morphological features, and the normalized vector of key morphological feature data is input into the neurons; the association calculation layer comprises a plurality of hidden layers, each layer has a plurality of neurons connected in a full connection manner, and ReLU is used as the activation function to implement association operation between features and events (e.g., matrix multiplication, nonlinear transformation); the result output layer comprises two output neurons, which respectively output the probability that the trajectory belongs to a collision event and an ocean wave impact event, and the Softmax activation function is used to ensure that the sum of probabilities is 1.

[0071] Step S138: Divide the data in the collision event feature correlation matrix and the impact event feature correlation matrix into a training data set and a validation data set. The proportion of the training data set to the total data volume is determined according to the total amount of data, so that the training data can cover the feature changes corresponding to different event parameters.

[0072] In this embodiment, the total data volume is the total amount of data in the collision and impact event feature correlation matrix (e.g., if the collision matrix has m data points and the impact matrix has n data points, the total data volume is m+n). The proportion of the training dataset is determined based on the total data volume. If the total data volume is greater than 1000 data points, the training set accounts for 80% (0.8×(m+n) data points); if it is less than 1000 data points, it accounts for 70%. When dividing the dataset, it is ensured that the training data covers the feature changes of different collision intensities, directions, wave levels, and durations. For example, a certain proportion of data is selected from each event parameter interval (e.g., collision intensity intervals p1-p2, p2-p3, wave levels 1-2, 2-3, etc.).

[0073] Step S139: Input the training data set into the feature input layer of the basic model, and process the feature data by calling the preset association operation algorithm through the association calculation layer, and output the event type probability corresponding to each training sample.

[0074] In this embodiment, each sample in the training dataset is a (key morphological feature vector, event type label), where the event type label is 0 (collision) or 1 (wave impact). The feature input layer passes the feature vector to the association calculation layer. The association calculation layer uses the forward propagation algorithm to sequentially process the hidden layers (e.g., the first layer of neurons calculates the dot product of the input vector and the weights, adds the bias, and then performs ReLU activation; the second layer repeats this process). Finally, the output layer calculates the outputs of the two neurons, which are then converted into probabilities using Softmax (e.g., the collision probability is P0, the wave impact probability is P1, and P0+P1=1).

[0075] Step S1310: Calculate the prediction error of the basic model based on the actual event types of the training samples and the event type probabilities output by the basic model. Adjust the operation parameters of the associated computation layer of the basic model through the backpropagation algorithm. After each parameter adjustment, input the validation dataset into the basic model for effect verification. Calculate the event type recognition accuracy of the basic model on the validation dataset. When the recognition accuracy of the basic model on the validation dataset remains stable for K consecutive iterations and reaches the preset accuracy requirement, stop parameter adjustment and obtain the initial trajectory event mapping model.

[0076] In this embodiment, the prediction error is calculated using the cross-entropy loss function: L = -Σ(y_i × log(P_i) + (1 - y_i) × log(1 - P_i)), where y_i is the actual label (0 or 1), and P_i is the model output probability. The backpropagation algorithm adjusts the operational parameters (weights and biases) of the associated computation layer based on the gradient of the loss function (calculated using the chain rule to measure the gradients of the weights and biases of each layer). After each parameter adjustment, the validation dataset is input into the model, and the recognition accuracy (number of correctly recognized samples / total number of validation samples) is calculated. When the accuracy change is less than 0.01 for K consecutive iterations (e.g., K = 10), and the accuracy reaches 90% or higher, the adjustment stops, and the initial trajectory event mapping model is obtained.

[0077] Step S1311: Collect the evolution trajectories of reference spectral features corresponding to new buoy collision events and wave impact events, and obtain new correlation matrix data.

[0078] In this embodiment, the new reference trajectory is derived from new collision and wave impact data collected during actual buoy operations, or new controlled experimental data. A spectral feature evolution trajectory is generated according to step S120, and new collision and impact event feature correlation matrix data is obtained through steps S133-S136.

[0079] Step S1312: Input the new correlation matrix data into the initial trajectory event mapping model, repeat the parameter adjustment and verification process, and iteratively optimize the initial trajectory event mapping model so that it can adapt to more diverse event scenarios, and finally obtain the trajectory event mapping model.

[0080] In this embodiment, the new correlation matrix data is used as training data, and the process of steps S139-S1310 is repeated to adjust the model parameters and verify them, so that the model can learn the feature-event correlation in the new scenario, improve the generalization ability, and finally obtain the optimized trajectory event mapping model.

[0081] Step S140: Based on the trajectory event mapping model, generate dynamic identification thresholds for collision events corresponding to buoy collision events and dynamic identification thresholds for impact events corresponding to wave impact events. Both the dynamic identification thresholds for collision events and the dynamic identification thresholds for impact events change over time and are adapted to the temporal characteristics of the spectral feature evolution trajectory.

[0082] Step S141: Call the association calculation layer of the trajectory event mapping model to extract all trajectory feature data associated with the buoy collision event in the trajectory event mapping model, and form the trajectory feature interval corresponding to the buoy collision event.

[0083] In this embodiment, the correlation calculation layer of the trajectory event mapping model stores trajectory feature data associated with collision events (such as the values ​​of key morphological features). By querying the model's parameters or output, these data are extracted, and their minimum and maximum values ​​are calculated to form trajectory feature intervals (such as the value range of feature F1 being [F1_min, F1_max], and feature F2 being [F2_min, F2_max], etc.).

[0084] Step S142: Call the association calculation layer of the trajectory event mapping model to extract all trajectory feature data associated with the wave impact event in the trajectory event mapping model, and form the trajectory feature interval corresponding to the wave impact event.

[0085] In this embodiment, the trajectory feature data associated with the wave impact event is extracted according to the method in step S141, and the minimum and maximum values ​​are statistically analyzed to form the trajectory feature interval of the impact event.

