Method and system for early warning of wheel boarding and leaving risk of pilot under severe sea condition
By acquiring historical data from pilot boats and piloted vessels, identifying overlapping influence areas, and performing real-time sea state characteristic matching and wave spectrum iterative analysis, the problem of early warning of risks for pilots boarding and disembarking vessels in severe sea conditions was solved, achieving precise risk monitoring and safety assurance.
Patent Information
- Application Number
- CN202511370640.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-09-24
- Publication Date
- 2025-11-11
- Estimated Expiration
- 2045-09-24
AI Technical Summary
In adverse sea conditions, traditional methods for monitoring the risks of boarding and disembarking vessels are insufficient to comprehensively analyze the overlapping influence areas of the pilot boat and the vessel being piloted, resulting in inaccurate risk warnings and failing to meet the safety requirements of ship pilotage operations.
By acquiring historical boarding and disembarking data of the target pilot boat and the piloted vessel, overlapping impact areas and severe sea state characteristics are identified. Combined with real-time sea state characteristic matching and multi-scale iterative analysis of wave spectrum, risks are assessed and early warning instructions are issued.
It enables accurate early warning of the risks of pilots boarding and disembarking ships in adverse sea conditions, improves the reliability and safety of monitoring, and provides a reliable basis for alarm devices.
Smart Images

Figure CN120932411A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of ship pilotage technology, and in particular to a method and system for early warning of risks for pilots boarding and disembarking ships in adverse sea conditions. Background Technology
[0002] In ship pilotage operations, pilots must board and disembark the guided vessel via pilot boat in adverse sea conditions, making their operational safety paramount. Current technologies primarily rely on single sea state parameters or ship motion data for risk monitoring during boarding and disembarkation, using simple alarm devices for early warning. However, due to the complex relative motion between the pilot boat and the guided vessel in adverse sea conditions, and the variable wave spectrum characteristics, traditional methods struggle to comprehensively analyze the dynamic risks in overlapping influence areas. This results in insufficiently accurate risk warnings during the pilot's boarding and disembarking process, failing to provide comprehensive and reliable data support for alarm devices and ultimately failing to meet the safety requirements of ship pilotage operations in adverse sea conditions. Summary of the Invention
[0003] This application provides a method and system for risk warning of pilots boarding and disembarking vessels in adverse sea conditions, which is used to solve the technical problem that traditional risk monitoring methods for boarding and disembarking vessels are difficult to warn of risks during the pilot boarding and disembarking process in adverse sea conditions based on alarm devices and relevant ship data.
[0004] The first aspect of this application provides a method for risk warning of pilot boarding and disembarking under severe sea conditions. The method includes: acquiring a historical set of pilot boarding and disembarking data for a target pilot vessel and a target piloted vessel; identifying overlapping influence areas of the target pilot vessel and the target piloted vessel under different severe sea conditions based on the historical pilot boarding and disembarking data set, and determining a set of overlapping influence areas and a corresponding set of severe sea condition features; performing real-time sea condition feature analysis on the target pilot vessel and the target piloted vessel according to preset motion parameters and preset sea condition environmental indicators to obtain real-time sea condition features; matching the real-time sea condition features with the set of severe sea condition features, and mapping and extracting the overlapping influence areas according to the matching results to obtain matched overlapping influence areas; continuously collecting regional wave spectrum data based on the matched overlapping influence areas, and performing iterative analysis based on the collection results to determine the target regional wave spectrum iterative features; determining whether the target regional wave spectrum iterative features meet preset requirements, and if not, obtaining a boarding and disembarking risk warning instruction.
[0005] A second aspect of this application provides a pilot boarding / disembarking risk warning system under severe sea conditions. The system includes: a boarding / disembarking data set acquisition module, used to acquire historical pilot boarding / disembarking data sets of the target pilot vessel and the target steered vessel; an overlapping influence area identification module, used to identify the overlapping influence areas of the target pilot vessel and the target steered vessel under different severe sea conditions based on the historical pilot boarding / disembarking data set, and determine the overlapping influence area set and the corresponding severe sea condition feature set; and a real-time sea state feature acquisition module, used to acquire real-time sea state features of the target pilot vessel and the target steered vessel according to preset motion parameters and preset sea state environmental indicators. The system performs sea state feature analysis to obtain real-time sea state features; a matching and overlapping influence area acquisition module matches the real-time sea state features with the severe sea state feature set, and maps and extracts the overlapping influence area based on the matching results to obtain the matching and overlapping influence area; a wave spectrum iterative feature acquisition module continuously collects regional wave spectrum data based on the matching and overlapping influence area, and performs iterative analysis based on the collection results to determine the wave spectrum iterative features of the target area; a wave spectrum iterative feature judgment module judges whether the wave spectrum iterative features of the target area meet the preset requirements, and if not, obtains a boarding and disembarking risk warning instruction.
[0006] One or more technical solutions provided in this application have at least the following technical effects or advantages: This application acquires historical boarding and disembarking data of pilot boats and piloted vessels, and establishes a correspondence between sea state and overlapping influence areas through aggregation and mapping. Combined with real-time sea state feature matching and multi-scale iterative analysis of wave spectra, it assesses risks and issues early warning commands, thereby accurately identifying the risks of pilots boarding and disembarking vessels in severe sea states. This provides a reliable basis for alarm devices, ensures the safety of boarding and disembarking operations, and achieves the technical effect of accurately warning of the risks of pilots boarding and disembarking vessels in severe sea states, thus improving the reliability of monitoring. Attached Figure Description
[0007] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0008] Figure 1 This is a flowchart illustrating the pilot boarding and disembarking risk warning method under severe sea conditions provided in this application embodiment.
[0009] Figure 2 This is a schematic diagram of the pilot boarding and disembarking risk warning system under severe sea conditions provided in this application embodiment.
[0010] Figure labeling: Module 1 for acquiring data set of boarding and disembarking vessels, Module 2 for identifying overlapping influence areas, Module 3 for acquiring real-time sea state features, Module 4 for acquiring matching overlapping influence areas, Module 5 for acquiring wave spectrum iterative features, and Module 6 for judging wave spectrum iterative features. Detailed Implementation
[0011] This application provides a method and system for risk warning of pilots boarding and disembarking vessels in adverse sea conditions, which is used to solve the technical problem that traditional risk monitoring methods for boarding and disembarking vessels are difficult to warn of risks during the pilot boarding and disembarking process in adverse sea conditions, based on alarm devices and relevant ship data.
[0012] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.
[0013] It should be noted that the terms "first," "second," etc., in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or modules not explicitly listed or inherent to such processes, methods, products, or devices.
[0014] Example 1, as Figure 1 As shown, a risk warning method for pilots boarding and disembarking vessels in adverse sea conditions is provided, wherein the method includes: Step A100: Obtain the historical pilot boarding and disembarking data set for the target pilot boat and the target piloted vessel.
[0015] In this embodiment, the pilot boat is a small vessel used to transport pilots to the piloted vessel, and it serves as a transport for pilots to board and disembark in adverse sea conditions. The piloted vessel is the target vessel that receives the pilot's boarding and requires the pilot's guidance to complete navigation operations.
[0016] Specifically, firstly, those skilled in the art collect raw data on the target pilot boat and the target vessel being piloted in past boarding and disembarking operations from the port pilotage management system database, the Automatic Identification System (AIS) historical trajectory database, and the archives of maritime meteorological monitoring stations. This data covers pilotage operation records over the past 10 years, with each record containing motion parameters such as pilot boat speed, pilot boat heading, and the speed and heading of the vessel being piloted. It also correlates these with sea state environmental indicators such as significant wave height, wave period, and wind direction at the corresponding time, forming a raw data set.
[0017] The raw data then undergoes a three-layer preprocessing process: The first layer uses timestamp alignment technology to synchronize the motion data and sea state indicators of the pilot boat and the piloted vessel to the same time series, with the error controlled within ±5 seconds; the second layer uses a density-based outlier detection algorithm (DBSCAN) to remove abnormal records where the pilot boat speed is greater than 25 knots or the wave height data exceeds 1.5 times the historical extreme value of the sea area, thus cleaning up invalid data; the third layer uses linear interpolation to fill in missing wave direction, current velocity, and other data, increasing the proportion of complete data records to 98%, and finally constructing a structured historical pilot boarding and disembarking data set.
