An AI-based method for predicting the impact force of fishing boats

Abstract

The present invention discloses an AI-based prediction method for the impact force of fishing boats, belonging to the technical field of fishing boat impacts, including: constructing a multi-ship motion state data set; calculating the relative distance and relative azimuth angle of each pair of fishing boats according to the multi-ship motion state data set, and extracting the distance change rate and azimuth angle change rate; if the relative distance of each pair of fishing boats is lower than the preset safe distance threshold, combining with the azimuth angle change rate to obtain the collision risk probability of each pair of fishing boats; predicting the impact force through the collision risk probability; generating the collision risk level of each ship according to the impact force value and the collision risk probability, and dynamically grouping the fishing boats to obtain an optimized formation structure; calculating the avoidance course adjustment angle and speed change amount of each ship according to the optimized formation structure, generating an emergency avoidance instruction, and obtaining the multi-ship cooperative motion state after execution. The present invention realizes the intelligent cooperative collision avoidance of fishing boat groups and significantly improves the safety of maritime navigation.

Description

Technical Field

[0001] The present invention belongs to the technical field of fishing boat impact prediction, and particularly relates to an AI-based fishing boat impact force prediction method. Background Art

[0002] As an important global economic and food pillar industry, the safe and efficient operation of marine fishery is crucial for ensuring food security and the safety of fishermen. However, when fishing boats operate in complex and changing marine environments, they frequently face the risk of collisions caused by multi-ship collaborative operations, which poses extremely high technical requirements for ship navigation and formation management.

[0003] Currently, fishing boat collision prevention mainly relies on traditional Automatic Identification Systems (AIS) and radar monitoring, but these methods have significant limitations in dealing with dynamic multi-ship scenarios. They often cannot accurately capture the relative azimuth changes of multiple ships in real time, making it difficult to accurately predict potential collision risks. Especially in dense fishing grounds or bad weather, insufficient data fusion and limited prediction capabilities lead to frequent misjudgments or missed judgments.

[0004] In view of the problems in the prior art, there is an urgent need to propose an AI-based fishing boat impact force prediction method, which integrates multi-source sensor data through artificial intelligence technology, real-time tracks the relative azimuth relationship of multiple ships, and then predicts the potential impact force. Summary of the Invention

[0005] To solve the above technical problems, the present invention proposes an AI-based fishing boat impact force prediction method to solve the problems existing in the above prior art.

[0006] To achieve the above object, the present invention provides an AI-based fishing boat impact force prediction method, and the method includes the following steps:

[0007] Obtain AIS, radar, and optical sensor data through multi-source sensors, and fuse them to generate a real-time multi-ship motion state data set;

[0008] According to the multi-ship motion state data set, calculate the relative distance and relative azimuth angle of each pair of fishing boats, and use time series modeling methods to extract the distance change rate and azimuth angle change rate;

[0009] If the relative distance of each pair of fishing boats is lower than a preset safety distance threshold, then combine the azimuth angle change rate and use a machine learning model to obtain the collision risk probability of each pair of fishing boats;

[0010] Predict the impact force through the collision risk probability using a dynamics model, and quantify the impact force value;

[0011] Generate the collision risk level of each ship according to the impact force value and the collision risk probability, and then use a clustering algorithm to dynamically group the fishing boats to obtain an optimized formation structure;

[0012] If the collision risk level is higher than the preset risk threshold, then according to the optimized formation structure, calculate the avoidance course adjustment angles and speed change amounts of each ship, and generate an emergency avoidance instruction.

[0013] Use a real-time communication protocol to send the emergency avoidance instruction to the navigation systems of each fishing boat, update the navigation parameters of each ship, and obtain the multi-ship cooperative motion state after execution.

[0014] Optionally, the method of obtaining AIS, radar and optical sensor data through multi-source sensors and fusing them to generate a real-time multi-ship motion state data set includes:

[0015] Collect AIS data, radar data and optical data through multi-source sensors and store them as an original data set.

[0016] Use the Kalman filtering algorithm to perform time synchronization processing on the original data set to obtain a time-aligned data set.

[0017] Complete the missing values in the time-aligned data set by the linear interpolation method to obtain a complete data set.

[0018] According to the complete data set, use the weighted average method to fuse AIS, radar and optical data to generate a fused real-time multi-ship motion state data set. The multi-ship motion state data set includes ship longitude and latitude, ship speed and ship course.

[0019] Optionally, the method of calculating the relative distance and relative azimuth angle of each pair of fishing boats according to the multi-ship motion state data set and using the time series modeling method to extract the distance change rate and azimuth angle change rate includes:

[0020] Obtain the longitude and latitude, speed and course of each pair of fishing boats from the multi-ship motion state data set, calculate the relative distance of each pair of fishing boats based on the Euclidean distance formula, and calculate the relative azimuth angle of each pair of fishing boats based on the arctangent function.

[0021] Use the time series analysis method to process the relative distance and relative azimuth angle, and calculate the distance change rate and azimuth angle change rate through a sliding window.

[0022] Optionally, if the relative distance of each pair of fishing boats is lower than the preset safe distance threshold, then combine the azimuth angle change rate and use a machine learning model to obtain the collision risk probability of each pair of fishing boats, including:

[0023] If the relative distance continues to decrease and is lower than the preset safe distance threshold, then generate a dynamic feature data set based on the relative distance and azimuth angle change rate of each pair of fishing boats.

[0024] Process the dynamic feature dataset through a sliding window method, calculate the continuous decreasing trend of the relative distance and the temporal features of the azimuth change rate, and obtain a set of temporal features;

[0025] Use the random forest algorithm to classify the set of temporal features, judge the collision risk of each pair of fishing boats, and generate a collision risk classification result;

[0026] Based on the collision risk classification result, combine the relative distance and the azimuth change rate, and predict the collision risk probability through a pre-trained logistic regression model to obtain the collision risk probability value of each pair of fishing boats.

[0027] Optionally, using the collision risk probability, predict the impact force by a dynamic model and quantify the impact force value, including:

[0028] Obtain the hull mass and relative speed of each pair of fishing boats from the multi-ship motion state dataset, and combine the collision risk probability to generate a feature dataset including mass parameters, speed parameters and risk probability;

[0029] Process the feature dataset through a sliding window method, calculate the temporal change trend of the relative speed and the distribution characteristics of the hull mass, and generate a set of temporal features;

[0030] Use the dynamic model to process the set of temporal features, calculate the potential impact force of each pair of fishing boats, and quantify the impact force value.