[0086] Step S143: Divide the trajectory feature interval of the buoy collision event into time dimensions. Divide the trajectory feature interval into multiple continuous time segments according to the preset time granularity. Each time segment corresponds to a continuous time range.

[0087] In this embodiment, the preset time granularity is consistent with the time interval of generating the spectral feature evolution trajectory (e.g., 0.01 seconds). The trajectory feature interval of the collision event is divided into multiple time segments according to the time granularity. Each segment corresponds to a continuous time range (e.g., from t0 to t0+0.01 seconds, t0+0.01 seconds to t0+0.02 seconds, etc.).

[0088] Step S144: Statistically analyze the distribution of trajectory feature data within each time segment, including the frequency of occurrence of feature data, the range of data concentration, and the degree of data dispersion.

[0089] In this embodiment, the frequency of feature data occurrence is statistically analyzed: within each time series segment, the number of times each feature value appears is counted, and the proportion of this number to the total data volume of that segment is calculated. Data concentration range statistics are also performed: the intervals where feature values ​​are mainly concentrated are determined (e.g., intervals with a frequency greater than 0.1, analyzed using histograms). Finally, data dispersion statistics are conducted: the variance or standard deviation of the feature values ​​is calculated; the larger the variance, the higher the dispersion.

[0090] Step S145: Calculate the average value of trajectory feature data within each time segment. This average value reflects the central tendency of the feature data of collision event trajectory characteristics within the time interval of that time segment.

[0091] In this embodiment, the average value is calculated by dividing the sum of all trajectory feature data within the time segment by the number of data points. For example, if there are n data points within a segment, with feature values ​​x1, x2, ..., xn, the average value is (x1 + x2 + ... + xn) / n.

[0092] Step S146: Calculate the difference between the maximum and minimum values ​​of the trajectory feature data within each time segment to obtain the range of change of the trajectory feature data within that time interval, reflecting the fluctuation of the feature data.

[0093] In this embodiment, the range of variation is the maximum value of the feature data within the segment minus the minimum value. For example, if the maximum value is x_max and the minimum value is x_min, the range of variation is x_max-x_min.

[0094] Step S147: Analyze the changing trend of the average value of trajectory feature data between adjacent time series segments. By calculating the difference between the average values ​​of adjacent time series segments, determine the upward or downward trend of the average value and the rate of change of the trend.

[0095] In this embodiment, the average value of adjacent segments 1 is μ1, and that of segment 2 is μ2, with a difference of Δμ = μ2 - μ1. If Δμ > 0, the trend is upward; otherwise, it is downward. The rate of trend change is Δμ divided by the time interval between adjacent segments (e.g., 0.01 seconds), i.e., the rate is Δμ / 0.01.

[0096] Step S148: Based on the characteristic data distribution, average value, range of variation and trend of each time series segment, determine the reasonable upper limit and lower limit of the fluctuation of the buoy collision event trajectory characteristics within that time series segment.

[0097] In this embodiment, the determination of the reasonable upper and lower limits of fluctuation takes into account multiple factors: - Distribution of characteristic data: If the data has a narrow central range and low dispersion, the fluctuation range can be small; otherwise, it will be large.

[0098] - Average value: The upper and lower limits are centered on the average value. For example, if the average value is μ, the fluctuation range is [μ-a×σ, μ+b×σ]. a and b are adjusted according to the distribution (e.g., under normal distribution, a=b=3, covering 99.7% of the data).

[0099] - Range of variation: When the range of variation is large, the fluctuation range should be appropriately expanded.

[0100] - Trend of change: If the average value rises, the upper limit of the fluctuation of subsequent segments can be adjusted upward with the trend, and the lower limit is the same.

[0101] For example, if the mean of a certain time series segment is μ, the standard deviation is σ, the data concentration range is [μ-2σ, μ+2σ], the variation range is 4σ, and the trend is upward (rate r), then the lower limit of the fluctuation of this segment is μ-2σ-r×Δt (Δt is the segment time interval), and the upper limit of the fluctuation is μ+2σ+r×Δt, ensuring that reasonable feature changes are covered.

[0102] Step S149: Arrange the reasonable upper and lower limits of fluctuation for each time series segment in chronological order to form a dynamic collision event identification threshold sequence that changes over time. Each threshold node in the dynamic collision event identification threshold sequence corresponds to the fluctuation range of a time series segment.

[0103] In this embodiment, the upper and lower limits of the fluctuation of each time segment are arranged in chronological order (such as t0-t0.01, t0.01-t0.02, etc.) to form a sequence. Each node contains a time interval and the corresponding fluctuation range (upper limit, lower limit), such as [(t0, t0.01), (L0, U0)], [(t0.01, t0.02), (L1, U1)], ...], where L is the lower limit and U is the upper limit.

[0104] Step S1410: Repeat the above steps of time dimension division, data statistics, average value calculation, range of change calculation, and trend analysis for the trajectory characteristic interval of the wave impact event to obtain the reasonable upper limit and reasonable lower limit of the trajectory characteristics of the impact event within each time segment.

[0105] In this embodiment, the trajectory feature range of the wave impact event is processed according to the method of steps S143-S148 to obtain the reasonable upper and lower limits of fluctuation for each time segment.

[0106] Step S1411: Arrange the reasonable upper and lower limits of fluctuation for each time segment of the impact event in chronological order to form a dynamic identification threshold sequence for the impact event that changes over time.

[0107] In this embodiment, the upper and lower limits of the fluctuation of the impact event are arranged according to the method of step S149 to form a dynamic identification threshold sequence.

[0108] Step S1412: Perform time granularity adaptation operation on the dynamic identification threshold sequence of collision events. Adjust the time interval of the dynamic identification threshold sequence of collision events according to the acquisition frequency of the real-time vibration signal of the buoy, so that the time node of the dynamic identification threshold sequence of collision events is synchronized with the acquisition time node of the real-time signal.