[0018] Through multi-source data acquisition, multi-layer cleaning and preprocessing, and classification and labeling, a historical data set covering various severe sea conditions and containing complete motion parameters and sea condition indicators was formed. This laid a precise data foundation for subsequent data-driven identification of overlapping influence areas and construction of risk warning models, enabling historical data to effectively support the mining and analysis of risk characteristics of pilotage operations under different severe sea conditions.
[0019] Step A200: Based on the historical pilot boarding and disembarking data set, identify the overlapping influence areas of the target pilot boat and the target piloted vessel under different severe sea conditions, and determine the set of overlapping influence areas and the corresponding set of severe sea condition characteristics.
[0020] In this embodiment, the overlapping influence area is the interaction area formed by the relative motion of the target pilot boat and the target guided vessel under different adverse sea conditions. Adverse sea conditions are complex sea conditions composed of sea state environmental indicators, as well as motion parameters such as the speed and heading of the guided vessel and the speed and heading of the pilot boat. When the combination of these parameters reaches a preset risk threshold, it constitutes an adverse sea condition scenario that threatens the pilot's boarding and disembarking operations.
[0021] Optionally, based on the historical pilot boarding and disembarking data set, extract the historical severe sea condition feature set using severe sea condition features as an index and aggregate it according to a preset similarity threshold. Extract the historical overlapping affected area set using the affected area as an index. By mapping the aggregated features to the historical areas and calculating the mean, determine the overlapping affected area set and the corresponding severe sea condition feature set. Specific steps are explained in detail in A210-A250.
[0022] Step A300: Perform real-time sea state characteristic analysis on the target pilot boat and the target guided vessel according to preset motion parameters and preset sea state environment indicators to obtain real-time sea state characteristics.
[0023] In one embodiment of this application, firstly, the motion parameters and sea state environmental data of the target pilot boat and the guided vessel are collected in real time through a ship sensor network (including GPS, Doppler current meters, weather stations, etc.). Specifically, the speeds of the pilot boat and the guided vessel are collected by a Doppler velocimeter at a frequency of 10Hz with an accuracy of 0.1 knots; the heading is obtained through a fiber optic gyroscope with an accuracy of 1°; the significant wave height is measured by a radar altimeter with a sampling interval of 5 seconds and an accuracy of 0.1 meters; environmental indicators such as wave period and wave direction are transmitted by weather buoys at 30-second intervals, with errors controlled within ±5%. For example, at a certain moment, the guided vessel's speed is collected as 12.5 knots and its heading as 240°, while the pilot boat's speed is collected as 8.3 knots and its heading as 270°. Simultaneously, environmental indicators show a significant wave height of 2.8 meters, a wave period of 7 seconds, a wind direction of 310°, and a surface current velocity of 1.2 knots.
[0024] The collected data enters the preprocessing stage. First, a Kalman filter is used to denoise dynamic parameters such as speed and heading, eliminating instantaneous outliers caused by ship rolling. Taking speed data as an example, a state-space model is established to control the random noise variance within 0.05 knots². Then, the DBSCAN algorithm is used to cluster the wave data, removing outliers with meaningful wave heights exceeding three times the historical mean. For example, if the historical mean meaningful wave height for a certain sea area is 1.5 meters, data points exceeding 4.5 meters are automatically marked as noise points. After preprocessing, data integrity is improved, providing a reliable foundation for feature analysis.
[0025] Next, the feature extraction stage begins. For motion parameters, the relative speed (vector synthesis) and heading angle between the pilot boat and the guided vessel are calculated. For example, the relative speed is calculated using the speed vector difference between the two vessels, and the heading angle is obtained using the cosine theorem. For sea state indicators, a sliding window with a window length of 10 minutes is used to calculate statistical quantities such as the mean of meaningful wave height and the median of wave period. The wave height time series is then converted into frequency domain features using Fourier transform to obtain the energy density distribution. Simultaneously, parameters such as the angle between the current direction and the vessel's heading, and the deviation of the surface current velocity from the actual vessel track are normalized and converted into feature values within the range of 0-1.
[0026] Finally, the preprocessed motion parameters and sea state index features are fused to construct a multi-dimensional real-time sea state feature vector. This vector contains eight dimensions: relative speed, heading angle, mean meaningful wave height, wave period, angle between wave direction and heading, wind direction and speed composite vector, surface current velocity influence factor, and water depth correction coefficient. For example, when the meaningful wave height is 2.8 meters and the wave period is 7 seconds, combined with the relative speed of the two ships at 4.2 knots and the heading angle at 30°, the feature vector [4.2,30,2.8,7,50,8.5,0.8,15.6] is generated. The wind direction and speed composite vector is calculated using a wind speed of 8.5 m / s and a wind direction of 310°. The surface current velocity influence factor is obtained by normalizing the offset of the track by a current velocity of 1.2 knots. The water depth of 15.6 meters is the real-time measurement value of the operating sea area.
[0027] By acquiring real-time data from multiple sensors, using Kalman filtering and DBSCAN for noise reduction, calculating feature parameters statistically, and fusing multi-dimensional data, a real-time feature vector containing motion state and sea state environment was constructed. This improved the accuracy of real-time sea state feature analysis and provided highly reliable input data for subsequent matching with historical severe sea state feature sets. This ensured the accuracy and timeliness of matching overlapping influence areas, thus laying a data foundation for pilot boarding and disembarking risk warning.
[0028] Step A400: Match the real-time sea state features with the severe sea state feature set, and map and extract the overlapping influence area according to the matching result to obtain the matched overlapping influence area.
[0029] Specifically, after constructing the real-time sea state features, the process proceeds to the feature matching and region mapping stage. First, an improved K-Nearest Neighbors (KNN) algorithm is used for feature matching, comparing the real-time feature vector with records in the historical severe sea state feature set. With K set to 15, similarity is calculated using Euclidean distance. For example, the distance between the real-time feature vector [4.2,30,2.8,7,50,8.5,0.8,15.6] and a historical record [4.5,28,3.0,6.8,47,9.0,0.75,16.0] is 0.43. After calculation, the historical records with the smallest distances are selected. To improve matching accuracy, different parameters are weighted: meaningful wave height and wave period have a weight of 0.3, relative speed and heading angle have a weight of 0.2, and other parameters have a weight of 0.1, thus improving the matching accuracy of severe sea state features.
[0030] After matching is complete, the overlapping influence area mapping stage begins. For the selected historical records, corresponding overlapping area coordinate sequences are extracted. Each sequence contains the relative positional changes of the pilot boat and the guided vessel during boarding and disembarking, with a sampling interval of 10 seconds. A spatiotemporal consistency verification algorithm is used to align the historical area coordinate sequences with the current vessel position. For example, the current relative position is calculated in real time using the vessel's GPS coordinates. If the lateral distance between the two vessels is 5 meters at a certain moment in the historical record, while the current measured distance is 4.8 meters, then the area is considered to have spatiotemporal consistency. After verification, records that do not conform to the current operational scenario are removed, and the remaining valid mapped areas are retained.
[0031] Subsequently, region fusion and optimization are performed. The remaining effective regions are fused using a weighted average method, with weights determined based on feature matching similarity. For example, the record with the highest similarity has a weight of 0.15, and the lowest is 0.05. During the fusion process, Gaussian smoothing is applied to the region boundaries to eliminate jagged edges caused by measurement errors, making the region contours more consistent with actual hydrodynamic characteristics. Simultaneously, a water depth correction factor is introduced; for example, a correction coefficient of 0.92 corresponds to a current water depth of 15.6 meters. This factor is used to shrink overlapping areas in shallow water zones, avoiding misjudgments caused by wave refraction effects.