[0031] Optionally, generating the collision risk level of each ship according to the impact force value and the collision risk probability, and using a clustering algorithm to dynamically group the fishing boats to obtain an optimized formation structure, including:

[0032] Use a statistical method to jointly analyze the impact force value and the collision risk probability. If the impact force value exceeds the preset impact threshold and the collision risk probability is higher than the preset risk threshold, it is determined as a high-risk collision scenario;

[0033] Dynamically group the fishing boats in the high-risk scenario set through the K-means clustering algorithm, generate the collision risk level of the fishing boats based on the collision risk probability and the impact force, and obtain a set of collision risk levels;

[0034] According to the set of collision risk levels, use a sorting method to divide the fishing boats into priorities to obtain an optimized formation structure.

[0035] Optionally, if the collision risk level is higher than the preset risk threshold, then according to the optimized formation structure, calculate the avoidance course adjustment angle and speed change amount of each ship, and generate an emergency avoidance instruction, including:

[0036] According to the optimized formation structure, a path planning algorithm is used to calculate the avoidance course adjustment angles of each fishing boat, and a set of course adjustment angles is obtained;

[0037] If the adjustment angles in the set of course adjustment angles exceed the preset range, the speed change amounts of each fishing boat are calculated through a dynamic model, and a set of speed change amounts is obtained;

[0038] A fusion method is used to jointly process the set of course adjustment angles and the set of speed change amounts to generate an emergency avoidance instruction that meets the safety distance constraint.

[0039] Optionally, after the emergency avoidance instruction is sent to the navigation system of each fishing boat by using a real-time communication protocol, the navigation parameters of each ship are updated, and after the multi-ship cooperative motion state after execution is obtained, it further includes:

[0040] According to the multi-ship cooperative motion state after execution, the relative distance and azimuth change rate of each pair of ships are repeatedly calculated to obtain an updated collision risk probability; if the updated collision risk probability is lower than the preset risk threshold, the current formation and navigation parameters are maintained, otherwise a new optimized formation structure is generated to obtain a new avoidance instruction.

[0041] The present invention also provides a computer device, including a memory, a processor, and a computer program stored on the memory, and the processor executes the computer program to implement the steps of the method.

[0042] The present invention also provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the method are implemented.

[0043] Compared with the prior art, the present invention has the following advantages and technical effects:

[0044] The present invention discloses an AI-based method for predicting the impact force of fishing boats. This method obtains the real-time multi-ship motion state through multi-source sensor fusion, calculates the relative motion characteristics between ships, and combines a machine learning model to predict the collision risk. Based on the collision risk and potential impact force, the present invention dynamically groups fishing boats to optimize the formation structure and generates an emergency avoidance instruction in high-risk situations. The instruction is sent to the navigation system of each ship through real-time communication to achieve multi-ship cooperative collision avoidance. The present invention also verifies the avoidance effect through iterative calculation and re-optimizes the formation structure when necessary, so as to continuously ensure navigation safety. This method effectively integrates technologies such as multi-source data fusion, machine learning prediction, and dynamic formation optimization, realizes the intelligent cooperative collision avoidance of fishing boat groups, and significantly improves the safety of maritime navigation. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] The accompanying drawings, which form a part of this application, are used to provide a further understanding of this application. The schematic embodiments and descriptions thereof of this application are used to explain this application and do not constitute an improper limitation to this application. In the drawings:

[0046] Figure 1 It is a flowchart of the method according to an embodiment of the present invention. Detailed implementation manners

[0047] It should be noted that, without conflict, the embodiments in this application and the features in the embodiments can be combined with each other. The following will refer to the accompanying drawings and combine with the embodiments to detail this application.

[0048] It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order from here.

[0049] Embodiment 1

[0050] As Figure 1 shown, in this embodiment, a method for predicting the impact force of fishing boats based on AI is provided. The method includes the following steps:

[0051] Obtain AIS, radar and optical sensor data through multi-source sensors, and fuse them to generate a real-time multi-ship motion state data set;

[0052] According to the multi-ship motion state data set, calculate the relative distance and relative azimuth angle of each pair of fishing boats, and adopt a time series modeling method to extract the distance change rate and azimuth angle change rate;

[0053] If the relative distance of each pair of fishing boats is lower than the preset safety distance threshold, then combine the azimuth angle change rate and use a machine learning model to obtain the collision risk probability of each pair of fishing boats;

[0054] Through the collision risk probability, use a dynamics model to predict the impact force and quantify the impact force value;

[0055] According to the impact force value and the collision risk probability, generate the collision risk level of each ship, and then use a clustering algorithm to dynamically group the fishing boats to obtain an optimized formation structure;

[0056] If the collision risk level is higher than the preset risk threshold, then according to the optimized formation structure, calculate the avoidance course adjustment angle and speed change amount of each ship, and generate an emergency avoidance instruction;

[0057] Adopt a real-time communication protocol to send the emergency avoidance instruction to the navigation system of each fishing boat, update the navigation parameters of each ship, and obtain the multi-ship cooperative motion state after execution.

[0058] As a specific implementation manner, it specifically includes the following steps:

[0059] S101. Obtain AIS, radar, and optical sensor data through multi-source sensors, fuse them to generate a real-time multi-ship motion state data set, including the longitude, latitude, speed, and heading of each ship, and obtain a unified multi-ship motion state description.

[0060] Collect AIS data, radar data, and optical data through multi-source sensors and store them as the original data set. Use the Kalman filter algorithm to perform time synchronization processing on the original data set to obtain a time-aligned data set. If there are missing data in the time-aligned data set, complete the missing values through linear interpolation to obtain a complete data set. According to the complete data set, fuse AIS, radar, and optical data using the weighted average method to generate a fused data set, including ship longitude, ship speed, and ship heading. Extract the motion states of each ship from the fused data set through cluster analysis to obtain a multi-ship motion state set. For the multi-ship motion state set, perform unified formatting processing to generate a unified multi-ship motion description. If there are abnormalities in the ship motion state in the unified description, screen them through a preset threshold to obtain the final multi-ship motion state description.