[0109] For example, step S14121: Obtain the acquisition parameters of the real-time vibration signal of the buoy, the acquisition parameters including the acquisition frequency of the real-time signal, and calculate the acquisition time interval of the real-time vibration signal of the buoy based on the acquisition frequency.

[0110] In this embodiment, the sampling frequency of the buoy's real-time vibration signal is determined by the sensor setting (e.g., 1000Hz), and the sampling time interval is 1 / sampling frequency (e.g., 1 / 1000 = 0.001 seconds).

[0111] Step S14122: Obtain the original time interval of the collision event dynamic recognition threshold sequence, and calculate the ratio coefficient between the original time interval and the acquisition time interval.

[0112] In this embodiment, the original time interval of the collision event dynamic identification threshold sequence is 0.01 seconds (as in step S143), the acquisition time interval is 0.001 seconds, and the scaling factor is 0.01 / 0.001=10.

[0113] Step S14123: When the scaling factor is greater than 1, determine the number of interpolation nodes to be inserted between adjacent original threshold nodes in the dynamic identification threshold sequence of the collision event according to the scaling factor, calculate the threshold value of each interpolation node using an interpolation algorithm, and arrange all interpolation nodes and original threshold nodes in chronological order to generate a new dynamic identification threshold sequence of the collision event with a time interval consistent with the acquisition time interval.

[0114] In this embodiment, the scaling factor is 10, indicating that the original time interval is 10 times the acquisition time interval. The time interval between adjacent original threshold nodes is 0.01 seconds, and the acquisition time interval is 0.001 seconds, therefore 10-1=9 interpolation nodes need to be inserted. The interpolation algorithm uses linear interpolation. For example, the time of original node 1 is t1, and the threshold is (L1, U1); the time of original node 2 is t2=t1+0.01 seconds, and the threshold is (L2, U2). The time of the interpolation node is t1+0.001×k seconds (k=1, 2, ..., 9), the lower threshold is L1+(L2-L1)×k / 10, and the upper threshold is U1+(U2-U1)×k / 10. The interpolation nodes and the original nodes are arranged in chronological order to generate a new sequence with a time interval of 0.001 seconds.

[0115] Step S14124: When the scaling factor is less than 1, determine the merging rule for merging consecutive original threshold nodes in the collision event dynamic identification threshold sequence according to the scaling factor, calculate the threshold value of each merged node according to the merging rule, and arrange all merged nodes in chronological order to generate a new collision event dynamic identification threshold sequence with a time interval consistent with the acquisition time interval.

[0116] (In this embodiment, the scaling factor is greater than 1, so this step is not performed for now and is only for explanation: If the scaling factor is 0.5, the data collection time interval is 0.001 seconds, and the original time interval is 0.0005 seconds. The merging rule is that every two original nodes are merged into one, the time of the merged node is the midpoint of the time of the two original nodes, and the threshold value is the average threshold value of the two original nodes.) Step S14125: When the scaling factor is equal to 1, the dynamic identification threshold sequence of the collision event remains unchanged.

[0117] (In this embodiment, the scaling factor is not equal to 1, so this step will not be performed for now.) Step S14126: Perform timestamp calibration on the collision event dynamic identification threshold sequence after interpolation, merging or keeping it unchanged, so that the timestamp of each threshold node in the calibrated sequence is consistent with the timestamp of the corresponding acquisition time point of the buoy's real-time vibration signal.

[0118] In this embodiment, timestamp calibration is achieved by adjusting the timestamp of the threshold node to match the acquisition time of the buoy's real-time signal (e.g., t=0, 0.001, 0.002, ... seconds). For example, if the interpolated node time is t1+0.001×k seconds, ensuring that t1 is the acquisition time of the buoy signal (e.g., t1=0 seconds), then the calibrated timestamp will be 0+0.001×k seconds, synchronized with the real-time signal acquisition time.

[0119] Step S14127: Verify the numerical validity of the calibrated collision event dynamic identification threshold sequence. If there are nodes whose values ​​exceed the preset reasonable range, re-execute the corresponding interpolation or merging operation for correction.

[0120] In this embodiment, the preset reasonable range is determined based on historical data and physical laws (e.g., the feature value of a collision event cannot be negative, or it cannot be less than a certain minimum value). If the lower limit of the threshold of a certain interpolation node is negative, the interpolation coefficient is recalculated and the threshold value is adjusted to ensure that it is within the reasonable range.

[0121] Step S14128: Store the verified collision event dynamic identification threshold sequence as an adapted threshold file and mark the corresponding buoy real-time signal acquisition parameters.

[0122] In this embodiment, the calibrated collision event dynamic recognition threshold sequence is stored as a binary file or a text file, and parameters such as the acquisition frequency (1000Hz) and acquisition time interval (0.001 seconds) are marked for easy subsequent retrieval.

[0123] Step S1413: Perform time granularity adaptation on the dynamic identification threshold sequence of impact events, extract the reasonable upper limit and reasonable lower limit of fluctuation for each time node in the dynamic identification threshold sequence of impact events, and calculate the median value of the reasonable upper limit and reasonable lower limit of fluctuation for each time node as the threshold benchmark value for that node.

[0124] In this embodiment, the dynamic identification threshold sequence of impact events is adapted to time granularity according to the method in step S1412. Then, the upper limit (U) and lower limit (L) of the fluctuation at each time node in the collision event threshold sequence are extracted, and the median value (L+U) / 2 is calculated as the threshold baseline value.

[0125] Step S1414: Adjust the magnitude of the reasonable upper and lower limits of fluctuation based on the threshold benchmark value and the dispersion of the characteristic data of the node.

[0126] In this embodiment, the dispersion of the feature data is represented by the standard deviation σ. If the dispersion is large (σ is large), the fluctuation range is increased (e.g., the upper limit is the benchmark value + 2σ, and the lower limit is the benchmark value - 2σ); if the dispersion is small (σ is small), the fluctuation range is decreased (e.g., the upper limit is the benchmark value + σ, and the lower limit is the benchmark value - σ).