[0032] Finally, the system generates a matching overlapping impact area. This area is presented as a polygonal coordinate sequence containing multiple key points, each with an accuracy of 0.1 meters. The wave energy density distribution (unit: kW / m²) is marked within the area; for example, the energy density in the central area is 12 kW / m², and in the edge area it is 5 kW / m². The risk level is visually displayed through color gradients. Simultaneously, the duration and area change rate of the area are calculated, providing a quantitative basis for subsequent risk assessment.
[0033] The KNN weighted matching algorithm enables accurate comparison between real-time features and historical data. Spatiotemporal consistency verification and Gaussian smoothing fusion technology are used to complete the mapping and extraction of overlapping areas. Combined with water depth correction and energy density analysis, a matching overlapping influence area with spatiotemporal continuity and physical significance is generated, which improves the accuracy of area extraction and provides a high-precision spatial analysis basis for the dynamic assessment of pilot boarding and disembarking risks under severe sea conditions.
[0034] Step A500: Continuously collect regional wave spectrum data based on the matching overlapping influence area, and perform iterative analysis based on the collection results to determine the wave spectrum iterative characteristics of the target area.
[0035] In this embodiment of the application, the wave spectrum is a mathematical function used to describe the distribution characteristics of ocean wave energy at different frequencies or wavelengths. That is, it is a distribution spectrum that characterizes the energy density of each frequency component after decomposing a random wave into a superposition of a series of simple harmonic waves of different frequencies.
[0036] Specifically, by continuously collecting data on the overlapping influence areas of the matching wave spectrum according to preset wave spectrum indicators such as frequency, energy density, wave height and wave period, an indicator sequence is formed. Then, a feature set is obtained through multi-scale identification. Finally, iterative analysis is performed to determine the wave spectrum iteration characteristics of the target area, so as to realize the dynamic monitoring and accurate characterization of wave characteristics under severe sea conditions. The specific steps are explained in detail in A510-A530.
[0037] Step A600: Determine whether the wave spectrum iteration characteristics of the target area meet the preset requirements. If not, obtain the boarding / disembarking risk warning instruction.
[0038] Specifically, firstly, the wave spectrum iterative characteristics of the target area are input into a pre-defined risk assessment model. This model is trained based on historical accident data, and the pre-defined requirements are determined through machine learning (random forest algorithm, 90% accuracy) of boarding and disembarking incident data over the past ten years. It includes safety thresholds for each feature: frequency drift rate ≤ ±0.015Hz / min, energy density gradient ≤ ±0.25kJ / m² / (Hz·min), wave height variation coefficient ≤ 0.3, etc. The thresholds are set at the 95th percentile of historical safe operation data to ensure that 95% of safe scenarios are covered.
[0039] Subsequently, threshold comparisons are performed using a parameter-by-parameter weighted judgment method. Taking frequency drift rate as an example, if the measured value of +0.02 Hz / min exceeds the safety threshold of +0.015 Hz / min, a single-item warning is triggered; if the energy density gradient of +0.3 kJ / m² / (Hz·min) exceeds +0.25 kJ / m² / (Hz·min), another single-item warning is triggered. Each warning item is assigned a weight based on its historical risk contribution, with frequency drift rate having a weight of 0.3 and energy density gradient having a weight of 0.25. When the weighted risk value exceeds the preset warning line of 0.5, a comprehensive risk judgment is initiated.
[0040] If one or more features fail to meet preset requirements, the system invokes the early warning rule engine to generate different levels of boarding / disembarking risk warning instructions based on the degree of feature deviation. For example, if the frequency drift rate exceeds the threshold by 10% and the energy density gradient exceeds the threshold by 20%, it is judged as a medium risk, and a suggestion to postpone boarding / disembarking is generated. If the wave height variation coefficient also exceeds 0.5, it is upgraded to a high risk, and an immediate stop operation instruction is generated. The instruction includes detailed risk parameters, such as frequency drift rate exceeding the limit by +30%; impact analysis, such as short-period waves exacerbating ship rolling; and operational suggestions, such as adjusting the pilot boat's course to a 60° angle with the wave direction. These suggestions are simultaneously pushed to the bridge via the ship's terminal in the form of audio-visual signals and text pop-ups.
[0041] By training with historical data to determine multi-dimensional safety thresholds, and by combining weighted judgment with a rule engine to verify feature compliance, the problem of high false negative rate and insufficient guidance of traditional single-parameter early warning is effectively solved. This provides a scientific and timely risk control basis for pilots to board and disembark vessels in adverse sea conditions.
[0042] Furthermore, step A200 in the method provided in this application embodiment includes: A210: Using the characteristics of severe sea conditions as an index, traverse the historical pilot boarding and disembarking data set to extract data and obtain the historical severe sea condition characteristic set.
[0043] A220: Based on a preset feature similarity threshold, the historical severe sea condition feature set is aggregated into similar categories to obtain multiple aggregated historical severe sea condition feature sets.
[0044] A230: Using the affected area as an index, traverse the historical pilot boarding and disembarking data set to extract data and obtain the set of historical overlapping affected areas.
[0045] A240: Based on the multiple aggregated historical severe sea state feature sets, the historical overlapping influence area set is mapped, and the mean of the mapping results is calculated to obtain the overlapping influence area set.
[0046] A250: Filter the multiple aggregated historical severe sea condition feature sets to determine the severe sea condition feature set, wherein each severe sea condition feature corresponds to an aggregated historical severe sea condition feature set.
[0047] Specifically, firstly, from the historical pilot boarding and disembarking data set, severe sea state characteristics, including sea state environmental indicators such as meaningful wave height, wave period, and wind direction, as well as motion parameters such as the speed of the piloted vessel and the heading of the pilot boat, were used as search keywords. The data time series was synchronized using timestamp alignment technology with an error of ±5 seconds. The DBSCAN algorithm (density-based spatial clustering with noise application algorithm) was used to remove abnormal records such as pilot boat speeds exceeding 25 knots. Linear interpolation was used to fill in missing wave direction data, ultimately extracting historical data containing complete parameters to form a historical severe sea state characteristic set. Different sea state scenarios in this set have been categorized and labeled, such as strong wind scenarios (wind speed ≥ level 10) and giant wave scenarios (meaningful wave height ≥ 4 meters). Each record includes the relative position coordinate sequence of the pilot boat and the piloted vessel (sampling interval 10 seconds) and key information such as the duration of the overlapping area.
[0048] Subsequently, the cosine similarity algorithm was used to calculate the feature vector distance of each data point in the historical severe sea state feature set. A preset feature similarity threshold of 0.8 was set, allowing for a 15% tolerance range for parameter fluctuations. Data with similarity higher than the threshold were grouped into the same category. For example, data with meaningful wave heights within the range of 3.0 ± 0.5 meters and wave periods within the range of 8 ± 1 seconds were aggregated into the short-period giant wave category, forming multiple aggregated historical severe sea state feature sets. This process was automated using the DBSCAN clustering algorithm, keeping the clustering error rate below 7%.
[0049] Simultaneously, using the affected area as an index, the relative position coordinate sequence of the pilot boat and the piloted vessel during the boarding and disembarking operation is extracted from the historical pilot boarding and disembarking data set. Each data record includes a UTC timestamp at 10-second intervals, the latitude and longitude of the pilot boat and the piloted vessel with an accuracy of ±0.0001°, and the coordinate data of the overlapping area polygon containing 8-15 vertices is generated by calculating the convex hull polygon of the real-time positions of the two vessels.
[0050] Since the original data may come from different coordinate systems, such as some data using UTM projection and some using local Mercator projection, it is necessary to unify them to the WGS84 geographic coordinate system through spatial coordinate transformation technology. The specific steps are as follows: First, the coordinate system of the original data is identified using a coordinate transformation parameter file (containing the reference ellipsoid parameters and projection parameters of each coordinate system). Then, the coordinate transformation is performed using a seven-parameter transformation method to convert the planar coordinates (unit: meters) to latitude and longitude coordinates (unit: degrees), with the transformation error controlled within ±0.0002°, corresponding to a ground distance of ±20 meters. After the transformation, Gaussian filtering is applied to the coordinates of all polygon vertices to remove noise, eliminating outliers that deviate from the population mean by more than two standard deviations, such as coordinate jumps caused by sensor malfunctions, thereby increasing the proportion of valid coordinate points.