[0061] Implementable, collect ship-related data through multi-source sensors. AIS data provides information such as ship identity, position, and speed, and is usually updated in seconds; radar data obtains the distance and azimuth of the ship through electromagnetic wave reflection and is suitable for long-distance detection; optical data uses a camera to capture ship images and is suitable for high-precision close-range identification. Assume that an AIS receiver, shore-based radar, and optical camera are deployed at a certain port to collect data of a cargo ship: the AIS reports a longitude of 120.5°E, a latitude of 30.2°N, and a speed of 10 knots; the radar detects a distance of 5 km and an azimuth angle of 30°; the optical data confirms the ship's contour. These data are stored as the original data set in the format of time stamps and sensor types.

[0062] In a possible implementation manner, the Kalman filter is used for time synchronization processing. Since the sampling frequencies of each sensor are different, such as AIS updates every 10 seconds, radar every 5 seconds, and optical every 2 seconds, they need to be aligned to a unified time axis. The Kalman filter smooths the data noise through prediction and update to generate a time-aligned data set. For example, unifiedly aligned to each second, the output position of the cargo ship at t = 10 seconds is 120.501°E, 30.201°N.

[0063] It should be noted that the Kalman filter can effectively reduce sensor noise and improve data consistency.

[0064] Furthermore, if there are missing values in the time-aligned dataset, such as no data from the radar at t = 12 seconds, linear interpolation can be used to fill in the gaps. Suppose the distances at t = 11 seconds and t = 13 seconds are 5.1 km and 5.3 km respectively, then the interpolated value at t = 12 seconds is 5.2 km. This method is simple and efficient, ensuring the continuity of the complete dataset and is suitable for real-time applications.

[0065] In a possible implementation, the weighted average method is used to fuse data, and the weights are based on sensor accuracy: the AIS position has high accuracy with a weight of 0.5; the radar distance is reliable with a weight of 0.3; the optical image is used for auxiliary verification with a weight of 0.2. After fusion, the longitude and latitude of the cargo ship at t = 10 seconds are 120.502°E, 30.202°N, the speed is 10.2 knots, and the heading is 350°. The fused dataset improves the positioning accuracy and reduces the error of a single sensor.

[0066] In one embodiment, cluster analysis is used to extract the ship motion states. The K-means algorithm is used to classify ships into states such as stationary, sailing, and turning based on longitude, latitude, speed, and heading. For example, a cargo ship with a speed of 10 knots and a stable heading is classified as the sailing state; another ship with a speed of 0 knots is classified as the stationary state. The clustering results form a set of multi-ship motion states, intuitively reflecting the dynamics of ships in the port. Then, unified formatting is performed to convert the motion states into a standard description, with the format "Ship ID - Time - State - Longitude, Latitude - Speed - Heading". For example, "Cargo1 - 10:00 - Sailing - 120.502°E, 30.202°N - 10.2 knots - 350°". This format is convenient for system analysis and sharing.

[0067] It can be understood that anomaly screening is achieved through preset thresholds. For example, a speed exceeding 30 knots or a sudden change in heading is considered an anomaly. Suppose a ship's speed suddenly increases to 35 knots, exceeding the threshold, it is marked as an anomaly and excluded. The final description only retains reliable data, such as the normal sailing state of the cargo ship. This process ensures data quality and supports port scheduling and safety monitoring.

[0068] In summary, this embodiment uses time synchronization to improve data consistency, interpolation to ensure continuity, fusion to improve positioning accuracy, clustering and anomaly screening to enhance the reliability of state analysis, and the final description to support efficient ship traffic management, reduce the collision risk, and improve port operation efficiency.

[0069] S102. According to the multi-ship motion state dataset, calculate the relative distance and relative azimuth angle of each pair of fishing boats, and use the time series modeling method to extract the distance change rate and azimuth angle change rate, obtaining the dynamic relative motion characteristics of each pair of ships.

[0070] Obtain the longitude, latitude, speed, and heading of each pair of fishing boats from the multi-vessel motion state dataset, and calculate the relative distance D and relative azimuth A of each pair of fishing boats. Among them, D is calculated through the Euclidean distance formula, and A is determined through the arctangent function based on the longitude and latitude differences. Use time series analysis methods to process the relative distance D and relative azimuth A, and calculate the distance change rate Rd and azimuth change rate Ra through a sliding window. Among them, Rd is the differential mean of D within the time window, and Ra is the differential mean of A within the time window, obtaining time series features. If there are missing values in the time series features, the missing Rd and Ra values are filled in through linear interpolation to generate a complete time series feature dataset. According to the complete time series feature dataset, use the K-means clustering algorithm to classify Rd and Ra, extract the dynamic relative motion patterns of each pair of boats, and obtain the classified motion feature set. Extract the dynamic features of each pair of fishing boats from the classified motion feature set, calculate the statistical indicators of Rd and Ra, including the mean and variance, and generate a dynamic feature description. If the statistical indicators of Rd or Ra in the dynamic feature description exceed the preset threshold, it is judged as an abnormal motion pattern, and the abnormal data is removed to obtain the final dynamic relative motion feature. Through unified formatting processing, store the final dynamic relative motion feature as a structured dataset, including boat pair combinations, relative distance, relative azimuth, distance change rate, and azimuth change rate, generating a multi-vessel dynamic relative motion description.

[0071] Implementable. When calculating the relative distance D and relative azimuth A of each pair of fishing boats based on the multi-vessel motion state dataset, it can be processed through longitude and latitude data. The relative distance D uses the Euclidean distance formula to calculate the spatial distance based on the longitude and latitude differences between the two boats. The relative azimuth A is determined through the arctangent function in combination with the longitude and latitude differences to determine the relative direction of the two boats. For example, fishing boat A is located at 120.3°E, 30.1°N, and fishing boat B is located at 120.4°E, 30.2°N. After calculation, D is approximately 15 kilometers and A is approximately 45°. This method intuitively reflects the spatial relationship between the two boats and is convenient for subsequent analysis.