[0127] Step S1415: Repeat the above-mentioned benchmark value calculation and upper and lower limit adjustment steps for the dynamic identification threshold sequence of impact events, and store the adjusted dynamic identification threshold sequence of collision events and dynamic identification threshold sequence of impact events as threshold files respectively.

[0128] In this embodiment, the impact event threshold sequence is processed according to the method of steps S1413-S1414, and after adjustment, the two sequences are stored as threshold files (such as collision_thresholds.dat and impact_thresholds.dat).

[0129] Step S150: Generate a corresponding real-time spectral feature evolution trajectory for the real-time vibration signal collected by the buoy in real time, and compare the real-time spectral feature evolution trajectory with the collision event dynamic identification threshold and the impact event dynamic identification threshold respectively, and obtain the buoy collision event identification result or the wave impact event identification result based on the comparison result.

[0130] Step S151: Establish a real-time communication link between the buoy and the data receiving terminal, and timestamp the real-time vibration signal transmitted to the data receiving terminal. Each signal data point is marked with the corresponding acquisition time to ensure the time accuracy of subsequent processing.

[0131] In this embodiment, the buoy and the data receiving terminal establish a real-time communication link via wireless communication (such as 4G or satellite communication), and the real-time vibration signal collected by the vibration sensor is transmitted through this link. The data receiving terminal marks the collection time for each signal data point (e.g., based on the receiving time combined with the buoy's clock calibration to ensure accurate timestamps), with the timestamp format being year-month-day hour:minute:second.millisecond.

[0132] Step S152: Divide the real-time vibration signal with timestamps into multiple continuous real-time signal segments according to a preset time interval. The preset time interval is consistent with the time interval when generating the historical spectrum feature evolution trajectory, so that the feature dimension of the real-time signal segment is consistent with the feature dimension of the historical spectrum feature evolution trajectory.

[0133] In this embodiment, the preset time interval is consistent with the time interval in step S121 (e.g., 0.01 seconds). The real-time vibration signal with timestamp is divided into multiple real-time signal segments according to this interval, with each segment corresponding to a time range of 0.01 seconds, ensuring consistency with the feature dimensions (time interval, data length, etc.) of the historical trajectory.

[0134] Step S153: Perform noise filtering on each real-time signal segment, and perform frequency domain transformation on each real-time signal segment after noise filtering to convert the real-time signal segment in the time domain into frequency domain data. During the transformation process, the frequency, amplitude and phase information of the signal are preserved.

[0135] In this embodiment, noise filtering employs a wavelet denoising algorithm. A suitable wavelet basis (e.g., db4) and a decomposition level (e.g., 3 levels) are selected to decompose the real-time signal segment, removing high-frequency noise components (e.g., components with energy less than 0.1% of the total energy). Then, frequency domain transformation is performed according to step S122, preserving frequency, amplitude, and phase information (e.g., the complex values ​​after FFT contain phase information).

[0136] Step S154: Extract the spectral features of each real-time signal segment from the frequency domain data. If any dimension of the real-time spectral feature data is missing, perform trend completion by using the corresponding dimension feature data of adjacent real-time signal segments. Then, arrange the completed real-time spectral feature data in time sequence according to the acquisition time of the real-time signal segments to form a preliminary real-time time sequence feature sequence.

[0137] In this embodiment, the spectral feature extraction follows the method in steps S123. If feature data of a certain dimension is missing (such as the energy percentage of a certain frequency range is lost), linear trend completion is used. Based on the corresponding feature data of the previous and next real-time signal segments, the missing value of the current segment is calculated (e.g., if the previous one is x_prev, the next one is x_next, the time interval is Δt, and the time difference between the current segment and the previous one is t, then the missing value is x_prev+(x_next-x_prev)×t / Δt). After completion, the data is arranged in the order of acquisition time to form a preliminary real-time time series feature sequence.

[0138] Step S155: In the preliminary real-time time series feature sequence, add a corresponding real-time acquisition time stamp to each real-time spectral feature data.

[0139] In this embodiment, each real-time spectrum feature data corresponds to the midpoint of the acquisition time of the real-time signal segment (e.g., the segment acquisition time is from t_start to t_end, and the midpoint is t_mid). A time stamp t_mid is added to the feature data to form the form (t_mid, feature data).

[0140] Step S156: Perform coherence processing on the real-time time-series feature sequence with time stamps, calculate the rate of change between adjacent feature data, and if the rate of change exceeds the preset coherence standard, insert transition feature data, perform smoothing operation on the real-time time-series feature sequence after inserting transition data, and connect the smoothed real-time time-series feature sequences in chronological order to form a real-time spectral feature evolution trajectory. This real-time spectral feature evolution trajectory is used to reflect the spectral feature changes corresponding to the buoy vibration signal.

[0141] In this embodiment, the coherence processing follows the method in step S126. The rate of change of adjacent feature data is calculated. If it exceeds a preset coherence standard (such as V_max), transitional feature data (linear interpolation) is inserted. Then, a moving average algorithm (such as a window size of 5) is used to smooth the sequence to obtain a smoothed real-time time series feature sequence, which is then connected in chronological order to form a real-time spectral feature evolution trajectory.

[0142] Step S157: Read the dynamic identification threshold sequence of collision events and the dynamic identification threshold sequence of impact events from the stored threshold file, so that the time granularity of the dynamic identification threshold sequence of collision events and the dynamic identification threshold sequence of impact events is consistent with the time granularity of the real-time spectral feature evolution trajectory.

[0143] In this embodiment, the threshold sequence is read from the threshold file (such as collision_thresholds.dat, impact_thresholds.dat) and its time granularity is adjusted (such as by interpolation or merging) to make it consistent with the time granularity of the real-time spectral feature evolution trajectory (such as 0.001 seconds).

[0144] Step S158: Extract the spectrum feature data corresponding to each time node in the real-time spectrum feature evolution trajectory to form a real-time feature data list. Each data entry contains a timestamp and the corresponding multi-dimensional feature value.