[0051] Finally, the processed polygon coordinate data is stored in chronological order of operation time, forming a historical set of overlapping impact areas. This set contains overlapping areas recorded over the past decade, each labeled with the operation start time, duration, and corresponding sea state characteristic tags, such as "high winds and waves" and "fast currents." All area coordinates are expressed in latitude and longitude using the WGS84 coordinate system, ensuring spatial comparability of overlapping areas across different times and sea areas. For example, the overlapping area in a giant wave scenario in January 2024 is a 10-vertex polygon composed of latitude and longitude sequences, with a calculated area of 215 square meters, precisely correlated with the tags "significant wave height 3.5 meters, wave period 6 seconds" in the sea state characteristic set for the same period.
[0052] Next, the obtained sets of severe sea state features (such as aggregate sets containing categories like "short-cycle giant waves" and "long-cycle strong winds") are used as mapping indexes to associate corresponding data in the historical overlapping influence area sets one by one. Taking the "short-cycle giant waves" category as an example, this set contains thousands of historical records, each corresponding to an overlapping area polygon composed of 10-15 latitude and longitude vertices, which have been uniformly projected to the WGS84 coordinate system with an accuracy of ±0.0001°.
[0053] For each aggregated sea state category, overlapping polygons of all associated records are extracted. First, the area of each polygon is calculated using the Gaussian area formula, converting latitude and longitude coordinates to planar distances before calculation, with an error of <1%. For example, the area of thousands of regions in the "Short-Period Giant Waves" category ranges from 180 to 260 square meters. After removing records with areas less than 150 square meters (outliers), the arithmetic mean of the remaining effective areas is calculated, yielding an average overlapping influence area of 220 square meters for this category. Simultaneously, the mean of the polygon vertex coordinates is calculated to obtain the geometric center coordinates of the overlapping area for this category, such as longitude 121.5° and latitude 31.2°, serving as representative values for the region's location.
[0054] Process all aggregated sea state categories using the steps described above, ultimately forming a set of overlapping influence areas containing multiple entries. Each entry includes the name of the corresponding sea state category, the average area of the overlapping region (in square meters), the geometric center coordinates (WGS84 latitude and longitude), and the number of associated historical records.
[0055] Finally, the mean shift algorithm is used to extract the center value of multiple aggregated historical severe sea state feature sets, and the core feature parameters of each set are selected (the specific steps are explained in detail in A251), thus forming a severe sea state feature set containing multiple typical severe sea state features. Each feature corresponds to an aggregated historical data set, and each feature is allowed a parameter tolerance range of ±5%.
[0056] By extracting multi-dimensional feature indexes, employing similarity aggregation algorithms with tolerance intervals, and performing spatial mapping and mean calculation, a structured classification of severe sea state features and overlapping impact areas in historical data was achieved. This resulted in a feature set that is both typical and fault-tolerant, making the classification of severe sea state scenarios more consistent with actual sea state fluctuation characteristics. This provides data model support that is both accurate and adaptable for subsequent real-time sea state feature matching and risk area identification, effectively solving the problem that feature classification in traditional methods is too rigid and cannot adapt to fluctuations in sea state parameters.
[0057] Furthermore, step A250 in the method provided in this application embodiment includes: A251: The mean shift is used to filter the multiple aggregated historical severe sea state feature sets, determine the center of each set, and summarize to obtain the severe sea state feature set.
[0058] In this embodiment, mean shift is a nonparametric clustering and pattern recognition method based on density gradient ascent. It works by iteratively calculating the mean vector in the neighborhood of a data point, causing the data point to drift to a region with higher density, and eventually converge to a local density peak.
[0059] Optionally, firstly, multiple aggregated historical severe sea state feature sets obtained by similarity aggregation can be used as input, such as a dataset containing multiple aggregated sets, each set storing multi-dimensional feature parameters such as wave height, wave period, and wind direction.
[0060] Subsequently, the mean-shift algorithm is applied iteratively to each aggregated set of historical severe sea state features. Taking three-dimensional features (sense wave height, wave period, and wind speed) as an example, the bandwidth parameter h is set to 1.5 times the standard deviation of the sense wave height (e.g., 0.4 meters); 1.2 times the standard deviation of the wave period (e.g., 0.8 seconds); and 1.0 times the standard deviation of the wind speed (e.g., 1.5 knots). For each data point in the set, the vector mean from it to all points in its neighborhood is calculated, and the data point is iteratively shifted towards the direction of maximum density until it converges to a local density peak. For example, in the set of short-period giant waves, after 15 iterations, all data points converge to the feature center point with a sense wave height of 3.2 meters, a wave period of 7.8 seconds, and a wind speed of 25 knots.
[0061] After calculating the mean shift for all aggregate sets, the converged density peak of each set is extracted as its central feature. Multiple aggregate sets converge to obtain a corresponding number of central features, with each dimension of the central feature located at the center of the data distribution of its corresponding aggregate set. Taking the wave height parameter as an example, the deviation of each central value from the mean of its corresponding set is less than 5%, ensuring the representativeness of the central features.
[0062] Finally, these central features are aggregated to form a set of severe sea state features. Each severe sea state feature in this set corresponds to an aggregated set of historical severe sea state features, such as short-period giant waves and long-period strong winds. Each parameter dimension of each feature is accompanied by a fluctuation tolerance range based on the mean drift bandwidth, so that the features can reflect typical sea states and adapt to actual fluctuations.
[0063] The mean-shift algorithm is used to extract density peaks from aggregated historical sea state features. The optimal representative point of each aggregate set is determined in the multidimensional feature space. The originally scattered massive historical data is transformed into multiple typical sea state features with clear central values and tolerance ranges. By using data dimensionality reduction and typical feature extraction, the efficiency and accuracy of subsequent real-time sea state feature matching are significantly improved, providing a concise and statistically representative benchmark feature library for risk warning models.
[0064] Furthermore, step A300 in the method provided in this application embodiment includes: A310: The preset motion parameters include at least the speed of the guided vessel, the course of the guided vessel, the speed of the pilot boat, and the course of the pilot boat.
[0065] Specifically, the speed of the guided vessel (in knots) is a key parameter for measuring how fast a vessel travels on the sea surface. Its function is to influence the dynamic frequency of the overlapping area between the guided vessel and the pilot boat by affecting their relative speeds, thereby impacting the stability of the wave spectrum characteristics. For example, when the speed of the guided vessel increases, the disturbance of the water flow around the hull intensifies, which may lead to an increase in wave energy density in the overlapping area, increasing the risk of boarding / disembarking.
[0066] The heading of the guided vessel (unit: degrees) determines the relative angle between the vessel and the waves. Its function is that different headings cause the vessel to experience wave forces in different directions, thus affecting the shape and area of the overlapping region. For example, when the angle between the guided vessel's heading and the wave direction is less than 30°, the vessel is significantly affected by transverse waves, and the variance of wave height fluctuations in the overlapping region may increase, requiring close monitoring.
[0067] Pilot speed: The speed at which the pilot boat approaches or moves away from the vessel being piloted (unit: knots), directly affecting the relative stability during docking. Excessive speed can make the pilot boat difficult to control in rough seas, complicate the relative motion trajectory in overlapping areas, and increase the difficulty of multi-scale identification of wave spectrum features; insufficient speed may prolong boarding and disembarking time, increasing the duration of exposure to high-risk sea conditions.
[0068] Pilot boat heading: This indicates the direction of the pilot boat's movement (unit: degrees) and determines the path the pilot boat takes to approach the vessel being piloted. Its function is to reduce the impact of cross waves on the pilot boat and minimize the range of irregular fluctuations in the overlapping area. For example, when the pilot boat's heading maintains a 45° angle with the wave direction, it reduces the lateral force of the waves on the hull, making the wave spectrum periodic characteristics of the overlapping area more stable and facilitating risk assessment.
[0069] By changing the relative motion of the two ships in real time, the dynamic changes of the overlapping area are affected, which becomes the core input for real-time sea state characteristic analysis and provides kinematic basis for wave spectrum data acquisition and risk warning command generation.