[0072] In a possible implementation, time series analysis processes the relative distance D and relative azimuth A through a sliding window, and calculates the distance change rate Rd and azimuth change rate Ra. The sliding window can be set to 10 seconds to statistically calculate the differential means of D and A within the window. For example, for fishing boats A and B, D is 15 kilometers at t = 10 seconds and 14.5 kilometers at t = 20 seconds, and Rd is -0.05 kilometers per second, indicating that the two boats are gradually approaching; A changes from 45° to 43°, and Ra is -0.2° per second, indicating a slight change in the azimuth. This analysis captures the dynamic trend and helps to judge the motion intention of the boats.

[0073] It should be noted that the missing values in the time series features are filled by linear interpolation. For example, if Rd is missing at t = 30 seconds, Rd is -0.05 km / s at t = 20 seconds, and Rd is -0.03 km / s at t = 40 seconds, then the interpolation value at t = 30 seconds is -0.04 km / s. Interpolation ensures data continuity, supports subsequent clustering analysis, and avoids pattern recognition biases caused by data missing.

[0074] Specifically, the K-means clustering algorithm classifies dynamic relative motion patterns based on Rd and Ra. Assume that the clustering is divided into three categories: approaching, moving away, and parallel. For example, a pair of ships with a negative Rd and Ra close to 0° is classified into the approaching pattern; a positive Rd indicates moving away. For fishing boats A and B with Rd = -0.05 km / s and Ra = -0.2° / s, they are classified into the approaching pattern after clustering. This classification intuitively reflects the relative motion law of the pair of ships and facilitates dynamic monitoring.

[0075] In one embodiment, the dynamic feature description is generated by calculating the mean and variance of Rd and Ra. For example, the mean of Rd for fishing boats A and B within 60 seconds is -0.04 km / s, the variance is 0.01, the mean of Ra is -0.15° / s, and the variance is 0.05. These statistical indicators quantify the motion stability, the mean reflects the trend, and the variance indicates the degree of fluctuation, which helps to evaluate the motion consistency.

[0076] Among them, the abnormal motion pattern is judged by comparing the statistical indicators of Rd or Ra with a preset threshold. For example, the threshold is set that the absolute value of Rd greater than 0.2 km / s is abnormal. If the Rd of a pair of ships is 0.3 km / s, exceeding the threshold, it is marked as abnormal and excluded. After excluding the abnormal data, reliable features are retained to ensure the accuracy of the analysis results.

[0077] The unified formatting process stores the dynamic relative motion features as a structured data set in the format of "pair of ships ID - time - D - A - Rd - Ra". Such as "fishing boat A - B - 10:00 - 15 km - 45° - -0.05 km / s - -0.2° / s". This format is convenient for the system to store and share, and supports dynamic monitoring and early warning in fishery management.

[0078] It can be understood that the above method forms a complete description of dynamic relative motion through multi-level processing, from distance and azimuth angle calculation to clustering and abnormal screening. This description provides reliable data support for fishing boat motion analysis and helps with maritime traffic management and safety guarantee.

[0079] S103. If the relative distance continues to decrease and is lower than the preset threshold D (D represents the safety distance threshold, unit: meter), then in combination with the azimuth angle change rate, a machine learning model is used to predict the collision possibility, and the collision risk probability of each pair of fishing boats is obtained.

[0080] If the relative distance continues to decrease and is lower than the preset safety distance threshold D (in meters, representing the minimum safety interval between two ships), then the relative distance and the azimuth change rate of each pair of fishing boats are obtained from the multi-ship motion state dataset to generate a dynamic feature dataset. The dynamic feature dataset is processed by a sliding window method to calculate the continuous decrease trend of the relative distance and the temporal features of the azimuth change rate, resulting in a temporal feature set. The random forest algorithm is used to classify the temporal feature set to determine whether each pair of fishing boats has a high collision risk, generating a collision risk classification result. If the collision risk classification result is high risk, then combining the relative distance and the azimuth change rate, the collision risk probability is predicted through a pre-trained logistic regression model to obtain the collision risk probability value of each pair of fishing boats.

[0081] Implementable, based on the multi-ship motion state dataset, the continuous decrease of the relative distance is a key indicator for judging the collision risk. The relative distance is calculated from the longitude and latitude data of two ships, reflecting the spatial proximity between the ships. If the relative distance is lower than the preset safety distance threshold, such as 500 meters, it indicates that the two ships may enter a high-risk area.

[0082] In a possible implementation, fishing boat A is located at 120.5°E, 30.2°N, and fishing boat B is located at 120.6°E, 30.3°N. The calculated relative distance is 400 meters, which is lower than the safety threshold, triggering subsequent dynamic feature extraction. This method intuitively judges potential risks through spatial positions, facilitating rapid response.

[0083] It should be noted that the generation of the dynamic feature dataset depends on the temporal analysis of the relative distance and the azimuth change rate. The azimuth change rate is calculated by the difference in azimuth at adjacent time points, reflecting the dynamic adjustment of the relative direction of the ships. For example, the azimuth of fishing boats A and B is 50° at t = 10 seconds and becomes 48° at t = 20 seconds, and the azimuth change rate is -0.2° / second, indicating a slight deflection in direction.

[0084] In one embodiment, the dynamic feature dataset includes the relative distance, the distance change rate, and the azimuth change rate, in the format of "ship pair ID - time - distance - azimuth change rate", such as "fishing boat A - B - 10:00 - 400 meters - -0.2° / second". This structured data facilitates system processing.

[0085] Specifically, the sliding window method processes the dynamic feature dataset to extract time-series features. The sliding window can be set to 20 seconds, and the mean and trend of the relative distance within the window are calculated. For example, the relative distance between fishing vessels A and B decreases from 400 meters to 350 meters from t = 10 seconds to t = 30 seconds, the trend is continuously decreasing, and the distance change rate is -2.5 m / s. The time-series features of the azimuth change rate are calculated through the mean and variance within the window, such as the mean being -0.15° / s and the variance being 0.02. This kind of analysis captures the dynamic laws of distance and direction and supports risk assessment.