[0145] In this embodiment, each time node (e.g., t=0, 0.001, 0.002, ... seconds) in the real-time spectrum feature evolution trajectory corresponds to a spectrum feature data (multi-dimensional, such as frequency distribution, amplitude change, and spectrum morphology feature values). These data are extracted and a list is formed, with each entry being (timestamp, [F1, F2, A1, A2, M1, M2, ...]).

[0146] Step S159: Extract threshold nodes corresponding to each timestamp in the real-time feature data list from the dynamic identification threshold sequence of collision events. Each threshold node contains the upper and lower limits of the fluctuation of the collision event at that timestamp.

[0147] In this embodiment, based on the timestamp of the real-time feature data list, the corresponding threshold node is found from the collision event threshold sequence to obtain the upper limit of fluctuation (U_collision) and the lower limit of fluctuation (L_collision) under that timestamp.

[0148] Step S1510: Compare the multi-dimensional feature values ​​of each time node in the real-time feature data list with the upper and lower limits of the fluctuation of the corresponding collision event threshold node.

[0149] In this embodiment, for each time point, the multi-dimensional feature values ​​(such as F1, F2, A1, A2, M1, M2, ...) are compared with the corresponding dimensions of U_collision and L_collision. For example, F1 is compared with the upper limit of the F1 dimension of U_collision and the lower limit of the F1 dimension of L_collision to determine whether F1 is within the range of [L_collision_F1, U_collision_F1]. Similarly, other dimensions are compared.

[0150] Step S1511: If all dimensional feature values ​​at any time node are between the upper and lower limits of the collision event fluctuation, then the real-time feature data at that time node is determined to meet the collision event threshold requirements; if at any time node, at least one dimensional feature value exceeds the upper limit of the collision event fluctuation or falls below the lower limit of the collision event fluctuation, then the real-time feature data at that time node is determined not to meet the collision event threshold requirements.

[0151] In this embodiment, the feature values ​​of all dimensions at each time point are judged. If all dimensions are within the upper and lower limits of fluctuation, the condition is satisfied; otherwise, the condition is not satisfied.

[0152] Step S1512: According to the above comparison method, judge all time nodes in the real-time feature data list one by one, count the number of time nodes that meet the collision event threshold requirements, and record them as the collision satisfaction number.

[0153] In this embodiment, all time nodes of the real-time feature data list are traversed, and the number of nodes that meet the collision threshold requirement is counted, denoted as N_collision.

[0154] Step S1513: Using the same comparison process, extract the threshold nodes corresponding to each timestamp in the real-time feature data list from the dynamic identification threshold sequence of the impact event. Each threshold node contains the upper and lower limits of the fluctuation of the impact event at that timestamp.

[0155] In this embodiment, the threshold nodes corresponding to the timestamps are extracted from the impact event threshold sequence according to the method in step S159 to obtain the upper limit of fluctuation (U_impact) and the lower limit of fluctuation (L_impact).

[0156] Step S1514: Compare the multi-dimensional feature values ​​of each time node in the real-time feature data list with the upper and lower limits of the fluctuation of the corresponding impact event threshold node.

[0157] In this embodiment, following the method in step S1510, the multi-dimensional feature values ​​of each time node are compared with the corresponding dimensions of U_impact and L_impact.

[0158] Step S1515: If all dimensional feature values ​​at any given time point are between the upper and lower limits of the shock event fluctuation, then the real-time feature data at that time point is determined to meet the shock event threshold requirements; if at any given time point at least one dimensional feature value exceeds the upper limit of the shock event fluctuation or falls below the lower limit of the shock event fluctuation, then the real-time feature data at that time point is determined not to meet the shock event threshold requirements.

[0159] In this embodiment, following the method in step S1511, it is determined whether each time node meets the impact threshold requirement. Step S1516: All time nodes in the real-time feature data list are determined one by one, and the number of time nodes that meet the impact event threshold requirement is counted and recorded as the impact satisfaction number.

[0160] In this embodiment, all time points are traversed, and the number of nodes that meet the impact threshold requirement is counted, denoted as N_impact.

[0161] Step S1517: Calculate the proportion of the number of collisions satisfied to the total number of time nodes in the real-time feature data list, denoted as the collision satisfaction ratio; calculate the proportion of the number of impacts satisfied to the total number of time nodes in the real-time feature data list, denoted as the impact satisfaction ratio.

[0162] In this embodiment, the total number of time nodes is N_total, the collision satisfaction ratio is N_collision / N_total, and the impact satisfaction ratio is N_impact / N_total.

[0163] Step S1518: Compare the collision satisfaction ratio with the impact satisfaction ratio. If the collision satisfaction ratio is greater than the impact satisfaction ratio and the collision satisfaction ratio reaches the preset ratio threshold, generate a buoy collision event identification result. If the impact satisfaction ratio is greater than the collision satisfaction ratio and the impact satisfaction ratio reaches the preset ratio threshold, generate a wave impact event identification result. If the collision satisfaction ratio and the impact satisfaction ratio are equal, or neither reaches the preset ratio threshold, re-extract and compare the real-time spectrum feature evolution trajectory. After re-extraction and comparison, based on the final ratio comparison result, generate an identification result report containing the event type, the number of time nodes that meet the ratio threshold, and the ratio threshold. Send the identification result report to the buoy monitoring terminal.

[0164] In this embodiment, the preset ratio threshold is 0.6 (i.e., 60% of the time nodes meet the requirements). If the collision satisfaction ratio (e.g., 0.7) is greater than the impact satisfaction ratio (e.g., 0.5) and ≥ 0.6, a collision event identification result is generated; if the impact satisfaction ratio (e.g., 0.7) is greater than the collision satisfaction ratio (e.g., 0.5) and ≥ 0.6, a wave impact event identification result is generated. If the two are equal (e.g., 0.5) or both are less than 0.6, the real-time spectrum feature evolution trajectory is re-extracted (e.g., re-acquired signals, re-processed), compared again, and a report is generated based on the final result (including event type, number of satisfactions, ratio, etc.), and sent to the buoy monitoring terminal (e.g., the server or terminal device of the monitoring center) via wireless communication.