[0070] Furthermore, step A300 in the method provided in this application embodiment includes: A320: The preset sea state environmental indicators include at least meaningful wave height, wave period, wave direction, wind direction, current direction, surface current velocity, and water depth.
[0071] Specifically, significant wave height refers to the average wave height (in meters) of the highest waves in the first 1 / 3 of the statistical period. Its function is to directly reflect the energy intensity of the waves. When the significant wave height exceeds 3 meters, the wave impact force in the overlapping area of the pilot boat and the guided vessel is significantly enhanced, which may cause the relative displacement of the vessel to exceed the safety threshold when boarding or disembarking. It is a core parameter for judging the severity of sea state.
[0072] Wave period: refers to the time interval (in seconds) between adjacent waves passing through a fixed point. Its function is to affect the continuity of waves and the frequency of ship rolling. Short-period waves (such as <8 seconds) will make the energy density distribution of the wave spectrum in the overlapping area more concentrated, and the relative motion amplitude fluctuation between the pilot boat and the piloted ship will be aggravated, increasing the difficulty of boarding and disembarking operations.
[0073] Wave direction: refers to the direction of wave propagation (unit: degrees). Its function is to determine the angle at which the wave affects the ship. When the angle between the wave direction and the pilot boat's course is less than 45°, the transverse wave effect in the overlapping area is significant, which may cause the pilot boat's lateral rolling amplitude to exceed the safe range. It is necessary to focus on analyzing the wave spectrum characteristics in this direction using a multi-scale identifier.
[0074] Wind direction: refers to the direction from which the wind is coming (unit: degrees), and together with wind speed, it affects wave generation. Its function is that strong winds (e.g., ≥10 on the Beaufort scale) intensify sea surface fluctuations and alter wave morphology in overlapping areas. For example, under headwind conditions, the drag on a pilot boat increases, and the variance of wave height fluctuations in the overlapping area may increase by more than 20%, requiring iterative analysis in conjunction with wind speed data.
[0075] Flow direction: refers to the direction of water flow (unit: degrees), which, along with surface velocity, affects the actual trajectory of a ship. Its function is that the angle between the flow direction and the ship's course causes a relative positional shift in the overlapping area. For example, when the angle between the flow direction and the pilot boat's course exceeds 90°, the surface velocity will cause the pilot boat to deviate from the intended docking area, requiring real-time sea state feature matching to adjust the mapping and extraction of the overlapping area.
[0076] Surface velocity: refers to the speed of water flow on the sea surface (unit: knots), which affects ship maneuverability and relative motion trajectory. Its function is that high flow velocities (e.g., >2 knots) increase the difficulty for pilot boats to approach the guided ship, resulting in a shortened or prolonged duration of the overlapping area, which in turn affects the integrity of wave spectrum data acquisition. It is necessary to dynamically adjust the receptive field parameters in multi-scale identification.
[0077] Water depth: refers to the depth of the operating sea area (unit: meters), affecting wave morphology and ship draft safety. Its function is that shallow water areas (e.g., <2 times the ship's draft) can cause wave refraction and shallowing effects (the physical phenomenon where wave morphology and energy distribution change as waves propagate from deep to shallow water), altering the wave spectrum frequency characteristics of overlapping areas and potentially causing abnormally high wave heights. Therefore, it is necessary to correct the safety threshold for wave spectrum iterative analysis using water depth data.
[0078] By influencing the wave spectrum characteristics of overlapping areas through multidimensional coupling, environmental basis data is provided for real-time sea state characteristic analysis, thereby supporting the dynamic judgment of wave spectrum iterative characteristics and the generation of risk warning instructions, forming a three-dimensional monitoring system for severe sea conditions.
[0079] Furthermore, step A500 in the method provided in this application embodiment includes: A510: Data is continuously collected from the matched overlapping influence area according to the preset wave spectrum index to obtain a regional wave spectrum index sequence, wherein the preset wave spectrum includes at least frequency, energy density, wave height and wave period.
[0080] A520: Perform multi-scale identification on the regional wave spectrum index sequence to obtain a set of regional wave spectrum features.
[0081] A530: Perform iterative analysis on the wave spectrum feature set of the region to determine the wave spectrum iterative features of the target region.
[0082] Specifically, firstly, wave data is continuously collected at a sampling frequency of 1 Hz using wave height meters, pressure sensors, and ADCP (Acoustic Doppler Current Profiler) deployed in the overlapping influence area. Each collection session lasts for 10 minutes, forming raw time-series data containing frequency, energy density, wave height, and wave period. Frequency describes the periodicity of the wave, typically expressed as the number of vibrations per second, measured in Hertz (Hz). Wave frequency is closely related to wavelength (wave period). Energy density represents the amount of wave energy per unit frequency range; the wave spectrum shows the energy distribution of the wave across different frequencies or periods. Wave height refers to the vertical height of the wave, typically used to describe its intensity. Wave period refers to the time from one wave crest to the next, usually a key characteristic of the wave. For example, one collection yielded 600 data sets, with an average wave height of 2.8 meters, a wave period of 7 seconds, and an energy density peaking at 3.2 kJ / m² / Hz at 0.3 Hz, thus forming the initial regional wave spectrum index sequence.
[0083] After collection, the data enters the multi-scale identification stage. The fluctuation variance of the frequency, energy density, wave height, and wave period of the regional wave spectrum index sequence is calculated and normalized. The identification scale set is determined based on the normalization value. The multi-scale recognizer is constructed using this scale as the receptive field. Then, the index sequence is identified on multiple scales to obtain the regional wave spectrum feature set. The specific steps are explained in detail in A521-A525.
[0084] Subsequently, iterative analysis is performed on the feature set. The regional wave spectrum feature set is serialized according to the identification scale. The initial iterative features are determined by iterating the first and second regional wave spectrum features. Then, the remaining features are iterated based on the initial features to finally obtain the target region wave spectrum iterative features. The specific steps are explained in detail in A531-A533.
[0085] By constructing a multi-dimensional wave spectrum index sequence through high-frequency sensor data acquisition, wavelet transform is used to separate wave features at different time scales, and a recursive algorithm is combined to dynamically track feature parameters, forming an iterative feature vector containing time-frequency domain information and changing trends. This improves the dynamic capture accuracy of wave spectrum features and effectively solves the problem that traditional single-point detection cannot reflect the trend of sea state evolution. It provides predictive wave dynamic characteristics for real-time assessment of boarding and disembarking risks.
[0086] Furthermore, step A520 in the method provided in this application embodiment includes: A521: Traverse the regional wave spectrum index sequence to identify the fluctuation variance of each type of regional wave spectrum index, and obtain the frequency fluctuation variance, energy density fluctuation variance, wave height fluctuation variance, and wave period fluctuation variance.
[0087] A522: Normalize the frequency fluctuation variance, energy density fluctuation variance, wave height fluctuation variance, and wave period fluctuation variance to obtain normalized values for the frequency fluctuation variance, energy density fluctuation variance, wave height fluctuation variance, and wave period fluctuation variance.
[0088] A523: Determine the set of identification scales based on the normalized values of frequency fluctuation variance, energy density fluctuation variance, wave height fluctuation variance, and wave period fluctuation variance.
[0089] A524: Construct a multi-scale recognizer by using the recognition scales in the set of recognition scales as the receptive fields of the network layer.
[0090] A525: The multi-scale recognizer is used to perform multi-scale recognition on the regional wave spectrum index sequence to obtain a set of regional wave spectrum features.
[0091] In this application embodiment, the multi-scale recognizer is an intelligent algorithm architecture that can simultaneously capture the features of data at different time or spatial scales. The principle is to design processing units with different receptive fields to extract multi-granular features from the data in parallel or serially, thereby more comprehensively describing the dynamic characteristics of complex systems.