[0086] In one embodiment, the random forest algorithm classifies the time-series feature set to determine the collision risk level. The random forest is integrated by multiple decision trees. Based on features such as the distance change rate and azimuth change rate, the pairs of vessels are classified into high, medium, and low risks. For example, the distance change rate between fishing vessels A and B is -2.5 m / s, and the variance of the azimuth change rate is 0.02, and the classification result is high risk. This method uses the combination of features to improve the classification accuracy.

[0087] In one embodiment, the high-risk pairs of vessels further predict the collision risk probability through a logistic regression model. The logistic regression model takes the relative distance and azimuth change rate as inputs and outputs a probability value. For example, the distance between fishing vessels A and B is 350 meters, and the distance change rate is -2.5 m / s, and the model predicts the collision probability to be 85%. This kind of quantitative result provides a basis for early warning.

[0088] It can be understood that the above method forms a complete risk assessment process through multi-level analysis, from distance monitoring to probability prediction. The implementation of each technical theme supports each other to ensure the rigor and practicality of the analysis and provides reliable support for fishery safety management.

[0089] S104. Calculate the potential impact force F (unit: Newton) through the collision risk probability and relative motion characteristics. Based on the hull mass m (unit: kilogram) and relative velocity v (unit: m / s), obtain the quantified impact force value.

[0090] Obtain the hull mass and relative velocity of each pair of fishing vessels from the fishing vessel motion state dataset. Combine the pre-established collision risk probability to generate a feature dataset containing mass parameters, velocity parameters, and risk probability, and obtain the feature dataset. Process the feature dataset through the sliding window method, calculate the time-series change trend of the relative velocity and the distribution characteristics of the hull mass, generate a time-series feature set, and obtain the time-series feature set. Use the dynamic model to process the time-series feature set, calculate the potential impact force of each pair of fishing vessels, and obtain the quantified impact force value.

[0091] Feasible. When obtaining the hull mass and relative speed of each pair of fishing vessels from the fishing vessel motion state dataset and generating the feature dataset in combination with the collision risk probability, it is necessary to clarify the data source and processing logic. The hull mass is usually extracted from the fishing vessel registration information, reflecting the physical attributes of the vessel. The relative speed is calculated from the speed and course difference between two vessels, reflecting the dynamic approaching characteristics.

[0092] In a possible implementation, the mass of fishing vessel C is 5000 kg, the mass of fishing vessel D is 7000 kg, the relative speed is 3 m / s, and the collision risk probability is 80%. These data are integrated into a feature dataset in the format of "vessel pair ID - time - mass - relative speed - risk probability", such as "Fishing vessel C - D - 10:00 - 5000 kg - 7000 kg - 3 m / s - 80%". This structured data is convenient for subsequent analysis and clearly reflects the dynamic and risk status of the vessel pair.

[0093] It should be noted that when using the sliding window method to process the feature dataset, it aims to capture the temporal variation of the relative speed and the mass distribution characteristics. The sliding window can be set to 30 seconds to analyze the mean value and change trend of the relative speed within the analysis window.

[0094] For example, the relative speed of fishing vessels C and D increases from 3 m / s to 4 m / s from t = 10 s to t = 40 s, with a trend of accelerating approach and a speed change rate of 0.033 m / s². The mass distribution characteristics are analyzed through the mean value and difference of the hull mass within the window. For example, the mean mass is 6000 kg and the difference is 2000 kg. This set of temporal characteristics provides a dynamic basis for kinetic analysis and highlights the synergistic effect of speed and mass.

[0095] Specifically, when using a kinetic model to process the set of temporal characteristics, a quantification value is calculated based on the potential impact force formula. The potential impact force reflects the energy intensity of the collision and is closely related to mass and speed.

[0096] In one embodiment, the relative speed of fishing vessels C and D is 4 m / s, and the smaller mass value of 5000 kg is taken, resulting in a higher impact force value, indicating a high-energy risk. In another example, the relative speed of fishing vessels E and F is 2 m / s, and the mass is 4000 kg, with a lower impact force and less risk. This quantitative analysis intuitively reflects the potential threats of different vessel pairs through the combination of mass and speed, facilitating priority ranking.

[0097] In one embodiment, the implementation of the kinetic model can adjust the analysis weight in combination with the risk probability. For example, for vessel pairs with a risk probability higher than 85%, their impact force values will be given priority and trigger a more stringent warning mechanism.

[0098] In one embodiment, the characteristic datasets of fishing vessels G and H show a mass of 6000 kg, a relative speed of 3.5 m / s, and a risk probability of 90%. The calculated impact force is relatively high, and the system marks it as high priority. This method improves the pertinence of risk assessment through multi-parameter fusion.

[0099] It can be understood that the above method forms a complete process from data to quantified risk through characteristic dataset generation, time-series feature extraction, and kinetic analysis. The implementation of each topic supports each other to ensure the comprehensiveness and practicality of the analysis.

[0100] For example, the structured design of the characteristic dataset facilitates sliding window processing, and the time-series features provide dynamic inputs for the kinetic model. The final quantification of the impact force provides an intuitive basis for risk management. This multi-level analysis is logically clear and adapts to the complex dynamic requirements of the fishing scenario.

[0101] S105. Generate the collision risk level of each vessel according to the quantified impact force value and the collision risk probability, and use a clustering algorithm to dynamically group the fishing vessels to obtain an optimized formation structure.

[0102] Obtain the relative speed characteristics and hull mass distribution of each pair of fishing vessels from the fishing vessel motion state dataset, and use a preprocessing method to generate a characteristic dataset containing time-series change trends to obtain a characteristic dataset. Process the characteristic dataset through a kinetic model, and calculate the quantified impact force of each pair of fishing vessels according to the formula F = mv² / 2, where F is the quantified impact force (unit: Newton), m is the hull mass (unit: kilogram), and v is the relative speed (unit: meter per second), to obtain a set of quantified impact forces. Use statistical methods to jointly analyze the set of quantified impact forces and the collision risk probability. If the quantified impact force exceeds the preset threshold and the risk probability is higher than the set value, determine it as a high-risk collision scenario to obtain a set of high-risk scenarios. Dynamically group the fishing vessels in the set of high-risk scenarios through the K-means clustering algorithm, generate the collision risk level of the fishing vessels according to the collision risk probability and the quantified impact force to obtain a set of collision risk levels. According to the set of collision risk levels, use a sorting method to prioritize the fishing vessels. If the collision risk level of a fishing vessel is high, adjust its motion trajectory first to obtain an adjusted trajectory set. Process the adjusted trajectory set through an optimization algorithm to generate a fishing vessel formation structure that meets the collision risk level constraints to obtain an optimized formation structure. Use a verification method to evaluate the optimized formation structure. If the distance between the fishing vessels in the optimized formation structure meets the preset safety threshold, determine it as the final formation plan to obtain the final formation plan.