[0165] Based on the same inventive concept, please refer to Figure 2The diagram shows a schematic block diagram of a buoy collision and wave impact identification system 100 based on spectrum features, provided in an embodiment of this application, for performing the above-described buoy collision and wave impact identification method based on spectrum features. The buoy collision and wave impact identification system 100 based on spectrum features may include a communication unit 110, a machine-readable storage medium 120, and a processor 130.

[0166] In this embodiment, both the machine-readable storage medium 120 and the processor 130 are located within the spectral feature-based buoy collision and wave impact identification system 100 and are separately configured. However, it should be understood that the machine-readable storage medium 120 may also be independent of the spectral feature-based buoy collision and wave impact identification system 100 and may be accessed by the processor 130 via a bus interface. Alternatively, the machine-readable storage medium 120 may also be integrated into the processor 130 and may communicate with external systems via the communication unit 110.

[0167] The processor 130 is the control center of the spectrum-based buoy collision and wave impact identification system 100. It connects various parts of the system via various interfaces and lines, and performs overall monitoring by running or executing software programs and / or modules stored in the machine-readable storage medium 120, and by calling data stored in the machine-readable storage medium 120, it executes various functions and processes data of the spectrum-based buoy collision and wave impact identification system 100. Optionally, the processor 130 may include one or more processing cores; for example, the processor 130 may integrate an application processor and a modem processor, wherein the application processor mainly handles the operating system, user interface, and applications, and the modem processor mainly handles wireless communication. It is understood that the modem processor may also not be integrated into the processor. The machine-readable storage medium 120 is used to store machine-executable instructions for executing the scheme of this application, and the processor 130 is used to execute the machine-executable instructions stored in the machine-readable storage medium 120 to implement the buoy collision and wave impact identification method based on spectral features provided in the aforementioned method embodiments.

[0168] It should be noted that, in order to simplify the description of the present invention and thus help to understand one or more embodiments of the invention, multiple features may sometimes be grouped into one embodiment, drawing or description thereof in the foregoing description of the embodiments of the present invention.

Claims

1. A method for identifying a buoy collision and a sea wave impact based on spectral features, characterized by, The method includes: The set of vibration signals generated by the buoy under actual operating conditions is obtained. The set of vibration signals includes vibration waveform data continuously collected by the buoy under different external forces. The vibration waveform data carries the mechanical response information of the buoy when it is hit by collision or wave impact. For each vibration waveform data in the vibration signal set, a corresponding spectral feature evolution trajectory is generated. The spectral feature evolution trajectory is composed of spectral features at multiple consecutive time points in chronological order. The spectral features at each time point include frequency distribution features, amplitude variation features, and spectral morphology features. Based on the correlation between the spectral feature evolution trajectory and known buoy collision events and wave impact events, a trajectory event mapping model is constructed. Based on the trajectory event mapping model, dynamic identification thresholds for collision events corresponding to buoy collision events and dynamic identification thresholds for impact events corresponding to wave impact events are generated. The real-time vibration signal collected by the buoy is used to generate a corresponding real-time spectral feature evolution trajectory. The real-time spectral feature evolution trajectory is compared with the collision event dynamic identification threshold and the impact event dynamic identification threshold respectively. The buoy collision event identification result or wave impact event identification result is obtained based on the comparison result. The construction of a trajectory event mapping model based on the correlation between the spectral feature evolution trajectory and known buoy collision events and wave impact events includes: Obtain the evolution trajectories of multiple collision reference spectral features corresponding to known buoy collision events; Obtain the evolution trajectory of multiple impact reference spectrum features corresponding to known wave impact events; Data cleaning operations are performed on the collision reference spectrum feature evolution trajectory and the impact reference spectrum feature evolution trajectory to extract the trajectory morphology features from the cleaned collision reference spectrum feature evolution trajectory. The trajectory morphology features of the impact reference spectrum feature evolution trajectory after cleaning are extracted. Cross-event difference quantization is performed on the trajectory morphology features of the collision reference spectrum feature evolution trajectory and the trajectory morphology features of the impact reference spectrum feature evolution trajectory to determine the key morphology features with the highest distinguishability of event types. The key morphological features of the collision reference spectrum feature evolution trajectory are correlated and quantified with the corresponding buoy collision events. Correlation weights are assigned to different key morphological features based on collision intensity and collision direction information to form a collision event feature correlation matrix. The key morphological features of the evolution trajectory of the impact reference spectrum are correlated and quantified with the corresponding wave impact events. Correlation weights are assigned to different key morphological features based on wave level and impact duration information to form an impact event feature correlation matrix. Based on the correlation matrix of collision event features and the correlation matrix of impact event features, a trajectory event mapping model is obtained through basic model training and iterative optimization.

2. The buoy collision and wave impact identification method based on spectral features according to claim 1, characterized in that, The step of generating a corresponding spectral feature evolution trajectory for each vibration waveform data in the vibration signal set includes: Perform time axis calibration on each vibration waveform data in the vibration signal set, and divide each calibrated vibration waveform data into multiple continuous signal segments according to a preset time interval; A quality screening operation is performed on each segmented signal to obtain valid signal segments. Then, a frequency domain transformation operation is performed on each valid signal segment to convert the valid signal segments in the time domain into frequency domain data. The spectral features of each effective signal segment are extracted from the frequency domain data. The spectral features include frequency distribution features, amplitude variation features, and spectral morphology features. The spectral features of each extracted valid signal segment are augmented to obtain augmented spectral feature data. The augmented spectral feature data are then sorted in chronological order according to the time sequence of the valid signal segments to form a preliminary time-series feature sequence. In the initial time-series feature sequence, the specific time node corresponding to each spectral feature is marked. The time node is the midpoint of the acquisition time of the effective signal segment to which the spectral feature belongs, forming the basic node of the trajectory. Perform temporal coherence operations on the basic nodes, calculate the characteristic change rate between adjacent basic nodes, and if the change rate exceeds the preset coherence standard, insert transition nodes by linear interpolation, and perform smoothness optimization operations on the trajectory containing the transition nodes to generate the spectral feature evolution trajectory.