[0092] In one embodiment, firstly, for the regional wave spectrum index sequence, such as 600 sets of sampled data within 10 minutes, with a frequency range of 0.05-2Hz and an energy density of 0.1-10kJ / m² / Hz, the variance of the fluctuation is calculated for each index. Taking the frequency index as an example, by calculating the mean squared difference between adjacent sampling points, the frequency fluctuation variance is found to be 0.015Hz²; after calculating the energy density using a sliding window with a window length of 10 seconds 20 times, the fluctuation variance is found to be 0.8kJ² / m². 4 / Hz²; the variance of the wave height sequence is 0.25㎡; the variance of the wave period is 0.6s², forming the initial variance set.
[0093] Subsequently, the Min-Max normalization method was used to map the variances to the [0,1] interval. The frequency fluctuation variance of 0.015Hz² was normalized to 0.75, corresponding to the preset interval 0-0.02Hz²; the energy density variance was 0.8kJ² / m³. 4 / Hz² normalized to 0.4, corresponding to the interval 0-2kJ² / m 4 / Hz²; the wave height variance of 0.25㎡ is normalized to 0.5, corresponding to the interval 0-0.5㎡; the wave period variance of 0.6s² is normalized to 0.6, corresponding to the interval 0-1s². The normalized values reflect the relative intensity of the fluctuations of each indicator. For example, a frequency fluctuation normalized value of 0.75 indicates that its fluctuation degree is higher than that of other indicators.
[0094] Next, a neural network model is constructed to determine the recognition scale set: the input layer is designed with 4 nodes, corresponding to the normalized variance values of frequency, energy density, wave height, and wave period, with a value range of [0,1]. Each node receives a single parameter value after Min-Max normalization, such as a normalized variance value of 0.75 for frequency fluctuation and 0.4 for energy density. The hidden layer has 8 nodes, using the ReLU activation function to introduce a nonlinear mapping. The input data is transformed through the weight matrix W1 (dimension 4×8) and the bias vector b1 to calculate... Where X is a 4-dimensional input vector. The output layer has 3 nodes, corresponding to large, medium, and small scale categories. A softmax activation function is used to output a normalized probability distribution. The final output is generated using a weight matrix W2 (8×3 dimensions) and a bias vector b2. Each element represents the probability of belonging to the corresponding scale category, such as [0.9,0.08,0.02] which represents a large-scale probability of 90%.
[0095] The model is trained using samples extracted from severe sea state data from the past decade. Each sample contains four normalized variance values and manually labeled scale tags, with large scales accounting for 30%, medium scales 50%, and small scales 20%. The tags are one-hot encoded, e.g., large scales are [1, 0, 0]. The cross-entropy loss function is used during training. ,in The model uses the Adam optimizer with a learning rate of 0.001, employing 32 samples per batch and iteratively training for 50 epochs. To prevent overfitting, a dropout layer is added after the hidden layer with a retention rate of 0.8, and the training and validation sets are divided into an 8:2 ratio, with the validation set loss monitored in real time. After training, the model achieves a classification accuracy of over 90% on the independent test set (2000 samples). For example, with an input frequency normalization of 0.75, energy density of 0.4, wave height of 0.5, and wave period of 0.6, the output is [0.92, 0.07, 0.01], accurately mapped to the large-scale encoding [1, 0, 0].
[0096] Next, a multi-scale recognizer is constructed based on the recognition scale, using a convolutional neural network architecture, the specific configuration of which is shown in Table 1. Large-scale convolutional kernels are applied to the frequency sequence to extract long-period fluctuation features, such as low-frequency components below 0.1 Hz; medium-scale convolutional kernels are applied to the energy density sequence to capture mid-frequency energy changes between 0.2 and 0.5 Hz. For example, in the wave height sequence, the medium-scale recognizer detects five consecutive sampling points with wave heights exceeding 3 meters, generating high-energy peak features.
[0097] Finally, the multi-scale recognizer extracts a set of multiple features from 600 sets of sampled data (regional wave spectrum index sequence): large-scale frequency features, mesoscale energy density features, mesoscale wave height features, mesoscale wave period features, etc. Each feature is accompanied by a corresponding scale label and confidence level, forming a structured set of regional wave spectrum features.
[0098] By quantifying the volatility intensity using volatility variance, normalizing the data to achieve cross-parameter comparison, and dynamically matching the identification scale using a mapping mechanism, a feature extraction model adapted to sea state fluctuation characteristics was constructed using a CNN multi-scale convolutional architecture. This improved the accuracy of multi-scale identification of wave spectrum features, effectively capturing wave change patterns at different time scales. It provided timely and complete feature inputs for subsequent iterative analysis, solving the problem that traditional single-scale analysis cannot adapt to multi-frequency sea state fluctuations.
[0099] Table 1: Multi-scale Recognizer Configuration Table Identification scale Receptive field (number of sampling points) Kernel type / size Feature extraction target Activation function Output feature dimension Large scale 20 3×3 convolution kernel (with padding 1) Long-period wave energy distribution (≤0.2Hz) ReLU Low-frequency energy ratio Mesoscale 10 2×2 convolution kernel (no padding) Mid-frequency wind and wave characteristics (0.2-0.5Hz) ReLU Energy density gradient Small scale 5 1×1 convolution kernel Transient changes in high-frequency breaking waves (≥0.5Hz) Sigmoid Wave height anomaly probability Furthermore, step A530 in the method provided in this application embodiment includes: A531: The regional wave spectrum feature set is serialized according to the recognition scale size corresponding to each regional wave spectrum feature in the regional wave spectrum feature set to obtain the regional wave spectrum feature sequence.
[0100] A532: Iterate the regional wave spectrum feature sequence from the first regional wave spectrum feature and the second regional wave spectrum feature to determine the first initial regional wave spectrum iterative feature.
[0101] A533: Iterate the remaining regional wave spectrum features in the set of regional wave spectrum features according to the first initial regional wave spectrum iterative features to obtain the target regional wave spectrum iterative features.
[0102] Optionally, the regional wave spectrum feature set obtained from multi-scale identification is first serialized. Based on the identification scale corresponding to each feature, with preset large scale = 20, medium scale = 10, and small scale = 5, the features are sorted from largest to smallest scale to form a regional wave spectrum feature sequence. For example, in a certain set, frequency fluctuation features correspond to large scale 20, energy density and wave height features correspond to medium scale 10, and wave period features correspond to small scale 5. After sorting, the first two elements of the sequence are the large-scale frequency feature and the medium-scale energy density feature, followed by the medium-scale wave height feature, the small-scale wave period feature, and so on.
[0103] After sorting, the first two features (wave spectrum features of the first and second regions) are extracted from the sequence and iterated. The specific steps are explained in detail in A532-1.
[0104] Subsequently, the remaining 10 features are iterated upon based on the initial iterative features. A recursive update model is used, where each iteration associates the current iterative feature with the next feature in a time series manner, and the weights are adjusted using exponential smoothing (smoothing coefficient α = 0.3). For example, the third feature is a mesoscale wave height feature, with a 25% probability of consecutive wave peaks exceeding 3 meters. During iteration, the new feature value is calculated as α × previous iteration value 0.272 + (1-α) × 25% = 0.266. Simultaneously, the correction of the smoothing coefficient for feature scale changes (mesoscale → mesoscale) is recorded. If the scale change exceeds one level, α is adjusted to 0.4. After multiple iterations, the target area wave spectrum iterative features, including core parameters such as frequency-energy coupling degree and wave height-period correlation trends, are finally obtained.
[0105] By constructing a time-series analysis framework by serializing features according to scale, and using weighted fusion and recursive smoothing algorithms to realize dynamic correlation between features, an iterative analysis model with scale dependence is constructed, which improves the accuracy of identifying the time-series correlation of wave spectrum features, effectively captures the evolution law of features at different scales over time, and provides predictive feature vectors containing multi-scale interactive information for the dynamic assessment of boarding and disembarking risks, solving the problem that traditional independent feature analysis cannot reflect the trend of sea state evolution.
[0106] Furthermore, step A532 in the method provided in this application embodiment includes: A532-1: Perform inner product mapping on the wave spectrum features of the first region and the wave spectrum features of the second region, and normalize the inner product mapping result to construct a mapping matrix. Use the mapping matrix to iterate on the wave spectrum features of the second region to obtain the wave spectrum iterative features of the first initial region.