[0103] Implementable. When extracting relative speed features and hull mass distribution from the fishing boat motion state dataset, a feature dataset containing time series change trends can be generated through preprocessing methods. The relative speed feature is calculated based on the fishing boat's speed and course difference, and the mass distribution is obtained from the registration information. In one possible implementation, the speeds of fishing boats A and B within 10 minutes are 5 m / s and 6 m / s respectively, and the course difference is 30 degrees. The calculated relative speed is 2.5 m / s. The mass distribution shows that A is 4000 kg and B is 6000 kg. The preprocessing uses a sliding window with a window size of 60 seconds to capture the trend of the relative speed rising from 2.5 m / s to 3 m / s, generating a feature dataset in the format of "boat pair ID - time - relative speed - mass mean - trend", such as "A - B - 10:00 - 2.5 m / s - 5000 kg - rising".

[0104] Specifically, when the kinetic model processes the feature dataset, it calculates the quantified impact force based on the formula.

[0105] It should be noted that the model takes a smaller mass and relative speed as inputs to ensure a conservative estimate. For example, the relative speed of fishing boats C and D is 3 m / s, and the mass is taken as 4000 kg to generate an impact force value, which is incorporated into the set of quantified impact forces. The impact force values of different boat pairs reflect the potential collision intensity, facilitating subsequent analysis.

[0106] In one embodiment, when jointly analyzing the set of quantified impact forces and the collision risk probability, an impact force threshold of 10000 N and a risk probability threshold of 75% are set. The impact force of fishing boats E and F is 12000 N, and the risk probability is 80%, which is determined as a high-risk scenario. In another example, the impact force of fishing boats G and H is 8000 N, and the risk probability is 70%, which does not reach the threshold, excluding the high risk. This screening logic is clear and highlights high-risk boat pairs.

[0107] In one embodiment, the K-means clustering algorithm groups the set of high-risk scenarios and generates risk levels based on the impact force and risk probability. For example, the impact force of fishing boats I and J is 15000 N, and the risk probability is 85%. After clustering, they are classified into the high-risk level; the impact force of fishing boats K and L is 11000 N, and the risk probability is 78%, which is classified into the medium-risk level. The clustering results form a set of collision risk levels, intuitively reflecting the threat degree of boat pairs.

[0108] It can be understood that the priority division is based on the set of risk levels, and a sorting method is used to adjust the trajectories. High-risk fishing boats such as I and J are given priority to adjust their headings and expand the spacing to 100 m, generating a set of adjusted trajectories. Low-risk boat pairs maintain their original trajectories. This method ensures that high-risk boat pairs are processed in a timely manner.

[0109] For example, when the optimization algorithm processes the trajectory set, it generates a formation structure based on the risk level constraint. After adjustment, the distance between fishing vessels M and N is 120 meters, meeting the safety requirements and being incorporated into the optimized formation structure. In another example, the distance between fishing vessels O and P is 80 meters, which does not meet the standard and needs further optimization. This iterative optimization ensures the rationality of the formation.

[0110] In a possible implementation, the verification method evaluates the optimized formation structure and checks whether the distance between fishing vessels meets the safety threshold of 100 meters. For example, the distance between fishing vessels Q and R is 130 meters, meeting the requirements and being determined as the final formation plan; the distance between fishing vessels S and T is 90 meters, and readjustment is required. This verification ensures the feasibility of the plan and adapts to the dynamic requirements of the fishing scenario.

[0111] S106. If the collision risk level is higher than the preset threshold R (R represents the risk level threshold, unitless), then according to the optimized formation structure, calculate the evasive course adjustment angle and speed change amount for each ship to obtain an emergency evasive instruction.

[0112] If the collision risk level is higher than the preset threshold R (R represents the risk level threshold, unitless), then according to the optimized formation structure, use the path planning algorithm to calculate the evasive course adjustment angle for each fishing vessel to obtain a set of course adjustment angles. If the adjustment angles in the set of course adjustment angles exceed the preset range, then calculate the speed change amount for each fishing vessel through the dynamic model to obtain a set of speed change amounts. Use a fusion method to jointly process the set of course adjustment angles and the set of speed change amounts to generate a set of emergency evasive instructions that meet the safety distance constraint, obtaining a set of emergency evasive instructions. Optimize the set of emergency evasive instructions through the dynamic trajectory planning method to generate the adjusted motion trajectories of each fishing vessel, obtaining a set of adjusted trajectories. According to the real-time data update method, verify the set of adjusted trajectories. If the relative position distribution between fishing vessels meets the safety distance constraint, then determine it as the final evasive plan, obtaining the final evasive plan.

[0113] Implementably, use the path planning algorithm to calculate the evasive course adjustment angle: use the A* algorithm, considering the turning radius of the fishing vessel and the sea area obstacles, to generate the adjustment angle. For example, fishing vessel E needs to adjust 15 degrees to the right, and fishing vessel F needs to adjust 10 degrees to the left, forming a set of course adjustment angles. If the adjustment angle exceeds the preset range, such as 20 degrees, then calculate the speed change amount through the dynamic model. The dynamic model takes the current speed and the ship mass as inputs to generate a speed adjustment plan. For example, fishing vessel G decelerates by 0.5 m / s to generate a set of speed change amounts.

[0114] It should be noted that the fusion method jointly processes the heading adjustment angle set and the speed change amount set to generate an emergency avoidance instruction set. The fusion method can adopt weighted decision-making, giving priority to ensuring heading adjustment and supplemented by speed changes. For example, fishing boat H adjusts its heading by 12 degrees and decelerates by 0.3 m / s, forming an instruction "Ship H - 10:05 - Heading 12 degrees - Deceleration 0.3 m / s", which is incorporated into the emergency avoidance instruction set.