3. The buoy collision and wave impact identification method based on spectral features according to claim 1, characterized in that, The trajectory event mapping model, obtained through basic model training and iterative optimization based on the collision event feature correlation matrix and the impact event feature correlation matrix, includes: A preset model framework is selected as the basic model. The basic model includes a feature input layer, an association calculation layer, and a result output layer. The feature input layer is used to receive key morphological feature data, the association calculation layer is used to realize the association operation between features and events, and the result output layer is used to output the event type probability corresponding to the trajectory. The data in the collision event feature correlation matrix and the impact event feature correlation matrix are divided into a training data set and a validation data set; The training dataset is input into the feature input layer of the base model. The feature data is processed by calling the preset association operation algorithm through the association calculation layer, and the event type probability corresponding to each training sample is output. Based on the actual event types of the training samples and the event type probabilities output by the basic model, the prediction error of the basic model is calculated. The operation parameters of the associated computation layer of the basic model are adjusted through the backpropagation algorithm. After each parameter adjustment, the validation dataset is input into the basic model for effect verification. The event type recognition accuracy of the basic model on the validation dataset is calculated. When the recognition accuracy of the basic model on the validation dataset remains stable for K consecutive iterations and reaches the preset accuracy requirement, the parameter adjustment is stopped, and the initial trajectory event mapping model is obtained. The evolution trajectories of reference spectral features corresponding to new buoy collision events and wave impact events are collected, and new correlation matrix data are obtained. The new correlation matrix data is input into the initial trajectory event mapping model, and the parameter adjustment and verification process is repeated to iteratively optimize the initial trajectory event mapping model, and finally the trajectory event mapping model is obtained.