[0107] In one embodiment, firstly, the previously serialized wave spectrum features of the first region, such as large-scale frequency features, represented as a vector [frequency fluctuation variance normalized value 0.75, low-frequency energy proportion 32%], and the wave spectrum features of the second region, such as mesoscale energy density features, represented as a vector [energy density fluctuation variance normalized value 0.4, 0.3Hz energy gradient + 0.2kJ / m² / Hz·s], are converted into 2D feature vectors. The feature correlation between the two is calculated through inner product operation, using the formula: This value reflects the synergy between the two features in the dimensions of volatility variance and energy change.
[0108] Subsequently, all possible feature inner product results are normalized, with the inner product range preset to [-1, 1], using the Min-Max normalization formula. Map 0.364 to 0.85 (assuming min = -0.5, max = 0.5), and fill the normalized result into an initial empty 2×2 matrix, with the rows and columns representing the first and second feature indices, to construct the mapping matrix. The diagonal of the matrix represents the self-correlation value of the features (1), and the off-diagonal represents the correlation degree between features.
[0109] A Graph Convolutional Network (GCN) is constructed based on a mapping matrix. Two features are treated as nodes in a graph structure, the mapping matrix serves as the adjacency matrix, and the node feature vectors are the initial inputs. The network architecture consists of two convolutional layers. The first layer uses the ReLU activation function, and the second layer has an output dimension of 1 (the dimension of the target iterative feature). A dropout rate of 0.5 is set to prevent overfitting. Taking the input data at a certain time as an example, after the first convolutional layer... ,in For adjacency matrices with self-loops, For degree matrix, The node feature matrix, The weight matrix is randomly initialized. After two layers of convolution operations, the output iteration result is the frequency-energy coupling factor, which comprehensively reflects the strength of the synergistic effect between large-scale frequency fluctuations and mesoscale energy changes.
[0110] The association strength is quantified by eigenvector inner product mapping, a structured mapping matrix is constructed by normalization, and nonlinear relationship modeling between features is achieved by combining graph convolutional networks. Initial iterative features containing interaction information are generated in the 2D feature space, which improves the association recognition accuracy of multi-scale features to over 90%. It effectively captures the feature coupling effect that is difficult to represent by traditional weighted fusion methods, and provides initial input containing structural association information for subsequent full feature iterative analysis. This solves the problem that linear fusion cannot model complex feature dependencies.
[0111] In summary, the pilot boarding and disembarking risk warning method under severe sea conditions provided in this application has the following technical effects: This application acquires historical boarding and disembarking data of pilot boats and piloted vessels, and establishes a correspondence between sea state and overlapping influence areas through aggregation, mapping, and other processing. Combined with real-time sea state feature matching and multi-scale iterative analysis of wave spectra, it judges risks and issues early warning commands, thereby accurately identifying the risks of pilots boarding and disembarking vessels in severe sea states. This makes the early warning results of the alarm device more reliable, achieving the technical effect of accurately warning of the risks of pilots boarding and disembarking vessels in severe sea states and improving the reliability of monitoring.
[0112] Example 2, as Figure 2 As shown, based on the same inventive concept as in Embodiment 1 above, this application provides a pilot boarding / disembarking risk warning system under adverse sea conditions, the system comprising: The boarding and disembarking data set acquisition module 1 is used to acquire the historical boarding and disembarking data set of the target pilot boat and the target piloted vessel.
[0113] The overlapping influence area identification module 2 identifies the overlapping influence areas of the target pilot boat and the target guided vessel under different severe sea conditions based on the historical pilot boarding and disembarking data set, and determines the overlapping influence area set and the corresponding severe sea condition feature set.
[0114] The real-time sea state feature acquisition module 3 is used to perform real-time sea state feature analysis on the target pilot boat and the target guided vessel according to preset motion parameters and preset sea state environment indicators, and obtain real-time sea state features.
[0115] The overlapping influence area acquisition module 4 matches the real-time sea state features with the severe sea state feature set, and maps and extracts the overlapping influence area according to the matching result to obtain the matching overlapping influence area.
[0116] Wave spectrum iterative feature acquisition module 5 is used to continuously collect regional wave spectrum data based on the matching overlapping influence area, and perform iterative analysis based on the collection results to determine the wave spectrum iterative features of the target area.
[0117] Wave spectrum iteration feature judgment module 6 is used to judge whether the wave spectrum iteration feature of the target area meets the preset requirements. If not, a boarding and disembarking risk warning instruction is obtained.
[0118] Furthermore, the overlapping influence region identification module 2 is used to perform the following steps: Using severe sea state features as an index, the historical pilot boarding and disembarking data set is traversed for data extraction to obtain a set of historical severe sea state features. Based on a preset feature similarity threshold, the historical severe sea state feature set is aggregated into multiple aggregated historical severe sea state feature sets. Using affected areas as an index, the historical pilot boarding and disembarking data set is traversed for data extraction to obtain a set of historically overlapping affected areas. The multiple aggregated historical severe sea state feature sets are mapped to the set of historically overlapping affected areas, and the mean of the mapping results is calculated to obtain a set of overlapping affected areas. The multiple aggregated historical severe sea state feature sets are then filtered to determine the severe sea state feature set, wherein each severe sea state feature corresponds to one aggregated historical severe sea state feature set.
[0119] Furthermore, the overlapping influence region identification module 2 is used to perform the following steps: The mean shift method is used to filter the multiple aggregated historical severe sea state feature sets, determine the center of each set, and summarize to obtain the severe sea state feature set.
[0120] Furthermore, the real-time sea state feature acquisition module 3 is used to perform the following steps: The preset motion parameters include at least the speed of the guided vessel, the course of the guided vessel, the speed of the pilot boat, and the course of the pilot boat.
[0121] Furthermore, the real-time sea state feature acquisition module 3 is used to perform the following steps: The preset sea state environmental indicators include at least the significant wave height, wave period, wave direction, wind direction, current direction, surface current velocity, and water depth.
[0122] Furthermore, the wave spectrum iterative feature acquisition module 5 is used to perform the following steps: Data is continuously collected from the matched overlapping influence area according to the preset wave spectrum index to obtain a regional wave spectrum index sequence, wherein the preset wave spectrum includes at least frequency, energy density, wave height and wave period; the regional wave spectrum index sequence is identified at multiple scales to obtain a regional wave spectrum feature set; the regional wave spectrum feature set is iteratively analyzed to determine the target regional wave spectrum iterative features.
[0123] Furthermore, the wave spectrum iterative feature acquisition module 5 is used to perform the following steps: The regional wave spectrum index sequence is traversed to identify the variance of fluctuations for each type of regional wave spectrum index, obtaining the variances of frequency fluctuations, energy density fluctuations, wave height fluctuations, and wave period fluctuations. These variances are then normalized to obtain normalized values. Based on the magnitudes of these normalized values, an identification scale set is determined. Each identification scale in the set is used as the receptive field of a network layer to construct a multi-scale recognizer. The multi-scale recognizer is then used to perform multi-scale identification on the regional wave spectrum index sequence to obtain a regional wave spectrum feature set.
[0124] Furthermore, the wave spectrum iterative feature acquisition module 5 is used to perform the following steps: According to the recognition scale size corresponding to each regional wave spectrum feature in the regional wave spectrum feature set, the regional wave spectrum feature set is serialized to obtain a regional wave spectrum feature sequence; the regional wave spectrum feature sequence is iterated from the first regional wave spectrum feature and the second regional wave spectrum feature to determine the first initial regional wave spectrum iterative feature; the remaining regional wave spectrum features in the regional wave spectrum feature set are iterated according to the first initial regional wave spectrum iterative feature to obtain the target regional wave spectrum iterative feature.
[0125] Furthermore, the wave spectrum iterative feature acquisition module 5 is used to perform the following steps: The wave spectrum features of the first region and the wave spectrum features of the second region are subjected to inner product mapping, and the inner product mapping results are normalized to construct a mapping matrix. The wave spectrum features of the second region are iterated using the mapping matrix to obtain the wave spectrum iterative features of the first initial region.