[0115] In one embodiment, the dynamic trajectory planning method optimizes the emergency avoidance instruction set to generate an adjusted trajectory set.

[0116] Implementable, spline curve fitting is used to smooth the track changes. For example, the adjusted trajectory of fishing boat I shows that its distance from fishing boat J has increased to 120 m, meeting the safety requirements. The real-time data update method verifies the adjusted trajectory set to check whether the distance between fishing boats meets the standard. For example, the distance between fishing boats K and L is 110 m, meeting the 100 m threshold, and is determined as the final avoidance plan; the distance between fishing boats M and N is 95 m, and re-planning is required.

[0117] It can be understood that the above method ensures the safe navigation of fishing boats in complex dynamic scenarios through multi-level analysis and optimization. The implementation methods of each technical theme are closely connected, forming a complete logical chain from data extraction to final plan verification to meet the real-time needs of the fishing scenario.

[0118] S107. Update the navigation parameters of each ship through the emergency avoidance instruction, and use the real-time communication protocol to send the instruction to the navigation system of each fishing boat to obtain the multi-ship cooperative motion state after execution. According to the multi-ship cooperative motion state after execution, recalculate the relative distance and the azimuth change rate, verify the effect of the avoidance instruction, and obtain the updated collision risk probability. If the updated collision risk probability is lower than the preset threshold P (P represents the safety probability threshold, without unit), then maintain the current formation and navigation parameters; otherwise, return to the dynamic grouping step, regenerate the optimized formation structure, and obtain a new avoidance instruction.

[0119] Update the navigation parameters of each fishing boat through emergency avoidance instructions, and send the instructions to the navigation system of each fishing boat using a real-time communication protocol to obtain collaborative motion state data. According to the collaborative motion state data, calculate the relative distance and azimuth change rate between each fishing boat to obtain a set of relative motion characteristics. Use a preprocessing method to filter the set of relative motion characteristics to generate a smooth motion feature dataset, and obtain a smooth feature dataset. Through a collision risk assessment model, based on the smooth feature dataset, calculate the collision risk probability of each fishing boat to obtain a set of collision risk probabilities. If any probability in the set of collision risk probabilities is higher than the preset threshold P (P represents the safety probability threshold, without unit), then trigger the dynamic grouping algorithm to generate an optimized formation structure and obtain optimized formation structure data. According to the optimized formation structure data, calculate the avoidance course adjustment angle and speed change amount of each fishing boat to generate a new set of avoidance instructions and obtain a set of avoidance instructions. Send the set of avoidance instructions to the navigation system of each fishing boat through the real-time communication protocol to update the collaborative motion state data and obtain the updated collaborative motion state data.

[0120] Implementable, the update of the emergency avoidance instruction involves applying the calculated course and speed adjustment parameters to the fishing boat navigation system. The instruction contains specific course angles and speed change amounts. For example, fishing boat A needs to adjust its course to 270 degrees and reduce its speed to 3 m / s. The real-time communication protocol uses a lightweight protocol based on MQTT to ensure low-latency instruction transmission. After receiving the instruction, the fishing boat navigation system automatically adjusts the rudder angle and engine power to update the navigation parameters. For example, fishing boat B receives the instruction "Adjust the course to 265 degrees and decelerate by 0.4 m / s", and the navigation system completes the parameter adjustment within 2 seconds, generates collaborative motion state data, and records the real-time position, speed, and course of each boat.

[0121] In a possible implementation, based on the collaborative motion state data, calculate the relative distance and azimuth change rate. The relative distance is calculated through the longitude and latitude of each fishing boat. For example, the distance between fishing boats C and D is 120 meters. The azimuth change rate reflects the relative motion trend between fishing boats. For example, the azimuth of fishing boat C relative to D changes 1.5 degrees per second. The set of relative motion characteristics contains "boat pair ID - time - distance - azimuth change rate", such as "C - D - 10:10 - 120 meters - 1.5 degrees / s". These data provide a basis for subsequent risk assessment.

[0122] It should be noted that the preprocessing method filters the set of relative motion characteristics to eliminate sensor noise.

[0123] In this embodiment, a Kalman filter is used to generate a smooth feature dataset. For example, the distance data between fishing boats E and F is smoothed from the original fluctuating value to a stable 110 meters, and the azimuth change rate is smoothed to 1.2 degrees / s. The smooth feature dataset ensures the accuracy of collision risk assessment.

[0124] Specifically, the collision risk assessment model calculates the collision risk probability based on the smooth feature dataset. The model combines the distance, speed difference, and azimuth rate of change to generate a probability value. For example, the distance between fishing vessels G and H is 90 meters, and the speed difference is 1.5 m / s. The model outputs a collision probability of 0.75, which is higher than the preset threshold of 0.6. The collision risk probability set records the probability values of all pairs of vessels, providing a basis for dynamic grouping.

[0125] In one embodiment, the dynamic grouping algorithm optimizes the formation structure according to the collision risk probability set. The algorithm considers the motion state of the fishing vessels and the sea area environment to generate a grouping scheme.

[0126] For example, fishing vessels I, J, and K are grouped together, maintaining an interval distance of 150 meters. The optimized formation structure data includes the vessel IDs and recommended spacings of each group, ensuring the overall navigation safety.

[0127] For example, based on the optimized formation structure data, the avoidance course adjustment angle and speed change amount are calculated. Fishing vessel L needs to adjust 10 degrees to the left and decelerate by 0.3 m / s; fishing vessel M adjusts 8 degrees to the right and maintains its speed. The avoidance instruction set is stored in the format of "vessel ID - time - course adjustment - speed change", such as "L - 10:15 - 10 degrees to the left - decelerate 0.3 m / s".

[0128] In one embodiment, the real-time communication protocol sends the avoidance instruction set to the navigation system to update the cooperative motion state data. After receiving the instruction, fishing vessel N adjusts its course to 280 degrees and reduces its speed to 4 m / s. The updated cooperative motion state data shows that the distance between fishing vessels N and O increases to 130 meters, meeting the safety requirements. This multi-level instruction generation and sending mechanism ensures the cooperative and safe navigation of fishing vessels in a dynamic scenario.