4. The buoy collision and wave impact identification method based on spectral features according to claim 1, characterized in that, The generation of dynamic identification thresholds for collision events corresponding to buoy collision events and dynamic identification thresholds for impact events corresponding to wave impact events based on the trajectory event mapping model includes: The associated calculation layer of the trajectory event mapping model is invoked to extract all trajectory feature data associated with the buoy collision event in the trajectory event mapping model, forming the trajectory feature interval corresponding to the buoy collision event; The associated calculation layer of the trajectory event mapping model is invoked to extract all trajectory feature data associated with the wave impact event in the trajectory event mapping model, forming the trajectory feature interval corresponding to the wave impact event; The trajectory feature interval of the buoy collision event is divided into time dimensions. According to the preset time granularity, the trajectory feature interval is divided into multiple continuous time segments, and each time segment corresponds to a continuous time range. The distribution of trajectory feature data within each time segment is statistically analyzed, including the frequency of occurrence of feature data, the range of data concentration, and the degree of data dispersion. Calculate the average value of trajectory feature data within each time segment. This average value reflects the central tendency of the feature data of collision event trajectory characteristics within the time interval of that time segment. Calculate the difference between the maximum and minimum values ​​of the trajectory feature data within each time segment to obtain the range of variation of the trajectory feature data within that time interval, reflecting the fluctuation of the feature data; Analyze the changing trend of the average value of trajectory feature data between adjacent time series segments, and determine the upward or downward trend and the rate of change of the average value by calculating the difference between the average values ​​of adjacent time series segments. Based on the characteristic data distribution, average value, range of variation and trend of each time series segment, determine the reasonable upper limit and lower limit of the fluctuation of the buoy collision event trajectory characteristics within that time series segment; The reasonable upper and lower limits of fluctuation for each time series segment are arranged in chronological order to form a dynamic collision event identification threshold sequence that changes over time. Each threshold node in this dynamic collision event identification threshold sequence corresponds to the fluctuation range of a time series segment. Repeat the above steps of time dimension division, data statistics, average value calculation, change range calculation, and trend analysis for the trajectory characteristic interval of the wave impact event to obtain the reasonable upper limit and reasonable lower limit of the trajectory characteristics of the impact event within each time segment; The reasonable upper and lower limits of fluctuation for each time segment of the impact event are arranged in chronological order to form a dynamic identification threshold sequence for impact events that changes over time. The dynamic identification threshold sequences for collision events and impact events are adapted to the time granularity and adjusted in magnitude, and then stored as threshold files. 5.The method of claim 4, wherein, The process of adapting the dynamic identification threshold sequences for collision events and impact events to a time granularity and adjusting their amplitude, and storing them as a threshold file, includes: The time granularity of the collision event dynamic identification threshold sequence is adapted by adjusting the time interval of the collision event dynamic identification threshold sequence according to the acquisition frequency of the buoy's real-time vibration signal, so that the time node of the collision event dynamic identification threshold sequence is synchronized with the acquisition time node of the real-time signal. A time granularity adaptation operation is performed on the dynamic identification threshold sequence of impact events to extract the reasonable upper and lower limits of fluctuation for each time node in the dynamic identification threshold sequence of impact events. The median value of the reasonable upper and lower limits of fluctuation for each time node is calculated as the threshold benchmark value for that node. Based on the threshold benchmark value and the dispersion of the characteristic data of the node, adjust the range of the reasonable upper and lower limits of fluctuation. Repeat the baseline calculation and upper and lower limit adjustment steps for the dynamic identification threshold sequence of impact events, and store the adjusted dynamic identification threshold sequence of collision events and dynamic identification threshold sequence of impact events as threshold files respectively. 6.The method of claim 1, wherein, The generation of the corresponding real-time spectral feature evolution trajectory from the real-time vibration signal acquired by the buoy includes: Establish a real-time communication link between the buoy and the data receiving terminal, and timestamp the real-time vibration signals transmitted to the data receiving terminal. Each signal data point is marked with the corresponding acquisition time to ensure the time accuracy of subsequent processing. The real-time vibration signal with timestamps is divided into multiple continuous real-time signal segments according to a preset time interval. Noise filtering is performed on each real-time signal segment, and frequency domain transformation is performed on each noise-filtered real-time signal segment to convert the real-time signal segment in the time domain into frequency domain data. During the transformation process, the frequency, amplitude and phase information of the signal are preserved. The spectral features of each real-time signal segment are extracted from the frequency domain data. If any dimension of the real-time spectral feature data is missing, it is completed by using the corresponding dimension feature data of adjacent real-time signal segments. The completed real-time spectral feature data are then arranged in time sequence according to the acquisition time of the real-time signal segments to form a preliminary real-time time sequence feature sequence. In the initial real-time time series feature sequence, a corresponding real-time acquisition time marker is added to each real-time spectral feature data. The real-time time-series feature sequence with time stamps is processed for coherence. The rate of change between adjacent feature data is calculated. If the rate of change exceeds the preset coherence standard, transitional feature data is inserted. The real-time time-series feature sequence after the insertion of transitional data is smoothed. The smoothed real-time time-series feature sequence is connected in chronological order to form a real-time spectral feature evolution trajectory. This real-time spectral feature evolution trajectory is used to reflect the spectral feature changes corresponding to the buoy vibration signal. 7.The method of claim 1, wherein, The step of comparing the real-time spectral feature evolution trajectory with the collision event dynamic identification threshold and the impact event dynamic identification threshold, respectively, and obtaining the buoy collision event identification result or the wave impact event identification result based on the comparison result, includes: Read the dynamic identification threshold sequence of collision event and the dynamic identification threshold sequence of impact event from the stored threshold file, so that the time granularity of the dynamic identification threshold sequence of collision event and the dynamic identification threshold sequence of impact event is consistent with the time granularity of the real-time spectral feature evolution trajectory. Extract the spectrum feature data corresponding to each time node in the real-time spectrum feature evolution trajectory to form a real-time feature data list. Each data entry contains a timestamp and the corresponding multi-dimensional feature values. Extract threshold nodes corresponding to each timestamp in the real-time feature data list from the dynamic identification threshold sequence of collision events. Each threshold node contains the upper and lower limits of the fluctuation of the collision event at that timestamp. The multi-dimensional feature values ​​of each time node in the real-time feature data list are compared with the upper and lower limits of the fluctuation of the corresponding collision event threshold node. If all dimensional feature values ​​at any given time point are between the upper and lower limits of the collision event fluctuation, then the real-time feature data at that time point is determined to meet the collision event threshold requirements. If at any given time point, the feature value of at least one dimension exceeds the upper limit of the collision event fluctuation or falls below the lower limit of the collision event fluctuation, then the real-time feature data of that time point is determined not to meet the collision event threshold requirements. Each time point in the real-time feature data list is judged one by one, and the number of time points that meet the collision event threshold requirements is counted and recorded as the collision satisfaction number. Using the same comparison process, threshold nodes corresponding to each timestamp in the real-time feature data list are extracted from the dynamic identification threshold sequence of impact events. Each threshold node contains the upper and lower limits of the fluctuation of the impact event at that timestamp. The multi-dimensional feature values ​​of each time node in the real-time feature data list are compared with the upper and lower limits of the fluctuation of the corresponding shock event threshold node. Based on the comparison results, it is determined whether the impact event threshold requirements are met. The number of impacts that meet the requirements is counted, the collision satisfaction ratio and the impact satisfaction ratio are calculated, and the final identification result is generated based on the ratio comparison results. 8.The method of claim 7, wherein, The process of determining whether the impact event threshold requirement is met based on the comparison results, counting the number of impacts that meet the threshold, calculating the collision satisfaction ratio and the impact satisfaction ratio, and generating the final identification result based on the ratio comparison results includes: If all dimensional feature values ​​at any given time point are between the upper and lower limits of the shock event fluctuation, then the real-time feature data at that time point is determined to meet the shock event threshold requirements. If at any given time point, the value of at least one dimension of the feature exceeds the upper limit of the shock event fluctuation or falls below the lower limit of the shock event fluctuation, then the real-time feature data of that time point is determined not to meet the shock event threshold requirements. Each time point in the real-time feature data list is judged one by one, and the number of time points that meet the threshold requirements of the impact event is counted and recorded as the impact satisfaction number. The proportion of collision satisfaction to the total number of time nodes in the real-time feature data list is denoted as the collision satisfaction ratio. The proportion of the number of impact satisfactions to the total number of time nodes in the real-time feature data list is denoted as the impact satisfaction ratio. The collision satisfaction ratio is compared with the impact satisfaction ratio. If the collision satisfaction ratio is greater than the impact satisfaction ratio and the collision satisfaction ratio reaches the preset ratio threshold, the buoy collision event recognition result is generated. If the impact satisfaction ratio is greater than the collision satisfaction ratio, and the impact satisfaction ratio reaches the preset ratio threshold, then the wave impact event identification result is generated. If the collision satisfaction ratio is equal to the impact satisfaction ratio, or if neither of them reaches the preset ratio threshold, the real-time spectrum feature evolution trajectory is re-extracted and compared. After re-extraction and comparison, based on the final ratio comparison result, an identification result report is generated, which includes the event type, the number of time nodes that meet the ratio threshold, and the ratio threshold. The identification result report is then sent to the buoy monitoring terminal.

9. A system for identifying collisions of buoys and impacts of waves based on spectral features, characterized by, include: processor; A machine-readable storage medium for storing machine-executable instructions of the processor; The processor is configured to execute the spectral feature-based buoy collision and wave impact identification method according to any one of claims 1 to 8 by executing the machine-executable instructions.

Images