[0126] The pilot boarding and disembarking risk warning system under severe sea conditions provided in this embodiment of the invention can execute the pilot boarding and disembarking risk warning method under severe sea conditions provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the execution method.
[0127] Although this application makes various references to certain modules in the system according to the embodiments of this application, any number of different modules can be used and run on user terminals and / or servers. The various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of each functional unit are only for easy distinction between each other and are not used to limit the scope of protection of this invention.
[0128] The specific embodiments described above do not constitute a limitation on the scope of protection of this application. Those skilled in the art should understand that various modifications, combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of this application. In some cases, the actions or steps described in this application can be performed in a different order than that shown in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
Claims
1. A method for early warning of pilot boarding and disembarking risks under adverse sea conditions, characterized in that, The method includes: Acquire the historical data set of pilot boarding and disembarking personnel for the target pilot boat and the target piloted vessel; Based on the historical pilot boarding and disembarking data set, the overlapping influence areas of the target pilot boat and the target guided vessel under different severe sea conditions are identified, and the set of overlapping influence areas and the corresponding set of severe sea condition characteristics are determined. Real-time sea state characteristics of the target pilot boat and the target guided vessel are analyzed according to preset motion parameters and preset sea state environment indicators to obtain real-time sea state characteristics. The real-time sea condition features are matched with the severe sea condition feature set, and the overlapping influence area is mapped and extracted according to the matching results to obtain the matching overlapping influence area. Based on the matching overlapping influence area, regional wave spectrum data is continuously collected, and iterative analysis is performed based on the collection results to determine the wave spectrum iterative characteristics of the target area. Determine whether the wave spectrum iteration characteristics of the target area meet the preset requirements. If not, obtain a risk warning instruction for boarding / leaving the ship.
2. The pilot boarding / disembarking risk warning method under adverse sea conditions as described in claim 1, characterized in that, Based on the historical pilot boarding and disembarking data set, the overlapping influence areas of the target pilot boat and the target guided vessel under different severe sea conditions are identified, and the set of overlapping influence areas and the corresponding set of severe sea condition characteristics are determined, including: Using the characteristics of severe sea conditions as an index, the historical pilot boarding and disembarking data set is traversed to extract data and obtain the historical severe sea condition characteristic set. According to a preset feature similarity threshold, the historical severe sea condition feature set is aggregated into similar categories to obtain multiple aggregated historical severe sea condition feature sets. Using the affected area as an index, the historical pilot boarding and disembarking data set is traversed to extract data and obtain the set of historical overlapping affected areas. The set of historical overlapping influence areas is mapped based on the multiple aggregated historical severe sea state feature sets, and the mean of the mapping results is calculated to obtain the set of overlapping influence areas. The multiple aggregated historical severe sea condition feature sets are filtered to determine the severe sea condition feature set, wherein each severe sea condition feature corresponds to an aggregated historical severe sea condition feature set.
3. The pilot boarding / disembarking risk warning method under adverse sea conditions as described in claim 2, characterized in that, The mean shift method is used to filter the multiple aggregated historical severe sea state feature sets, determine the center of each set, and summarize to obtain the severe sea state feature set.
4. The pilot boarding / disembarking risk warning method under adverse sea conditions as described in claim 1, characterized in that, The preset motion parameters include at least the speed of the guided vessel, the course of the guided vessel, the speed of the pilot boat, and the course of the pilot boat.
5. The pilot boarding and disembarking risk warning method under adverse sea conditions as described in claim 1, characterized in that, The preset sea state environmental indicators include at least meaningful wave height, wave period, wave direction, wind direction, current direction, surface current velocity, and water depth.
6. The pilot boarding / disembarking risk warning method under adverse sea conditions as described in claim 1, characterized in that, Based on the matching and overlapping influence area, regional wave spectrum data is continuously collected, and iterative analysis is performed based on the collection results to determine the wave spectrum iterative characteristics of the target area, including: Data is continuously collected from the matched overlapping influence area according to the preset wave spectrum index to obtain a regional wave spectrum index sequence, wherein the preset wave spectrum includes at least frequency, energy density, wave height and wave period; Multi-scale identification is performed on the regional wave spectrum index sequence to obtain a set of regional wave spectrum features; Iterative analysis is performed on the wave spectrum feature set of the region to determine the wave spectrum iterative features of the target region.
7. The pilot boarding / disembarking risk warning method under adverse sea conditions as described in claim 6, characterized in that, Multi-scale identification is performed on the regional wave spectrum index sequence to obtain a set of regional wave spectrum features, including: The fluctuation variance of each type of regional wave spectrum index is identified by traversing the regional wave spectrum index sequence to obtain the frequency fluctuation variance, energy density fluctuation variance, wave height fluctuation variance, and wave period fluctuation variance. The frequency fluctuation variance, energy density fluctuation variance, wave height fluctuation variance, and wave period fluctuation variance are normalized to obtain normalized values for frequency fluctuation variance, energy density fluctuation variance, wave height fluctuation variance, and wave period fluctuation variance. The set of identification scales is determined based on the normalized values of frequency fluctuation variance, energy density fluctuation variance, wave height fluctuation variance, and wave period fluctuation variance. A multi-scale recognizer is constructed by using the recognition scales in the set of recognition scales as the receptive fields of the network layer. The multi-scale recognizer is used to perform multi-scale recognition on the regional wave spectrum index sequence to obtain a set of regional wave spectrum features.
8. The pilot boarding / disembarking risk warning method under adverse sea conditions as described in claim 7, characterized in that, include: According to the recognition scale size corresponding to each regional wave spectrum feature in the regional wave spectrum feature set, the regional wave spectrum feature set is serialized to obtain the regional wave spectrum feature sequence. The regional wave spectrum feature sequence is iterated from the first regional wave spectrum feature and the second regional wave spectrum feature to determine the first initial regional wave spectrum iterative feature; Based on the first initial regional wave spectrum iterative features, the remaining regional wave spectrum features in the regional wave spectrum feature set are iterated to obtain the target regional wave spectrum iterative features.
9. The pilot boarding and disembarking risk warning method under adverse sea conditions as described in claim 8, characterized in that, The wave spectrum features of the first region and the wave spectrum features of the second region are subjected to inner product mapping, and the inner product mapping results are normalized to construct a mapping matrix. The wave spectrum features of the second region are iterated using the mapping matrix to obtain the wave spectrum iterative features of the first initial region.
10. A pilot boarding / disembarking risk warning system under adverse sea conditions, characterized in that: The system is used to implement the pilot boarding and disembarking risk warning method under adverse sea conditions as described in any one of claims 1-9, the system comprising: The boarding and disembarking data acquisition module is used to acquire historical pilot boarding and disembarking data sets for the target pilot boat and the target piloted vessel. The overlapping influence area identification module identifies the overlapping influence areas of the target pilot boat and the target guided vessel under different severe sea conditions based on the historical pilot boarding and disembarking data set, and determines the set of overlapping influence areas and the corresponding set of severe sea condition features. The real-time sea state feature acquisition module is used to perform real-time sea state feature analysis on the target pilot boat and the target guided vessel according to preset motion parameters and preset sea state environment indicators, and obtain real-time sea state features. The overlapping influence area acquisition module matches the real-time sea state features with the severe sea state feature set, and maps and extracts the overlapping influence area based on the matching results to obtain the matching overlapping influence area. The wave spectrum iterative feature acquisition module is used to continuously collect regional wave spectrum data based on the matching overlapping influence area, and perform iterative analysis based on the collection results to determine the wave spectrum iterative features of the target area. The wave spectrum iteration feature judgment module is used to determine whether the wave spectrum iteration feature of the target area meets the preset requirements. If not, a boarding and disembarking risk warning instruction is obtained.
Citation Information
Patent Citations
Digital-intelligent dual-drive real-time wave parameter identification and optimization method
CN117034758A
Method for calculating ship collision risk distribution in port channel water area
CN118095855A
Tug configuration method based on AIS piloting data
CN120372164A
Large-scale ocean data buoy safety monitoring system and method based on multi-source fusion
CN120482253A
Danger warning and emergency response system and method
US6084510A