[0129] Embodiment Two

[0130] This embodiment also provides a computer device, including a memory, a processor, and a computer program stored on the memory. The processor executes the computer program to implement the steps of the method.

[0131] Embodiment Three

[0132] This embodiment also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the method are implemented.

[0133] The above is only a preferred specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed by the present application should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A fishing vessel impact force prediction method based on AI, characterized in that: The method comprises the following steps: Acquire AIS, radar, and optical sensor data through multi-source sensors and fuse them to generate a real-time multi-ship motion status dataset; Based on the multi-vessel motion state dataset, the relative distance and relative azimuth of each pair of fishing vessels were calculated, and the distance change rate and azimuth change rate were extracted using a time series modeling method. If the relative distance between each pair of fishing vessels is lower than the preset safety distance threshold, the collision risk probability of each pair of fishing vessels is obtained using a machine learning model combined with the azimuth angle change rate; Based on the collision risk probability, the dynamic model is used to predict the impact force and quantify the impact force value; Based on the impact force value and collision risk probability, the collision risk level of each ship is generated, and then the fishing vessels are dynamically grouped using a clustering algorithm to obtain the optimized formation structure; If the collision risk level is higher than the preset risk threshold, the avoidance course adjustment angle and speed change of each ship are calculated based on the optimized formation structure, and emergency avoidance instructions are generated; A real-time communication protocol is used to send emergency avoidance instructions to the navigation systems of each fishing vessel, update the navigation parameters of each vessel, and obtain the multi-vessel coordinated motion status after execution; The collision risk probability is used to predict the impact force using a dynamic model and quantify the impact force value, including: The hull mass and relative speed of each pair of fishing vessels are obtained from the multi-vessel motion state dataset, and combined with the collision risk probability, a feature dataset containing mass parameters, speed parameters and risk probability is generated; The feature data set is processed by the sliding window method to calculate the time series variation trend of relative speed and the distribution characteristics of hull mass, and generate a time series feature set; The dynamic model is used to process the time series feature set, calculate the potential impact force of each pair of fishing vessels, and quantify the impact force value; The method generates a collision risk level for each vessel based on the impact force value and the collision risk probability, and dynamically groups the fishing vessels using a clustering algorithm to obtain an optimized formation structure, including: A statistical method is used to jointly analyze the impact force value and the collision risk probability. If the impact force value exceeds the preset impact threshold and the collision risk probability is higher than the preset risk threshold, it is determined to be a high-risk collision scenario; The fishing vessels in the high-risk scenario set are dynamically grouped using the K-means clustering algorithm, and the collision risk level of the fishing vessels is generated according to the collision risk probability and impact force, thus obtaining a collision risk level set. According to the collision risk level set, the fishing vessels are prioritized using a ranking method to obtain the optimized formation structure. If the collision risk level is higher than the preset risk threshold, the avoidance course adjustment angle and speed change of each ship are calculated according to the optimized formation structure, and an emergency avoidance instruction is generated, including: According to the optimized formation structure, the path planning algorithm is used to calculate the avoidance heading adjustment angle of each fishing vessel and obtain the heading adjustment angle set; If the adjustment angle in the course adjustment angle set exceeds the preset range, the speed change of each fishing boat is calculated by the dynamic model to obtain a speed change set; A fusion method is used to jointly process the heading adjustment angle set and the speed change set to generate emergency avoidance instructions that meet the safety distance constraints.

2. The method according to claim 1, characterized in that The method of acquiring AIS, radar and optical sensor data through multi-source sensors and fusing them to generate a real-time multi-ship motion status dataset includes: Collect AIS data, radar data, and optical data through multi-source sensors and store them as raw data sets; The Kalman filter algorithm is used to perform time synchronization processing on the original data set to obtain a time-aligned data set; The missing values in the time-aligned dataset are filled by linear interpolation to obtain a complete dataset; Based on the complete data set, the AIS, radar and optical data are fused using a weighted average method to generate a fused real-time multi-ship motion state data set. The multi-ship motion state data set includes the ship's latitude and longitude, ship speed and ship heading.

3. The method according to claim 1, characterized in that The method of calculating the relative distance and relative azimuth of each pair of fishing vessels based on the multi-vessel motion state dataset and extracting the distance change rate and azimuth change rate using a time series modeling method includes: The latitude and longitude, speed and heading of each pair of fishing vessels are obtained from the multi-vessel motion status dataset. The relative distance of each pair of fishing vessels is calculated based on the Euclidean distance formula, and the relative azimuth of each pair of fishing vessels is calculated based on the inverse tangent function. The time series analysis method is used to process the relative distance and relative azimuth, and the distance change rate and azimuth change rate are calculated through a sliding window.

4. The method according to claim 1, wherein If the relative distance between each pair of fishing vessels is lower than the preset safety distance threshold, a machine learning model is used to obtain the collision risk probability of each pair of fishing vessels in combination with the azimuth angle change rate, including: If the relative distance continues to decrease and falls below the preset safety distance threshold, a dynamic feature dataset is generated based on the relative distance and azimuth change rate of each pair of fishing vessels; The dynamic feature dataset is processed by the sliding window method, and the time series features of the continuous decreasing trend of the relative distance and the azimuth angle change rate are calculated to obtain the time series feature set; The random forest algorithm is used to classify the time series feature set, determine the collision risk of each pair of fishing vessels, and generate collision risk classification results; Based on the collision risk classification results, combined with the relative distance and azimuth angle change rate, the collision risk probability is predicted through a pre-trained logistic regression model to obtain the collision risk probability value of each pair of fishing vessels.

5. The method according to claim 1, wherein After the emergency avoidance instruction is sent to the navigation system of each fishing vessel using the real-time communication protocol, the navigation parameters of each vessel are updated, and the multi-vessel coordinated motion state after execution is obtained, the method further includes: According to the multi-ship coordinated motion state after execution, the relative distance and azimuth change rate of each ship pair are repeatedly calculated to obtain the updated collision risk probability; if the updated collision risk probability is lower than the preset risk threshold, the current formation and navigation parameters are maintained; otherwise, the optimized formation structure is regenerated and a new avoidance instruction is obtained.

6. A computer device comprising a memory, a processor, and a computer program stored in the memory, wherein: The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 5.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.

Images